diff options
Diffstat (limited to 'src/gallium/drivers')
208 files changed, 47397 insertions, 0 deletions
diff --git a/src/gallium/drivers/cell/Makefile b/src/gallium/drivers/cell/Makefile new file mode 100644 index 0000000000..47aef7b05f --- /dev/null +++ b/src/gallium/drivers/cell/Makefile @@ -0,0 +1,12 @@ +# Cell Gallium driver Makefile + + +default: + ( cd spu ; make ) + ( cd ppu ; make ) + + + +clean: + ( cd spu ; make clean ) + ( cd ppu ; make clean ) diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h new file mode 100644 index 0000000000..4de514c358 --- /dev/null +++ b/src/gallium/drivers/cell/common.h @@ -0,0 +1,220 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Types and tokens which are common to the SPU and PPU code. + */ + + +#ifndef CELL_COMMON_H +#define CELL_COMMON_H + +#include "pipe/p_compiler.h" +#include "pipe/p_util.h" +#include "pipe/p_format.h" + + +/** The standard assert macro doesn't seem to work reliably */ +#define ASSERT(x) \ + if (!(x)) { \ + ubyte *p = NULL; \ + fprintf(stderr, "%s:%d: %s(): assertion %s failed.\n", \ + __FILE__, __LINE__, __FUNCTION__, #x); \ + *p = 0; \ + exit(1); \ + } + + +/** for sanity checking */ +#define ASSERT_ALIGN16(ptr) \ + ASSERT((((unsigned long) (ptr)) & 0xf) == 0); + + +/** round up value to next multiple of 4 */ +#define ROUNDUP4(k) (((k) + 0x3) & ~0x3) + +/** round up value to next multiple of 8 */ +#define ROUNDUP8(k) (((k) + 0x7) & ~0x7) + +/** round up value to next multiple of 16 */ +#define ROUNDUP16(k) (((k) + 0xf) & ~0xf) + + +#define CELL_MAX_SPUS 6 + +#define TILE_SIZE 32 + + +/** + * The low byte of a mailbox word contains the command opcode. + * Remaining higher bytes are command specific. + */ +#define CELL_CMD_OPCODE_MASK 0xff + +#define CELL_CMD_EXIT 1 +#define CELL_CMD_CLEAR_SURFACE 2 +#define CELL_CMD_FINISH 3 +#define CELL_CMD_RENDER 4 +#define CELL_CMD_BATCH 5 +#define CELL_CMD_RELEASE_VERTS 6 +#define CELL_CMD_STATE_FRAMEBUFFER 10 +#define CELL_CMD_STATE_DEPTH_STENCIL 11 +#define CELL_CMD_STATE_SAMPLER 12 +#define CELL_CMD_STATE_TEXTURE 13 +#define CELL_CMD_STATE_VERTEX_INFO 14 +#define CELL_CMD_STATE_VIEWPORT 15 +#define CELL_CMD_STATE_VS_ARRAY_INFO 16 +#define CELL_CMD_STATE_BLEND 17 +#define CELL_CMD_VS_EXECUTE 18 + + +#define CELL_NUM_BUFFERS 4 +#define CELL_BUFFER_SIZE (4*1024) /**< 16KB would be the max */ + +#define CELL_BUFFER_STATUS_FREE 10 +#define CELL_BUFFER_STATUS_USED 20 + + + +/** + * Tell SPUs about the framebuffer size, location + */ +struct cell_command_framebuffer +{ + uint64_t opcode; /**< CELL_CMD_FRAMEBUFFER */ + int width, height; + void *color_start, *depth_start; + enum pipe_format color_format, depth_format; +}; + + +/** + * Clear framebuffer to the given value/color. + */ +struct cell_command_clear_surface +{ + uint64_t opcode; /**< CELL_CMD_CLEAR_SURFACE */ + uint surface; /**< Temporary: 0=color, 1=Z */ + uint value; +}; + + +/** + * Array info used by the vertex shader's vertex puller. + */ +struct cell_array_info +{ + uint64_t base; /**< Base address of the 0th element. */ + uint attr; /**< Attribute that this state is for. */ + uint pitch; /**< Byte pitch from one entry to the next. */ + uint format; /**< Pipe format of each entry. */ +} ALIGN16_ATTRIB; + + +struct cell_shader_info +{ + unsigned num_outputs; + + uint64_t declarations; + unsigned num_declarations; + uint64_t instructions; + unsigned num_instructions; + uint64_t uniforms; + uint64_t immediates; + unsigned num_immediates; +} ALIGN16_ATTRIB; + + +#define SPU_VERTS_PER_BATCH 64 +struct cell_command_vs +{ + uint64_t opcode; /**< CELL_CMD_VS_EXECUTE */ + struct cell_shader_info shader; + unsigned num_elts; + unsigned elts[SPU_VERTS_PER_BATCH]; + uint64_t vOut[SPU_VERTS_PER_BATCH]; + float plane[12][4]; + unsigned nr_planes; + unsigned nr_attrs; +} ALIGN16_ATTRIB; + + +struct cell_command_render +{ + uint64_t opcode; /**< CELL_CMD_RENDER */ + uint prim_type; /**< PIPE_PRIM_x */ + uint num_verts; + uint vertex_size; /**< bytes per vertex */ + uint num_indexes; + uint vertex_buf; /**< which cell->buffer[] contains the vertex data */ + float xmin, ymin, xmax, ymax; /* XXX another dummy field */ + uint min_index; + boolean inline_verts; +}; + + +struct cell_command_release_verts +{ + uint64_t opcode; /**< CELL_CMD_RELEASE_VERTS */ + uint vertex_buf; /**< in [0, CELL_NUM_BUFFERS-1] */ +}; + + +struct cell_command_texture +{ + void *start; /**< Address in main memory */ + uint width, height; +}; + + +/** XXX unions don't seem to work */ +/* XXX this should go away; all commands should be placed in batch buffers */ +struct cell_command +{ +#if 0 + struct cell_command_framebuffer fb; + struct cell_command_clear_surface clear; + struct cell_command_render render; +#endif + struct cell_command_vs vs; +} ALIGN16_ATTRIB; + + +/** This is the object passed to spe_create_thread() */ +struct cell_init_info +{ + unsigned id; + unsigned num_spus; + struct cell_command *cmd; + + /** Buffers for command batches, vertex/index data */ + ubyte *buffers[CELL_NUM_BUFFERS]; + uint *buffer_status; /**< points at cell_context->buffer_status */ +} ALIGN16_ATTRIB; + + +#endif /* CELL_COMMON_H */ diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile new file mode 100644 index 0000000000..50060f5cd3 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/Makefile @@ -0,0 +1,76 @@ +# Gallium3D Cell driver: PPU code + +# This makefile builds the g3dcell.a library which gets pulled into +# the main libGL.so library + + +TOP = ../../../../.. +include $(TOP)/configs/linux-cell + + +#PROG = gl4 + +CELL_LIB = libcell.a + +SPU_CODE_MODULE = ../spu/g3d_spu.a + + +SOURCES = \ + cell_batch.c \ + cell_clear.c \ + cell_context.c \ + cell_draw_arrays.c \ + cell_flush.c \ + cell_state_blend.c \ + cell_state_clip.c \ + cell_state_derived.c \ + cell_state_emit.c \ + cell_state_fs.c \ + cell_state_rasterizer.c \ + cell_state_sampler.c \ + cell_state_surface.c \ + cell_state_vertex.c \ + cell_spu.c \ + cell_surface.c \ + cell_texture.c \ + cell_vbuf.c \ + cell_vertex_shader.c \ + cell_winsys.c + + +OBJECTS = $(SOURCES:.c=.o) \ + +INCLUDE_DIRS = -I$(TOP)/src/mesa + + +.c.o: + $(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@ + + + +default: $(CELL_LIB) + + +$(CELL_LIB): $(OBJECTS) $(SPU_CODE_MODULE) +# ar -ru $(CELL_LIB) $(OBJECTS) $(SPU_CODE_MODULE) + ar -ru $(CELL_LIB) $(OBJECTS) + +#$(PROG): $(PPU_OBJECTS) +# $(CC) -o $(PROG) $(PPU_OBJECTS) $(SPU_CODE_MODULE) $(PPU_LFLAGS) + + + +clean: + rm -f *.o *~ $(CELL_LIB) + + + +depend: $(SOURCES) + rm -f depend + touch depend + $(MKDEP) $(MKDEP_OPTIONS) $(INCLUDE_DIRS) $(SOURCES) 2> /dev/null + +include depend + + + diff --git a/src/gallium/drivers/cell/ppu/cell_batch.c b/src/gallium/drivers/cell/ppu/cell_batch.c new file mode 100644 index 0000000000..f45e5f25b6 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_batch.c @@ -0,0 +1,217 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "cell_context.h" +#include "cell_batch.h" +#include "cell_spu.h" + + + +uint +cell_get_empty_buffer(struct cell_context *cell) +{ + uint buf = 0, tries = 0; + + /* Find a buffer that's marked as free by all SPUs */ + while (1) { + uint spu, num_free = 0; + + for (spu = 0; spu < cell->num_spus; spu++) { + if (cell->buffer_status[spu][buf][0] == CELL_BUFFER_STATUS_FREE) { + num_free++; + + if (num_free == cell->num_spus) { + /* found a free buffer, now mark status as used */ + for (spu = 0; spu < cell->num_spus; spu++) { + cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED; + } + /* + printf("PPU: ALLOC BUFFER %u\n", buf); + */ + return buf; + } + } + else { + break; + } + } + + /* try next buf */ + buf = (buf + 1) % CELL_NUM_BUFFERS; + + tries++; + if (tries == 100) { + /* + printf("PPU WAITING for buffer...\n"); + */ + } + } +} + + +void +cell_batch_flush(struct cell_context *cell) +{ + static boolean flushing = FALSE; + uint batch = cell->cur_batch; + const uint size = cell->buffer_size[batch]; + uint spu, cmd_word; + + assert(!flushing); + + if (size == 0) + return; + + flushing = TRUE; + + assert(batch < CELL_NUM_BUFFERS); + + /* + printf("cell_batch_dispatch: buf %u at %p, size %u\n", + batch, &cell->batch_buffer[batch][0], size); + */ + + /* + * Build "BATCH" command and sent to all SPUs. + */ + cmd_word = CELL_CMD_BATCH | (batch << 8) | (size << 16); + + for (spu = 0; spu < cell->num_spus; spu++) { + assert(cell->buffer_status[spu][batch][0] == CELL_BUFFER_STATUS_USED); + send_mbox_message(cell_global.spe_contexts[spu], cmd_word); + } + + /* When the SPUs are done copying the buffer into their locals stores + * they'll write a BUFFER_STATUS_FREE message into the buffer_status[] + * array indicating that the PPU can re-use the buffer. + */ + + batch = cell_get_empty_buffer(cell); + + cell->buffer_size[batch] = 0; /* empty */ + cell->cur_batch = batch; + + flushing = FALSE; +} + + +uint +cell_batch_free_space(const struct cell_context *cell) +{ + uint free = CELL_BUFFER_SIZE - cell->buffer_size[cell->cur_batch]; + return free; +} + + +/** + * Append data to current batch. + */ +void +cell_batch_append(struct cell_context *cell, const void *data, uint bytes) +{ + uint size; + + ASSERT(bytes % 8 == 0); + ASSERT(bytes <= CELL_BUFFER_SIZE); + ASSERT(cell->cur_batch >= 0); + +#ifdef ASSERT + { + uint spu; + for (spu = 0; spu < cell->num_spus; spu++) { + ASSERT(cell->buffer_status[spu][cell->cur_batch][0] + == CELL_BUFFER_STATUS_USED); + } + } +#endif + + size = cell->buffer_size[cell->cur_batch]; + + if (size + bytes > CELL_BUFFER_SIZE) { + cell_batch_flush(cell); + size = 0; + } + + ASSERT(size + bytes <= CELL_BUFFER_SIZE); + + memcpy(cell->buffer[cell->cur_batch] + size, data, bytes); + + cell->buffer_size[cell->cur_batch] = size + bytes; +} + + +void * +cell_batch_alloc(struct cell_context *cell, uint bytes) +{ + return cell_batch_alloc_aligned(cell, bytes, 1); +} + + +void * +cell_batch_alloc_aligned(struct cell_context *cell, uint bytes, + uint alignment) +{ + void *pos; + uint size, padbytes; + + ASSERT(bytes % 8 == 0); + ASSERT(bytes <= CELL_BUFFER_SIZE); + ASSERT(alignment > 0); + ASSERT(cell->cur_batch >= 0); + +#ifdef ASSERT + { + uint spu; + for (spu = 0; spu < cell->num_spus; spu++) { + ASSERT(cell->buffer_status[spu][cell->cur_batch][0] + == CELL_BUFFER_STATUS_USED); + } + } +#endif + + size = cell->buffer_size[cell->cur_batch]; + + padbytes = (alignment - (size % alignment)) % alignment; + + if (padbytes + size + bytes > CELL_BUFFER_SIZE) { + cell_batch_flush(cell); + size = 0; + } + else { + size += padbytes; + } + + ASSERT(size % alignment == 0); + ASSERT(size + bytes <= CELL_BUFFER_SIZE); + + pos = (void *) (cell->buffer[cell->cur_batch] + size); + + cell->buffer_size[cell->cur_batch] = size + bytes; + + return pos; +} diff --git a/src/gallium/drivers/cell/ppu/cell_batch.h b/src/gallium/drivers/cell/ppu/cell_batch.h new file mode 100644 index 0000000000..a6eee0a8b1 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_batch.h @@ -0,0 +1,58 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef CELL_BATCH_H +#define CELL_BATCH_H + +#include "pipe/p_compiler.h" + + +struct cell_context; + + +extern uint +cell_get_empty_buffer(struct cell_context *cell); + +extern void +cell_batch_flush(struct cell_context *cell); + +extern uint +cell_batch_free_space(const struct cell_context *cell); + +extern void +cell_batch_append(struct cell_context *cell, const void *data, uint bytes); + +extern void * +cell_batch_alloc(struct cell_context *cell, uint bytes); + +extern void * +cell_batch_alloc_aligned(struct cell_context *cell, uint bytes, + uint alignment); + + +#endif /* CELL_BATCH_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c new file mode 100644 index 0000000000..07b908eec5 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_clear.c @@ -0,0 +1,76 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Authors + * Brian Paul + */ + +#include <stdio.h> +#include <assert.h> +#include <stdint.h> +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "pipe/cell/common.h" +#include "cell_clear.h" +#include "cell_context.h" +#include "cell_batch.h" +#include "cell_flush.h" +#include "cell_spu.h" + + +void +cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue) +{ + struct cell_context *cell = cell_context(pipe); + uint surfIndex; + + if (cell->dirty) + cell_update_derived(cell); + + + if (!cell->cbuf_map[0]) + cell->cbuf_map[0] = pipe_surface_map(ps); + + if (ps == cell->framebuffer.zsbuf) { + surfIndex = 1; + } + else { + surfIndex = 0; + } + + + { + struct cell_command_clear_surface *clr + = (struct cell_command_clear_surface *) + cell_batch_alloc(cell, sizeof(*clr)); + clr->opcode = CELL_CMD_CLEAR_SURFACE; + clr->surface = surfIndex; + clr->value = clearValue; + } +} diff --git a/src/gallium/drivers/cell/ppu/cell_clear.h b/src/gallium/drivers/cell/ppu/cell_clear.h new file mode 100644 index 0000000000..ff47d43f4c --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_clear.h @@ -0,0 +1,43 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef CELL_CLEAR_H +#define CELL_CLEAR_H + + +struct pipe_context; +struct pipe_surface; + + +extern void +cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue); + + + +#endif /* CELL_CLEAR_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c new file mode 100644 index 0000000000..bbe1fd7a11 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_context.c @@ -0,0 +1,287 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Authors + * Brian Paul + */ + + +#include <stdio.h> + +#include "pipe/p_defines.h" +#include "pipe/p_format.h" +#include "pipe/p_util.h" +#include "pipe/p_winsys.h" +#include "pipe/cell/common.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_private.h" +#include "cell_clear.h" +#include "cell_context.h" +#include "cell_draw_arrays.h" +#include "cell_flush.h" +#include "cell_render.h" +#include "cell_state.h" +#include "cell_surface.h" +#include "cell_spu.h" +#include "cell_texture.h" +#include "cell_vbuf.h" + + + +static boolean +cell_is_format_supported( struct pipe_context *pipe, + enum pipe_format format, uint type ) +{ + /*struct cell_context *cell = cell_context( pipe );*/ + + switch (type) { + case PIPE_TEXTURE: + /* cell supports all texture formats, XXX for now anyway */ + return TRUE; + case PIPE_SURFACE: + /* cell supports all (off-screen) surface formats, XXX for now */ + return TRUE; + default: + assert(0); + return FALSE; + } +} + + +static int cell_get_param(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return 8; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 1; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 0; + case PIPE_CAP_POINT_SPRITE: + return 1; + case PIPE_CAP_MAX_RENDER_TARGETS: + return 1; + case PIPE_CAP_OCCLUSION_QUERY: + return 1; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 12; /* max 2Kx2K */ + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 8; /* max 128x128x128 */ + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 12; /* max 2Kx2K */ + default: + return 0; + } +} + +static float cell_get_paramf(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 255.0; /* arbitrary */ + + case PIPE_CAP_MAX_POINT_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 255.0; /* arbitrary */ + + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 0.0; + + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; /* arbitrary */ + + default: + return 0; + } +} + + +static const char * +cell_get_name( struct pipe_context *pipe ) +{ + return "Cell"; +} + +static const char * +cell_get_vendor( struct pipe_context *pipe ) +{ + return "Tungsten Graphics, Inc."; +} + + + +static void +cell_destroy_context( struct pipe_context *pipe ) +{ + struct cell_context *cell = cell_context(pipe); + + cell_spu_exit(cell); + + align_free(cell); +} + + +static struct draw_context * +cell_draw_create(struct cell_context *cell) +{ + struct draw_context *draw = draw_create(); + + if (getenv("GALLIUM_CELL_VS")) { + /* plug in SPU-based vertex transformation code */ + draw->shader_queue_flush = cell_vertex_shader_queue_flush; + draw->driver_private = cell; + } + + return draw; +} + + +struct pipe_context * +cell_create_context(struct pipe_winsys *winsys, struct cell_winsys *cws) +{ + struct cell_context *cell; + uint spu, buf; + + /* some fields need to be 16-byte aligned, so align the whole object */ + cell = (struct cell_context*) align_malloc(sizeof(struct cell_context), 16); + if (!cell) + return NULL; + + memset(cell, 0, sizeof(*cell)); + + cell->winsys = cws; + cell->pipe.winsys = winsys; + cell->pipe.destroy = cell_destroy_context; + + /* queries */ + cell->pipe.is_format_supported = cell_is_format_supported; + cell->pipe.get_name = cell_get_name; + cell->pipe.get_vendor = cell_get_vendor; + cell->pipe.get_param = cell_get_param; + cell->pipe.get_paramf = cell_get_paramf; + + + /* state setters */ + cell->pipe.create_blend_state = cell_create_blend_state; + cell->pipe.bind_blend_state = cell_bind_blend_state; + cell->pipe.delete_blend_state = cell_delete_blend_state; + + cell->pipe.create_sampler_state = cell_create_sampler_state; + cell->pipe.bind_sampler_state = cell_bind_sampler_state; + cell->pipe.delete_sampler_state = cell_delete_sampler_state; + + cell->pipe.create_depth_stencil_alpha_state = cell_create_depth_stencil_alpha_state; + cell->pipe.bind_depth_stencil_alpha_state = cell_bind_depth_stencil_alpha_state; + cell->pipe.delete_depth_stencil_alpha_state = cell_delete_depth_stencil_alpha_state; + + cell->pipe.create_rasterizer_state = cell_create_rasterizer_state; + cell->pipe.bind_rasterizer_state = cell_bind_rasterizer_state; + cell->pipe.delete_rasterizer_state = cell_delete_rasterizer_state; + + cell->pipe.create_fs_state = cell_create_fs_state; + cell->pipe.bind_fs_state = cell_bind_fs_state; + cell->pipe.delete_fs_state = cell_delete_fs_state; + + cell->pipe.create_vs_state = cell_create_vs_state; + cell->pipe.bind_vs_state = cell_bind_vs_state; + cell->pipe.delete_vs_state = cell_delete_vs_state; + + cell->pipe.set_blend_color = cell_set_blend_color; + cell->pipe.set_clip_state = cell_set_clip_state; + cell->pipe.set_constant_buffer = cell_set_constant_buffer; + + cell->pipe.set_framebuffer_state = cell_set_framebuffer_state; + + cell->pipe.set_polygon_stipple = cell_set_polygon_stipple; + cell->pipe.set_scissor_state = cell_set_scissor_state; + cell->pipe.set_viewport_state = cell_set_viewport_state; + + cell->pipe.set_vertex_buffer = cell_set_vertex_buffer; + cell->pipe.set_vertex_element = cell_set_vertex_element; + + cell->pipe.draw_arrays = cell_draw_arrays; + cell->pipe.draw_elements = cell_draw_elements; + + cell->pipe.clear = cell_clear_surface; + cell->pipe.flush = cell_flush; + + /* textures */ + cell->pipe.texture_create = cell_texture_create; + cell->pipe.texture_release = cell_texture_release; + cell->pipe.get_tex_surface = cell_get_tex_surface; + + cell->pipe.set_sampler_texture = cell_set_sampler_texture; + +#if 0 + cell->pipe.begin_query = cell_begin_query; + cell->pipe.end_query = cell_end_query; + cell->pipe.wait_query = cell_wait_query; +#endif + + cell_init_surface_functions(cell); + + cell->draw = cell_draw_create(cell); + + cell_init_vbuf(cell); + draw_set_rasterize_stage(cell->draw, cell->vbuf); + + /* + * SPU stuff + */ + cell->num_spus = 6; + + cell_start_spus(cell); + + /* init command, vertex/index buffer info */ + for (buf = 0; buf < CELL_NUM_BUFFERS; buf++) { + cell->buffer_size[buf] = 0; + + /* init batch buffer status values, + * mark 0th buffer as used, rest as free. + */ + for (spu = 0; spu < cell->num_spus; spu++) { + if (buf == 0) + cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED; + else + cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_FREE; + } + } + + return &cell->pipe; +} diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h new file mode 100644 index 0000000000..3b63419b5e --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_context.h @@ -0,0 +1,135 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef CELL_CONTEXT_H +#define CELL_CONTEXT_H + + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/draw/draw_vertex.h" +#include "pipe/draw/draw_vbuf.h" +#include "cell_winsys.h" +#include "pipe/cell/common.h" + + +struct cell_vbuf_render; + +struct cell_vertex_shader_state +{ + struct pipe_shader_state shader; + void *draw_data; +}; + + +struct cell_fragment_shader_state +{ + struct pipe_shader_state shader; + void *data; +}; + + +struct cell_context +{ + struct pipe_context pipe; + + struct cell_winsys *winsys; + + const struct pipe_blend_state *blend; + const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + const struct pipe_depth_stencil_alpha_state *depth_stencil; + const struct pipe_rasterizer_state *rasterizer; + const struct cell_vertex_shader_state *vs; + const struct cell_fragment_shader_state *fs; + + struct pipe_blend_color blend_color; + struct pipe_clip_state clip; + struct pipe_constant_buffer constants[2]; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct cell_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX]; + struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX]; + + ubyte *cbuf_map[PIPE_MAX_COLOR_BUFS]; + ubyte *zsbuf_map; + + struct pipe_surface *tex_surf; + uint *tex_map; + + uint dirty; + + /** The primitive drawing context */ + struct draw_context *draw; + struct draw_stage *render_stage; + + /** For post-transformed vertex buffering: */ + struct cell_vbuf_render *vbuf_render; + struct draw_stage *vbuf; + + struct vertex_info vertex_info; + + /** Mapped constant buffers */ + void *mapped_constants[PIPE_SHADER_TYPES]; + + + uint num_spus; + + /** Buffers for command batches, vertex/index data */ + uint buffer_size[CELL_NUM_BUFFERS]; + ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE] ALIGN16_ATTRIB; + + int cur_batch; /**< which buffer is being filled w/ commands */ + + /** [4] to ensure 16-byte alignment for each status word */ + uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4] ALIGN16_ATTRIB; + +}; + + + + +static INLINE struct cell_context * +cell_context(struct pipe_context *pipe) +{ + return (struct cell_context *) pipe; +} + + +extern struct pipe_context * +cell_create_context(struct pipe_winsys *ws, struct cell_winsys *cws); + +extern void +cell_vertex_shader_queue_flush(struct draw_context *draw); + + + + +#endif /* CELL_CONTEXT_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c new file mode 100644 index 0000000000..717cd8370f --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c @@ -0,0 +1,164 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + * Keith Whitwell + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_context.h" +#include "pipe/p_winsys.h" + +#include "cell_context.h" +#include "cell_draw_arrays.h" +#include "cell_state.h" + +#include "pipe/draw/draw_context.h" + + + +static void +cell_map_constant_buffers(struct cell_context *sp) +{ + struct pipe_winsys *ws = sp->pipe.winsys; + uint i; + for (i = 0; i < 2; i++) { + if (sp->constants[i].size) + sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + } + + draw_set_mapped_constant_buffer(sp->draw, + sp->mapped_constants[PIPE_SHADER_VERTEX]); +} + +static void +cell_unmap_constant_buffers(struct cell_context *sp) +{ + struct pipe_winsys *ws = sp->pipe.winsys; + uint i; + for (i = 0; i < 2; i++) { + if (sp->constants[i].size) + ws->buffer_unmap(ws, sp->constants[i].buffer); + sp->mapped_constants[i] = NULL; + } +} + + +boolean +cell_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count) +{ + return cell_draw_elements(pipe, NULL, 0, mode, start, count); +} + + + +/** + * Draw vertex arrays, with optional indexing. + * Basically, map the vertex buffers (and drawing surfaces), then hand off + * the drawing to the 'draw' module. + * + * XXX should the element buffer be specified/bound with a separate function? + */ +boolean +cell_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count) +{ + struct cell_context *sp = cell_context(pipe); + struct draw_context *draw = sp->draw; + unsigned i; + + /* first, check that the primitive is not malformed. It is the + * state tracker's responsibility to do send only correctly formed + * primitives down. It currently isn't doing that though... + */ +#if 1 + count = draw_trim_prim( mode, count ); +#else + if (!draw_validate_prim( mode, count )) + assert(0); +#endif + + if (sp->dirty) + cell_update_derived( sp ); + +#if 0 + cell_map_surfaces(sp); +#endif + cell_map_constant_buffers(sp); + + /* + * Map vertex buffers + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (sp->vertex_buffer[i].buffer) { + void *buf = pipe->winsys->buffer_map(pipe->winsys, + sp->vertex_buffer[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_vertex_buffer(draw, i, buf); + } + } + /* Map index buffer, if present */ + if (indexBuffer) { + void *mapped_indexes = pipe->winsys->buffer_map(pipe->winsys, + indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); + } + else { + /* no index/element buffer */ + draw_set_mapped_element_buffer(draw, 0, NULL); + } + + + /* draw! */ + draw_arrays(draw, mode, start, count); + + /* + * unmap vertex/index buffers - will cause draw module to flush + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (sp->vertex_buffer[i].buffer) { + draw_set_mapped_vertex_buffer(draw, i, NULL); + pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer); + } + } + if (indexBuffer) { + draw_set_mapped_element_buffer(draw, 0, NULL); + pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + } + + /* Note: leave drawing surfaces mapped */ + cell_unmap_constant_buffers(sp); + + return TRUE; +} diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.h b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h new file mode 100644 index 0000000000..d5df4aa05f --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h @@ -0,0 +1,42 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_DRAW_ARRAYS_H +#define CELL_DRAW_ARRAYS_H + + +boolean cell_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count); + +boolean cell_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count); + + + +#endif /* CELL_DRAW_ARRAYS_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_flush.c b/src/gallium/drivers/cell/ppu/cell_flush.c new file mode 100644 index 0000000000..f62bc4650c --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_flush.c @@ -0,0 +1,84 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "cell_context.h" +#include "cell_batch.h" +#include "cell_flush.h" +#include "cell_spu.h" +#include "cell_render.h" +#include "pipe/draw/draw_context.h" + + +void +cell_flush(struct pipe_context *pipe, unsigned flags) +{ + struct cell_context *cell = cell_context(pipe); + + if (flags & PIPE_FLUSH_SWAPBUFFERS) + flags |= PIPE_FLUSH_WAIT; + + draw_flush( cell->draw ); + cell_flush_int(pipe, flags); +} + + +/** internal flush */ +void +cell_flush_int(struct pipe_context *pipe, unsigned flags) +{ + static boolean flushing = FALSE; /* recursion catcher */ + struct cell_context *cell = cell_context(pipe); + uint i; + + ASSERT(!flushing); + flushing = TRUE; + + if (flags & PIPE_FLUSH_WAIT) { + uint64_t *cmd = (uint64_t *) cell_batch_alloc(cell, sizeof(uint64_t)); + *cmd = CELL_CMD_FINISH; + } + + cell_batch_flush(cell); + +#if 0 + /* Send CMD_FINISH to all SPUs */ + for (i = 0; i < cell->num_spus; i++) { + send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_FINISH); + } +#endif + + if (flags & PIPE_FLUSH_WAIT) { + /* Wait for ack */ + for (i = 0; i < cell->num_spus; i++) { + uint k = wait_mbox_message(cell_global.spe_contexts[i]); + assert(k == CELL_CMD_FINISH); + } + } + + flushing = FALSE; +} diff --git a/src/gallium/drivers/cell/ppu/cell_flush.h b/src/gallium/drivers/cell/ppu/cell_flush.h new file mode 100644 index 0000000000..eda351b1cb --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_flush.h @@ -0,0 +1,38 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef CELL_FLUSH +#define CELL_FLUSH + +extern void +cell_flush(struct pipe_context *pipe, unsigned flags); + +extern void +cell_flush_int(struct pipe_context *pipe, unsigned flags); + +#endif diff --git a/src/gallium/drivers/cell/ppu/cell_render.c b/src/gallium/drivers/cell/ppu/cell_render.c new file mode 100644 index 0000000000..4ab277a4b2 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_render.c @@ -0,0 +1,210 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief Last stage of 'draw' pipeline: send tris to SPUs. + * \author Brian Paul + */ + +#include "cell_context.h" +#include "cell_render.h" +#include "cell_spu.h" +#include "pipe/p_util.h" +#include "pipe/draw/draw_private.h" + + +struct render_stage { + struct draw_stage stage; /**< This must be first (base class) */ + + struct cell_context *cell; +}; + + +static INLINE struct render_stage * +render_stage(struct draw_stage *stage) +{ + return (struct render_stage *) stage; +} + + +static void render_begin( struct draw_stage *stage ) +{ +#if 0 + struct render_stage *render = render_stage(stage); + struct cell_context *sp = render->cell; + const struct pipe_shader_state *fs = &render->cell->fs->shader; + render->quad.nr_attrs = render->cell->nr_frag_attrs; + + render->firstFpInput = fs->input_semantic_name[0]; + + sp->quad.first->begin(sp->quad.first); +#endif +} + + +static void render_end( struct draw_stage *stage ) +{ +} + + +static void reset_stipple_counter( struct draw_stage *stage ) +{ + struct render_stage *render = render_stage(stage); + /*render->cell->line_stipple_counter = 0;*/ +} + + +static void +render_point(struct draw_stage *stage, struct prim_header *prim) +{ +} + + +static void +render_line(struct draw_stage *stage, struct prim_header *prim) +{ +} + + +/** Write a vertex into the prim buffer */ +static void +save_vertex(struct cell_prim_buffer *buf, uint pos, + const struct vertex_header *vert) +{ + uint attr, j; + + for (attr = 0; attr < 2; attr++) { + for (j = 0; j < 4; j++) { + buf->vertex[pos][attr][j] = vert->data[attr][j]; + } + } + + /* update bounding box */ + if (vert->data[0][0] < buf->xmin) + buf->xmin = vert->data[0][0]; + if (vert->data[0][0] > buf->xmax) + buf->xmax = vert->data[0][0]; + if (vert->data[0][1] < buf->ymin) + buf->ymin = vert->data[0][1]; + if (vert->data[0][1] > buf->ymax) + buf->ymax = vert->data[0][1]; +} + + +static void +render_tri(struct draw_stage *stage, struct prim_header *prim) +{ + struct render_stage *rs = render_stage(stage); + struct cell_context *cell = rs->cell; + struct cell_prim_buffer *buf = &cell->prim_buffer; + uint i; + + if (buf->num_verts + 3 > CELL_MAX_VERTS) { + cell_flush_prim_buffer(cell); + } + + i = buf->num_verts; + assert(i+2 <= CELL_MAX_VERTS); + save_vertex(buf, i+0, prim->v[0]); + save_vertex(buf, i+1, prim->v[1]); + save_vertex(buf, i+2, prim->v[2]); + buf->num_verts += 3; +} + + +/** + * Send the a RENDER command to all SPUs to have them render the prims + * in the current prim_buffer. + */ +void +cell_flush_prim_buffer(struct cell_context *cell) +{ + uint i; + + if (cell->prim_buffer.num_verts == 0) + return; + + for (i = 0; i < cell->num_spus; i++) { + struct cell_command_render *render = &cell_global.command[i].render; + render->prim_type = PIPE_PRIM_TRIANGLES; + render->num_verts = cell->prim_buffer.num_verts; + render->vertex_size = cell->vertex_info->size * 4; + render->xmin = cell->prim_buffer.xmin; + render->ymin = cell->prim_buffer.ymin; + render->xmax = cell->prim_buffer.xmax; + render->ymax = cell->prim_buffer.ymax; + render->vertex_data = &cell->prim_buffer.vertex; + ASSERT_ALIGN16(render->vertex_data); + send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_RENDER); + } + + cell->prim_buffer.num_verts = 0; + + cell->prim_buffer.xmin = 1e100; + cell->prim_buffer.ymin = 1e100; + cell->prim_buffer.xmax = -1e100; + cell->prim_buffer.ymax = -1e100; + + /* XXX temporary, need to double-buffer the prim buffer until we get + * a real command buffer/list system. + */ + cell_flush(&cell->pipe, 0x0); +} + + + +static void render_destroy( struct draw_stage *stage ) +{ + FREE( stage ); +} + + +/** + * Create a new draw/render stage. This will be plugged into the + * draw module as the last pipeline stage. + */ +struct draw_stage *cell_draw_render_stage( struct cell_context *cell ) +{ + struct render_stage *render = CALLOC_STRUCT(render_stage); + + render->cell = cell; + render->stage.draw = cell->draw; + render->stage.begin = render_begin; + render->stage.point = render_point; + render->stage.line = render_line; + render->stage.tri = render_tri; + render->stage.end = render_end; + render->stage.reset_stipple_counter = reset_stipple_counter; + render->stage.destroy = render_destroy; + + /* + render->quad.coef = render->coef; + render->quad.posCoef = &render->posCoef; + */ + + return &render->stage; +} diff --git a/src/gallium/drivers/cell/ppu/cell_render.h b/src/gallium/drivers/cell/ppu/cell_render.h new file mode 100644 index 0000000000..826dcbafeb --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_render.h @@ -0,0 +1,39 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_RENDER_H +#define CELL_RENDER_H + +struct cell_context; +struct draw_stage; + +extern void +cell_flush_prim_buffer(struct cell_context *cell); + +extern struct draw_stage *cell_draw_render_stage( struct cell_context *cell ); + +#endif /* CELL_RENDER_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_spu.c b/src/gallium/drivers/cell/ppu/cell_spu.c new file mode 100644 index 0000000000..7c83a47e57 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_spu.c @@ -0,0 +1,155 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <pthread.h> + +#include "cell_spu.h" +#include "pipe/p_format.h" +#include "pipe/p_state.h" +#include "pipe/cell/common.h" + + +/* +helpful headers: +/opt/ibm/cell-sdk/prototype/src/include/ppu/cbe_mfc.h +*/ + + +struct cell_global_info cell_global; + + +/** + * Write a 1-word message to the given SPE mailbox. + */ +void +send_mbox_message(spe_context_ptr_t ctx, unsigned int msg) +{ + spe_in_mbox_write(ctx, &msg, 1, SPE_MBOX_ALL_BLOCKING); +} + + +/** + * Wait for a 1-word message to arrive in given mailbox. + */ +uint +wait_mbox_message(spe_context_ptr_t ctx) +{ + do { + unsigned data; + int count = spe_out_mbox_read(ctx, &data, 1); + + if (count == 1) { + return data; + } + + if (count < 0) { + /* error */ ; + } + } while (1); +} + + +static void *cell_thread_function(void *arg) +{ + struct cell_init_info *init = (struct cell_init_info *) arg; + unsigned entry = SPE_DEFAULT_ENTRY; + + ASSERT_ALIGN16(init); + + if (spe_context_run(cell_global.spe_contexts[init->id], &entry, 0, + init, NULL, NULL) < 0) { + fprintf(stderr, "spe_context_run() failed\n"); + exit(1); + } + + pthread_exit(NULL); +} + + +/** + * Create the SPU threads + */ +void +cell_start_spus(struct cell_context *cell) +{ + uint i, j; + + assert(cell->num_spus <= MAX_SPUS); + + ASSERT_ALIGN16(&cell_global.command[0]); + ASSERT_ALIGN16(&cell_global.command[1]); + + ASSERT_ALIGN16(&cell_global.inits[0]); + ASSERT_ALIGN16(&cell_global.inits[1]); + + for (i = 0; i < cell->num_spus; i++) { + cell_global.inits[i].id = i; + cell_global.inits[i].num_spus = cell->num_spus; + cell_global.inits[i].cmd = &cell_global.command[i]; + for (j = 0; j < CELL_NUM_BUFFERS; j++) { + cell_global.inits[i].buffers[j] = cell->buffer[j]; + } + cell_global.inits[i].buffer_status = &cell->buffer_status[0][0][0]; + + cell_global.spe_contexts[i] = spe_context_create(0, NULL); + if (!cell_global.spe_contexts[i]) { + fprintf(stderr, "spe_context_create() failed\n"); + exit(1); + } + + if (spe_program_load(cell_global.spe_contexts[i], &g3d_spu)) { + fprintf(stderr, "spe_program_load() failed\n"); + exit(1); + } + + pthread_create(&cell_global.spe_threads[i], NULL, &cell_thread_function, + &cell_global.inits[i]); + } +} + + +/** + * Tell all the SPUs to stop/exit. + */ +void +cell_spu_exit(struct cell_context *cell) +{ + uint i; + + for (i = 0; i < cell->num_spus; i++) { + send_mbox_message(cell_global.spe_contexts[i], CELL_CMD_EXIT); + } + + /* wait for threads to exit */ + for (i = 0; i < cell->num_spus; i++) { + void *value; + pthread_join(cell_global.spe_threads[i], &value); + cell_global.spe_threads[i] = 0; + cell_global.spe_contexts[i] = 0; + } +} diff --git a/src/gallium/drivers/cell/ppu/cell_spu.h b/src/gallium/drivers/cell/ppu/cell_spu.h new file mode 100644 index 0000000000..19eff94f96 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_spu.h @@ -0,0 +1,82 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_SPU +#define CELL_SPU + + +#include <libspe2.h> +#include <libmisc.h> +#include "pipe/cell/common.h" + +#include "cell_context.h" + + +#define MAX_SPUS 8 + +/** + * Global vars, for now anyway. + */ +struct cell_global_info +{ + /** + * SPU/SPE handles, etc + */ + spe_context_ptr_t spe_contexts[MAX_SPUS]; + pthread_t spe_threads[MAX_SPUS]; + + /** + * Data sent to SPUs + */ + struct cell_init_info inits[MAX_SPUS]; + struct cell_command command[MAX_SPUS]; +}; + + +extern struct cell_global_info cell_global; + + +/** This is the handle for the actual SPE code */ +extern spe_program_handle_t g3d_spu; + + +extern void +send_mbox_message(spe_context_ptr_t ctx, unsigned int msg); + +extern uint +wait_mbox_message(spe_context_ptr_t ctx); + + +extern void +cell_start_spus(struct cell_context *cell); + + +extern void +cell_spu_exit(struct cell_context *cell); + + +#endif /* CELL_SPU */ diff --git a/src/gallium/drivers/cell/ppu/cell_state.h b/src/gallium/drivers/cell/ppu/cell_state.h new file mode 100644 index 0000000000..3a71ba14fa --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state.h @@ -0,0 +1,115 @@ + + +#ifndef CELL_STATE_H +#define CELL_STATE_H + + +#define CELL_NEW_VIEWPORT 0x1 +#define CELL_NEW_RASTERIZER 0x2 +#define CELL_NEW_FS 0x4 +#define CELL_NEW_BLEND 0x8 +#define CELL_NEW_CLIP 0x10 +#define CELL_NEW_SCISSOR 0x20 +#define CELL_NEW_STIPPLE 0x40 +#define CELL_NEW_FRAMEBUFFER 0x80 +#define CELL_NEW_ALPHA_TEST 0x100 +#define CELL_NEW_DEPTH_STENCIL 0x200 +#define CELL_NEW_SAMPLER 0x400 +#define CELL_NEW_TEXTURE 0x800 +#define CELL_NEW_VERTEX 0x1000 +#define CELL_NEW_VS 0x2000 +#define CELL_NEW_CONSTANTS 0x4000 +#define CELL_NEW_VERTEX_INFO 0x8000 + + + +extern void +cell_set_framebuffer_state( struct pipe_context *, + const struct pipe_framebuffer_state * ); + + + +extern void * +cell_create_blend_state(struct pipe_context *, const struct pipe_blend_state *); +extern void cell_bind_blend_state(struct pipe_context *, void *); +extern void cell_delete_blend_state(struct pipe_context *, void *); + +extern void cell_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ); + + +void * +cell_create_sampler_state(struct pipe_context *, + const struct pipe_sampler_state *); + +extern void +cell_bind_sampler_state(struct pipe_context *, unsigned, void *); + +extern void +cell_delete_sampler_state(struct pipe_context *, void *); + + +extern void * +cell_create_depth_stencil_alpha_state(struct pipe_context *, + const struct pipe_depth_stencil_alpha_state *); + +extern void +cell_bind_depth_stencil_alpha_state(struct pipe_context *, void *); + +extern void +cell_delete_depth_stencil_alpha_state(struct pipe_context *, void *); + + +void *cell_create_fs_state(struct pipe_context *, + const struct pipe_shader_state *); +void cell_bind_fs_state(struct pipe_context *, void *); +void cell_delete_fs_state(struct pipe_context *, void *); +void *cell_create_vs_state(struct pipe_context *, + const struct pipe_shader_state *); +void cell_bind_vs_state(struct pipe_context *, void *); +void cell_delete_vs_state(struct pipe_context *, void *); + + +void * +cell_create_rasterizer_state(struct pipe_context *, + const struct pipe_rasterizer_state *); +void cell_bind_rasterizer_state(struct pipe_context *, void *); +void cell_delete_rasterizer_state(struct pipe_context *, void *); + + +void cell_set_clip_state( struct pipe_context *, + const struct pipe_clip_state * ); + +void cell_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf); + +void cell_set_polygon_stipple( struct pipe_context *, + const struct pipe_poly_stipple * ); + +void +cell_set_sampler_texture(struct pipe_context *pipe, + unsigned sampler, + struct pipe_texture *texture); + +void cell_set_scissor_state( struct pipe_context *, + const struct pipe_scissor_state * ); + +void cell_set_texture_state( struct pipe_context *, + unsigned unit, struct pipe_texture * ); + +void cell_set_vertex_element(struct pipe_context *, + unsigned index, + const struct pipe_vertex_element *); + +void cell_set_vertex_buffer(struct pipe_context *, + unsigned index, + const struct pipe_vertex_buffer *); + +void cell_set_viewport_state( struct pipe_context *, + const struct pipe_viewport_state * ); + + +void cell_update_derived( struct cell_context *softpipe ); + +#endif diff --git a/src/gallium/drivers/cell/ppu/cell_state_blend.c b/src/gallium/drivers/cell/ppu/cell_state_blend.c new file mode 100644 index 0000000000..4fc60548c8 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_blend.c @@ -0,0 +1,109 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "pipe/p_util.h" +#include "pipe/draw/draw_context.h" +#include "cell_context.h" +#include "cell_state.h" + + + +void * +cell_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + return mem_dup(blend, sizeof(*blend)); +} + + +void +cell_bind_blend_state(struct pipe_context *pipe, void *blend) +{ + struct cell_context *cell = cell_context(pipe); + + draw_flush(cell->draw); + + cell->blend = (const struct pipe_blend_state *)blend; + + cell->dirty |= CELL_NEW_BLEND; +} + + +void +cell_delete_blend_state(struct pipe_context *pipe, void *blend) +{ + FREE(blend); +} + + +void +cell_set_blend_color(struct pipe_context *pipe, + const struct pipe_blend_color *blend_color) +{ + struct cell_context *cell = cell_context(pipe); + + draw_flush(cell->draw); + + cell->blend_color = *blend_color; + + cell->dirty |= CELL_NEW_BLEND; +} + + + + +void * +cell_create_depth_stencil_alpha_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + return mem_dup(depth_stencil, sizeof(*depth_stencil)); +} + + +void +cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct cell_context *cell = cell_context(pipe); + + draw_flush(cell->draw); + + cell->depth_stencil + = (const struct pipe_depth_stencil_alpha_state *) depth_stencil; + + cell->dirty |= CELL_NEW_DEPTH_STENCIL; +} + + +void +cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth) +{ + FREE(depth); +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_clip.c b/src/gallium/drivers/cell/ppu/cell_state_clip.c new file mode 100644 index 0000000000..4f43665941 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_clip.c @@ -0,0 +1,84 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "cell_context.h" +#include "cell_state.h" +#include "pipe/draw/draw_context.h" + + +void cell_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct cell_context *cell = cell_context(pipe); + + /* pass the clip state to the draw module */ + draw_set_clip_state(cell->draw, clip); +} + + + +/* Called when driver state tracker notices changes to the viewport + * matrix: + */ +void cell_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct cell_context *cell = cell_context(pipe); + + cell->viewport = *viewport; /* struct copy */ + cell->dirty |= CELL_NEW_VIEWPORT; + + /* pass the viewport info to the draw module */ + draw_set_viewport_state(cell->draw, viewport); + + /* Using tnl/ and vf/ modules is temporary while getting started. + * Full pipe will have vertex shader, vertex fetch of its own. + */ +} + + +void cell_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct cell_context *cell = cell_context(pipe); + + memcpy( &cell->scissor, scissor, sizeof(*scissor) ); + cell->dirty |= CELL_NEW_SCISSOR; +} + + +void cell_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ + struct cell_context *cell = cell_context(pipe); + + memcpy( &cell->poly_stipple, stipple, sizeof(*stipple) ); + cell->dirty |= CELL_NEW_STIPPLE; +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_derived.c b/src/gallium/drivers/cell/ppu/cell_state_derived.c new file mode 100644 index 0000000000..56daf5dfde --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_derived.c @@ -0,0 +1,192 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_vertex.h" +#include "cell_context.h" +#include "cell_batch.h" +#include "cell_state.h" +#include "cell_state_emit.h" + + +static int +find_vs_output(const struct pipe_shader_state *vs, + uint semantic_name, + uint semantic_index) +{ + uint i; + for (i = 0; i < vs->num_outputs; i++) { + if (vs->output_semantic_name[i] == semantic_name && + vs->output_semantic_index[i] == semantic_index) + return i; + } + return -1; +} + + +/** + * Determine how to map vertex program outputs to fragment program inputs. + * Basically, this will be used when computing the triangle interpolation + * coefficients from the post-transform vertex attributes. + */ +static void +calculate_vertex_layout( struct cell_context *cell ) +{ + const struct pipe_shader_state *vs = &cell->vs->shader; + const struct pipe_shader_state *fs = &cell->fs->shader; + const enum interp_mode colorInterp + = cell->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR; + struct vertex_info *vinfo = &cell->vertex_info; + uint i; + int src; + +#if 0 + if (cell->vbuf) { + /* if using the post-transform vertex buffer, tell draw_vbuf to + * simply emit the whole post-xform vertex as-is: + */ + struct vertex_info *vinfo_vbuf = &cell->vertex_info_vbuf; + vinfo_vbuf->num_attribs = 0; + draw_emit_vertex_attr(vinfo_vbuf, EMIT_ALL, INTERP_NONE, 0); + vinfo_vbuf->size = 4 * vs->num_outputs + sizeof(struct vertex_header)/4; + } +#endif + + /* reset vinfo */ + vinfo->num_attribs = 0; + + /* we always want to emit vertex pos */ + src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0); + assert(src >= 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src); + + + /* + * Loop over fragment shader inputs, searching for the matching output + * from the vertex shader. + */ + for (i = 0; i < fs->num_inputs; i++) { + switch (fs->input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + /* already done above */ + break; + + case TGSI_SEMANTIC_COLOR: + src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, + fs->input_semantic_index[i]); + assert(src >= 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src); + break; + + case TGSI_SEMANTIC_FOG: + src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0); +#if 1 + if (src < 0) /* XXX temp hack, try demos/fogcoord.c with this */ + src = 0; +#endif + assert(src >= 0); + draw_emit_vertex_attr(vinfo, EMIT_1F, INTERP_PERSPECTIVE, src); + break; + + case TGSI_SEMANTIC_GENERIC: + /* this includes texcoords and varying vars */ + src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC, + fs->input_semantic_index[i]); + assert(src >= 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src); + break; + + default: + assert(0); + } + } + + draw_compute_vertex_size(vinfo); + + /* XXX only signal this if format really changes */ + cell->dirty |= CELL_NEW_VERTEX_INFO; +} + + +#if 0 +/** + * Recompute cliprect from scissor bounds, scissor enable and surface size. + */ +static void +compute_cliprect(struct cell_context *sp) +{ + unsigned surfWidth, surfHeight; + + if (sp->framebuffer.num_cbufs > 0) { + surfWidth = sp->framebuffer.cbufs[0]->width; + surfHeight = sp->framebuffer.cbufs[0]->height; + } + else { + /* no surface? */ + surfWidth = sp->scissor.maxx; + surfHeight = sp->scissor.maxy; + } + + if (sp->rasterizer->scissor) { + /* clip to scissor rect */ + sp->cliprect.minx = MAX2(sp->scissor.minx, 0); + sp->cliprect.miny = MAX2(sp->scissor.miny, 0); + sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth); + sp->cliprect.maxy = MIN2(sp->scissor.maxy, surfHeight); + } + else { + /* clip to surface bounds */ + sp->cliprect.minx = 0; + sp->cliprect.miny = 0; + sp->cliprect.maxx = surfWidth; + sp->cliprect.maxy = surfHeight; + } +} +#endif + + + +void cell_update_derived( struct cell_context *cell ) +{ + if (cell->dirty & (CELL_NEW_RASTERIZER | + CELL_NEW_FS | + CELL_NEW_VS)) + calculate_vertex_layout( cell ); + +#if 0 + if (cell->dirty & (CELL_NEW_SCISSOR | + CELL_NEW_DEPTH_STENCIL_ALPHA | + CELL_NEW_FRAMEBUFFER)) + compute_cliprect(cell); +#endif + + cell_emit_state(cell); + + cell->dirty = 0; +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c new file mode 100644 index 0000000000..5d2a786449 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c @@ -0,0 +1,103 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_util.h" +#include "cell_context.h" +#include "cell_state.h" +#include "cell_state_emit.h" +#include "cell_batch.h" +#include "cell_texture.h" + + +static void +emit_state_cmd(struct cell_context *cell, uint cmd, + const void *state, uint state_size) +{ + uint64_t *dst = (uint64_t *) + cell_batch_alloc(cell, ROUNDUP8(sizeof(uint64_t) + state_size)); + *dst = cmd; + memcpy(dst + 1, state, state_size); +} + + + +void +cell_emit_state(struct cell_context *cell) +{ + if (cell->dirty & CELL_NEW_FRAMEBUFFER) { + struct pipe_surface *cbuf = cell->framebuffer.cbufs[0]; + struct pipe_surface *zbuf = cell->framebuffer.zsbuf; + struct cell_command_framebuffer *fb + = cell_batch_alloc(cell, sizeof(*fb)); + fb->opcode = CELL_CMD_STATE_FRAMEBUFFER; + fb->color_start = cell->cbuf_map[0]; + fb->color_format = cbuf->format; + fb->depth_start = cell->zsbuf_map; + fb->depth_format = zbuf ? zbuf->format : PIPE_FORMAT_NONE; + fb->width = cell->framebuffer.cbufs[0]->width; + fb->height = cell->framebuffer.cbufs[0]->height; + } + + if (cell->dirty & CELL_NEW_BLEND) { + emit_state_cmd(cell, CELL_CMD_STATE_BLEND, + cell->blend, + sizeof(struct pipe_blend_state)); + } + + if (cell->dirty & CELL_NEW_DEPTH_STENCIL) { + emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL, + cell->depth_stencil, + sizeof(struct pipe_depth_stencil_alpha_state)); + } + + if (cell->dirty & CELL_NEW_SAMPLER) { + emit_state_cmd(cell, CELL_CMD_STATE_SAMPLER, + cell->sampler[0], sizeof(struct pipe_sampler_state)); + } + + if (cell->dirty & CELL_NEW_TEXTURE) { + struct cell_command_texture texture; + if (cell->texture[0]) { + texture.start = cell->texture[0]->tiled_data; + texture.width = cell->texture[0]->base.width[0]; + texture.height = cell->texture[0]->base.height[0]; + } + else { + texture.start = NULL; + texture.width = 0; + texture.height = 0; + } + + emit_state_cmd(cell, CELL_CMD_STATE_TEXTURE, + &texture, sizeof(struct cell_command_texture)); + } + + if (cell->dirty & CELL_NEW_VERTEX_INFO) { + emit_state_cmd(cell, CELL_CMD_STATE_VERTEX_INFO, + &cell->vertex_info, sizeof(struct vertex_info)); + } +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.h b/src/gallium/drivers/cell/ppu/cell_state_emit.h new file mode 100644 index 0000000000..59f8affe8d --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_emit.h @@ -0,0 +1,36 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_STATE_EMIT_H +#define CELL_STATE_EMIT_H + + +extern void +cell_emit_state(struct cell_context *cell); + + +#endif /* CELL_STATE_EMIT_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_state_fs.c b/src/gallium/drivers/cell/ppu/cell_state_fs.c new file mode 100644 index 0000000000..3f46a87d18 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_fs.c @@ -0,0 +1,171 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/draw/draw_context.h" +#if 0 +#include "pipe/p_shader_tokens.h" +#include "pipe/llvm/gallivm.h" +#include "pipe/tgsi/util/tgsi_dump.h" +#include "pipe/tgsi/exec/tgsi_sse2.h" +#endif + +#include "cell_context.h" +#include "cell_state.h" + + +void * +cell_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + /*struct cell_context *cell = cell_context(pipe);*/ + struct cell_fragment_shader_state *state; + + state = CALLOC_STRUCT(cell_fragment_shader_state); + if (!state) + return NULL; + + state->shader = *templ; + +#if 0 + if (cell->dump_fs) { + tgsi_dump(state->shader.tokens, 0); + } + +#if defined(__i386__) || defined(__386__) + if (cell->use_sse) { + x86_init_func( &state->sse2_program ); + tgsi_emit_sse2_fs( state->shader.tokens, &state->sse2_program ); + } +#endif + +#ifdef MESA_LLVM + state->llvm_prog = 0; + if (!gallivm_global_cpu_engine()) { + gallivm_cpu_engine_create(state->llvm_prog); + } + else + gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog); +#endif +#endif + + return state; +} + + +void +cell_bind_fs_state(struct pipe_context *pipe, void *fs) +{ + struct cell_context *cell = cell_context(pipe); + + cell->fs = (struct cell_fragment_shader_state *) fs; + + cell->dirty |= CELL_NEW_FS; +} + + +void +cell_delete_fs_state(struct pipe_context *pipe, void *fs) +{ + struct cell_fragment_shader_state *state = + (struct cell_fragment_shader_state *) fs; + + FREE( state ); +} + + +void * +cell_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct cell_context *cell = cell_context(pipe); + struct cell_vertex_shader_state *state; + + state = CALLOC_STRUCT(cell_vertex_shader_state); + if (!state) + return NULL; + + state->shader = *templ; + + state->draw_data = draw_create_vertex_shader(cell->draw, &state->shader); + if (state->draw_data == NULL) { + FREE( state ); + return NULL; + } + + return state; +} + + +void +cell_bind_vs_state(struct pipe_context *pipe, void *vs) +{ + struct cell_context *cell = cell_context(pipe); + + cell->vs = (const struct cell_vertex_shader_state *) vs; + + draw_bind_vertex_shader(cell->draw, cell->vs->draw_data); + + cell->dirty |= CELL_NEW_VS; +} + + +void +cell_delete_vs_state(struct pipe_context *pipe, void *vs) +{ + struct cell_context *cell = cell_context(pipe); + + struct cell_vertex_shader_state *state = + (struct cell_vertex_shader_state *) vs; + + draw_delete_vertex_shader(cell->draw, state->draw_data); + FREE( state ); +} + + +void +cell_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct cell_context *cell = cell_context(pipe); + struct pipe_winsys *ws = pipe->winsys; + + assert(shader < PIPE_SHADER_TYPES); + assert(index == 0); + + /* note: reference counting */ + pipe_buffer_reference(ws, + &cell->constants[shader].buffer, + buf->buffer); + cell->constants[shader].size = buf->size; + + cell->dirty |= CELL_NEW_CONSTANTS; +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c b/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c new file mode 100644 index 0000000000..d8128ece54 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c @@ -0,0 +1,106 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/draw/draw_context.h" +#include "cell_context.h" +#include "cell_state.h" + + + +struct spu_rasterizer_state +{ + unsigned flatshade:1; +#if 0 + unsigned light_twoside:1; + unsigned front_winding:2; /**< PIPE_WINDING_x */ + unsigned cull_mode:2; /**< PIPE_WINDING_x */ + unsigned fill_cw:2; /**< PIPE_POLYGON_MODE_x */ + unsigned fill_ccw:2; /**< PIPE_POLYGON_MODE_x */ + unsigned offset_cw:1; + unsigned offset_ccw:1; +#endif + unsigned scissor:1; + unsigned poly_smooth:1; + unsigned poly_stipple_enable:1; + unsigned point_smooth:1; +#if 0 + unsigned point_sprite:1; + unsigned point_size_per_vertex:1; /**< size computed in vertex shader */ +#endif + unsigned multisample:1; /* XXX maybe more ms state in future */ + unsigned line_smooth:1; + unsigned line_stipple_enable:1; + unsigned line_stipple_factor:8; /**< [1..256] actually */ + unsigned line_stipple_pattern:16; +#if 0 + unsigned bypass_clipping:1; +#endif + unsigned origin_lower_left:1; /**< Is (0,0) the lower-left corner? */ + + float line_width; + float point_size; /**< used when no per-vertex size */ +#if 0 + float offset_units; + float offset_scale; + ubyte sprite_coord_mode[PIPE_MAX_SHADER_OUTPUTS]; /**< PIPE_SPRITE_COORD_ */ +#endif +}; + + + +void * +cell_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *setup) +{ + struct pipe_rasterizer_state *state + = MALLOC(sizeof(struct pipe_rasterizer_state)); + memcpy(state, setup, sizeof(struct pipe_rasterizer_state)); + return state; +} + + +void +cell_bind_rasterizer_state(struct pipe_context *pipe, void *setup) +{ + struct cell_context *cell = cell_context(pipe); + + /* pass-through to draw module */ + draw_set_rasterizer_state(cell->draw, setup); + + cell->rasterizer = (struct pipe_rasterizer_state *)setup; + + cell->dirty |= CELL_NEW_RASTERIZER; +} + + +void +cell_delete_rasterizer_state(struct pipe_context *pipe, void *rasterizer) +{ + FREE(rasterizer); +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_sampler.c b/src/gallium/drivers/cell/ppu/cell_state_sampler.c new file mode 100644 index 0000000000..ade6cc8338 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_sampler.c @@ -0,0 +1,84 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: + * Brian Paul + */ + +#include "pipe/p_util.h" +#include "pipe/draw/draw_context.h" +#include "cell_context.h" +#include "cell_state.h" +#include "cell_texture.h" + + +void * +cell_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + return mem_dup(sampler, sizeof(*sampler)); +} + +void +cell_bind_sampler_state(struct pipe_context *pipe, + unsigned unit, void *sampler) +{ + struct cell_context *cell = cell_context(pipe); + + draw_flush(cell->draw); + + assert(unit < PIPE_MAX_SAMPLERS); + cell->sampler[unit] = (struct pipe_sampler_state *)sampler; + + cell->dirty |= CELL_NEW_SAMPLER; +} + + +void +cell_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + FREE( sampler ); +} + + + +void +cell_set_sampler_texture(struct pipe_context *pipe, + unsigned sampler, + struct pipe_texture *texture) +{ + struct cell_context *cell = cell_context(pipe); + + draw_flush(cell->draw); + + cell->texture[sampler] = texture; + + cell_update_texture_mapping(cell); + + cell->dirty |= CELL_NEW_TEXTURE; +} diff --git a/src/gallium/drivers/cell/ppu/cell_state_surface.c b/src/gallium/drivers/cell/ppu/cell_state_surface.c new file mode 100644 index 0000000000..287610b76b --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_surface.c @@ -0,0 +1,71 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_inlines.h" +#include "cell_context.h" +#include "cell_state.h" + + +void +cell_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct cell_context *cell = cell_context(pipe); + + if (1 /*memcmp(&cell->framebuffer, fb, sizeof(*fb))*/) { + struct pipe_surface *csurf = fb->cbufs[0]; + struct pipe_surface *zsurf = fb->zsbuf; + uint i; + + /* unmap old surfaces */ + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) { + if (cell->framebuffer.cbufs[i] && cell->cbuf_map[i]) { + pipe_surface_unmap(cell->framebuffer.cbufs[i]); + cell->cbuf_map[i] = NULL; + } + } + + if (cell->framebuffer.zsbuf && cell->zsbuf_map) { + pipe_surface_unmap(cell->framebuffer.zsbuf); + cell->zsbuf_map = NULL; + } + + /* update my state */ + cell->framebuffer = *fb; + + /* map new surfaces */ + if (csurf) + cell->cbuf_map[0] = pipe_surface_map(csurf); + + if (zsurf) + cell->zsbuf_map = pipe_surface_map(zsurf); + + cell->dirty |= CELL_NEW_FRAMEBUFFER; + } +} + diff --git a/src/gallium/drivers/cell/ppu/cell_state_vertex.c b/src/gallium/drivers/cell/ppu/cell_state_vertex.c new file mode 100644 index 0000000000..0f01e920f9 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_state_vertex.c @@ -0,0 +1,63 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "cell_context.h" +#include "cell_state.h" + +#include "pipe/draw/draw_context.h" + + +void +cell_set_vertex_element(struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_element *attrib) +{ + struct cell_context *cell = cell_context(pipe); + assert(index < PIPE_ATTRIB_MAX); + cell->vertex_element[index] = *attrib; /* struct copy */ + cell->dirty |= CELL_NEW_VERTEX; + + draw_set_vertex_element(cell->draw, index, attrib); +} + + +void +cell_set_vertex_buffer(struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_buffer *buffer) +{ + struct cell_context *cell = cell_context(pipe); + assert(index < PIPE_ATTRIB_MAX); + cell->vertex_buffer[index] = *buffer; /* struct copy */ + cell->dirty |= CELL_NEW_VERTEX; + + draw_set_vertex_buffer(cell->draw, index, buffer); +} diff --git a/src/gallium/drivers/cell/ppu/cell_surface.c b/src/gallium/drivers/cell/ppu/cell_surface.c new file mode 100644 index 0000000000..fca93e4742 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_surface.c @@ -0,0 +1,179 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/util/p_tile.h" +#include "cell_context.h" +#include "cell_surface.h" + + +/* Upload data to a rectangular sub-region. Lots of choices how to do this: + * + * - memcpy by span to current destination + * - upload data as new buffer and blit + * + * Currently always memcpy. + */ +static void +cell_surface_data(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + const void *src, unsigned src_pitch, + unsigned srcx, unsigned srcy, + unsigned width, unsigned height) +{ + pipe_copy_rect(pipe_surface_map(dst), + dst->cpp, + dst->pitch, + dstx, dsty, width, height, src, src_pitch, srcx, srcy); + + pipe_surface_unmap(dst); +} + + +static void +cell_surface_copy(struct pipe_context *pipe, + unsigned do_flip, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + struct pipe_surface *src, + unsigned srcx, unsigned srcy, + unsigned width, unsigned height) +{ + assert( dst->cpp == src->cpp ); + + pipe_copy_rect(pipe_surface_map(dst), + dst->cpp, + dst->pitch, + dstx, dsty, + width, height, + pipe_surface_map(src), + do_flip ? -src->pitch : src->pitch, + srcx, do_flip ? 1 - srcy - height : srcy); + + pipe_surface_unmap(src); + pipe_surface_unmap(dst); +} + + +static void * +get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y) +{ + return (char *)dst_map + (y * dst->pitch + x) * dst->cpp; +} + + +#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8)) + + +/** + * Fill a rectangular sub-region. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +static void +cell_surface_fill(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, unsigned value) +{ + unsigned i, j; + void *dst_map = pipe_surface_map(dst); + + assert(dst->pitch > 0); + assert(width <= dst->pitch); + + switch (dst->cpp) { + case 1: + { + ubyte *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + memset(row, value, width); + row += dst->pitch; + } + } + break; + case 2: + { + ushort *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = (ushort) value; + row += dst->pitch; + } + } + break; + case 4: + { + unsigned *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = value; + row += dst->pitch; + } + } + break; + case 8: + { + /* expand the 4-byte clear value to an 8-byte value */ + ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty); + ushort val0 = UBYTE_TO_USHORT((value >> 0) & 0xff); + ushort val1 = UBYTE_TO_USHORT((value >> 8) & 0xff); + ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff); + ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff); + val0 = (val0 << 8) | val0; + val1 = (val1 << 8) | val1; + val2 = (val2 << 8) | val2; + val3 = (val3 << 8) | val3; + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) { + row[j*4+0] = val0; + row[j*4+1] = val1; + row[j*4+2] = val2; + row[j*4+3] = val3; + } + row += dst->pitch * 4; + } + } + break; + default: + assert(0); + break; + } + + pipe_surface_unmap( dst ); +} + + +void +cell_init_surface_functions(struct cell_context *cell) +{ + cell->pipe.surface_copy = cell_surface_copy; + cell->pipe.surface_fill = cell_surface_fill; +} diff --git a/src/gallium/drivers/cell/ppu/cell_surface.h b/src/gallium/drivers/cell/ppu/cell_surface.h new file mode 100644 index 0000000000..9e58f32944 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_surface.h @@ -0,0 +1,42 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef CELL_SURFACE_H +#define CELL_SURFACE_H + + +struct cell_context; + + +extern void +cell_init_surface_functions(struct cell_context *cell); + + +#endif /* SP_SURFACE_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c new file mode 100644 index 0000000000..c8ef36002f --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_texture.c @@ -0,0 +1,252 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + * Michel Dänzer <michel@tungstengraphics.com> + */ + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "pipe/p_winsys.h" + +#include "cell_context.h" +#include "cell_state.h" +#include "cell_texture.h" + + +/* Simple, maximally packed layout. + */ + +static unsigned minify( unsigned d ) +{ + return MAX2(1, d>>1); +} + + +static void +cell_texture_layout(struct cell_texture * spt) +{ + struct pipe_texture *pt = &spt->base; + unsigned level; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + + spt->buffer_size = 0; + + for ( level = 0 ; level <= pt->last_level ; level++ ) { + pt->width[level] = width; + pt->height[level] = height; + pt->depth[level] = depth; + + spt->level_offset[level] = spt->buffer_size; + + spt->buffer_size += ((pt->compressed) ? MAX2(1, height/4) : height) * + ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) * + width * pt->cpp; + + width = minify(width); + height = minify(height); + depth = minify(depth); + } +} + + +struct pipe_texture * +cell_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat) +{ + struct cell_texture *spt = CALLOC_STRUCT(cell_texture); + if (!spt) + return NULL; + + spt->base = *templat; + + cell_texture_layout(spt); + + spt->buffer = pipe->winsys->buffer_create(pipe->winsys, 32, + PIPE_BUFFER_USAGE_PIXEL, + spt->buffer_size); + + if (!spt->buffer) { + FREE(spt); + return NULL; + } + + return &spt->base; +} + + +void +cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt) +{ + if (!*pt) + return; + + /* + DBG("%s %p refcount will be %d\n", + __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); + */ + if (--(*pt)->refcount <= 0) { + struct cell_texture *spt = cell_texture(*pt); + + /* + DBG("%s deleting %p\n", __FUNCTION__, (void *) spt); + */ + + pipe_buffer_reference(pipe->winsys, &spt->buffer, NULL); + + FREE(spt); + } + *pt = NULL; +} + + +/** + * Called via pipe->get_tex_surface() + */ +struct pipe_surface * +cell_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice) +{ + struct cell_texture *spt = cell_texture(pt); + struct pipe_surface *ps; + + ps = pipe->winsys->surface_alloc(pipe->winsys); + if (ps) { + assert(ps->refcount); + assert(ps->winsys); + pipe_buffer_reference(pipe->winsys, &ps->buffer, spt->buffer); + ps->format = pt->format; + ps->cpp = pt->cpp; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->pitch = ps->width; + ps->offset = spt->level_offset[level]; + + if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) { + ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) * + (pt->compressed ? ps->height/4 : ps->height) * + ps->width * ps->cpp; + } else { + assert(face == 0); + assert(zslice == 0); + } + } + return ps; +} + + + +static void +tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src) +{ + const uint tile_size2 = tile_size * tile_size; + const uint h_t = h / tile_size, w_t = w / tile_size; + + uint it, jt; /* tile counters */ + uint i, j; /* intra-tile counters */ + + for (it = 0; it < h_t; it++) { + for (jt = 0; jt < w_t; jt++) { + /* fill in tile (i, j) */ + uint *tdst = dst + (it * w_t + jt) * tile_size2; + for (i = 0; i < tile_size; i++) { + for (j = 0; j < tile_size; j++) { + const uint srci = it * tile_size + i; + const uint srcj = jt * tile_size + j; + *tdst++ = src[srci * h + srcj]; + } + } + } + } +} + + + +/** + * Convert linear texture image data to tiled format for SPU usage. + */ +static void +cell_tile_texture(struct cell_context *cell, + struct cell_texture *texture) +{ + uint face = 0, level = 0, zslice = 0; + struct pipe_surface *surf; + const uint w = texture->base.width[0], h = texture->base.height[0]; + const uint *src; + + /* temporary restrictions: */ + assert(w >= TILE_SIZE); + assert(h >= TILE_SIZE); + assert(w % TILE_SIZE == 0); + assert(h % TILE_SIZE == 0); + + surf = cell_get_tex_surface(&cell->pipe, &texture->base, face, level, zslice); + ASSERT(surf); + + src = (const uint *) pipe_surface_map(surf); + + if (texture->tiled_data) { + align_free(texture->tiled_data); + } + texture->tiled_data = align_malloc(w * h * 4, 16); + + tile_copy_data(w, h, TILE_SIZE, texture->tiled_data, src); + + pipe_surface_unmap(surf); + + pipe_surface_reference(&surf, NULL); +} + + + +void +cell_update_texture_mapping(struct cell_context *cell) +{ + uint face = 0, level = 0, zslice = 0; + + if (cell->texture[0]) + cell_tile_texture(cell, cell->texture[0]); +#if 0 + if (cell->tex_surf && cell->tex_map) { + pipe_surface_unmap(cell->tex_surf); + cell->tex_map = NULL; + } + + /* XXX free old surface */ + + cell->tex_surf = cell_get_tex_surface(&cell->pipe, + &cell->texture[0]->base, + face, level, zslice); + + cell->tex_map = pipe_surface_map(cell->tex_surf); +#endif +} diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h new file mode 100644 index 0000000000..0264fed88e --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_texture.h @@ -0,0 +1,80 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_TEXTURE_H +#define CELL_TEXTURE_H + + +struct pipe_context; +struct pipe_texture; + + +/** + * Subclass of pipe_texture + */ +struct cell_texture +{ + struct pipe_texture base; + + unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; + unsigned long buffer_size; + + void *tiled_data; /* XXX this may be temporary */ /*ALIGN16*/ +}; + + +/** cast wrapper */ +static INLINE struct cell_texture * +cell_texture(struct pipe_texture *pt) +{ + return (struct cell_texture *) pt; +} + + + +extern struct pipe_texture * +cell_texture_create(struct pipe_context *pipe, + const struct pipe_texture *templat); + +extern void +cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt); + +extern struct pipe_surface * +cell_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice); + + +extern void +cell_update_texture_mapping(struct cell_context *cell); + + +#endif /* CELL_TEXTURE */ diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c new file mode 100644 index 0000000000..e9fafe492e --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c @@ -0,0 +1,294 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Authors + * Brian Paul + */ + + +#include "cell_batch.h" +#include "cell_context.h" +#include "cell_flush.h" +#include "cell_spu.h" +#include "cell_vbuf.h" +#include "pipe/draw/draw_vbuf.h" + + +/** Allow vertex data to be inlined after RENDER command */ +#define ALLOW_INLINE_VERTS 1 + + +/** + * Subclass of vbuf_render because we need a cell_context pointer in + * a few places. + */ +struct cell_vbuf_render +{ + struct vbuf_render base; + struct cell_context *cell; + uint prim; /**< PIPE_PRIM_x */ + uint vertex_size; /**< in bytes */ + void *vertex_buffer; /**< just for debug, really */ + uint vertex_buf; /**< in [0, CELL_NUM_BUFFERS-1] */ +}; + + +/** cast wrapper */ +static struct cell_vbuf_render * +cell_vbuf_render(struct vbuf_render *vbr) +{ + return (struct cell_vbuf_render *) vbr; +} + + + +static const struct vertex_info * +cell_vbuf_get_vertex_info(struct vbuf_render *vbr) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + return &cvbr->cell->vertex_info; +} + + +static void * +cell_vbuf_allocate_vertices(struct vbuf_render *vbr, + ushort vertex_size, ushort nr_vertices) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + /*printf("Alloc verts %u * %u\n", vertex_size, nr_vertices);*/ + + assert(cvbr->vertex_buf == ~0); + cvbr->vertex_buf = cell_get_empty_buffer(cvbr->cell); + cvbr->vertex_buffer = cvbr->cell->buffer[cvbr->vertex_buf]; + cvbr->vertex_size = vertex_size; + return cvbr->vertex_buffer; +} + + +static void +cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, + unsigned vertex_size, unsigned vertices_used) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + struct cell_context *cell = cvbr->cell; + + /* + printf("%s vertex_buf = %u count = %u\n", + __FUNCTION__, cvbr->vertex_buf, vertices_used); + */ + + /* Tell SPUs they can release the vert buf */ + if (cvbr->vertex_buf != ~0U) { + struct cell_command_release_verts *release + = (struct cell_command_release_verts *) + cell_batch_alloc(cell, sizeof(struct cell_command_release_verts)); + release->opcode = CELL_CMD_RELEASE_VERTS; + release->vertex_buf = cvbr->vertex_buf; + } + + cvbr->vertex_buf = ~0; + cell_flush_int(&cell->pipe, 0x0); + + assert(vertices == cvbr->vertex_buffer); + cvbr->vertex_buffer = NULL; +} + + + +static void +cell_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + cvbr->prim = prim; + /*printf("cell_set_prim %u\n", prim);*/ +} + + +static void +cell_vbuf_draw(struct vbuf_render *vbr, + const ushort *indices, + uint nr_indices) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + struct cell_context *cell = cvbr->cell; + float xmin, ymin, xmax, ymax; + uint i; + uint nr_vertices = 0, min_index = ~0; + const void *vertices = cvbr->vertex_buffer; + const uint vertex_size = cvbr->vertex_size; + + for (i = 0; i < nr_indices; i++) { + if (indices[i] > nr_vertices) + nr_vertices = indices[i]; + if (indices[i] < min_index) + min_index = indices[i]; + } + nr_vertices++; + +#if 0 + /*if (min_index > 0)*/ + printf("%s min_index = %u\n", __FUNCTION__, min_index); +#endif + +#if 0 + printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u\n", + nr_indices, nr_vertices); + printf(" "); + for (i = 0; i < nr_indices; i += 3) { + printf("%u %u %u, ", indices[i+0], indices[i+1], indices[i+2]); + } + printf("\n"); +#elif 0 + printf("cell_vbuf_draw() nr_indices = %u nr_verts = %u indexes = [%u %u %u ...]\n", + nr_indices, nr_vertices, + indices[0], indices[1], indices[2]); + printf("ind space = %u, vert space = %u, space = %u\n", + nr_indices * 2, + nr_vertices * 4 * cell->vertex_info.size, + cell_batch_free_space(cell)); +#endif + + /* compute x/y bounding box */ + xmin = ymin = 1e50; + xmax = ymax = -1e50; + for (i = min_index; i < nr_vertices; i++) { + const float *v = (float *) ((ubyte *) vertices + i * vertex_size); + if (v[0] < xmin) + xmin = v[0]; + if (v[0] > xmax) + xmax = v[0]; + if (v[1] < ymin) + ymin = v[1]; + if (v[1] > ymax) + ymax = v[1]; + } +#if 0 + printf("PPU Bounds %g, %g .. %g, %g\n", xmin, ymin, xmax, ymax); + fflush(stdout); +#endif + + if (cvbr->prim != PIPE_PRIM_TRIANGLES) + return; /* only render tris for now */ + + /* build/insert batch RENDER command */ + { + const uint index_bytes = ROUNDUP8(nr_indices * 2); + const uint vertex_bytes = nr_vertices * 4 * cell->vertex_info.size; + const uint batch_size = sizeof(struct cell_command_render) + index_bytes; + + struct cell_command_render *render + = (struct cell_command_render *) + cell_batch_alloc(cell, batch_size); + + render->opcode = CELL_CMD_RENDER; + render->prim_type = cvbr->prim; + + render->num_indexes = nr_indices; + render->min_index = min_index; + + /* append indices after render command */ + memcpy(render + 1, indices, nr_indices * 2); + + /* if there's room, append vertices after the indices, else leave + * vertices in the original/separate buffer. + */ + render->vertex_size = 4 * cell->vertex_info.size; + render->num_verts = nr_vertices; + if (ALLOW_INLINE_VERTS && + min_index == 0 && + vertex_bytes + 16 <= cell_batch_free_space(cell)) { + /* vertex data inlined, after indices, at 16-byte boundary */ + void *dst = cell_batch_alloc_aligned(cell, vertex_bytes, 16); + memcpy(dst, vertices, vertex_bytes); + render->inline_verts = TRUE; + render->vertex_buf = ~0; + } + else { + /* vertex data in separate buffer */ + render->inline_verts = FALSE; + ASSERT(cvbr->vertex_buf >= 0); + render->vertex_buf = cvbr->vertex_buf; + } + + render->xmin = xmin; + render->ymin = ymin; + render->xmax = xmax; + render->ymax = ymax; + } + +#if 0 + /* helpful for debug */ + cell_flush_int(&cell->pipe, PIPE_FLUSH_WAIT); +#endif +} + + +static void +cell_vbuf_destroy(struct vbuf_render *vbr) +{ + struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr); + cvbr->cell->vbuf_render = NULL; + FREE(cvbr); +} + + +/** + * Initialize the post-transform vertex buffer information for the given + * context. + */ +void +cell_init_vbuf(struct cell_context *cell) +{ + assert(cell->draw); + + cell->vbuf_render = CALLOC_STRUCT(cell_vbuf_render); + + /* The max number of indexes is what can fix into a batch buffer, + * minus the render and release-verts commands. + */ + cell->vbuf_render->base.max_indices + = (CELL_BUFFER_SIZE + - sizeof(struct cell_command_render) + - sizeof(struct cell_command_release_verts)) + / sizeof(ushort); + cell->vbuf_render->base.max_vertex_buffer_bytes = CELL_BUFFER_SIZE; + + cell->vbuf_render->base.get_vertex_info = cell_vbuf_get_vertex_info; + cell->vbuf_render->base.allocate_vertices = cell_vbuf_allocate_vertices; + cell->vbuf_render->base.set_primitive = cell_vbuf_set_primitive; + cell->vbuf_render->base.draw = cell_vbuf_draw; + cell->vbuf_render->base.release_vertices = cell_vbuf_release_vertices; + cell->vbuf_render->base.destroy = cell_vbuf_destroy; + + cell->vbuf_render->cell = cell; +#if 1 + cell->vbuf_render->vertex_buf = ~0; +#endif + + cell->vbuf = draw_vbuf_stage(cell->draw, &cell->vbuf_render->base); +} diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.h b/src/gallium/drivers/cell/ppu/cell_vbuf.h new file mode 100644 index 0000000000..d265cbf770 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_vbuf.h @@ -0,0 +1,38 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef CELL_VBUF_H +#define CELL_VBUF_H + + +struct cell_context; + +extern void +cell_init_vbuf(struct cell_context *cell); + + +#endif /* CELL_VBUF_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c new file mode 100644 index 0000000000..80dd500b34 --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c @@ -0,0 +1,120 @@ +/* + * (C) Copyright IBM Corporation 2008 + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + */ + +/** + * \file cell_vertex_shader.c + * Vertex shader interface routines for Cell. + * + * \author Ian Romanick <idr@us.ibm.com> + */ + +#include "pipe/p_defines.h" +#include "pipe/p_context.h" +#include "pipe/p_winsys.h" + +#include "cell_context.h" +#include "cell_draw_arrays.h" +#include "cell_spu.h" +#include "cell_batch.h" + +#include "pipe/cell/common.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_private.h" + +/** + * Run the vertex shader on all vertices in the vertex queue. + * Called by the draw module when the vertx cache needs to be flushed. + */ +void +cell_vertex_shader_queue_flush(struct draw_context *draw) +{ + struct cell_context *const cell = + (struct cell_context *) draw->driver_private; + struct cell_command_vs *const vs = &cell_global.command[0].vs; + uint64_t *batch; + struct cell_array_info *array_info; + unsigned i, j; + + assert(draw->vs.queue_nr != 0); + + /* XXX: do this on statechange: + */ + draw_update_vertex_fetch(draw); + + for (i = 0; i < draw->vertex_fetch.nr_attrs; i++) { + batch = cell_batch_alloc(cell, sizeof(batch[0]) + sizeof(*array_info)); + + batch[0] = CELL_CMD_STATE_VS_ARRAY_INFO; + + array_info = (struct cell_array_info *) &batch[1]; + assert(draw->vertex_fetch.src_ptr[i] != NULL); + array_info->base = (uintptr_t) draw->vertex_fetch.src_ptr[i]; + array_info->attr = i; + array_info->pitch = draw->vertex_fetch.pitch[i]; + array_info->format = draw->vertex_element[i].src_format; + } + + batch = cell_batch_alloc(cell, sizeof(batch[0]) + + sizeof(struct pipe_viewport_state)); + batch[0] = CELL_CMD_STATE_VIEWPORT; + (void) memcpy(&batch[1], &draw->viewport, + sizeof(struct pipe_viewport_state)); + + cell_batch_flush(cell); + + vs->opcode = CELL_CMD_VS_EXECUTE; + vs->shader.num_outputs = draw->num_vs_outputs; + vs->shader.declarations = (uintptr_t) draw->machine.Declarations; + vs->shader.num_declarations = draw->machine.NumDeclarations; + vs->shader.instructions = (uintptr_t) draw->machine.Instructions; + vs->shader.num_instructions = draw->machine.NumInstructions; + vs->shader.uniforms = (uintptr_t) draw->user.constants; + vs->shader.immediates = (uintptr_t) draw->machine.Imms; + vs->shader.num_immediates = draw->machine.ImmLimit / 4; + vs->nr_attrs = draw->vertex_fetch.nr_attrs; + + (void) memcpy(vs->plane, draw->plane, sizeof(draw->plane)); + vs->nr_planes = draw->nr_planes; + + for (i = 0; i < draw->vs.queue_nr; i += SPU_VERTS_PER_BATCH) { + const unsigned n = MIN2(SPU_VERTS_PER_BATCH, draw->vs.queue_nr - i); + + for (j = 0; j < n; j++) { + vs->elts[j] = draw->vs.queue[i + j].elt; + vs->vOut[j] = (uintptr_t) draw->vs.queue[i + j].dest; + } + + for (/* empty */; j < SPU_VERTS_PER_BATCH; j++) { + vs->elts[j] = vs->elts[0]; + vs->vOut[j] = vs->vOut[0]; + } + + vs->num_elts = n; + send_mbox_message(cell_global.spe_contexts[0], CELL_CMD_VS_EXECUTE); + + cell_flush_int(& cell->pipe, PIPE_FLUSH_WAIT); + } + + draw->vs.queue_nr = 0; +} diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.c b/src/gallium/drivers/cell/ppu/cell_winsys.c new file mode 100644 index 0000000000..ebabce3c8f --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_winsys.c @@ -0,0 +1,40 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_util.h" +#include "cell_winsys.h" + + +struct cell_winsys * +cell_get_winsys(uint format) +{ + struct cell_winsys *cws = CALLOC_STRUCT(cell_winsys); + if (cws) + cws->preferredFormat = format; + return cws; +} diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.h b/src/gallium/drivers/cell/ppu/cell_winsys.h new file mode 100644 index 0000000000..ae2af5696b --- /dev/null +++ b/src/gallium/drivers/cell/ppu/cell_winsys.h @@ -0,0 +1,50 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef CELL_WINSYS_H +#define CELL_WINSYS_H + +#include "pipe/p_compiler.h" + + +/** + * Very simple winsys at this time. + * Will probably eventually add SPU control info. + */ +struct cell_winsys +{ + uint preferredFormat; +}; + + +extern struct cell_winsys * +cell_get_winsys(uint format); + + + +#endif diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile new file mode 100644 index 0000000000..f202971d73 --- /dev/null +++ b/src/gallium/drivers/cell/spu/Makefile @@ -0,0 +1,72 @@ +# Gallium3D Cell driver: SPU code + +# This makefile builds the g3d_spu.a file that's linked into the +# PPU code/library. + + +TOP = ../../../../.. +include $(TOP)/configs/linux-cell + + +PROG = g3d + +PROG_SPU = $(PROG)_spu +PROG_SPU_A = $(PROG)_spu.a +PROG_SPU_EMBED_O = $(PROG)_spu-embed.o + + +SOURCES = \ + spu_main.c \ + spu_blend.c \ + spu_render.c \ + spu_texture.c \ + spu_tile.c \ + spu_tri.c \ + spu_exec.c \ + spu_util.c \ + spu_vertex_fetch.c \ + spu_vertex_shader.c + +SPU_OBJECTS = $(SOURCES:.c=.o) \ + +SPU_ASM_OUT = $(SOURCES:.c=.s) \ + +INCLUDE_DIRS = -I$(TOP)/src/mesa + + +.c.o: + $(SPU_CC) $(SPU_CFLAGS) -c $< + +.c.s: + $(SPU_CC) $(SPU_CFLAGS) -S $< + + +# The .a file will be linked into the main/PPU executable +default: $(PROG_SPU_A) + +$(PROG_SPU_A): $(PROG_SPU_EMBED_O) + $(SPU_AR) $(SPU_AR_FLAGS) $(PROG_SPU_A) $(PROG_SPU_EMBED_O) + +$(PROG_SPU_EMBED_O): $(PROG_SPU) + $(SPU_EMBED) $(SPU_EMBED_FLAGS) $(PROG_SPU) $(PROG_SPU) $(PROG_SPU_EMBED_O) + +$(PROG_SPU): $(SPU_OBJECTS) + $(SPU_CC) -o $(PROG_SPU) $(SPU_OBJECTS) $(SPU_LFLAGS) + + + +asmfiles: $(SPU_ASM_OUT) + + +clean: + rm -f *~ *.o *.a *.d *.s $(PROG_SPU) + + + +depend: $(SOURCES) + rm -f depend + touch depend + $(MKDEP) $(MKDEP_OPTIONS) $(INCLUDE_DIRS) $(SOURCES) 2> /dev/null + +include depend + diff --git a/src/gallium/drivers/cell/spu/spu_blend.c b/src/gallium/drivers/cell/spu/spu_blend.c new file mode 100644 index 0000000000..23ec0eeb45 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_blend.c @@ -0,0 +1,62 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "spu_main.h" +#include "spu_blend.h" +#include "spu_colorpack.h" + + +void +blend_quad(uint itx, uint ity, vector float colors[4]) +{ + /* simple SRC_ALPHA, ONE_MINUS_SRC_ALPHA blending */ + vector float fbc00 = spu_unpack_color(spu.ctile.ui[ity][itx]); + vector float fbc01 = spu_unpack_color(spu.ctile.ui[ity][itx+1]); + vector float fbc10 = spu_unpack_color(spu.ctile.ui[ity+1][itx]); + vector float fbc11 = spu_unpack_color(spu.ctile.ui[ity+1][itx+1]); + + vector float alpha00 = spu_splats(spu_extract(colors[0], 3)); + vector float alpha01 = spu_splats(spu_extract(colors[1], 3)); + vector float alpha10 = spu_splats(spu_extract(colors[2], 3)); + vector float alpha11 = spu_splats(spu_extract(colors[3], 3)); + + vector float one_minus_alpha00 = spu_sub(spu_splats(1.0f), alpha00); + vector float one_minus_alpha01 = spu_sub(spu_splats(1.0f), alpha01); + vector float one_minus_alpha10 = spu_sub(spu_splats(1.0f), alpha10); + vector float one_minus_alpha11 = spu_sub(spu_splats(1.0f), alpha11); + + colors[0] = spu_add(spu_mul(colors[0], alpha00), + spu_mul(fbc00, one_minus_alpha00)); + colors[1] = spu_add(spu_mul(colors[1], alpha01), + spu_mul(fbc01, one_minus_alpha01)); + colors[2] = spu_add(spu_mul(colors[2], alpha10), + spu_mul(fbc10, one_minus_alpha10)); + colors[3] = spu_add(spu_mul(colors[3], alpha11), + spu_mul(fbc11, one_minus_alpha11)); +} + diff --git a/src/gallium/drivers/cell/spu/spu_blend.h b/src/gallium/drivers/cell/spu/spu_blend.h new file mode 100644 index 0000000000..2b594b578b --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_blend.h @@ -0,0 +1,37 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef SPU_BLEND_H +#define SPU_BLEND_H + + +extern void +blend_quad(uint itx, uint ity, vector float colors[4]); + + +#endif /* SPU_BLEND_H */ diff --git a/src/gallium/drivers/cell/spu/spu_colorpack.h b/src/gallium/drivers/cell/spu/spu_colorpack.h new file mode 100644 index 0000000000..e9fee8a3a6 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_colorpack.h @@ -0,0 +1,110 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + + +#ifndef SPU_COLORPACK_H +#define SPU_COLORPACK_H + + +#include <spu_intrinsics.h> + + +static INLINE unsigned int +spu_pack_R8G8B8A8(vector float rgba) +{ + vector unsigned int out = spu_convtu(rgba, 32); + + out = spu_shuffle(out, out, ((vector unsigned char) { + 0, 4, 8, 12, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0 }) ); + + return spu_extract(out, 0); +} + + +static INLINE unsigned int +spu_pack_A8R8G8B8(vector float rgba) +{ + vector unsigned int out = spu_convtu(rgba, 32); + out = spu_shuffle(out, out, ((vector unsigned char) { + 12, 0, 4, 8, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0}) ); + return spu_extract(out, 0); +} + + +static INLINE unsigned int +spu_pack_B8G8R8A8(vector float rgba) +{ + vector unsigned int out = spu_convtu(rgba, 32); + out = spu_shuffle(out, out, ((vector unsigned char) { + 8, 4, 0, 12, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0}) ); + return spu_extract(out, 0); +} + + +static INLINE unsigned int +spu_pack_color_shuffle(vector float rgba, vector unsigned char shuffle) +{ + vector unsigned int out = spu_convtu(rgba, 32); + out = spu_shuffle(out, out, shuffle); + return spu_extract(out, 0); +} + + +static INLINE vector float +spu_unpack_color(uint color) +{ + vector unsigned int color_u4 = spu_splats(color); + color_u4 = spu_shuffle(color_u4, color_u4, + ((vector unsigned char) { + 0, 0, 0, 0, + 5, 5, 5, 5, + 10, 10, 10, 10, + 15, 15, 15, 15}) ); + return spu_convtf(color_u4, 32); +} + + +static INLINE vector float +spu_unpack_A8R8G8B8(uint color) +{ + vector unsigned int color_u4 = spu_splats(color); + color_u4 = spu_shuffle(color_u4, color_u4, + ((vector unsigned char) { + 5, 5, 5, 5, + 10, 10, 10, 10, + 15, 15, 15, 15, + 0, 0, 0, 0}) ); + + return spu_convtf(color_u4, 32); +} + + +#endif /* SPU_COLORPACK_H */ diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c new file mode 100644 index 0000000000..e51008b9b3 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_exec.c @@ -0,0 +1,1948 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * TGSI interpretor/executor. + * + * Flow control information: + * + * Since we operate on 'quads' (4 pixels or 4 vertices in parallel) + * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special + * care since a condition may be true for some quad components but false + * for other components. + * + * We basically execute all statements (even if they're in the part of + * an IF/ELSE clause that's "not taken") and use a special mask to + * control writing to destination registers. This is the ExecMask. + * See store_dest(). + * + * The ExecMask is computed from three other masks (CondMask, LoopMask and + * ContMask) which are controlled by the flow control instructions (namely: + * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT). + * + * + * Authors: + * Michal Krol + * Brian Paul + */ + +#include <libmisc.h> +#include <spu_mfcio.h> +#include <transpose_matrix4x4.h> +#include <simdmath/ceilf4.h> +#include <simdmath/cosf4.h> +#include <simdmath/divf4.h> +#include <simdmath/floorf4.h> +#include <simdmath/log2f4.h> +#include <simdmath/powf4.h> +#include <simdmath/sinf4.h> +#include <simdmath/sqrtf4.h> +#include <simdmath/truncf4.h> + +#include "pipe/p_compiler.h" +#include "pipe/p_state.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" +#include "pipe/tgsi/util/tgsi_util.h" +#include "spu_exec.h" +#include "spu_main.h" +#include "spu_vertex_shader.h" + +#define TILE_TOP_LEFT 0 +#define TILE_TOP_RIGHT 1 +#define TILE_BOTTOM_LEFT 2 +#define TILE_BOTTOM_RIGHT 3 + +/* + * Shorthand locations of various utility registers (_I = Index, _C = Channel) + */ +#define TEMP_0_I TGSI_EXEC_TEMP_00000000_I +#define TEMP_0_C TGSI_EXEC_TEMP_00000000_C +#define TEMP_7F_I TGSI_EXEC_TEMP_7FFFFFFF_I +#define TEMP_7F_C TGSI_EXEC_TEMP_7FFFFFFF_C +#define TEMP_80_I TGSI_EXEC_TEMP_80000000_I +#define TEMP_80_C TGSI_EXEC_TEMP_80000000_C +#define TEMP_FF_I TGSI_EXEC_TEMP_FFFFFFFF_I +#define TEMP_FF_C TGSI_EXEC_TEMP_FFFFFFFF_C +#define TEMP_1_I TGSI_EXEC_TEMP_ONE_I +#define TEMP_1_C TGSI_EXEC_TEMP_ONE_C +#define TEMP_2_I TGSI_EXEC_TEMP_TWO_I +#define TEMP_2_C TGSI_EXEC_TEMP_TWO_C +#define TEMP_128_I TGSI_EXEC_TEMP_128_I +#define TEMP_128_C TGSI_EXEC_TEMP_128_C +#define TEMP_M128_I TGSI_EXEC_TEMP_MINUS_128_I +#define TEMP_M128_C TGSI_EXEC_TEMP_MINUS_128_C +#define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I +#define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C +#define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I +#define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C +#define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I +#define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C +#define TEMP_R0 TGSI_EXEC_TEMP_R0 + +#define FOR_EACH_CHANNEL(CHAN)\ + for (CHAN = 0; CHAN < 4; CHAN++) + +#define IS_CHANNEL_ENABLED(INST, CHAN)\ + ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN))) + +#define IS_CHANNEL_ENABLED2(INST, CHAN)\ + ((INST).FullDstRegisters[1].DstRegister.WriteMask & (1 << (CHAN))) + +#define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\ + FOR_EACH_CHANNEL( CHAN )\ + if (IS_CHANNEL_ENABLED( INST, CHAN )) + +#define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\ + FOR_EACH_CHANNEL( CHAN )\ + if (IS_CHANNEL_ENABLED2( INST, CHAN )) + + +/** The execution mask depends on the conditional mask and the loop mask */ +#define UPDATE_EXEC_MASK(MACH) \ + MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask + + +#define CHAN_X 0 +#define CHAN_Y 1 +#define CHAN_Z 2 +#define CHAN_W 3 + + + +/** + * Initialize machine state by expanding tokens to full instructions, + * allocating temporary storage, setting up constants, etc. + * After this, we can call spu_exec_machine_run() many times. + */ +void +spu_exec_machine_init(struct spu_exec_machine *mach, + uint numSamplers, + struct spu_sampler *samplers, + unsigned processor) +{ + qword zero; + qword not_zero; + uint i; + + mach->Samplers = samplers; + mach->Processor = processor; + mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS]; + + zero = si_xor(zero, zero); + not_zero = si_xori(zero, 0xff); + + /* Setup constants. */ + mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q = zero; + mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].q = not_zero; + mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].q = si_shli(not_zero, -1); + mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].q = si_shli(not_zero, 31); + + mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q = (qword) spu_splats(1.0f); + mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].q = (qword) spu_splats(2.0f); + mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].q = (qword) spu_splats(128.0f); + mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].q = (qword) spu_splats(-128.0f); +} + + +static INLINE qword +micro_abs(qword src) +{ + return si_rotmi(si_shli(src, 1), -1); +} + +static INLINE qword +micro_ceil(qword src) +{ + return (qword) _ceilf4((vec_float4) src); +} + +static INLINE qword +micro_cos(qword src) +{ + return (qword) _cosf4((vec_float4) src); +} + +static const qword br_shuf = { + TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1, + TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3, + TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1, + TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3, + TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1, + TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3, + TILE_BOTTOM_RIGHT + 0, TILE_BOTTOM_RIGHT + 1, + TILE_BOTTOM_RIGHT + 2, TILE_BOTTOM_RIGHT + 3, +}; + +static const qword bl_shuf = { + TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1, + TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3, + TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1, + TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3, + TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1, + TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3, + TILE_BOTTOM_LEFT + 0, TILE_BOTTOM_LEFT + 1, + TILE_BOTTOM_LEFT + 2, TILE_BOTTOM_LEFT + 3, +}; + +static const qword tl_shuf = { + TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1, + TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3, + TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1, + TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3, + TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1, + TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3, + TILE_TOP_LEFT + 0, TILE_TOP_LEFT + 1, + TILE_TOP_LEFT + 2, TILE_TOP_LEFT + 3, +}; + +static qword +micro_ddx(qword src) +{ + qword bottom_right = si_shufb(src, src, br_shuf); + qword bottom_left = si_shufb(src, src, bl_shuf); + + return si_fs(bottom_right, bottom_left); +} + +static qword +micro_ddy(qword src) +{ + qword top_left = si_shufb(src, src, tl_shuf); + qword bottom_left = si_shufb(src, src, bl_shuf); + + return si_fs(top_left, bottom_left); +} + +static INLINE qword +micro_div(qword src0, qword src1) +{ + return (qword) _divf4((vec_float4) src0, (vec_float4) src1); +} + +static qword +micro_flr(qword src) +{ + return (qword) _floorf4((vec_float4) src); +} + +static qword +micro_frc(qword src) +{ + return si_fs(src, (qword) _floorf4((vec_float4) src)); +} + +static INLINE qword +micro_ge(qword src0, qword src1) +{ + return si_or(si_fceq(src0, src1), si_fcgt(src0, src1)); +} + +static qword +micro_lg2(qword src) +{ + return (qword) _log2f4((vec_float4) src); +} + +static INLINE qword +micro_lt(qword src0, qword src1) +{ + const qword tmp = si_or(si_fceq(src0, src1), si_fcgt(src0, src1)); + + return si_xori(tmp, 0xff); +} + +static INLINE qword +micro_max(qword src0, qword src1) +{ + return si_selb(src1, src0, si_fcgt(src0, src1)); +} + +static INLINE qword +micro_min(qword src0, qword src1) +{ + return si_selb(src0, src1, si_fcgt(src0, src1)); +} + +static qword +micro_neg(qword src) +{ + return si_xor(src, (qword) spu_splats(0x80000000)); +} + +static qword +micro_set_sign(qword src) +{ + return si_or(src, (qword) spu_splats(0x80000000)); +} + +static qword +micro_pow(qword src0, qword src1) +{ + return (qword) _powf4((vec_float4) src0, (vec_float4) src1); +} + +static qword +micro_rnd(qword src) +{ + const qword half = (qword) spu_splats(0.5f); + + /* May be able to use _roundf4. There may be some difference, though. + */ + return (qword) _floorf4((vec_float4) si_fa(src, half)); +} + +static INLINE qword +micro_ishr(qword src0, qword src1) +{ + return si_rotma(src0, si_sfi(src1, 0)); +} + +static qword +micro_trunc(qword src) +{ + return (qword) _truncf4((vec_float4) src); +} + +static qword +micro_sin(qword src) +{ + return (qword) _sinf4((vec_float4) src); +} + +static INLINE qword +micro_sqrt(qword src) +{ + return (qword) _sqrtf4((vec_float4) src); +} + +static void +fetch_src_file_channel( + const struct spu_exec_machine *mach, + const uint file, + const uint swizzle, + const union spu_exec_channel *index, + union spu_exec_channel *chan ) +{ + switch( swizzle ) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + switch( file ) { + case TGSI_FILE_CONSTANT: { + unsigned char buffer[32] ALIGN16_ATTRIB; + unsigned i; + + for (i = 0; i < 4; i++) { + const float *ptr = mach->Consts[index->i[i]]; + const uint64_t addr = (uint64_t)(uintptr_t) ptr; + const unsigned size = ((addr & 0x0f) == 0) ? 16 : 32; + + mfc_get(buffer, addr & ~0x0f, size, TAG_VERTEX_BUFFER, 0, 0); + wait_on_mask(1 << TAG_VERTEX_BUFFER); + + (void) memcpy(& chan->f[i], &buffer[(addr & 0x0f) + + (sizeof(float) * swizzle)], sizeof(float)); + } + break; + } + + case TGSI_FILE_INPUT: + chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0]; + chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1]; + chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2]; + chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3]; + break; + + case TGSI_FILE_TEMPORARY: + chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0]; + chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1]; + chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2]; + chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3]; + break; + + case TGSI_FILE_IMMEDIATE: + assert( index->i[0] < (int) mach->ImmLimit ); + assert( index->i[1] < (int) mach->ImmLimit ); + assert( index->i[2] < (int) mach->ImmLimit ); + assert( index->i[3] < (int) mach->ImmLimit ); + + chan->f[0] = mach->Imms[index->i[0]][swizzle]; + chan->f[1] = mach->Imms[index->i[1]][swizzle]; + chan->f[2] = mach->Imms[index->i[2]][swizzle]; + chan->f[3] = mach->Imms[index->i[3]][swizzle]; + break; + + case TGSI_FILE_ADDRESS: + chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0]; + chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1]; + chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2]; + chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3]; + break; + + case TGSI_FILE_OUTPUT: + /* vertex/fragment output vars can be read too */ + chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0]; + chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1]; + chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2]; + chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3]; + break; + + default: + assert( 0 ); + } + break; + + case TGSI_EXTSWIZZLE_ZERO: + *chan = mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]; + break; + + case TGSI_EXTSWIZZLE_ONE: + *chan = mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]; + break; + + default: + assert( 0 ); + } +} + +static void +fetch_source( + const struct spu_exec_machine *mach, + union spu_exec_channel *chan, + const struct tgsi_full_src_register *reg, + const uint chan_index ) +{ + union spu_exec_channel index; + uint swizzle; + + index.i[0] = + index.i[1] = + index.i[2] = + index.i[3] = reg->SrcRegister.Index; + + if (reg->SrcRegister.Indirect) { + union spu_exec_channel index2; + union spu_exec_channel indir_index; + + index2.i[0] = + index2.i[1] = + index2.i[2] = + index2.i[3] = reg->SrcRegisterInd.Index; + + swizzle = tgsi_util_get_src_register_swizzle(®->SrcRegisterInd, + CHAN_X); + fetch_src_file_channel( + mach, + reg->SrcRegisterInd.File, + swizzle, + &index2, + &indir_index ); + + index.q = si_a(index.q, indir_index.q); + } + + if( reg->SrcRegister.Dimension ) { + switch( reg->SrcRegister.File ) { + case TGSI_FILE_INPUT: + index.q = si_mpyi(index.q, 17); + break; + case TGSI_FILE_CONSTANT: + index.q = si_shli(index.q, 12); + break; + default: + assert( 0 ); + } + + index.i[0] += reg->SrcRegisterDim.Index; + index.i[1] += reg->SrcRegisterDim.Index; + index.i[2] += reg->SrcRegisterDim.Index; + index.i[3] += reg->SrcRegisterDim.Index; + + if (reg->SrcRegisterDim.Indirect) { + union spu_exec_channel index2; + union spu_exec_channel indir_index; + + index2.i[0] = + index2.i[1] = + index2.i[2] = + index2.i[3] = reg->SrcRegisterDimInd.Index; + + swizzle = tgsi_util_get_src_register_swizzle( ®->SrcRegisterDimInd, CHAN_X ); + fetch_src_file_channel( + mach, + reg->SrcRegisterDimInd.File, + swizzle, + &index2, + &indir_index ); + + index.q = si_a(index.q, indir_index.q); + } + } + + swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index ); + fetch_src_file_channel( + mach, + reg->SrcRegister.File, + swizzle, + &index, + chan ); + + switch (tgsi_util_get_full_src_register_sign_mode( reg, chan_index )) { + case TGSI_UTIL_SIGN_CLEAR: + chan->q = micro_abs(chan->q); + break; + + case TGSI_UTIL_SIGN_SET: + chan->q = micro_set_sign(chan->q); + break; + + case TGSI_UTIL_SIGN_TOGGLE: + chan->q = micro_neg(chan->q); + break; + + case TGSI_UTIL_SIGN_KEEP: + break; + } + + if (reg->SrcRegisterExtMod.Complement) { + chan->q = si_fs(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, chan->q); + } +} + +static void +store_dest( + struct spu_exec_machine *mach, + const union spu_exec_channel *chan, + const struct tgsi_full_dst_register *reg, + const struct tgsi_full_instruction *inst, + uint chan_index ) +{ + union spu_exec_channel *dst; + + switch( reg->DstRegister.File ) { + case TGSI_FILE_NULL: + return; + + case TGSI_FILE_OUTPUT: + dst = &mach->Outputs[mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] + + reg->DstRegister.Index].xyzw[chan_index]; + break; + + case TGSI_FILE_TEMPORARY: + dst = &mach->Temps[reg->DstRegister.Index].xyzw[chan_index]; + break; + + case TGSI_FILE_ADDRESS: + dst = &mach->Addrs[reg->DstRegister.Index].xyzw[chan_index]; + break; + + default: + assert( 0 ); + return; + } + + switch (inst->Instruction.Saturate) + { + case TGSI_SAT_NONE: + if (mach->ExecMask & 0x1) + dst->i[0] = chan->i[0]; + if (mach->ExecMask & 0x2) + dst->i[1] = chan->i[1]; + if (mach->ExecMask & 0x4) + dst->i[2] = chan->i[2]; + if (mach->ExecMask & 0x8) + dst->i[3] = chan->i[3]; + break; + + case TGSI_SAT_ZERO_ONE: + /* XXX need to obey ExecMask here */ + dst->q = micro_max(chan->q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q); + dst->q = micro_min(dst->q, mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q); + break; + + case TGSI_SAT_MINUS_PLUS_ONE: + assert( 0 ); + break; + + default: + assert( 0 ); + } +} + +#define FETCH(VAL,INDEX,CHAN)\ + fetch_source (mach, VAL, &inst->FullSrcRegisters[INDEX], CHAN) + +#define STORE(VAL,INDEX,CHAN)\ + store_dest (mach, VAL, &inst->FullDstRegisters[INDEX], inst, CHAN ) + + +/** + * Execute ARB-style KIL which is predicated by a src register. + * Kill fragment if any of the four values is less than zero. + */ +static void +exec_kilp(struct spu_exec_machine *mach, + const struct tgsi_full_instruction *inst) +{ + uint uniquemask; + uint chan_index; + uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ + union spu_exec_channel r[1]; + + /* This mask stores component bits that were already tested. Note that + * we test if the value is less than zero, so 1.0 and 0.0 need not to be + * tested. */ + uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE); + + for (chan_index = 0; chan_index < 4; chan_index++) + { + uint swizzle; + uint i; + + /* unswizzle channel */ + swizzle = tgsi_util_get_full_src_register_extswizzle ( + &inst->FullSrcRegisters[0], + chan_index); + + /* check if the component has not been already tested */ + if (uniquemask & (1 << swizzle)) + continue; + uniquemask |= 1 << swizzle; + + FETCH(&r[0], 0, chan_index); + for (i = 0; i < 4; i++) + if (r[0].f[i] < 0.0f) + kilmask |= 1 << i; + } + + mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; +} + + +/* + * Fetch a texel using STR texture coordinates. + */ +static void +fetch_texel( struct spu_sampler *sampler, + const union spu_exec_channel *s, + const union spu_exec_channel *t, + const union spu_exec_channel *p, + float lodbias, /* XXX should be float[4] */ + union spu_exec_channel *r, + union spu_exec_channel *g, + union spu_exec_channel *b, + union spu_exec_channel *a ) +{ + qword rgba[4]; + qword out[4]; + + sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, (float *) rgba); + + _transpose_matrix4x4(out, rgba); + r->q = out[0]; + g->q = out[1]; + b->q = out[2]; + a->q = out[3]; +} + + +static void +exec_tex(struct spu_exec_machine *mach, + const struct tgsi_full_instruction *inst, + boolean biasLod) +{ + const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; + union spu_exec_channel r[8]; + uint chan_index; + float lodBias; + + /* printf("Sampler %u unit %u\n", sampler, unit); */ + + switch (inst->InstructionExtTexture.Texture) { + case TGSI_TEXTURE_1D: + + FETCH(&r[0], 0, CHAN_X); + + switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) { + case TGSI_EXTSWIZZLE_W: + FETCH(&r[1], 0, CHAN_W); + r[0].q = micro_div(r[0].q, r[1].q); + break; + + case TGSI_EXTSWIZZLE_ONE: + break; + + default: + assert (0); + } + + if (biasLod) { + FETCH(&r[1], 0, CHAN_W); + lodBias = r[2].f[0]; + } + else + lodBias = 0.0; + + fetch_texel(&mach->Samplers[unit], + &r[0], NULL, NULL, lodBias, /* S, T, P, BIAS */ + &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ + break; + + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: + + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 0, CHAN_Z); + + switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) { + case TGSI_EXTSWIZZLE_W: + FETCH(&r[3], 0, CHAN_W); + r[0].q = micro_div(r[0].q, r[3].q); + r[1].q = micro_div(r[1].q, r[3].q); + r[2].q = micro_div(r[2].q, r[3].q); + break; + + case TGSI_EXTSWIZZLE_ONE: + break; + + default: + assert (0); + } + + if (biasLod) { + FETCH(&r[3], 0, CHAN_W); + lodBias = r[3].f[0]; + } + else + lodBias = 0.0; + + fetch_texel(&mach->Samplers[unit], + &r[0], &r[1], &r[2], lodBias, /* inputs */ + &r[0], &r[1], &r[2], &r[3]); /* outputs */ + break; + + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: + + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 0, CHAN_Z); + + switch (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide) { + case TGSI_EXTSWIZZLE_W: + FETCH(&r[3], 0, CHAN_W); + r[0].q = micro_div(r[0].q, r[3].q); + r[1].q = micro_div(r[1].q, r[3].q); + r[2].q = micro_div(r[2].q, r[3].q); + break; + + case TGSI_EXTSWIZZLE_ONE: + break; + + default: + assert (0); + } + + if (biasLod) { + FETCH(&r[3], 0, CHAN_W); + lodBias = r[3].f[0]; + } + else + lodBias = 0.0; + + fetch_texel(&mach->Samplers[unit], + &r[0], &r[1], &r[2], lodBias, + &r[0], &r[1], &r[2], &r[3]); + break; + + default: + assert (0); + } + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[chan_index], 0, chan_index ); + } +} + + + +static void +constant_interpolation( + struct spu_exec_machine *mach, + unsigned attrib, + unsigned chan ) +{ + unsigned i; + + for( i = 0; i < QUAD_SIZE; i++ ) { + mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan]; + } +} + +static void +linear_interpolation( + struct spu_exec_machine *mach, + unsigned attrib, + unsigned chan ) +{ + const float x = mach->QuadPos.xyzw[0].f[0]; + const float y = mach->QuadPos.xyzw[1].f[0]; + const float dadx = mach->InterpCoefs[attrib].dadx[chan]; + const float dady = mach->InterpCoefs[attrib].dady[chan]; + const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; + mach->Inputs[attrib].xyzw[chan].f[0] = a0; + mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx; + mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady; + mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady; +} + +static void +perspective_interpolation( + struct spu_exec_machine *mach, + unsigned attrib, + unsigned chan ) +{ + const float x = mach->QuadPos.xyzw[0].f[0]; + const float y = mach->QuadPos.xyzw[1].f[0]; + const float dadx = mach->InterpCoefs[attrib].dadx[chan]; + const float dady = mach->InterpCoefs[attrib].dady[chan]; + const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; + const float *w = mach->QuadPos.xyzw[3].f; + /* divide by W here */ + mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0]; + mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1]; + mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2]; + mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3]; +} + + +typedef void (* interpolation_func)( + struct spu_exec_machine *mach, + unsigned attrib, + unsigned chan ); + +static void +exec_declaration(struct spu_exec_machine *mach, + const struct tgsi_full_declaration *decl) +{ + if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) { + if( decl->Declaration.File == TGSI_FILE_INPUT ) { + unsigned first, last, mask; + interpolation_func interp; + + assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE ); + + first = decl->u.DeclarationRange.First; + last = decl->u.DeclarationRange.Last; + mask = decl->Declaration.UsageMask; + + switch( decl->Interpolation.Interpolate ) { + case TGSI_INTERPOLATE_CONSTANT: + interp = constant_interpolation; + break; + + case TGSI_INTERPOLATE_LINEAR: + interp = linear_interpolation; + break; + + case TGSI_INTERPOLATE_PERSPECTIVE: + interp = perspective_interpolation; + break; + + default: + assert( 0 ); + } + + if( mask == TGSI_WRITEMASK_XYZW ) { + unsigned i, j; + + for( i = first; i <= last; i++ ) { + for( j = 0; j < NUM_CHANNELS; j++ ) { + interp( mach, i, j ); + } + } + } + else { + unsigned i, j; + + for( j = 0; j < NUM_CHANNELS; j++ ) { + if( mask & (1 << j) ) { + for( i = first; i <= last; i++ ) { + interp( mach, i, j ); + } + } + } + } + } + } +} + +static void +exec_instruction( + struct spu_exec_machine *mach, + const struct tgsi_full_instruction *inst, + int *pc ) +{ + uint chan_index; + union spu_exec_channel r[8]; + + (*pc)++; + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_ARL: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = si_cflts(r[0].q, 0); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_MOV: + /* TGSI_OPCODE_SWZ */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_LIT: + if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { + FETCH( &r[0], 0, CHAN_X ); + if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { + r[0].q = micro_max(r[0].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q); + STORE( &r[0], 0, CHAN_Y ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { + FETCH( &r[1], 0, CHAN_Y ); + r[1].q = micro_max(r[1].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q); + + FETCH( &r[2], 0, CHAN_W ); + r[2].q = micro_min(r[2].q, mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].q); + r[2].q = micro_max(r[2].q, mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].q); + r[1].q = micro_pow(r[1].q, r[2].q); + + /* r0 = (r0 > 0.0) ? r1 : 0.0 + */ + r[0].q = si_fcgt(r[0].q, mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q); + r[0].q = si_selb(mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q, r[1].q, + r[0].q); + STORE( &r[0], 0, CHAN_Z ); + } + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W ); + } + break; + + case TGSI_OPCODE_RCP: + /* TGSI_OPCODE_RECIP */ + FETCH( &r[0], 0, CHAN_X ); + r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q); + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_RSQ: + /* TGSI_OPCODE_RECIPSQRT */ + FETCH( &r[0], 0, CHAN_X ); + r[0].q = micro_sqrt(r[0].q); + r[0].q = micro_div(mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].q, r[0].q); + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_EXP: + assert (0); + break; + + case TGSI_OPCODE_LOG: + assert (0); + break; + + case TGSI_OPCODE_MUL: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) + { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + + r[0].q = si_fm(r[0].q, r[1].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_ADD: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = si_fa(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_DP3: + /* TGSI_OPCODE_DOT3 */ + FETCH( &r[0], 0, CHAN_X ); + FETCH( &r[1], 1, CHAN_X ); + r[0].q = si_fm(r[0].q, r[1].q); + + FETCH( &r[1], 0, CHAN_Y ); + FETCH( &r[2], 1, CHAN_Y ); + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + + FETCH( &r[1], 0, CHAN_Z ); + FETCH( &r[2], 1, CHAN_Z ); + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_DP4: + /* TGSI_OPCODE_DOT4 */ + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 1, CHAN_X); + + r[0].q = si_fm(r[0].q, r[1].q); + + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 1, CHAN_Y); + + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FETCH(&r[1], 0, CHAN_Z); + FETCH(&r[2], 1, CHAN_Z); + + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FETCH(&r[1], 0, CHAN_W); + FETCH(&r[2], 1, CHAN_W); + + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_DST: + if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { + FETCH( &r[0], 0, CHAN_Y ); + FETCH( &r[1], 1, CHAN_Y); + r[0].q = si_fm(r[0].q, r[1].q); + STORE( &r[0], 0, CHAN_Y ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { + FETCH( &r[0], 0, CHAN_Z ); + STORE( &r[0], 0, CHAN_Z ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { + FETCH( &r[0], 1, CHAN_W ); + STORE( &r[0], 0, CHAN_W ); + } + break; + + case TGSI_OPCODE_MIN: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + + r[0].q = micro_min(r[0].q, r[1].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_MAX: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + + r[0].q = micro_max(r[0].q, r[1].q); + + STORE(&r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SLT: + /* TGSI_OPCODE_SETLT */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + + r[0].q = micro_ge(r[0].q, r[1].q); + r[0].q = si_xori(r[0].q, 0xff); + + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SGE: + /* TGSI_OPCODE_SETGE */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = micro_ge(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_MAD: + /* TGSI_OPCODE_MADD */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + FETCH( &r[2], 2, chan_index ); + r[0].q = si_fma(r[0].q, r[1].q, r[2].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SUB: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + + r[0].q = si_fs(r[0].q, r[1].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_LERP: + /* TGSI_OPCODE_LRP */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + FETCH(&r[2], 2, chan_index); + + r[1].q = si_fs(r[1].q, r[2].q); + r[0].q = si_fma(r[0].q, r[1].q, r[2].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_CND: + assert (0); + break; + + case TGSI_OPCODE_CND0: + assert (0); + break; + + case TGSI_OPCODE_DOT2ADD: + /* TGSI_OPCODE_DP2A */ + assert (0); + break; + + case TGSI_OPCODE_INDEX: + assert (0); + break; + + case TGSI_OPCODE_NEGATE: + assert (0); + break; + + case TGSI_OPCODE_FRAC: + /* TGSI_OPCODE_FRC */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_frc(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_CLAMP: + assert (0); + break; + + case TGSI_OPCODE_FLOOR: + /* TGSI_OPCODE_FLR */ + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_flr(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_ROUND: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_rnd(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_EXPBASE2: + /* TGSI_OPCODE_EX2 */ + FETCH(&r[0], 0, CHAN_X); + + r[0].q = micro_pow(mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].q, r[0].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_LOGBASE2: + /* TGSI_OPCODE_LG2 */ + FETCH( &r[0], 0, CHAN_X ); + r[0].q = micro_lg2(r[0].q); + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_POWER: + /* TGSI_OPCODE_POW */ + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 1, CHAN_X); + + r[0].q = micro_pow(r[0].q, r[1].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_CROSSPRODUCT: + /* TGSI_OPCODE_XPD */ + FETCH(&r[0], 0, CHAN_Y); + FETCH(&r[1], 1, CHAN_Z); + FETCH(&r[3], 0, CHAN_Z); + FETCH(&r[4], 1, CHAN_Y); + + /* r2 = (r0 * r1) - (r3 * r5) + */ + r[2].q = si_fm(r[3].q, r[5].q); + r[2].q = si_fms(r[0].q, r[1].q, r[2].q); + + if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) { + STORE( &r[2], 0, CHAN_X ); + } + + FETCH(&r[2], 1, CHAN_X); + FETCH(&r[5], 0, CHAN_X); + + /* r3 = (r3 * r2) - (r1 * r5) + */ + r[1].q = si_fm(r[1].q, r[5].q); + r[3].q = si_fms(r[3].q, r[2].q, r[1].q); + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) { + STORE( &r[3], 0, CHAN_Y ); + } + + /* r5 = (r5 * r4) - (r0 * r2) + */ + r[0].q = si_fm(r[0].q, r[2].q); + r[5].q = si_fms(r[5].q, r[4].q, r[0].q); + + if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) { + STORE( &r[5], 0, CHAN_Z ); + } + + if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W ); + } + break; + + case TGSI_OPCODE_MULTIPLYMATRIX: + assert (0); + break; + + case TGSI_OPCODE_ABS: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + + r[0].q = micro_abs(r[0].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_RCC: + assert (0); + break; + + case TGSI_OPCODE_DPH: + FETCH(&r[0], 0, CHAN_X); + FETCH(&r[1], 1, CHAN_X); + + r[0].q = si_fm(r[0].q, r[1].q); + + FETCH(&r[1], 0, CHAN_Y); + FETCH(&r[2], 1, CHAN_Y); + + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FETCH(&r[1], 0, CHAN_Z); + FETCH(&r[2], 1, CHAN_Z); + + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FETCH(&r[1], 1, CHAN_W); + + r[0].q = si_fa(r[0].q, r[1].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_COS: + FETCH(&r[0], 0, CHAN_X); + + r[0].q = micro_cos(r[0].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_DDX: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_ddx(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_DDY: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_ddy(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_KILP: + exec_kilp (mach, inst); + break; + + case TGSI_OPCODE_KIL: + /* for enabled ExecMask bits, set the killed bit */ + mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= mach->ExecMask; + break; + + case TGSI_OPCODE_PK2H: + assert (0); + break; + + case TGSI_OPCODE_PK2US: + assert (0); + break; + + case TGSI_OPCODE_PK4B: + assert (0); + break; + + case TGSI_OPCODE_PK4UB: + assert (0); + break; + + case TGSI_OPCODE_RFL: + assert (0); + break; + + case TGSI_OPCODE_SEQ: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + + r[0].q = si_fceq(r[0].q, r[1].q); + + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SFL: + assert (0); + break; + + case TGSI_OPCODE_SGT: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = si_fcgt(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SIN: + FETCH( &r[0], 0, CHAN_X ); + r[0].q = micro_sin(r[0].q); + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SLE: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + + r[0].q = si_fcgt(r[0].q, r[1].q); + r[0].q = si_xori(r[0].q, 0xff); + + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SNE: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + + r[0].q = si_fceq(r[0].q, r[1].q); + r[0].q = si_xori(r[0].q, 0xff); + + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_STR: + assert (0); + break; + + case TGSI_OPCODE_TEX: + /* simple texture lookup */ + /* src[0] = texcoord */ + /* src[1] = sampler unit */ + exec_tex(mach, inst, FALSE); + break; + + case TGSI_OPCODE_TXB: + /* Texture lookup with lod bias */ + /* src[0] = texcoord (src[0].w = load bias) */ + /* src[1] = sampler unit */ + exec_tex(mach, inst, TRUE); + break; + + case TGSI_OPCODE_TXD: + /* Texture lookup with explict partial derivatives */ + /* src[0] = texcoord */ + /* src[1] = d[strq]/dx */ + /* src[2] = d[strq]/dy */ + /* src[3] = sampler unit */ + assert (0); + break; + + case TGSI_OPCODE_TXL: + /* Texture lookup with explit LOD */ + /* src[0] = texcoord (src[0].w = load bias) */ + /* src[1] = sampler unit */ + exec_tex(mach, inst, TRUE); + break; + + case TGSI_OPCODE_UP2H: + assert (0); + break; + + case TGSI_OPCODE_UP2US: + assert (0); + break; + + case TGSI_OPCODE_UP4B: + assert (0); + break; + + case TGSI_OPCODE_UP4UB: + assert (0); + break; + + case TGSI_OPCODE_X2D: + assert (0); + break; + + case TGSI_OPCODE_ARA: + assert (0); + break; + + case TGSI_OPCODE_ARR: + assert (0); + break; + + case TGSI_OPCODE_BRA: + assert (0); + break; + + case TGSI_OPCODE_CAL: + /* skip the call if no execution channels are enabled */ + if (mach->ExecMask) { + /* do the call */ + + /* push the Cond, Loop, Cont stacks */ + assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); + mach->CondStack[mach->CondStackTop++] = mach->CondMask; + assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; + assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + mach->ContStack[mach->ContStackTop++] = mach->ContMask; + + assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); + mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; + + /* note that PC was already incremented above */ + mach->CallStack[mach->CallStackTop++] = *pc; + *pc = inst->InstructionExtLabel.Label; + } + break; + + case TGSI_OPCODE_RET: + mach->FuncMask &= ~mach->ExecMask; + UPDATE_EXEC_MASK(mach); + + if (mach->ExecMask == 0x0) { + /* really return now (otherwise, keep executing */ + + if (mach->CallStackTop == 0) { + /* returning from main() */ + *pc = -1; + return; + } + *pc = mach->CallStack[--mach->CallStackTop]; + + /* pop the Cond, Loop, Cont stacks */ + assert(mach->CondStackTop > 0); + mach->CondMask = mach->CondStack[--mach->CondStackTop]; + assert(mach->LoopStackTop > 0); + mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; + assert(mach->ContStackTop > 0); + mach->ContMask = mach->ContStack[--mach->ContStackTop]; + assert(mach->FuncStackTop > 0); + mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; + + UPDATE_EXEC_MASK(mach); + } + break; + + case TGSI_OPCODE_SSG: + assert (0); + break; + + case TGSI_OPCODE_CMP: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH(&r[0], 0, chan_index); + FETCH(&r[1], 1, chan_index); + FETCH(&r[2], 2, chan_index); + + /* r0 = (r0 < 0.0) ? r1 : r2 + */ + r[3].q = si_xor(r[3].q, r[3].q); + r[0].q = micro_lt(r[0].q, r[3].q); + r[0].q = si_selb(r[1].q, r[2].q, r[0].q); + + STORE(&r[0], 0, chan_index); + } + break; + + case TGSI_OPCODE_SCS: + if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) { + FETCH( &r[0], 0, CHAN_X ); + } + if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) ) { + r[1].q = micro_cos(r[0].q); + STORE( &r[1], 0, CHAN_X ); + } + if( IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) { + r[1].q = micro_sin(r[0].q); + STORE( &r[1], 0, CHAN_Y ); + } + if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) { + STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z ); + } + if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) { + STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W ); + } + break; + + case TGSI_OPCODE_NRM: + assert (0); + break; + + case TGSI_OPCODE_DIV: + assert( 0 ); + break; + + case TGSI_OPCODE_DP2: + FETCH( &r[0], 0, CHAN_X ); + FETCH( &r[1], 1, CHAN_X ); + r[0].q = si_fm(r[0].q, r[1].q); + + FETCH( &r[1], 0, CHAN_Y ); + FETCH( &r[2], 1, CHAN_Y ); + r[0].q = si_fma(r[1].q, r[2].q, r[0].q); + + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_IF: + /* push CondMask */ + assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); + mach->CondStack[mach->CondStackTop++] = mach->CondMask; + FETCH( &r[0], 0, CHAN_X ); + /* update CondMask */ + if( ! r[0].u[0] ) { + mach->CondMask &= ~0x1; + } + if( ! r[0].u[1] ) { + mach->CondMask &= ~0x2; + } + if( ! r[0].u[2] ) { + mach->CondMask &= ~0x4; + } + if( ! r[0].u[3] ) { + mach->CondMask &= ~0x8; + } + UPDATE_EXEC_MASK(mach); + /* Todo: If CondMask==0, jump to ELSE */ + break; + + case TGSI_OPCODE_ELSE: + /* invert CondMask wrt previous mask */ + { + uint prevMask; + assert(mach->CondStackTop > 0); + prevMask = mach->CondStack[mach->CondStackTop - 1]; + mach->CondMask = ~mach->CondMask & prevMask; + UPDATE_EXEC_MASK(mach); + /* Todo: If CondMask==0, jump to ENDIF */ + } + break; + + case TGSI_OPCODE_ENDIF: + /* pop CondMask */ + assert(mach->CondStackTop > 0); + mach->CondMask = mach->CondStack[--mach->CondStackTop]; + UPDATE_EXEC_MASK(mach); + break; + + case TGSI_OPCODE_END: + /* halt execution */ + *pc = -1; + break; + + case TGSI_OPCODE_REP: + assert (0); + break; + + case TGSI_OPCODE_ENDREP: + assert (0); + break; + + case TGSI_OPCODE_PUSHA: + assert (0); + break; + + case TGSI_OPCODE_POPA: + assert (0); + break; + + case TGSI_OPCODE_CEIL: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_ceil(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_I2F: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = si_csflt(r[0].q, 0); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_NOT: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = si_xorbi(r[0].q, 0xff); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_TRUNC: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + r[0].q = micro_trunc(r[0].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SHL: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + + r[0].q = si_shl(r[0].q, r[1].q); + + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SHR: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = micro_ishr(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_AND: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = si_and(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_OR: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = si_or(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_MOD: + assert (0); + break; + + case TGSI_OPCODE_XOR: + FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) { + FETCH( &r[0], 0, chan_index ); + FETCH( &r[1], 1, chan_index ); + r[0].q = si_xor(r[0].q, r[1].q); + STORE( &r[0], 0, chan_index ); + } + break; + + case TGSI_OPCODE_SAD: + assert (0); + break; + + case TGSI_OPCODE_TXF: + assert (0); + break; + + case TGSI_OPCODE_TXQ: + assert (0); + break; + + case TGSI_OPCODE_EMIT: + mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += 16; + mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++; + break; + + case TGSI_OPCODE_ENDPRIM: + mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]++; + mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0; + break; + + case TGSI_OPCODE_LOOP: + /* fall-through (for now) */ + case TGSI_OPCODE_BGNLOOP2: + /* push LoopMask and ContMasks */ + assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; + assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); + mach->ContStack[mach->ContStackTop++] = mach->ContMask; + break; + + case TGSI_OPCODE_ENDLOOP: + /* fall-through (for now at least) */ + case TGSI_OPCODE_ENDLOOP2: + /* Restore ContMask, but don't pop */ + assert(mach->ContStackTop > 0); + mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; + if (mach->LoopMask) { + /* repeat loop: jump to instruction just past BGNLOOP */ + *pc = inst->InstructionExtLabel.Label + 1; + } + else { + /* exit loop: pop LoopMask */ + assert(mach->LoopStackTop > 0); + mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; + /* pop ContMask */ + assert(mach->ContStackTop > 0); + mach->ContMask = mach->ContStack[--mach->ContStackTop]; + } + UPDATE_EXEC_MASK(mach); + break; + + case TGSI_OPCODE_BRK: + /* turn off loop channels for each enabled exec channel */ + mach->LoopMask &= ~mach->ExecMask; + /* Todo: if mach->LoopMask == 0, jump to end of loop */ + UPDATE_EXEC_MASK(mach); + break; + + case TGSI_OPCODE_CONT: + /* turn off cont channels for each enabled exec channel */ + mach->ContMask &= ~mach->ExecMask; + /* Todo: if mach->LoopMask == 0, jump to end of loop */ + UPDATE_EXEC_MASK(mach); + break; + + case TGSI_OPCODE_BGNSUB: + /* no-op */ + break; + + case TGSI_OPCODE_ENDSUB: + /* no-op */ + break; + + case TGSI_OPCODE_NOISE1: + assert( 0 ); + break; + + case TGSI_OPCODE_NOISE2: + assert( 0 ); + break; + + case TGSI_OPCODE_NOISE3: + assert( 0 ); + break; + + case TGSI_OPCODE_NOISE4: + assert( 0 ); + break; + + case TGSI_OPCODE_NOP: + break; + + default: + assert( 0 ); + } +} + + +/** + * Run TGSI interpreter. + * \return bitmask of "alive" quad components + */ +uint +spu_exec_machine_run( struct spu_exec_machine *mach ) +{ + uint i; + int pc = 0; + + mach->CondMask = 0xf; + mach->LoopMask = 0xf; + mach->ContMask = 0xf; + mach->FuncMask = 0xf; + mach->ExecMask = 0xf; + + mach->CondStackTop = 0; /* temporarily subvert this assertion */ + assert(mach->CondStackTop == 0); + assert(mach->LoopStackTop == 0); + assert(mach->ContStackTop == 0); + assert(mach->CallStackTop == 0); + + mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0; + mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0; + + if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) { + mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0; + mach->Primitives[0] = 0; + } + + + /* execute declarations (interpolants) */ + if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) { + for (i = 0; i < mach->NumDeclarations; i++) { + uint8_t buffer[sizeof(struct tgsi_full_declaration) + 32] ALIGN16_ATTRIB; + struct tgsi_full_declaration decl; + unsigned long decl_addr = (unsigned long) (mach->Declarations+i); + unsigned size = ((sizeof(decl) + (decl_addr & 0x0f) + 0x0f) & ~0x0f); + + mfc_get(buffer, decl_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0); + wait_on_mask(1 << TAG_INSTRUCTION_FETCH); + + memcpy(& decl, buffer + (decl_addr & 0x0f), sizeof(decl)); + exec_declaration( mach, &decl ); + } + } + + /* execute instructions, until pc is set to -1 */ + while (pc != -1) { + uint8_t buffer[sizeof(struct tgsi_full_instruction) + 32] ALIGN16_ATTRIB; + struct tgsi_full_instruction inst; + unsigned long inst_addr = (unsigned long) (mach->Instructions + pc); + unsigned size = ((sizeof(inst) + (inst_addr & 0x0f) + 0x0f) & ~0x0f); + + assert(pc < mach->NumInstructions); + mfc_get(buffer, inst_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0); + wait_on_mask(1 << TAG_INSTRUCTION_FETCH); + + memcpy(& inst, buffer + (inst_addr & 0x0f), sizeof(inst)); + exec_instruction( mach, & inst, &pc ); + } + +#if 0 + /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */ + if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { + /* + * Scale back depth component. + */ + for (i = 0; i < 4; i++) + mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF; + } +#endif + + return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0]; +} + + diff --git a/src/gallium/drivers/cell/spu/spu_exec.h b/src/gallium/drivers/cell/spu/spu_exec.h new file mode 100644 index 0000000000..b4c7661ef6 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_exec.h @@ -0,0 +1,172 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#if !defined SPU_EXEC_H +#define SPU_EXEC_H + +#include "pipe/p_compiler.h" +#include "pipe/tgsi/exec/tgsi_exec.h" + +#if defined __cplusplus +extern "C" { +#endif + +/** + * Registers may be treated as float, signed int or unsigned int. + */ +union spu_exec_channel +{ + float f[QUAD_SIZE]; + int i[QUAD_SIZE]; + unsigned u[QUAD_SIZE]; + qword q; +}; + +/** + * A vector[RGBA] of channels[4 pixels] + */ +struct spu_exec_vector +{ + union spu_exec_channel xyzw[NUM_CHANNELS]; +}; + +/** + * For fragment programs, information for computing fragment input + * values from plane equation of the triangle/line. + */ +struct spu_interp_coef +{ + float a0[NUM_CHANNELS]; /* in an xyzw layout */ + float dadx[NUM_CHANNELS]; + float dady[NUM_CHANNELS]; +}; + + +struct softpipe_tile_cache; /**< Opaque to TGSI */ + +/** + * Information for sampling textures, which must be implemented + * by code outside the TGSI executor. + */ +struct spu_sampler +{ + const struct pipe_sampler_state *state; + struct pipe_texture *texture; + /** Get samples for four fragments in a quad */ + void (*get_samples)(struct spu_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]); + void *pipe; /*XXX temporary*/ + struct softpipe_tile_cache *cache; +}; + + +/** + * Run-time virtual machine state for executing TGSI shader. + */ +struct spu_exec_machine +{ + /* + * 32 program temporaries + * 4 internal temporaries + * 1 address + */ + struct spu_exec_vector Temps[TGSI_EXEC_NUM_TEMPS + + TGSI_EXEC_NUM_ADDRS + 1] + ALIGN16_ATTRIB; + + struct spu_exec_vector *Addrs; + + struct spu_sampler *Samplers; + + float Imms[TGSI_EXEC_NUM_IMMEDIATES][4]; + unsigned ImmLimit; + float (*Consts)[4]; + struct spu_exec_vector *Inputs; + struct spu_exec_vector *Outputs; + unsigned Processor; + + /* GEOMETRY processor only. */ + unsigned *Primitives; + + /* FRAGMENT processor only. */ + const struct spu_interp_coef *InterpCoefs; + struct spu_exec_vector QuadPos; + + /* Conditional execution masks */ + uint CondMask; /**< For IF/ELSE/ENDIF */ + uint LoopMask; /**< For BGNLOOP/ENDLOOP */ + uint ContMask; /**< For loop CONT statements */ + uint FuncMask; /**< For function calls */ + uint ExecMask; /**< = CondMask & LoopMask */ + + /** Condition mask stack (for nested conditionals) */ + uint CondStack[TGSI_EXEC_MAX_COND_NESTING]; + int CondStackTop; + + /** Loop mask stack (for nested loops) */ + uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING]; + int LoopStackTop; + + /** Loop continue mask stack (see comments in tgsi_exec.c) */ + uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING]; + int ContStackTop; + + /** Function execution mask stack (for executing subroutine code) */ + uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING]; + int FuncStackTop; + + /** Function call stack for saving/restoring the program counter */ + uint CallStack[TGSI_EXEC_MAX_CALL_NESTING]; + int CallStackTop; + + struct tgsi_full_instruction *Instructions; + uint NumInstructions; + + struct tgsi_full_declaration *Declarations; + uint NumDeclarations; +}; + + +extern void +spu_exec_machine_init(struct spu_exec_machine *mach, + uint numSamplers, + struct spu_sampler *samplers, + unsigned processor); + +extern uint +spu_exec_machine_run( struct spu_exec_machine *mach ); + + +#if defined __cplusplus +} /* extern "C" */ +#endif + +#endif /* SPU_EXEC_H */ diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c new file mode 100644 index 0000000000..e375197fe6 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_main.c @@ -0,0 +1,567 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/* main() for Cell SPU code */ + + +#include <stdio.h> +#include <libmisc.h> + +#include "spu_main.h" +#include "spu_render.h" +#include "spu_texture.h" +#include "spu_tile.h" +//#include "spu_test.h" +#include "spu_vertex_shader.h" +#include "pipe/cell/common.h" +#include "pipe/p_defines.h" + + +/* +helpful headers: +/usr/lib/gcc/spu/4.1.1/include/spu_mfcio.h +/opt/ibm/cell-sdk/prototype/sysroot/usr/include/libmisc.h +*/ + +boolean Debug = FALSE; + +struct spu_global spu; + +struct spu_vs_context draw; + +/** + * Tell the PPU that this SPU has finished copying a buffer to + * local store and that it may be reused by the PPU. + * This is done by writting a 16-byte batch-buffer-status block back into + * main memory (in cell_context->buffer_status[]). + */ +static void +release_buffer(uint buffer) +{ + /* Evidently, using less than a 16-byte status doesn't work reliably */ + static const uint status[4] ALIGN16_ATTRIB + = {CELL_BUFFER_STATUS_FREE, 0, 0, 0}; + + const uint index = 4 * (spu.init.id * CELL_NUM_BUFFERS + buffer); + uint *dst = spu.init.buffer_status + index; + + ASSERT(buffer < CELL_NUM_BUFFERS); + + mfc_put((void *) &status, /* src in local memory */ + (unsigned int) dst, /* dst in main memory */ + sizeof(status), /* size */ + TAG_MISC, /* tag is unimportant */ + 0, /* tid */ + 0 /* rid */); +} + + +/** + * For tiles whose status is TILE_STATUS_CLEAR, write solid-filled + * tiles back to the main framebuffer. + */ +static void +really_clear_tiles(uint surfaceIndex) +{ + const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles; + uint i; + + if (surfaceIndex == 0) { + clear_c_tile(&spu.ctile); + + for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) { + uint tx = i % spu.fb.width_tiles; + uint ty = i / spu.fb.width_tiles; + if (spu.ctile_status[ty][tx] == TILE_STATUS_CLEAR) { + put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0); + } + } + } + else { + clear_z_tile(&spu.ztile); + + for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) { + uint tx = i % spu.fb.width_tiles; + uint ty = i / spu.fb.width_tiles; + if (spu.ztile_status[ty][tx] == TILE_STATUS_CLEAR) + put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 1); + } + } + +#if 0 + wait_on_mask(1 << TAG_SURFACE_CLEAR); +#endif +} + + +static void +cmd_clear_surface(const struct cell_command_clear_surface *clear) +{ + const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles; + uint i; + + if (Debug) + printf("SPU %u: CLEAR SURF %u to 0x%08x\n", spu.init.id, + clear->surface, clear->value); + +#define CLEAR_OPT 1 +#if CLEAR_OPT + /* set all tile's status to CLEAR */ + if (clear->surface == 0) { + memset(spu.ctile_status, TILE_STATUS_CLEAR, sizeof(spu.ctile_status)); + spu.fb.color_clear_value = clear->value; + } + else { + memset(spu.ztile_status, TILE_STATUS_CLEAR, sizeof(spu.ztile_status)); + spu.fb.depth_clear_value = clear->value; + } + return; +#endif + + if (clear->surface == 0) { + spu.fb.color_clear_value = clear->value; + clear_c_tile(&spu.ctile); + } + else { + spu.fb.depth_clear_value = clear->value; + clear_z_tile(&spu.ztile); + } + + /* + printf("SPU: %s num=%d w=%d h=%d\n", + __FUNCTION__, num_tiles, spu.fb.width_tiles, spu.fb.height_tiles); + */ + + for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) { + uint tx = i % spu.fb.width_tiles; + uint ty = i / spu.fb.width_tiles; + if (clear->surface == 0) + put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0); + else + put_tile(tx, ty, &spu.ztile, TAG_SURFACE_CLEAR, 1); + /* XXX we don't want this here, but it fixes bad tile results */ + } + +#if 0 + wait_on_mask(1 << TAG_SURFACE_CLEAR); +#endif + + if (Debug) + printf("SPU %u: CLEAR SURF done\n", spu.init.id); +} + + +static void +cmd_release_verts(const struct cell_command_release_verts *release) +{ + if (Debug) + printf("SPU %u: RELEASE VERTS %u\n", + spu.init.id, release->vertex_buf); + ASSERT(release->vertex_buf != ~0U); + release_buffer(release->vertex_buf); +} + + +static void +cmd_state_framebuffer(const struct cell_command_framebuffer *cmd) +{ + if (Debug) + printf("SPU %u: FRAMEBUFFER: %d x %d at %p, cformat 0x%x zformat 0x%x\n", + spu.init.id, + cmd->width, + cmd->height, + cmd->color_start, + cmd->color_format, + cmd->depth_format); + + ASSERT_ALIGN16(cmd->color_start); + ASSERT_ALIGN16(cmd->depth_start); + + spu.fb.color_start = cmd->color_start; + spu.fb.depth_start = cmd->depth_start; + spu.fb.color_format = cmd->color_format; + spu.fb.depth_format = cmd->depth_format; + spu.fb.width = cmd->width; + spu.fb.height = cmd->height; + spu.fb.width_tiles = (spu.fb.width + TILE_SIZE - 1) / TILE_SIZE; + spu.fb.height_tiles = (spu.fb.height + TILE_SIZE - 1) / TILE_SIZE; + + if (spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM) + spu.fb.zsize = 4; + else if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) + spu.fb.zsize = 2; + else + spu.fb.zsize = 0; + + if (spu.fb.color_format == PIPE_FORMAT_A8R8G8B8_UNORM) + spu.color_shuffle = ((vector unsigned char) { + 12, 0, 4, 8, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0}); + else if (spu.fb.color_format == PIPE_FORMAT_B8G8R8A8_UNORM) + spu.color_shuffle = ((vector unsigned char) { + 8, 4, 0, 12, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0}); + else + ASSERT(0); +} + + +static void +cmd_state_blend(const struct pipe_blend_state *state) +{ + if (Debug) + printf("SPU %u: BLEND: enabled %d\n", + spu.init.id, + state->blend_enable); + + memcpy(&spu.blend, state, sizeof(*state)); +} + + +static void +cmd_state_depth_stencil(const struct pipe_depth_stencil_alpha_state *state) +{ + if (Debug) + printf("SPU %u: DEPTH_STENCIL: ztest %d\n", + spu.init.id, + state->depth.enabled); + + memcpy(&spu.depth_stencil, state, sizeof(*state)); +} + + +static void +cmd_state_sampler(const struct pipe_sampler_state *state) +{ + if (Debug) + printf("SPU %u: SAMPLER\n", + spu.init.id); + + memcpy(&spu.sampler[0], state, sizeof(*state)); + if (spu.sampler[0].min_img_filter == PIPE_TEX_FILTER_LINEAR) + spu.sample_texture = sample_texture_bilinear; + else + spu.sample_texture = sample_texture_nearest; +} + + +static void +cmd_state_texture(const struct cell_command_texture *texture) +{ + if (Debug) + printf("SPU %u: TEXTURE at %p size %u x %u\n", + spu.init.id, texture->start, texture->width, texture->height); + + memcpy(&spu.texture, texture, sizeof(*texture)); + spu.tex_size = (vector float) + { spu.texture.width, spu.texture.height, 0.0, 0.0}; + spu.tex_size_mask = (vector unsigned int) + { spu.texture.width - 1, spu.texture.height - 1, 0, 0 }; +} + + +static void +cmd_state_vertex_info(const struct vertex_info *vinfo) +{ + if (Debug) { + printf("SPU %u: VERTEX_INFO num_attribs=%u\n", spu.init.id, + vinfo->num_attribs); + } + ASSERT(vinfo->num_attribs >= 1); + ASSERT(vinfo->num_attribs <= 8); + memcpy(&spu.vertex_info, vinfo, sizeof(*vinfo)); +} + + +static void +cmd_state_vs_array_info(const struct cell_array_info *vs_info) +{ + const unsigned attr = vs_info->attr; + + ASSERT(attr < PIPE_ATTRIB_MAX); + draw.vertex_fetch.src_ptr[attr] = vs_info->base; + draw.vertex_fetch.pitch[attr] = vs_info->pitch; + draw.vertex_fetch.format[attr] = vs_info->format; + draw.vertex_fetch.dirty = 1; +} + + +static void +cmd_finish(void) +{ + if (Debug) + printf("SPU %u: FINISH\n", spu.init.id); + really_clear_tiles(0); + /* wait for all outstanding DMAs to finish */ + mfc_write_tag_mask(~0); + mfc_read_tag_status_all(); + /* send mbox message to PPU */ + spu_write_out_mbox(CELL_CMD_FINISH); +} + + +/** + * Execute a batch of commands + * The opcode param encodes the location of the buffer and its size. + */ +static void +cmd_batch(uint opcode) +{ + const uint buf = (opcode >> 8) & 0xff; + uint size = (opcode >> 16); + uint64_t buffer[CELL_BUFFER_SIZE / 8] ALIGN16_ATTRIB; + const unsigned usize = size / sizeof(buffer[0]); + uint pos; + + if (Debug) + printf("SPU %u: BATCH buffer %u, len %u, from %p\n", + spu.init.id, buf, size, spu.init.buffers[buf]); + + ASSERT((opcode & CELL_CMD_OPCODE_MASK) == CELL_CMD_BATCH); + + ASSERT_ALIGN16(spu.init.buffers[buf]); + + size = ROUNDUP16(size); + + ASSERT_ALIGN16(spu.init.buffers[buf]); + + mfc_get(buffer, /* dest */ + (unsigned int) spu.init.buffers[buf], /* src */ + size, + TAG_BATCH_BUFFER, + 0, /* tid */ + 0 /* rid */); + wait_on_mask(1 << TAG_BATCH_BUFFER); + + /* Tell PPU we're done copying the buffer to local store */ + if (Debug) + printf("SPU %u: release batch buf %u\n", spu.init.id, buf); + release_buffer(buf); + + for (pos = 0; pos < usize; /* no incr */) { + switch (buffer[pos]) { + case CELL_CMD_STATE_FRAMEBUFFER: + { + struct cell_command_framebuffer *fb + = (struct cell_command_framebuffer *) &buffer[pos]; + cmd_state_framebuffer(fb); + pos += sizeof(*fb) / 8; + } + break; + case CELL_CMD_CLEAR_SURFACE: + { + struct cell_command_clear_surface *clr + = (struct cell_command_clear_surface *) &buffer[pos]; + cmd_clear_surface(clr); + pos += sizeof(*clr) / 8; + } + break; + case CELL_CMD_RENDER: + { + struct cell_command_render *render + = (struct cell_command_render *) &buffer[pos]; + uint pos_incr; + cmd_render(render, &pos_incr); + pos += pos_incr; + } + break; + case CELL_CMD_RELEASE_VERTS: + { + struct cell_command_release_verts *release + = (struct cell_command_release_verts *) &buffer[pos]; + cmd_release_verts(release); + pos += sizeof(*release) / 8; + } + break; + case CELL_CMD_FINISH: + cmd_finish(); + pos += 1; + break; + case CELL_CMD_STATE_BLEND: + cmd_state_blend((struct pipe_blend_state *) + &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct pipe_blend_state)) / 8); + break; + case CELL_CMD_STATE_DEPTH_STENCIL: + cmd_state_depth_stencil((struct pipe_depth_stencil_alpha_state *) + &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct pipe_depth_stencil_alpha_state)) / 8); + break; + case CELL_CMD_STATE_SAMPLER: + cmd_state_sampler((struct pipe_sampler_state *) &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct pipe_sampler_state)) / 8); + break; + case CELL_CMD_STATE_TEXTURE: + cmd_state_texture((struct cell_command_texture *) &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct cell_command_texture)) / 8); + break; + case CELL_CMD_STATE_VERTEX_INFO: + cmd_state_vertex_info((struct vertex_info *) &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct vertex_info)) / 8); + break; + case CELL_CMD_STATE_VIEWPORT: + (void) memcpy(& draw.viewport, &buffer[pos+1], + sizeof(struct pipe_viewport_state)); + pos += (1 + ROUNDUP8(sizeof(struct pipe_viewport_state)) / 8); + break; + case CELL_CMD_STATE_VS_ARRAY_INFO: + cmd_state_vs_array_info((struct cell_array_info *) &buffer[pos+1]); + pos += (1 + ROUNDUP8(sizeof(struct cell_array_info)) / 8); + break; + default: + printf("SPU %u: bad opcode: 0x%llx\n", spu.init.id, buffer[pos]); + ASSERT(0); + break; + } + } + + if (Debug) + printf("SPU %u: BATCH complete\n", spu.init.id); +} + + +/** + * Temporary/simple main loop for SPEs: Get a command, execute it, repeat. + */ +static void +main_loop(void) +{ + struct cell_command cmd; + int exitFlag = 0; + + if (Debug) + printf("SPU %u: Enter main loop\n", spu.init.id); + + ASSERT((sizeof(struct cell_command) & 0xf) == 0); + ASSERT_ALIGN16(&cmd); + + while (!exitFlag) { + unsigned opcode; + int tag = 0; + + if (Debug) + printf("SPU %u: Wait for cmd...\n", spu.init.id); + + /* read/wait from mailbox */ + opcode = (unsigned int) spu_read_in_mbox(); + + if (Debug) + printf("SPU %u: got cmd 0x%x\n", spu.init.id, opcode); + + /* command payload */ + mfc_get(&cmd, /* dest */ + (unsigned int) spu.init.cmd, /* src */ + sizeof(struct cell_command), /* bytes */ + tag, + 0, /* tid */ + 0 /* rid */); + wait_on_mask( 1 << tag ); + + /* + * NOTE: most commands should be contained in a batch buffer + */ + + switch (opcode & CELL_CMD_OPCODE_MASK) { + case CELL_CMD_EXIT: + if (Debug) + printf("SPU %u: EXIT\n", spu.init.id); + exitFlag = 1; + break; + case CELL_CMD_VS_EXECUTE: + spu_execute_vertex_shader(&draw, &cmd.vs); + break; + case CELL_CMD_BATCH: + cmd_batch(opcode); + break; + default: + printf("Bad opcode!\n"); + } + + } + + if (Debug) + printf("SPU %u: Exit main loop\n", spu.init.id); +} + + + +static void +one_time_init(void) +{ + memset(spu.ctile_status, TILE_STATUS_DEFINED, sizeof(spu.ctile_status)); + memset(spu.ztile_status, TILE_STATUS_DEFINED, sizeof(spu.ztile_status)); + invalidate_tex_cache(); +} + + + +/* In some versions of the SDK the SPE main takes 'unsigned long' as a + * parameter. In others it takes 'unsigned long long'. Use a define to + * select between the two. + */ +#ifdef SPU_MAIN_PARAM_LONG_LONG +typedef unsigned long long main_param_t; +#else +typedef unsigned long main_param_t; +#endif + +/** + * SPE entrypoint. + */ +int +main(main_param_t speid, main_param_t argp) +{ + int tag = 0; + + (void) speid; + + ASSERT(sizeof(tile_t) == TILE_SIZE * TILE_SIZE * 4); + ASSERT(sizeof(struct cell_command_render) % 8 == 0); + + one_time_init(); + + if (Debug) + printf("SPU: main() speid=%lu\n", speid); + + mfc_get(&spu.init, /* dest */ + (unsigned int) argp, /* src */ + sizeof(struct cell_init_info), /* bytes */ + tag, + 0, /* tid */ + 0 /* rid */); + wait_on_mask( 1 << tag ); + +#if 0 + if (spu.init.id==0) + spu_test_misc(); +#endif + + main_loop(); + + return 0; +} diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h new file mode 100644 index 0000000000..1710a17512 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_main.h @@ -0,0 +1,177 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SPU_MAIN_H +#define SPU_MAIN_H + + +#include <spu_mfcio.h> + +#include "pipe/cell/common.h" +#include "pipe/draw/draw_vertex.h" +#include "pipe/p_state.h" + + + +#define MAX_WIDTH 1024 +#define MAX_HEIGHT 1024 + + +typedef union { + ushort us[TILE_SIZE][TILE_SIZE]; + uint ui[TILE_SIZE][TILE_SIZE]; + vector unsigned short us8[TILE_SIZE/2][TILE_SIZE/4]; + vector unsigned int ui4[TILE_SIZE/2][TILE_SIZE/2]; +} tile_t; + + +#define TILE_STATUS_CLEAR 1 +#define TILE_STATUS_DEFINED 2 /**< defined in FB, but not in local store */ +#define TILE_STATUS_CLEAN 3 /**< in local store, but not changed */ +#define TILE_STATUS_DIRTY 4 /**< modified locally, but not put back yet */ +#define TILE_STATUS_GETTING 5 /**< mfc_get() called but not yet arrived */ + + +struct spu_framebuffer { + void *color_start; /**< addr of color surface in main memory */ + void *depth_start; /**< addr of depth surface in main memory */ + enum pipe_format color_format; + enum pipe_format depth_format; + uint width, height; /**< size in pixels */ + uint width_tiles, height_tiles; /**< width and height in tiles */ + + uint color_clear_value; + uint depth_clear_value; + + uint zsize; /**< 0, 2 or 4 bytes per Z */ +} ALIGN16_ATTRIB; + + +/** + * All SPU global/context state will be in singleton object of this type: + */ +struct spu_global +{ + struct cell_init_info init; + + struct spu_framebuffer fb; + struct pipe_blend_state blend_stencil; + struct pipe_depth_stencil_alpha_state depth_stencil; + struct pipe_blend_state blend; + struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS]; + struct cell_command_texture texture; + + struct vertex_info vertex_info; + + /* XXX more state to come */ + + + /** current color and Z tiles */ + tile_t ctile ALIGN16_ATTRIB; + tile_t ztile ALIGN16_ATTRIB; + + /** Current tiles' status */ + ubyte cur_ctile_status, cur_ztile_status; + + /** Status of all tiles in framebuffer */ + ubyte ctile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB; + ubyte ztile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB; + + + /** for converting RGBA to PIPE_FORMAT_x colors */ + vector unsigned char color_shuffle; + + vector float tex_size; + vector unsigned int tex_size_mask; /**< == int(size - 1) */ + + vector float (*sample_texture)(vector float texcoord); + +} ALIGN16_ATTRIB; + + +extern struct spu_global spu; +extern boolean Debug; + + + + +/* DMA TAGS */ + +#define TAG_SURFACE_CLEAR 10 +#define TAG_VERTEX_BUFFER 11 +#define TAG_READ_TILE_COLOR 12 +#define TAG_READ_TILE_Z 13 +#define TAG_WRITE_TILE_COLOR 14 +#define TAG_WRITE_TILE_Z 15 +#define TAG_INDEX_BUFFER 16 +#define TAG_BATCH_BUFFER 17 +#define TAG_MISC 18 +#define TAG_TEXTURE_TILE 19 +#define TAG_INSTRUCTION_FETCH 20 + + + +static INLINE void +wait_on_mask(unsigned tagMask) +{ + mfc_write_tag_mask( tagMask ); + /* wait for completion of _any_ DMAs specified by tagMask */ + mfc_read_tag_status_any(); +} + + +static INLINE void +wait_on_mask_all(unsigned tagMask) +{ + mfc_write_tag_mask( tagMask ); + /* wait for completion of _any_ DMAs specified by tagMask */ + mfc_read_tag_status_all(); +} + + + + + +static INLINE void +memset16(ushort *d, ushort value, uint count) +{ + uint i; + for (i = 0; i < count; i++) + d[i] = value; +} + + +static INLINE void +memset32(uint *d, uint value, uint count) +{ + uint i; + for (i = 0; i < count; i++) + d[i] = value; +} + + +#endif /* SPU_MAIN_H */ diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c new file mode 100644 index 0000000000..932fb500b3 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_render.c @@ -0,0 +1,301 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <stdio.h> +#include <libmisc.h> +#include <spu_mfcio.h> + +#include "spu_main.h" +#include "spu_render.h" +#include "spu_tri.h" +#include "spu_tile.h" +#include "pipe/cell/common.h" + + + +/** + * Given a rendering command's bounding box (in pixels) compute the + * location of the corresponding screen tile bounding box. + */ +static INLINE void +tile_bounding_box(const struct cell_command_render *render, + uint *txmin, uint *tymin, + uint *box_num_tiles, uint *box_width_tiles) +{ +#if 0 + /* Debug: full-window bounding box */ + uint txmax = spu.fb.width_tiles - 1; + uint tymax = spu.fb.height_tiles - 1; + *txmin = 0; + *tymin = 0; + *box_num_tiles = spu.fb.width_tiles * spu.fb.height_tiles; + *box_width_tiles = spu.fb.width_tiles; + (void) render; + (void) txmax; + (void) tymax; +#else + uint txmax, tymax, box_height_tiles; + + *txmin = (uint) render->xmin / TILE_SIZE; + *tymin = (uint) render->ymin / TILE_SIZE; + txmax = (uint) render->xmax / TILE_SIZE; + tymax = (uint) render->ymax / TILE_SIZE; + if (txmax >= spu.fb.width_tiles) + txmax = spu.fb.width_tiles-1; + if (tymax >= spu.fb.height_tiles) + tymax = spu.fb.height_tiles-1; + *box_width_tiles = txmax - *txmin + 1; + box_height_tiles = tymax - *tymin + 1; + *box_num_tiles = *box_width_tiles * box_height_tiles; +#endif +#if 0 + printf("SPU %u: bounds: %g, %g ... %g, %g\n", spu.init.id, + render->xmin, render->ymin, render->xmax, render->ymax); + printf("SPU %u: tiles: %u, %u .. %u, %u\n", + spu.init.id, *txmin, *tymin, txmax, tymax); + ASSERT(render->xmin <= render->xmax); + ASSERT(render->ymin <= render->ymax); +#endif +} + + +/** Check if the tile at (tx,ty) belongs to this SPU */ +static INLINE boolean +my_tile(uint tx, uint ty) +{ + return (spu.fb.width_tiles * ty + tx) % spu.init.num_spus == spu.init.id; +} + + +/** + * Start fetching non-clear color/Z tiles from main memory + */ +static INLINE void +get_cz_tiles(uint tx, uint ty) +{ + if (spu.depth_stencil.depth.enabled) { + if (spu.cur_ztile_status != TILE_STATUS_CLEAR) { + //printf("SPU %u: getting Z tile %u, %u\n", spu.init.id, tx, ty); + get_tile(tx, ty, &spu.ztile, TAG_READ_TILE_Z, 1); + spu.cur_ztile_status = TILE_STATUS_GETTING; + } + } + + if (spu.cur_ctile_status != TILE_STATUS_CLEAR) { + //printf("SPU %u: getting C tile %u, %u\n", spu.init.id, tx, ty); + get_tile(tx, ty, &spu.ctile, TAG_READ_TILE_COLOR, 0); + spu.cur_ctile_status = TILE_STATUS_GETTING; + } +} + + +/** + * Start putting dirty color/Z tiles back to main memory + */ +static INLINE void +put_cz_tiles(uint tx, uint ty) +{ + if (spu.cur_ztile_status == TILE_STATUS_DIRTY) { + /* tile was modified and needs to be written back */ + //printf("SPU %u: put dirty Z tile %u, %u\n", spu.init.id, tx, ty); + put_tile(tx, ty, &spu.ztile, TAG_WRITE_TILE_Z, 1); + spu.cur_ztile_status = TILE_STATUS_DEFINED; + } + else if (spu.cur_ztile_status == TILE_STATUS_GETTING) { + /* tile was never used */ + spu.cur_ztile_status = TILE_STATUS_DEFINED; + //printf("SPU %u: put getting Z tile %u, %u\n", spu.init.id, tx, ty); + } + + if (spu.cur_ctile_status == TILE_STATUS_DIRTY) { + /* tile was modified and needs to be written back */ + //printf("SPU %u: put dirty C tile %u, %u\n", spu.init.id, tx, ty); + put_tile(tx, ty, &spu.ctile, TAG_WRITE_TILE_COLOR, 0); + spu.cur_ctile_status = TILE_STATUS_DEFINED; + } + else if (spu.cur_ctile_status == TILE_STATUS_GETTING) { + /* tile was never used */ + spu.cur_ctile_status = TILE_STATUS_DEFINED; + //printf("SPU %u: put getting C tile %u, %u\n", spu.init.id, tx, ty); + } +} + + +/** + * Wait for 'put' of color/z tiles to complete. + */ +static INLINE void +wait_put_cz_tiles(void) +{ + wait_on_mask(1 << TAG_WRITE_TILE_COLOR); + if (spu.depth_stencil.depth.enabled) { + wait_on_mask(1 << TAG_WRITE_TILE_Z); + } +} + + +/** + * Render primitives + * \param pos_incr returns value indicating how may words to skip after + * this command in the batch buffer + */ +void +cmd_render(const struct cell_command_render *render, uint *pos_incr) +{ + /* we'll DMA into these buffers */ + ubyte vertex_data[CELL_BUFFER_SIZE] ALIGN16_ATTRIB; + const uint vertex_size = render->vertex_size; /* in bytes */ + /*const*/ uint total_vertex_bytes = render->num_verts * vertex_size; + uint index_bytes; + const ubyte *vertices; + const ushort *indexes; + uint i, j; + + + if (Debug) { + printf("SPU %u: RENDER prim %u, num_vert=%u num_ind=%u " + "inline_vert=%u\n", + spu.init.id, + render->prim_type, + render->num_verts, + render->num_indexes, + render->inline_verts); + + /* + printf(" bound: %g, %g .. %g, %g\n", + render->xmin, render->ymin, render->xmax, render->ymax); + */ + } + + ASSERT(sizeof(*render) % 4 == 0); + ASSERT(total_vertex_bytes % 16 == 0); + ASSERT(render->prim_type == PIPE_PRIM_TRIANGLES); + ASSERT(render->num_indexes % 3 == 0); + + + /* indexes are right after the render command in the batch buffer */ + indexes = (const ushort *) (render + 1); + index_bytes = ROUNDUP8(render->num_indexes * 2); + *pos_incr = index_bytes / 8 + sizeof(*render) / 8; + + + if (render->inline_verts) { + /* Vertices are after indexes in batch buffer at next 16-byte addr */ + vertices = (const ubyte *) render + (*pos_incr * 8); + vertices = (const ubyte *) align_pointer((void *) vertices, 16); + ASSERT_ALIGN16(vertices); + *pos_incr = ((vertices + total_vertex_bytes) - (ubyte *) render) / 8; + } + else { + /* Begin DMA fetch of vertex buffer */ + ubyte *src = spu.init.buffers[render->vertex_buf]; + ubyte *dest = vertex_data; + + /* skip vertex data we won't use */ +#if 01 + src += render->min_index * vertex_size; + dest += render->min_index * vertex_size; + total_vertex_bytes -= render->min_index * vertex_size; +#endif + ASSERT(total_vertex_bytes % 16 == 0); + ASSERT_ALIGN16(dest); + ASSERT_ALIGN16(src); + + mfc_get(dest, /* in vertex_data[] array */ + (unsigned int) src, /* src in main memory */ + total_vertex_bytes, /* size */ + TAG_VERTEX_BUFFER, + 0, /* tid */ + 0 /* rid */); + + vertices = vertex_data; + + wait_on_mask(1 << TAG_VERTEX_BUFFER); + } + + + /** + ** find tiles which intersect the prim bounding box + **/ + uint txmin, tymin, box_width_tiles, box_num_tiles; + tile_bounding_box(render, &txmin, &tymin, + &box_num_tiles, &box_width_tiles); + + + /* make sure any pending clears have completed */ + wait_on_mask(1 << TAG_SURFACE_CLEAR); /* XXX temporary */ + + + /** + ** loop over tiles, rendering tris + **/ + for (i = 0; i < box_num_tiles; i++) { + const uint tx = txmin + i % box_width_tiles; + const uint ty = tymin + i / box_width_tiles; + + ASSERT(tx < spu.fb.width_tiles); + ASSERT(ty < spu.fb.height_tiles); + + if (!my_tile(tx, ty)) + continue; + + spu.cur_ctile_status = spu.ctile_status[ty][tx]; + spu.cur_ztile_status = spu.ztile_status[ty][tx]; + + get_cz_tiles(tx, ty); + + uint drawn = 0; + + /* loop over tris */ + for (j = 0; j < render->num_indexes; j += 3) { + const float *v0, *v1, *v2; + + v0 = (const float *) (vertices + indexes[j+0] * vertex_size); + v1 = (const float *) (vertices + indexes[j+1] * vertex_size); + v2 = (const float *) (vertices + indexes[j+2] * vertex_size); + + drawn += tri_draw(v0, v1, v2, tx, ty); + } + + //printf("SPU %u: drew %u of %u\n", spu.init.id, drawn, render->num_indexes/3); + + /* write color/z tiles back to main framebuffer, if dirtied */ + put_cz_tiles(tx, ty); + + wait_put_cz_tiles(); /* XXX seems unnecessary... */ + + spu.ctile_status[ty][tx] = spu.cur_ctile_status; + spu.ztile_status[ty][tx] = spu.cur_ztile_status; + } + + if (Debug) + printf("SPU %u: RENDER done\n", + spu.init.id); +} + + diff --git a/src/gallium/drivers/cell/spu/spu_render.h b/src/gallium/drivers/cell/spu/spu_render.h new file mode 100644 index 0000000000..fbcdc5ec31 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_render.h @@ -0,0 +1,38 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef SPU_RENDER_H +#define SPU_RENDER_H + +#include "pipe/cell/common.h" + +extern void +cmd_render(const struct cell_command_render *render, uint *pos_incr); + +#endif /* SPU_RENDER_H */ + diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c new file mode 100644 index 0000000000..3962aaa4a9 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_texture.c @@ -0,0 +1,217 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_compiler.h" +#include "spu_main.h" +#include "spu_texture.h" +#include "spu_tile.h" +#include "spu_colorpack.h" + + +/** + * Number of texture tiles to cache. + * Note that this will probably be the largest consumer of SPU local store/ + * memory for this driver! + */ +#define CACHE_SIZE 16 + +static tile_t tex_tiles[CACHE_SIZE] ALIGN16_ATTRIB; + +static vector unsigned int tex_tile_xy[CACHE_SIZE]; + + + +/** + * Mark all tex cache entries as invalid. + */ +void +invalidate_tex_cache(void) +{ + /* XXX memset? */ + uint i; + for (i = 0; i < CACHE_SIZE; i++) { + tex_tile_xy[i] = ((vector unsigned int) { ~0U, ~0U, ~0U, ~0U }); + } +} + + +/** + * Return the cache pos/index which corresponds to tile (tx,ty) + */ +static INLINE uint +cache_pos(vector unsigned int txty) +{ + uint pos = (spu_extract(txty,0) + spu_extract(txty,1) * 4) % CACHE_SIZE; + return pos; +} + + +/** + * Make sure the tile for texel (i,j) is present, return its position/index + * in the cache. + */ +static uint +get_tex_tile(vector unsigned int ij) +{ + /* tile address: tx,ty */ + const vector unsigned int txty = spu_rlmask(ij, -5); /* divide by 32 */ + const uint pos = cache_pos(txty); + + if ((spu_extract(tex_tile_xy[pos], 0) != spu_extract(txty, 0)) || + (spu_extract(tex_tile_xy[pos], 1) != spu_extract(txty, 1))) { + + /* texture cache miss, fetch tile from main memory */ + const uint tiles_per_row = spu.texture.width / TILE_SIZE; + const uint bytes_per_tile = sizeof(tile_t); + const void *src = (const ubyte *) spu.texture.start + + (spu_extract(txty,1) * tiles_per_row + spu_extract(txty,0)) * bytes_per_tile; + + printf("SPU %u: tex cache miss at %d, %d pos=%u old=%d,%d\n", + spu.init.id, + spu_extract(txty,0), + spu_extract(txty,1), + pos, + spu_extract(tex_tile_xy[pos],0), + spu_extract(tex_tile_xy[pos],1)); + + ASSERT_ALIGN16(tex_tiles[pos].ui); + ASSERT_ALIGN16(src); + + mfc_get(tex_tiles[pos].ui, /* dest */ + (unsigned int) src, + bytes_per_tile, /* size */ + TAG_TEXTURE_TILE, + 0, /* tid */ + 0 /* rid */); + + wait_on_mask(1 << TAG_TEXTURE_TILE); + + tex_tile_xy[pos] = txty; + } + else { +#if 0 + printf("SPU %u: tex cache HIT at %d, %d\n", + spu.init.id, tx, ty); +#endif + } + + return pos; +} + + +/** + * Get texture sample at texcoord. + * XXX this is extremely primitive for now. + */ +vector float +sample_texture_nearest(vector float texcoord) +{ + vector float tc = spu_mul(texcoord, spu.tex_size); + vector unsigned int itc = spu_convtu(tc, 0); /* convert to int */ + itc = spu_and(itc, spu.tex_size_mask); /* mask (GL_REPEAT) */ + vector unsigned int ij = spu_and(itc, TILE_SIZE-1); /* intra tile addr */ + uint pos = get_tex_tile(itc); + uint texel = tex_tiles[pos].ui[spu_extract(ij, 1)][spu_extract(ij, 0)]; + return spu_unpack_A8R8G8B8(texel); +} + + +vector float +sample_texture_bilinear(vector float texcoord) +{ + static const vector unsigned int offset10 = {1, 0, 0, 0}; + static const vector unsigned int offset01 = {0, 1, 0, 0}; + + vector float tc = spu_mul(texcoord, spu.tex_size); + tc = spu_add(tc, spu_splats(-0.5f)); /* half texel bias */ + + /* integer texcoords S,T: */ + vector unsigned int itc00 = spu_convtu(tc, 0); /* convert to int */ + vector unsigned int itc01 = spu_add(itc00, offset01); + vector unsigned int itc10 = spu_add(itc00, offset10); + vector unsigned int itc11 = spu_add(itc10, offset01); + + /* mask (GL_REPEAT) */ + itc00 = spu_and(itc00, spu.tex_size_mask); + itc01 = spu_and(itc01, spu.tex_size_mask); + itc10 = spu_and(itc10, spu.tex_size_mask); + itc11 = spu_and(itc11, spu.tex_size_mask); + + /* intra tile addr */ + vector unsigned int ij00 = spu_and(itc00, TILE_SIZE-1); + vector unsigned int ij01 = spu_and(itc01, TILE_SIZE-1); + vector unsigned int ij10 = spu_and(itc10, TILE_SIZE-1); + vector unsigned int ij11 = spu_and(itc11, TILE_SIZE-1); + + /* get tile cache positions */ + uint pos00 = get_tex_tile(itc00); + uint pos01, pos10, pos11; + if ((spu_extract(ij00, 0) < TILE_SIZE-1) && + (spu_extract(ij00, 1) < TILE_SIZE-1)) { + /* all texels are in the same tile */ + pos01 = pos10 = pos11 = pos00; + } + else { + pos01 = get_tex_tile(itc01); + pos10 = get_tex_tile(itc10); + pos11 = get_tex_tile(itc11); + } + + /* get texels from tiles and convert to float[4] */ + vector float texel00 = spu_unpack_A8R8G8B8(tex_tiles[pos00].ui[spu_extract(ij00, 1)][spu_extract(ij00, 0)]); + vector float texel01 = spu_unpack_A8R8G8B8(tex_tiles[pos01].ui[spu_extract(ij01, 1)][spu_extract(ij01, 0)]); + vector float texel10 = spu_unpack_A8R8G8B8(tex_tiles[pos10].ui[spu_extract(ij10, 1)][spu_extract(ij10, 0)]); + vector float texel11 = spu_unpack_A8R8G8B8(tex_tiles[pos11].ui[spu_extract(ij11, 1)][spu_extract(ij11, 0)]); + + /* Compute weighting factors in [0,1] + * Multiply texcoord by 1024, AND with 1023, convert back to float. + */ + vector float tc1024 = spu_mul(tc, spu_splats(1024.0f)); + vector signed int itc1024 = spu_convts(tc1024, 0); + itc1024 = spu_and(itc1024, spu_splats((1 << 10) - 1)); + vector float weight = spu_convtf(itc1024, 10); + + /* smeared frac and 1-frac */ + vector float sfrac = spu_splats(spu_extract(weight, 0)); + vector float tfrac = spu_splats(spu_extract(weight, 1)); + vector float sfrac1 = spu_sub(spu_splats(1.0f), sfrac); + vector float tfrac1 = spu_sub(spu_splats(1.0f), tfrac); + + /* multiply the samples (colors) by the S/T weights */ + texel00 = spu_mul(spu_mul(texel00, sfrac1), tfrac1); + texel10 = spu_mul(spu_mul(texel10, sfrac ), tfrac1); + texel01 = spu_mul(spu_mul(texel01, sfrac1), tfrac ); + texel11 = spu_mul(spu_mul(texel11, sfrac ), tfrac ); + + /* compute sum of weighted samples */ + vector float texel_sum = spu_add(texel00, texel01); + texel_sum = spu_add(texel_sum, texel10); + texel_sum = spu_add(texel_sum, texel11); + + return texel_sum; +} diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h new file mode 100644 index 0000000000..95eb87080f --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_texture.h @@ -0,0 +1,47 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SPU_TEXTURE_H +#define SPU_TEXTURE_H + + +#include "pipe/p_compiler.h" + + +extern void +invalidate_tex_cache(void); + + +extern vector float +sample_texture_nearest(vector float texcoord); + + +extern vector float +sample_texture_bilinear(vector float texcoord); + + +#endif /* SPU_TEXTURE_H */ diff --git a/src/gallium/drivers/cell/spu/spu_tile.c b/src/gallium/drivers/cell/spu/spu_tile.c new file mode 100644 index 0000000000..12dc246328 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_tile.c @@ -0,0 +1,83 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + + +#include "spu_tile.h" +#include "spu_main.h" + + +void +get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf) +{ + const uint offset = ty * spu.fb.width_tiles + tx; + const uint bytesPerTile = TILE_SIZE * TILE_SIZE * (zBuf ? spu.fb.zsize : 4); + const ubyte *src = zBuf ? spu.fb.depth_start : spu.fb.color_start; + + src += offset * bytesPerTile; + + ASSERT(tx < spu.fb.width_tiles); + ASSERT(ty < spu.fb.height_tiles); + ASSERT_ALIGN16(tile); + /* + printf("get_tile: dest: %p src: 0x%x size: %d\n", + tile, (unsigned int) src, bytesPerTile); + */ + mfc_get(tile->ui, /* dest in local memory */ + (unsigned int) src, /* src in main memory */ + bytesPerTile, + tag, + 0, /* tid */ + 0 /* rid */); +} + + +void +put_tile(uint tx, uint ty, const tile_t *tile, int tag, int zBuf) +{ + const uint offset = ty * spu.fb.width_tiles + tx; + const uint bytesPerTile = TILE_SIZE * TILE_SIZE * (zBuf ? spu.fb.zsize : 4); + ubyte *dst = zBuf ? spu.fb.depth_start : spu.fb.color_start; + + dst += offset * bytesPerTile; + + ASSERT(tx < spu.fb.width_tiles); + ASSERT(ty < spu.fb.height_tiles); + ASSERT_ALIGN16(tile); + /* + printf("SPU %u: put_tile: src: %p dst: 0x%x size: %d\n", + spu.init.id, + tile, (unsigned int) dst, bytesPerTile); + */ + mfc_put((void *) tile->ui, /* src in local memory */ + (unsigned int) dst, /* dst in main memory */ + bytesPerTile, + tag, + 0, /* tid */ + 0 /* rid */); +} + diff --git a/src/gallium/drivers/cell/spu/spu_tile.h b/src/gallium/drivers/cell/spu/spu_tile.h new file mode 100644 index 0000000000..e53340a55a --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_tile.h @@ -0,0 +1,73 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SPU_TILE_H +#define SPU_TILE_H + + +#include <libmisc.h> +#include <spu_mfcio.h> +#include "spu_main.h" +#include "pipe/cell/common.h" + + + +void +get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf); + +void +put_tile(uint tx, uint ty, const tile_t *tile, int tag, int zBuf); + + + +static INLINE void +clear_c_tile(tile_t *ctile) +{ + memset32((uint*) ctile->ui, + spu.fb.color_clear_value, + TILE_SIZE * TILE_SIZE); +} + + +static INLINE void +clear_z_tile(tile_t *ztile) +{ + if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) { + memset16((ushort*) ztile->us, + spu.fb.depth_clear_value, + TILE_SIZE * TILE_SIZE); + } + else { + ASSERT(spu.fb.depth_format == PIPE_FORMAT_Z32_UNORM); + memset32((uint*) ztile->ui, + spu.fb.depth_clear_value, + TILE_SIZE * TILE_SIZE); + } +} + + +#endif /* SPU_TILE_H */ diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c new file mode 100644 index 0000000000..be9624cf7d --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_tri.c @@ -0,0 +1,926 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Triangle rendering within a tile. + */ + +#include "pipe/p_compiler.h" +#include "pipe/p_format.h" +#include "pipe/p_util.h" +#include "spu_blend.h" +#include "spu_colorpack.h" +#include "spu_main.h" +#include "spu_texture.h" +#include "spu_tile.h" +#include "spu_tri.h" + +#include "spu_ztest.h" + + +/** Masks are uint[4] vectors with each element being 0 or 0xffffffff */ +typedef vector unsigned int mask_t; + +typedef union +{ + vector float v; + float f[4]; +} float4; + + +/** + * Simplified types taken from other parts of Gallium + */ +struct vertex_header { + vector float data[1]; +}; + + + +/* XXX fix this */ +#undef CEILF +#define CEILF(X) ((float) (int) ((X) + 0.99999)) + + +#define QUAD_TOP_LEFT 0 +#define QUAD_TOP_RIGHT 1 +#define QUAD_BOTTOM_LEFT 2 +#define QUAD_BOTTOM_RIGHT 3 +#define MASK_TOP_LEFT (1 << QUAD_TOP_LEFT) +#define MASK_TOP_RIGHT (1 << QUAD_TOP_RIGHT) +#define MASK_BOTTOM_LEFT (1 << QUAD_BOTTOM_LEFT) +#define MASK_BOTTOM_RIGHT (1 << QUAD_BOTTOM_RIGHT) +#define MASK_ALL 0xf + + +#define DEBUG_VERTS 0 + +/** + * Triangle edge info + */ +struct edge { + float dx; /**< X(v1) - X(v0), used only during setup */ + float dy; /**< Y(v1) - Y(v0), used only during setup */ + float dxdy; /**< dx/dy */ + float sx, sy; /**< first sample point coord */ + int lines; /**< number of lines on this edge */ +}; + + +struct interp_coef +{ + float4 a0; + float4 dadx; + float4 dady; +}; + + +/** + * Triangle setup info (derived from draw_stage). + * Also used for line drawing (taking some liberties). + */ +struct setup_stage { + + /* Vertices are just an array of floats making up each attribute in + * turn. Currently fixed at 4 floats, but should change in time. + * Codegen will help cope with this. + */ + const struct vertex_header *vmax; + const struct vertex_header *vmid; + const struct vertex_header *vmin; + const struct vertex_header *vprovoke; + + struct edge ebot; + struct edge etop; + struct edge emaj; + + float oneoverarea; + + uint tx, ty; + + int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy; + +#if 0 + struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; +#else + struct interp_coef coef[PIPE_MAX_SHADER_INPUTS]; +#endif + +#if 0 + struct quad_header quad; +#endif + + struct { + int left[2]; /**< [0] = row0, [1] = row1 */ + int right[2]; + int y; + unsigned y_flags; + unsigned mask; /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */ + } span; +}; + + + +static struct setup_stage setup; + + + + +#if 0 +/** + * Basically a cast wrapper. + */ +static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) +{ + return (struct setup_stage *)stage; +} +#endif + +#if 0 +/** + * Clip setup.quad against the scissor/surface bounds. + */ +static INLINE void +quad_clip(struct setup_stage *setup) +{ + const struct pipe_scissor_state *cliprect = &setup.softpipe->cliprect; + const int minx = (int) cliprect->minx; + const int maxx = (int) cliprect->maxx; + const int miny = (int) cliprect->miny; + const int maxy = (int) cliprect->maxy; + + if (setup.quad.x0 >= maxx || + setup.quad.y0 >= maxy || + setup.quad.x0 + 1 < minx || + setup.quad.y0 + 1 < miny) { + /* totally clipped */ + setup.quad.mask = 0x0; + return; + } + if (setup.quad.x0 < minx) + setup.quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); + if (setup.quad.y0 < miny) + setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); + if (setup.quad.x0 == maxx - 1) + setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); + if (setup.quad.y0 == maxy - 1) + setup.quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); +} +#endif + +#if 0 +/** + * Emit a quad (pass to next stage) with clipping. + */ +static INLINE void +clip_emit_quad(struct setup_stage *setup) +{ + quad_clip(setup); + if (setup.quad.mask) { + struct softpipe_context *sp = setup.softpipe; + sp->quad.first->run(sp->quad.first, &setup.quad); + } +} +#endif + +/** + * Evaluate attribute coefficients (plane equations) to compute + * attribute values for the four fragments in a quad. + * Eg: four colors will be compute. + */ +static INLINE void +eval_coeff(uint slot, float x, float y, vector float result[4]) +{ + switch (spu.vertex_info.interp_mode[slot]) { + case INTERP_CONSTANT: + result[QUAD_TOP_LEFT] = + result[QUAD_TOP_RIGHT] = + result[QUAD_BOTTOM_LEFT] = + result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0.v; + break; + + case INTERP_LINEAR: + /* fall-through, for now */ + default: + { + register vector float dadx = setup.coef[slot].dadx.v; + register vector float dady = setup.coef[slot].dady.v; + register vector float topLeft + = spu_add(setup.coef[slot].a0.v, + spu_add(spu_mul(spu_splats(x), dadx), + spu_mul(spu_splats(y), dady))); + + result[QUAD_TOP_LEFT] = topLeft; + result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx); + result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady); + result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady); + } + } +} + + +static INLINE vector float +eval_z(float x, float y) +{ + const uint slot = 0; + const float dzdx = setup.coef[slot].dadx.f[2]; + const float dzdy = setup.coef[slot].dady.f[2]; + const float topLeft = setup.coef[slot].a0.f[2] + x * dzdx + y * dzdy; + const vector float topLeftv = spu_splats(topLeft); + const vector float derivs = (vector float) { 0.0, dzdx, dzdy, dzdx + dzdy }; + return spu_add(topLeftv, derivs); +} + + +static INLINE mask_t +do_depth_test(int x, int y, mask_t quadmask) +{ + float4 zvals; + mask_t mask; + + zvals.v = eval_z((float) x, (float) y); + + if (spu.fb.depth_format == PIPE_FORMAT_Z16_UNORM) { + int ix = (x - setup.cliprect_minx) / 4; + int iy = (y - setup.cliprect_miny) / 2; + mask = spu_z16_test_less(zvals.v, &spu.ztile.us8[iy][ix], x>>1, quadmask); + } + else { + int ix = (x - setup.cliprect_minx) / 2; + int iy = (y - setup.cliprect_miny) / 2; + mask = spu_z32_test_less(zvals.v, &spu.ztile.ui4[iy][ix], quadmask); + } + + if (spu_extract(spu_orx(mask), 0)) + spu.cur_ztile_status = TILE_STATUS_DIRTY; + + return mask; +} + + +/** + * Emit a quad (pass to next stage). No clipping is done. + * Note: about 1/5 to 1/7 of the time, mask is zero and this function + * should be skipped. But adding the test for that slows things down + * overall. + */ +static INLINE void +emit_quad( int x, int y, mask_t mask ) +{ +#if 0 + struct softpipe_context *sp = setup.softpipe; + setup.quad.x0 = x; + setup.quad.y0 = y; + setup.quad.mask = mask; + sp->quad.first->run(sp->quad.first, &setup.quad); +#else + + if (spu.depth_stencil.depth.enabled) { + mask = do_depth_test(x, y, mask); + } + + /* If any bits in mask are set... */ + if (spu_extract(spu_orx(mask), 0)) { + const int ix = x - setup.cliprect_minx; + const int iy = y - setup.cliprect_miny; + const vector unsigned char shuffle = spu.color_shuffle; + vector float colors[4]; + + spu.cur_ctile_status = TILE_STATUS_DIRTY; + + if (spu.texture.start) { + /* texture mapping */ + vector float texcoords[4]; + eval_coeff(2, (float) x, (float) y, texcoords); + + if (spu_extract(mask, 0)) + colors[0] = spu.sample_texture(texcoords[0]); + if (spu_extract(mask, 1)) + colors[1] = spu.sample_texture(texcoords[1]); + if (spu_extract(mask, 2)) + colors[2] = spu.sample_texture(texcoords[2]); + if (spu_extract(mask, 3)) + colors[3] = spu.sample_texture(texcoords[3]); + } + else { + /* simple shading */ + eval_coeff(1, (float) x, (float) y, colors); + } + +#if 1 + if (spu.blend.blend_enable) + blend_quad(ix % TILE_SIZE, iy % TILE_SIZE, colors); +#endif + + if (spu_extract(mask, 0)) + spu.ctile.ui[iy][ix] = spu_pack_color_shuffle(colors[0], shuffle); + if (spu_extract(mask, 1)) + spu.ctile.ui[iy][ix+1] = spu_pack_color_shuffle(colors[1], shuffle); + if (spu_extract(mask, 2)) + spu.ctile.ui[iy+1][ix] = spu_pack_color_shuffle(colors[2], shuffle); + if (spu_extract(mask, 3)) + spu.ctile.ui[iy+1][ix+1] = spu_pack_color_shuffle(colors[3], shuffle); + +#if 0 + /* SIMD_Z with swizzled color buffer (someday) */ + vector unsigned int uicolors = *((vector unsigned int *) &colors); + spu.ctile.ui4[iy/2][ix/2] = spu_sel(spu.ctile.ui4[iy/2][ix/2], uicolors, mask); +#endif + } + +#endif +} + + +/** + * Given an X or Y coordinate, return the block/quad coordinate that it + * belongs to. + */ +static INLINE int block( int x ) +{ + return x & ~1; +} + + +/** + * Compute mask which indicates which pixels in the 2x2 quad are actually inside + * the triangle's bounds. + * The mask is a uint4 vector and each element will be 0 or 0xffffffff. + */ +static INLINE mask_t calculate_mask( int x ) +{ + /* This is a little tricky. + * Use & instead of && to avoid branches. + * Use negation to convert true/false to ~0/0 values. + */ + mask_t mask; + mask = spu_insert(-((x >= setup.span.left[0]) & (x < setup.span.right[0])), mask, 0); + mask = spu_insert(-((x+1 >= setup.span.left[0]) & (x+1 < setup.span.right[0])), mask, 1); + mask = spu_insert(-((x >= setup.span.left[1]) & (x < setup.span.right[1])), mask, 2); + mask = spu_insert(-((x+1 >= setup.span.left[1]) & (x+1 < setup.span.right[1])), mask, 3); + return mask; +} + + +/** + * Render a horizontal span of quads + */ +static void flush_spans( void ) +{ + int minleft, maxright; + int x; + + switch (setup.span.y_flags) { + case 0x3: + /* both odd and even lines written (both quad rows) */ + minleft = MIN2(setup.span.left[0], setup.span.left[1]); + maxright = MAX2(setup.span.right[0], setup.span.right[1]); + break; + + case 0x1: + /* only even line written (quad top row) */ + minleft = setup.span.left[0]; + maxright = setup.span.right[0]; + break; + + case 0x2: + /* only odd line written (quad bottom row) */ + minleft = setup.span.left[1]; + maxright = setup.span.right[1]; + break; + + default: + return; + } + + + /* OK, we're very likely to need the tile data now. + * clear or finish waiting if needed. + */ + if (spu.cur_ctile_status == TILE_STATUS_GETTING) { + /* wait for mfc_get() to complete */ + //printf("SPU: %u: waiting for ctile\n", spu.init.id); + wait_on_mask(1 << TAG_READ_TILE_COLOR); + spu.cur_ctile_status = TILE_STATUS_CLEAN; + } + else if (spu.cur_ctile_status == TILE_STATUS_CLEAR) { + //printf("SPU %u: clearing C tile %u, %u\n", spu.init.id, setup.tx, setup.ty); + clear_c_tile(&spu.ctile); + spu.cur_ctile_status = TILE_STATUS_DIRTY; + } + ASSERT(spu.cur_ctile_status != TILE_STATUS_DEFINED); + + if (spu.depth_stencil.depth.enabled) { + if (spu.cur_ztile_status == TILE_STATUS_GETTING) { + /* wait for mfc_get() to complete */ + //printf("SPU: %u: waiting for ztile\n", spu.init.id); + wait_on_mask(1 << TAG_READ_TILE_Z); + spu.cur_ztile_status = TILE_STATUS_CLEAN; + } + else if (spu.cur_ztile_status == TILE_STATUS_CLEAR) { + //printf("SPU %u: clearing Z tile %u, %u\n", spu.init.id, setup.tx, setup.ty); + clear_z_tile(&spu.ztile); + spu.cur_ztile_status = TILE_STATUS_DIRTY; + } + ASSERT(spu.cur_ztile_status != TILE_STATUS_DEFINED); + } + + /* XXX this loop could be moved into the above switch cases and + * calculate_mask() could be simplified a bit... + */ + for (x = block(minleft); x <= block(maxright); x += 2) { +#if 1 + emit_quad( x, setup.span.y, calculate_mask( x ) ); +#endif + } + + setup.span.y = 0; + setup.span.y_flags = 0; + setup.span.right[0] = 0; + setup.span.right[1] = 0; +} + +#if DEBUG_VERTS +static void print_vertex(const struct vertex_header *v) +{ + int i; + fprintf(stderr, "Vertex: (%p)\n", v); + for (i = 0; i < setup.quad.nr_attrs; i++) { + fprintf(stderr, " %d: %f %f %f %f\n", i, + v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]); + } +} +#endif + + +static boolean setup_sort_vertices(const struct vertex_header *v0, + const struct vertex_header *v1, + const struct vertex_header *v2) +{ + +#if DEBUG_VERTS + fprintf(stderr, "Triangle:\n"); + print_vertex(v0); + print_vertex(v1); + print_vertex(v2); +#endif + + setup.vprovoke = v2; + + /* determine bottom to top order of vertices */ + { + float y0 = spu_extract(v0->data[0], 1); + float y1 = spu_extract(v1->data[0], 1); + float y2 = spu_extract(v2->data[0], 1); + if (y0 <= y1) { + if (y1 <= y2) { + /* y0<=y1<=y2 */ + setup.vmin = v0; + setup.vmid = v1; + setup.vmax = v2; + } + else if (y2 <= y0) { + /* y2<=y0<=y1 */ + setup.vmin = v2; + setup.vmid = v0; + setup.vmax = v1; + } + else { + /* y0<=y2<=y1 */ + setup.vmin = v0; + setup.vmid = v2; + setup.vmax = v1; + } + } + else { + if (y0 <= y2) { + /* y1<=y0<=y2 */ + setup.vmin = v1; + setup.vmid = v0; + setup.vmax = v2; + } + else if (y2 <= y1) { + /* y2<=y1<=y0 */ + setup.vmin = v2; + setup.vmid = v1; + setup.vmax = v0; + } + else { + /* y1<=y2<=y0 */ + setup.vmin = v1; + setup.vmid = v2; + setup.vmax = v0; + } + } + } + + /* Check if triangle is completely outside the tile bounds */ + if (spu_extract(setup.vmin->data[0], 1) > setup.cliprect_maxy) + return FALSE; + if (spu_extract(setup.vmax->data[0], 1) < setup.cliprect_miny) + return FALSE; + if (spu_extract(setup.vmin->data[0], 0) < setup.cliprect_minx && + spu_extract(setup.vmid->data[0], 0) < setup.cliprect_minx && + spu_extract(setup.vmax->data[0], 0) < setup.cliprect_minx) + return FALSE; + if (spu_extract(setup.vmin->data[0], 0) > setup.cliprect_maxx && + spu_extract(setup.vmid->data[0], 0) > setup.cliprect_maxx && + spu_extract(setup.vmax->data[0], 0) > setup.cliprect_maxx) + return FALSE; + + setup.ebot.dx = spu_extract(setup.vmid->data[0], 0) - spu_extract(setup.vmin->data[0], 0); + setup.ebot.dy = spu_extract(setup.vmid->data[0], 1) - spu_extract(setup.vmin->data[0], 1); + setup.emaj.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmin->data[0], 0); + setup.emaj.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmin->data[0], 1); + setup.etop.dx = spu_extract(setup.vmax->data[0], 0) - spu_extract(setup.vmid->data[0], 0); + setup.etop.dy = spu_extract(setup.vmax->data[0], 1) - spu_extract(setup.vmid->data[0], 1); + + /* + * Compute triangle's area. Use 1/area to compute partial + * derivatives of attributes later. + * + * The area will be the same as prim->det, but the sign may be + * different depending on how the vertices get sorted above. + * + * To determine whether the primitive is front or back facing we + * use the prim->det value because its sign is correct. + */ + { + const float area = (setup.emaj.dx * setup.ebot.dy - + setup.ebot.dx * setup.emaj.dy); + + setup.oneoverarea = 1.0f / area; + /* + _mesa_printf("%s one-over-area %f area %f det %f\n", + __FUNCTION__, setup.oneoverarea, area, prim->det ); + */ + } + +#if 0 + /* We need to know if this is a front or back-facing triangle for: + * - the GLSL gl_FrontFacing fragment attribute (bool) + * - two-sided stencil test + */ + setup.quad.facing = (prim->det > 0.0) ^ (setup.softpipe->rasterizer->front_winding == PIPE_WINDING_CW); +#endif + + return TRUE; +} + + +/** + * Compute a0 for a constant-valued coefficient (GL_FLAT shading). + * The value value comes from vertex->data[slot]. + * The result will be put into setup.coef[slot].a0. + * \param slot which attribute slot + */ +static INLINE void +const_coeff(uint slot) +{ + setup.coef[slot].dadx.v = (vector float) {0.0, 0.0, 0.0, 0.0}; + setup.coef[slot].dady.v = (vector float) {0.0, 0.0, 0.0, 0.0}; + setup.coef[slot].a0.v = setup.vprovoke->data[slot]; +} + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a triangle. + */ +static INLINE void +tri_linear_coeff(uint slot, uint firstComp, uint lastComp) +{ + uint i; + const float *vmin_d = (float *) &setup.vmin->data[slot]; + const float *vmid_d = (float *) &setup.vmid->data[slot]; + const float *vmax_d = (float *) &setup.vmax->data[slot]; + const float x = spu_extract(setup.vmin->data[0], 0) - 0.5f; + const float y = spu_extract(setup.vmin->data[0], 1) - 0.5f; + + for (i = firstComp; i < lastComp; i++) { + float botda = vmid_d[i] - vmin_d[i]; + float majda = vmax_d[i] - vmin_d[i]; + float a = setup.ebot.dy * majda - botda * setup.emaj.dy; + float b = setup.emaj.dx * botda - majda * setup.ebot.dx; + + ASSERT(slot < PIPE_MAX_SHADER_INPUTS); + + setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; + setup.coef[slot].dady.f[i] = b * setup.oneoverarea; + + /* calculate a0 as the value which would be sampled for the + * fragment at (0,0), taking into account that we want to sample at + * pixel centers, in other words (0.5, 0.5). + * + * this is neat but unfortunately not a good way to do things for + * triangles with very large values of dadx or dady as it will + * result in the subtraction and re-addition from a0 of a very + * large number, which means we'll end up loosing a lot of the + * fractional bits and precision from a0. the way to fix this is + * to define a0 as the sample at a pixel center somewhere near vmin + * instead - i'll switch to this later. + */ + setup.coef[slot].a0.f[i] = (vmin_d[i] - + (setup.coef[slot].dadx.f[i] * x + + setup.coef[slot].dady.f[i] * y)); + } + + /* + _mesa_printf("attr[%d].%c: %f dx:%f dy:%f\n", + slot, "xyzw"[i], + setup.coef[slot].a0[i], + setup.coef[slot].dadx.f[i], + setup.coef[slot].dady.f[i]); + */ +} + + +/** + * As above, but interp setup all four vector components. + */ +static INLINE void +tri_linear_coeff4(uint slot) +{ + const vector float vmin_d = setup.vmin->data[slot]; + const vector float vmid_d = setup.vmid->data[slot]; + const vector float vmax_d = setup.vmax->data[slot]; + const vector float xxxx = spu_splats(spu_extract(setup.vmin->data[0], 0) - 0.5f); + const vector float yyyy = spu_splats(spu_extract(setup.vmin->data[0], 1) - 0.5f); + + vector float botda = vmid_d - vmin_d; + vector float majda = vmax_d - vmin_d; + + vector float a = spu_sub(spu_mul(spu_splats(setup.ebot.dy), majda), + spu_mul(botda, spu_splats(setup.emaj.dy))); + vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda), + spu_mul(majda, spu_splats(setup.ebot.dx))); + + setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneoverarea)); + setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneoverarea)); + + vector float tempx = spu_mul(setup.coef[slot].dadx.v, xxxx); + vector float tempy = spu_mul(setup.coef[slot].dady.v, yyyy); + + setup.coef[slot].a0.v = spu_sub(vmin_d, spu_add(tempx, tempy)); +} + + + +#if 0 +/** + * Compute a0, dadx and dady for a perspective-corrected interpolant, + * for a triangle. + * We basically multiply the vertex value by 1/w before computing + * the plane coefficients (a0, dadx, dady). + * Later, when we compute the value at a particular fragment position we'll + * divide the interpolated value by the interpolated W at that fragment. + */ +static void tri_persp_coeff( unsigned slot, + unsigned i ) +{ + /* premultiply by 1/w: + */ + float mina = setup.vmin->data[slot][i] * setup.vmin->data[0][3]; + float mida = setup.vmid->data[slot][i] * setup.vmid->data[0][3]; + float maxa = setup.vmax->data[slot][i] * setup.vmax->data[0][3]; + + float botda = mida - mina; + float majda = maxa - mina; + float a = setup.ebot.dy * majda - botda * setup.emaj.dy; + float b = setup.emaj.dx * botda - majda * setup.ebot.dx; + + /* + printf("tri persp %d,%d: %f %f %f\n", slot, i, + setup.vmin->data[slot][i], + setup.vmid->data[slot][i], + setup.vmax->data[slot][i] + ); + */ + + assert(slot < PIPE_MAX_SHADER_INPUTS); + assert(i <= 3); + + setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; + setup.coef[slot].dady.f[i] = b * setup.oneoverarea; + setup.coef[slot].a0.f[i] = (mina - + (setup.coef[slot].dadx.f[i] * (setup.vmin->data[0][0] - 0.5f) + + setup.coef[slot].dady.f[i] * (setup.vmin->data[0][1] - 0.5f))); +} +#endif + + +/** + * Compute the setup.coef[] array dadx, dady, a0 values. + * Must be called after setup.vmin,vmid,vmax,vprovoke are initialized. + */ +static void setup_tri_coefficients(void) +{ +#if 1 + uint i; + + for (i = 0; i < spu.vertex_info.num_attribs; i++) { + switch (spu.vertex_info.interp_mode[i]) { + case INTERP_NONE: + break; + case INTERP_POS: + /*tri_linear_coeff(i, 2, 3);*/ + /* XXX interp W if PERSPECTIVE... */ + tri_linear_coeff4(i); + break; + case INTERP_CONSTANT: + const_coeff(i); + break; + case INTERP_LINEAR: + tri_linear_coeff4(i); + break; + case INTERP_PERSPECTIVE: + tri_linear_coeff4(i); /* temporary */ + break; + default: + ASSERT(0); + } + } +#else + ASSERT(spu.vertex_info.interp_mode[0] == INTERP_POS); + ASSERT(spu.vertex_info.interp_mode[1] == INTERP_LINEAR || + spu.vertex_info.interp_mode[1] == INTERP_CONSTANT); + tri_linear_coeff(0, 2, 3); /* slot 0, z */ + tri_linear_coeff(1, 0, 4); /* slot 1, color */ +#endif +} + + +static void setup_tri_edges(void) +{ + float vmin_x = spu_extract(setup.vmin->data[0], 0) + 0.5f; + float vmid_x = spu_extract(setup.vmid->data[0], 0) + 0.5f; + + float vmin_y = spu_extract(setup.vmin->data[0], 1) - 0.5f; + float vmid_y = spu_extract(setup.vmid->data[0], 1) - 0.5f; + float vmax_y = spu_extract(setup.vmax->data[0], 1) - 0.5f; + + setup.emaj.sy = CEILF(vmin_y); + setup.emaj.lines = (int) CEILF(vmax_y - setup.emaj.sy); + setup.emaj.dxdy = setup.emaj.dx / setup.emaj.dy; + setup.emaj.sx = vmin_x + (setup.emaj.sy - vmin_y) * setup.emaj.dxdy; + + setup.etop.sy = CEILF(vmid_y); + setup.etop.lines = (int) CEILF(vmax_y - setup.etop.sy); + setup.etop.dxdy = setup.etop.dx / setup.etop.dy; + setup.etop.sx = vmid_x + (setup.etop.sy - vmid_y) * setup.etop.dxdy; + + setup.ebot.sy = CEILF(vmin_y); + setup.ebot.lines = (int) CEILF(vmid_y - setup.ebot.sy); + setup.ebot.dxdy = setup.ebot.dx / setup.ebot.dy; + setup.ebot.sx = vmin_x + (setup.ebot.sy - vmin_y) * setup.ebot.dxdy; +} + + +/** + * Render the upper or lower half of a triangle. + * Scissoring/cliprect is applied here too. + */ +static void subtriangle( struct edge *eleft, + struct edge *eright, + unsigned lines ) +{ + const int minx = setup.cliprect_minx; + const int maxx = setup.cliprect_maxx; + const int miny = setup.cliprect_miny; + const int maxy = setup.cliprect_maxy; + int y, start_y, finish_y; + int sy = (int)eleft->sy; + + ASSERT((int)eleft->sy == (int) eright->sy); + + /* clip top/bottom */ + start_y = sy; + finish_y = sy + lines; + + if (start_y < miny) + start_y = miny; + + if (finish_y > maxy) + finish_y = maxy; + + start_y -= sy; + finish_y -= sy; + + /* + _mesa_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y); + */ + + for (y = start_y; y < finish_y; y++) { + + /* avoid accumulating adds as floats don't have the precision to + * accurately iterate large triangle edges that way. luckily we + * can just multiply these days. + * + * this is all drowned out by the attribute interpolation anyway. + */ + int left = (int)(eleft->sx + y * eleft->dxdy); + int right = (int)(eright->sx + y * eright->dxdy); + + /* clip left/right */ + if (left < minx) + left = minx; + if (right > maxx) + right = maxx; + + if (left < right) { + int _y = sy + y; + if (block(_y) != setup.span.y) { + flush_spans(); + setup.span.y = block(_y); + } + + setup.span.left[_y&1] = left; + setup.span.right[_y&1] = right; + setup.span.y_flags |= 1<<(_y&1); + } + } + + + /* save the values so that emaj can be restarted: + */ + eleft->sx += lines * eleft->dxdy; + eright->sx += lines * eright->dxdy; + eleft->sy += lines; + eright->sy += lines; +} + + +/** + * Draw triangle into tile at (tx, ty) (tile coords) + * The tile data should have already been fetched. + */ +boolean +tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) +{ + setup.tx = tx; + setup.ty = ty; + + /* set clipping bounds to tile bounds */ + setup.cliprect_minx = tx * TILE_SIZE; + setup.cliprect_miny = ty * TILE_SIZE; + setup.cliprect_maxx = (tx + 1) * TILE_SIZE; + setup.cliprect_maxy = (ty + 1) * TILE_SIZE; + + if (!setup_sort_vertices((struct vertex_header *) v0, + (struct vertex_header *) v1, + (struct vertex_header *) v2)) { + return FALSE; /* totally clipped */ + } + + setup_tri_coefficients(); + setup_tri_edges(); + + setup.span.y = 0; + setup.span.y_flags = 0; + setup.span.right[0] = 0; + setup.span.right[1] = 0; + /* setup.span.z_mode = tri_z_mode( setup.ctx ); */ + + /* init_constant_attribs( setup ); */ + + if (setup.oneoverarea < 0.0) { + /* emaj on left: + */ + subtriangle( &setup.emaj, &setup.ebot, setup.ebot.lines ); + subtriangle( &setup.emaj, &setup.etop, setup.etop.lines ); + } + else { + /* emaj on right: + */ + subtriangle( &setup.ebot, &setup.emaj, setup.ebot.lines ); + subtriangle( &setup.etop, &setup.emaj, setup.etop.lines ); + } + + flush_spans(); + + return TRUE; +} diff --git a/src/gallium/drivers/cell/spu/spu_tri.h b/src/gallium/drivers/cell/spu/spu_tri.h new file mode 100644 index 0000000000..aa694dd7c9 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_tri.h @@ -0,0 +1,37 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef SPU_TRI_H +#define SPU_TRI_H + + +extern boolean +tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty); + + +#endif /* SPU_TRI_H */ diff --git a/src/gallium/drivers/cell/spu/spu_util.c b/src/gallium/drivers/cell/spu/spu_util.c new file mode 100644 index 0000000000..ac373240c1 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_util.c @@ -0,0 +1,165 @@ +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" +//#include "tgsi_build.h" +#include "pipe/tgsi/util/tgsi_util.h" + +unsigned +tgsi_util_get_src_register_swizzle( + const struct tgsi_src_register *reg, + unsigned component ) +{ + switch( component ) { + case 0: + return reg->SwizzleX; + case 1: + return reg->SwizzleY; + case 2: + return reg->SwizzleZ; + case 3: + return reg->SwizzleW; + default: + assert( 0 ); + } + return 0; +} + +unsigned +tgsi_util_get_src_register_extswizzle( + const struct tgsi_src_register_ext_swz *reg, + unsigned component ) +{ + switch( component ) { + case 0: + return reg->ExtSwizzleX; + case 1: + return reg->ExtSwizzleY; + case 2: + return reg->ExtSwizzleZ; + case 3: + return reg->ExtSwizzleW; + default: + assert( 0 ); + } + return 0; +} + +unsigned +tgsi_util_get_full_src_register_extswizzle( + const struct tgsi_full_src_register *reg, + unsigned component ) +{ + unsigned swizzle; + + /* + * First, calculate the extended swizzle for a given channel. This will give + * us either a channel index into the simple swizzle or a constant 1 or 0. + */ + swizzle = tgsi_util_get_src_register_extswizzle( + ®->SrcRegisterExtSwz, + component ); + + assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X); + assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y); + assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z); + assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W); + assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W); + assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W); + + /* + * Second, calculate the simple swizzle for the unswizzled channel index. + * Leave the constants intact, they are not affected by the simple swizzle. + */ + if( swizzle <= TGSI_SWIZZLE_W ) { + swizzle = tgsi_util_get_src_register_swizzle( + ®->SrcRegister, + component ); + } + + return swizzle; +} + +unsigned +tgsi_util_get_src_register_extnegate( + const struct tgsi_src_register_ext_swz *reg, + unsigned component ) +{ + switch( component ) { + case 0: + return reg->NegateX; + case 1: + return reg->NegateY; + case 2: + return reg->NegateZ; + case 3: + return reg->NegateW; + default: + assert( 0 ); + } + return 0; +} + +void +tgsi_util_set_src_register_extnegate( + struct tgsi_src_register_ext_swz *reg, + unsigned negate, + unsigned component ) +{ + switch( component ) { + case 0: + reg->NegateX = negate; + break; + case 1: + reg->NegateY = negate; + break; + case 2: + reg->NegateZ = negate; + break; + case 3: + reg->NegateW = negate; + break; + default: + assert( 0 ); + } +} + +unsigned +tgsi_util_get_full_src_register_sign_mode( + const struct tgsi_full_src_register *reg, + unsigned component ) +{ + unsigned sign_mode; + + if( reg->SrcRegisterExtMod.Absolute ) { + /* Consider only the post-abs negation. */ + + if( reg->SrcRegisterExtMod.Negate ) { + sign_mode = TGSI_UTIL_SIGN_SET; + } + else { + sign_mode = TGSI_UTIL_SIGN_CLEAR; + } + } + else { + /* Accumulate the three negations. */ + + unsigned negate; + + negate = reg->SrcRegister.Negate; + if( tgsi_util_get_src_register_extnegate( ®->SrcRegisterExtSwz, component ) ) { + negate = !negate; + } + if( reg->SrcRegisterExtMod.Negate ) { + negate = !negate; + } + + if( negate ) { + sign_mode = TGSI_UTIL_SIGN_TOGGLE; + } + else { + sign_mode = TGSI_UTIL_SIGN_KEEP; + } + } + + return sign_mode; +} diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c new file mode 100644 index 0000000000..45e3c26c00 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c @@ -0,0 +1,673 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include <spu_mfcio.h> +#include <transpose_matrix4x4.h> + +#include "pipe/p_util.h" +#include "pipe/p_state.h" +#include "pipe/p_shader_tokens.h" +#include "spu_exec.h" +#include "spu_vertex_shader.h" +#include "spu_main.h" + +#define CACHE_NAME attribute +#define CACHED_TYPE qword +#define CACHE_TYPE CACHE_TYPE_RO +#define CACHE_SET_TAGID(set) TAG_VERTEX_BUFFER +#define CACHE_LOG2NNWAY 2 +#define CACHE_LOG2NSETS 6 +#include <cache-api.h> + +/* Yes folks, this is ugly. + */ +#undef CACHE_NWAY +#undef CACHE_NSETS +#define CACHE_NAME attribute +#define CACHE_NWAY 4 +#define CACHE_NSETS (1U << 6) + + +#define DRAW_DBG 0 + +static const qword fetch_shuffle_data[] = { + /* Shuffle used by CVT_64_FLOAT + */ + { + 0x00, 0x01, 0x02, 0x03, 0x10, 0x11, 0x12, 0x13, + 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, + }, + + /* Shuffle used by CVT_8_USCALED and CVT_8_SSCALED + */ + { + 0x00, 0x80, 0x80, 0x80, 0x01, 0x80, 0x80, 0x80, + 0x02, 0x80, 0x80, 0x80, 0x03, 0x80, 0x80, 0x80, + }, + + /* Shuffle used by CVT_16_USCALED and CVT_16_SSCALED + */ + { + 0x00, 0x01, 0x80, 0x80, 0x02, 0x03, 0x80, 0x80, + 0x04, 0x05, 0x80, 0x80, 0x06, 0x07, 0x80, 0x80, + }, + + /* High value shuffle used by trans4x4. + */ + { + 0x00, 0x01, 0x02, 0x03, 0x10, 0x11, 0x12, 0x13, + 0x04, 0x05, 0x06, 0x07, 0x14, 0x15, 0x16, 0x17 + }, + + /* Low value shuffle used by trans4x4. + */ + { + 0x08, 0x09, 0x0A, 0x0B, 0x18, 0x19, 0x1A, 0x1B, + 0x0C, 0x0D, 0x0E, 0x0F, 0x1C, 0x1D, 0x1E, 0x1F + } +}; + + +static INLINE void +trans4x4(qword row0, qword row1, qword row2, qword row3, qword *out, + const qword *shuffle) +{ + qword t1 = si_shufb(row0, row2, shuffle[3]); + qword t2 = si_shufb(row0, row2, shuffle[4]); + qword t3 = si_shufb(row1, row3, shuffle[3]); + qword t4 = si_shufb(row1, row3, shuffle[4]); + + out[0] = si_shufb(t1, t3, shuffle[3]); + out[1] = si_shufb(t1, t3, shuffle[4]); + out[2] = si_shufb(t2, t4, shuffle[3]); + out[3] = si_shufb(t2, t4, shuffle[4]); +} + + +/** + * Fetch between 1 and 32 bytes from an unaligned address + */ +static INLINE void +fetch_unaligned(qword *dst, unsigned ea, unsigned size) +{ + qword tmp[4]; + const int shift = ea & 0x0f; + const unsigned aligned_start_ea = ea & ~0x0f; + const unsigned aligned_end_ea = (ea + size) & ~0x0f; + const unsigned num_entries = ((aligned_end_ea - aligned_start_ea) / 16) + 1; + unsigned i; + + + if (shift == 0) { + /* Data is already aligned. Fetch directly into the destination buffer. + */ + for (i = 0; i < num_entries; i++) { + dst[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16)); + } + } else { + /* Fetch data from the cache to the local buffer. + */ + for (i = 0; i < num_entries; i++) { + tmp[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16)); + } + + + /* Fix the alignment of the data and write to the destination buffer. + */ + for (i = 0; i < ((size + 15) / 16); i++) { + dst[i] = si_or((qword) spu_slqwbyte(tmp[i], shift), + (qword) spu_rlmaskqwbyte(tmp[i + 1], shift - 16)); + } + } +} + + +#define CVT_32_FLOAT(q, s) (*(q)) + +static INLINE qword +CVT_64_FLOAT(const qword *qw, const qword *shuffle) +{ + qword a = si_frds(qw[0]); + qword b = si_frds(si_rotqbyi(qw[0], 8)); + qword c = si_frds(qw[1]); + qword d = si_frds(si_rotqbyi(qw[1], 8)); + + qword ab = si_shufb(a, b, shuffle[0]); + qword cd = si_shufb(c, d, si_rotqbyi(shuffle[0], 8)); + + return si_or(ab, cd); +} + + +static INLINE qword +CVT_8_USCALED(const qword *qw, const qword *shuffle) +{ + return si_cuflt(si_shufb(*qw, *qw, shuffle[1]), 0); +} + + +static INLINE qword +CVT_16_USCALED(const qword *qw, const qword *shuffle) +{ + return si_cuflt(si_shufb(*qw, *qw, shuffle[2]), 0); +} + + +static INLINE qword +CVT_32_USCALED(const qword *qw, const qword *shuffle) +{ + (void) shuffle; + return si_cuflt(*qw, 0); +} + +static INLINE qword +CVT_8_SSCALED(const qword *qw, const qword *shuffle) +{ + return si_csflt(si_shufb(*qw, *qw, shuffle[1]), 0); +} + + +static INLINE qword +CVT_16_SSCALED(const qword *qw, const qword *shuffle) +{ + return si_csflt(si_shufb(*qw, *qw, shuffle[2]), 0); +} + + +static INLINE qword +CVT_32_SSCALED(const qword *qw, const qword *shuffle) +{ + (void) shuffle; + return si_csflt(*qw, 0); +} + + +static INLINE qword +CVT_8_UNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 255.0f); + return si_fm(CVT_8_USCALED(qw, shuffle), scale); +} + + +static INLINE qword +CVT_16_UNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 65535.0f); + return si_fm(CVT_16_USCALED(qw, shuffle), scale); +} + + +static INLINE qword +CVT_32_UNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 4294967295.0f); + return si_fm(CVT_32_USCALED(qw, shuffle), scale); +} + + +static INLINE qword +CVT_8_SNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 127.0f); + return si_fm(CVT_8_SSCALED(qw, shuffle), scale); +} + + +static INLINE qword +CVT_16_SNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 32767.0f); + return si_fm(CVT_16_SSCALED(qw, shuffle), scale); +} + + +static INLINE qword +CVT_32_SNORM(const qword *qw, const qword *shuffle) +{ + const qword scale = (qword) spu_splats(1.0f / 2147483647.0f); + return si_fm(CVT_32_SSCALED(qw, shuffle), scale); +} + +#define SZ_4 si_il(0U) +#define SZ_3 si_fsmbi(0x000f) +#define SZ_2 si_fsmbi(0x00ff) +#define SZ_1 si_fsmbi(0x0fff) + +/** + * Fetch a float[4] vertex attribute from memory, doing format/type + * conversion as needed. + * + * This is probably needed/dupliocated elsewhere, eg format + * conversion, texture sampling etc. + */ +#define FETCH_ATTRIB( NAME, SZ, CVT, N ) \ +static void \ +fetch_##NAME(qword *out, const qword *in, qword defaults, \ + const qword *shuffle) \ +{ \ + qword tmp[4]; \ + \ + tmp[0] = si_selb(CVT(in + (0 * N), shuffle), defaults, SZ); \ + tmp[1] = si_selb(CVT(in + (1 * N), shuffle), defaults, SZ); \ + tmp[2] = si_selb(CVT(in + (2 * N), shuffle), defaults, SZ); \ + tmp[3] = si_selb(CVT(in + (3 * N), shuffle), defaults, SZ); \ + trans4x4(tmp[0], tmp[1], tmp[2], tmp[3], out, shuffle); \ +} + + +FETCH_ATTRIB( R64G64B64A64_FLOAT, SZ_4, CVT_64_FLOAT, 2 ) +FETCH_ATTRIB( R64G64B64_FLOAT, SZ_3, CVT_64_FLOAT, 2 ) +FETCH_ATTRIB( R64G64_FLOAT, SZ_2, CVT_64_FLOAT, 2 ) +FETCH_ATTRIB( R64_FLOAT, SZ_1, CVT_64_FLOAT, 2 ) + +FETCH_ATTRIB( R32G32B32A32_FLOAT, SZ_4, CVT_32_FLOAT, 1 ) +FETCH_ATTRIB( R32G32B32_FLOAT, SZ_3, CVT_32_FLOAT, 1 ) +FETCH_ATTRIB( R32G32_FLOAT, SZ_2, CVT_32_FLOAT, 1 ) +FETCH_ATTRIB( R32_FLOAT, SZ_1, CVT_32_FLOAT, 1 ) + +FETCH_ATTRIB( R32G32B32A32_USCALED, SZ_4, CVT_32_USCALED, 1 ) +FETCH_ATTRIB( R32G32B32_USCALED, SZ_3, CVT_32_USCALED, 1 ) +FETCH_ATTRIB( R32G32_USCALED, SZ_2, CVT_32_USCALED, 1 ) +FETCH_ATTRIB( R32_USCALED, SZ_1, CVT_32_USCALED, 1 ) + +FETCH_ATTRIB( R32G32B32A32_SSCALED, SZ_4, CVT_32_SSCALED, 1 ) +FETCH_ATTRIB( R32G32B32_SSCALED, SZ_3, CVT_32_SSCALED, 1 ) +FETCH_ATTRIB( R32G32_SSCALED, SZ_2, CVT_32_SSCALED, 1 ) +FETCH_ATTRIB( R32_SSCALED, SZ_1, CVT_32_SSCALED, 1 ) + +FETCH_ATTRIB( R32G32B32A32_UNORM, SZ_4, CVT_32_UNORM, 1 ) +FETCH_ATTRIB( R32G32B32_UNORM, SZ_3, CVT_32_UNORM, 1 ) +FETCH_ATTRIB( R32G32_UNORM, SZ_2, CVT_32_UNORM, 1 ) +FETCH_ATTRIB( R32_UNORM, SZ_1, CVT_32_UNORM, 1 ) + +FETCH_ATTRIB( R32G32B32A32_SNORM, SZ_4, CVT_32_SNORM, 1 ) +FETCH_ATTRIB( R32G32B32_SNORM, SZ_3, CVT_32_SNORM, 1 ) +FETCH_ATTRIB( R32G32_SNORM, SZ_2, CVT_32_SNORM, 1 ) +FETCH_ATTRIB( R32_SNORM, SZ_1, CVT_32_SNORM, 1 ) + +FETCH_ATTRIB( R16G16B16A16_USCALED, SZ_4, CVT_16_USCALED, 1 ) +FETCH_ATTRIB( R16G16B16_USCALED, SZ_3, CVT_16_USCALED, 1 ) +FETCH_ATTRIB( R16G16_USCALED, SZ_2, CVT_16_USCALED, 1 ) +FETCH_ATTRIB( R16_USCALED, SZ_1, CVT_16_USCALED, 1 ) + +FETCH_ATTRIB( R16G16B16A16_SSCALED, SZ_4, CVT_16_SSCALED, 1 ) +FETCH_ATTRIB( R16G16B16_SSCALED, SZ_3, CVT_16_SSCALED, 1 ) +FETCH_ATTRIB( R16G16_SSCALED, SZ_2, CVT_16_SSCALED, 1 ) +FETCH_ATTRIB( R16_SSCALED, SZ_1, CVT_16_SSCALED, 1 ) + +FETCH_ATTRIB( R16G16B16A16_UNORM, SZ_4, CVT_16_UNORM, 1 ) +FETCH_ATTRIB( R16G16B16_UNORM, SZ_3, CVT_16_UNORM, 1 ) +FETCH_ATTRIB( R16G16_UNORM, SZ_2, CVT_16_UNORM, 1 ) +FETCH_ATTRIB( R16_UNORM, SZ_1, CVT_16_UNORM, 1 ) + +FETCH_ATTRIB( R16G16B16A16_SNORM, SZ_4, CVT_16_SNORM, 1 ) +FETCH_ATTRIB( R16G16B16_SNORM, SZ_3, CVT_16_SNORM, 1 ) +FETCH_ATTRIB( R16G16_SNORM, SZ_2, CVT_16_SNORM, 1 ) +FETCH_ATTRIB( R16_SNORM, SZ_1, CVT_16_SNORM, 1 ) + +FETCH_ATTRIB( R8G8B8A8_USCALED, SZ_4, CVT_8_USCALED, 1 ) +FETCH_ATTRIB( R8G8B8_USCALED, SZ_3, CVT_8_USCALED, 1 ) +FETCH_ATTRIB( R8G8_USCALED, SZ_2, CVT_8_USCALED, 1 ) +FETCH_ATTRIB( R8_USCALED, SZ_1, CVT_8_USCALED, 1 ) + +FETCH_ATTRIB( R8G8B8A8_SSCALED, SZ_4, CVT_8_SSCALED, 1 ) +FETCH_ATTRIB( R8G8B8_SSCALED, SZ_3, CVT_8_SSCALED, 1 ) +FETCH_ATTRIB( R8G8_SSCALED, SZ_2, CVT_8_SSCALED, 1 ) +FETCH_ATTRIB( R8_SSCALED, SZ_1, CVT_8_SSCALED, 1 ) + +FETCH_ATTRIB( R8G8B8A8_UNORM, SZ_4, CVT_8_UNORM, 1 ) +FETCH_ATTRIB( R8G8B8_UNORM, SZ_3, CVT_8_UNORM, 1 ) +FETCH_ATTRIB( R8G8_UNORM, SZ_2, CVT_8_UNORM, 1 ) +FETCH_ATTRIB( R8_UNORM, SZ_1, CVT_8_UNORM, 1 ) + +FETCH_ATTRIB( R8G8B8A8_SNORM, SZ_4, CVT_8_SNORM, 1 ) +FETCH_ATTRIB( R8G8B8_SNORM, SZ_3, CVT_8_SNORM, 1 ) +FETCH_ATTRIB( R8G8_SNORM, SZ_2, CVT_8_SNORM, 1 ) +FETCH_ATTRIB( R8_SNORM, SZ_1, CVT_8_SNORM, 1 ) + +FETCH_ATTRIB( A8R8G8B8_UNORM, SZ_4, CVT_8_UNORM, 1 ) + + + +static spu_fetch_func get_fetch_func( enum pipe_format format ) +{ + switch (format) { + case PIPE_FORMAT_R64_FLOAT: + return fetch_R64_FLOAT; + case PIPE_FORMAT_R64G64_FLOAT: + return fetch_R64G64_FLOAT; + case PIPE_FORMAT_R64G64B64_FLOAT: + return fetch_R64G64B64_FLOAT; + case PIPE_FORMAT_R64G64B64A64_FLOAT: + return fetch_R64G64B64A64_FLOAT; + + case PIPE_FORMAT_R32_FLOAT: + return fetch_R32_FLOAT; + case PIPE_FORMAT_R32G32_FLOAT: + return fetch_R32G32_FLOAT; + case PIPE_FORMAT_R32G32B32_FLOAT: + return fetch_R32G32B32_FLOAT; + case PIPE_FORMAT_R32G32B32A32_FLOAT: + return fetch_R32G32B32A32_FLOAT; + + case PIPE_FORMAT_R32_UNORM: + return fetch_R32_UNORM; + case PIPE_FORMAT_R32G32_UNORM: + return fetch_R32G32_UNORM; + case PIPE_FORMAT_R32G32B32_UNORM: + return fetch_R32G32B32_UNORM; + case PIPE_FORMAT_R32G32B32A32_UNORM: + return fetch_R32G32B32A32_UNORM; + + case PIPE_FORMAT_R32_USCALED: + return fetch_R32_USCALED; + case PIPE_FORMAT_R32G32_USCALED: + return fetch_R32G32_USCALED; + case PIPE_FORMAT_R32G32B32_USCALED: + return fetch_R32G32B32_USCALED; + case PIPE_FORMAT_R32G32B32A32_USCALED: + return fetch_R32G32B32A32_USCALED; + + case PIPE_FORMAT_R32_SNORM: + return fetch_R32_SNORM; + case PIPE_FORMAT_R32G32_SNORM: + return fetch_R32G32_SNORM; + case PIPE_FORMAT_R32G32B32_SNORM: + return fetch_R32G32B32_SNORM; + case PIPE_FORMAT_R32G32B32A32_SNORM: + return fetch_R32G32B32A32_SNORM; + + case PIPE_FORMAT_R32_SSCALED: + return fetch_R32_SSCALED; + case PIPE_FORMAT_R32G32_SSCALED: + return fetch_R32G32_SSCALED; + case PIPE_FORMAT_R32G32B32_SSCALED: + return fetch_R32G32B32_SSCALED; + case PIPE_FORMAT_R32G32B32A32_SSCALED: + return fetch_R32G32B32A32_SSCALED; + + case PIPE_FORMAT_R16_UNORM: + return fetch_R16_UNORM; + case PIPE_FORMAT_R16G16_UNORM: + return fetch_R16G16_UNORM; + case PIPE_FORMAT_R16G16B16_UNORM: + return fetch_R16G16B16_UNORM; + case PIPE_FORMAT_R16G16B16A16_UNORM: + return fetch_R16G16B16A16_UNORM; + + case PIPE_FORMAT_R16_USCALED: + return fetch_R16_USCALED; + case PIPE_FORMAT_R16G16_USCALED: + return fetch_R16G16_USCALED; + case PIPE_FORMAT_R16G16B16_USCALED: + return fetch_R16G16B16_USCALED; + case PIPE_FORMAT_R16G16B16A16_USCALED: + return fetch_R16G16B16A16_USCALED; + + case PIPE_FORMAT_R16_SNORM: + return fetch_R16_SNORM; + case PIPE_FORMAT_R16G16_SNORM: + return fetch_R16G16_SNORM; + case PIPE_FORMAT_R16G16B16_SNORM: + return fetch_R16G16B16_SNORM; + case PIPE_FORMAT_R16G16B16A16_SNORM: + return fetch_R16G16B16A16_SNORM; + + case PIPE_FORMAT_R16_SSCALED: + return fetch_R16_SSCALED; + case PIPE_FORMAT_R16G16_SSCALED: + return fetch_R16G16_SSCALED; + case PIPE_FORMAT_R16G16B16_SSCALED: + return fetch_R16G16B16_SSCALED; + case PIPE_FORMAT_R16G16B16A16_SSCALED: + return fetch_R16G16B16A16_SSCALED; + + case PIPE_FORMAT_R8_UNORM: + return fetch_R8_UNORM; + case PIPE_FORMAT_R8G8_UNORM: + return fetch_R8G8_UNORM; + case PIPE_FORMAT_R8G8B8_UNORM: + return fetch_R8G8B8_UNORM; + case PIPE_FORMAT_R8G8B8A8_UNORM: + return fetch_R8G8B8A8_UNORM; + + case PIPE_FORMAT_R8_USCALED: + return fetch_R8_USCALED; + case PIPE_FORMAT_R8G8_USCALED: + return fetch_R8G8_USCALED; + case PIPE_FORMAT_R8G8B8_USCALED: + return fetch_R8G8B8_USCALED; + case PIPE_FORMAT_R8G8B8A8_USCALED: + return fetch_R8G8B8A8_USCALED; + + case PIPE_FORMAT_R8_SNORM: + return fetch_R8_SNORM; + case PIPE_FORMAT_R8G8_SNORM: + return fetch_R8G8_SNORM; + case PIPE_FORMAT_R8G8B8_SNORM: + return fetch_R8G8B8_SNORM; + case PIPE_FORMAT_R8G8B8A8_SNORM: + return fetch_R8G8B8A8_SNORM; + + case PIPE_FORMAT_R8_SSCALED: + return fetch_R8_SSCALED; + case PIPE_FORMAT_R8G8_SSCALED: + return fetch_R8G8_SSCALED; + case PIPE_FORMAT_R8G8B8_SSCALED: + return fetch_R8G8B8_SSCALED; + case PIPE_FORMAT_R8G8B8A8_SSCALED: + return fetch_R8G8B8A8_SSCALED; + + case PIPE_FORMAT_A8R8G8B8_UNORM: + return fetch_A8R8G8B8_UNORM; + + case 0: + return NULL; /* not sure why this is needed */ + + default: + assert(0); + return NULL; + } +} + + +static unsigned get_vertex_size( enum pipe_format format ) +{ + switch (format) { + case PIPE_FORMAT_R64_FLOAT: + return 8; + case PIPE_FORMAT_R64G64_FLOAT: + return 2 * 8; + case PIPE_FORMAT_R64G64B64_FLOAT: + return 3 * 8; + case PIPE_FORMAT_R64G64B64A64_FLOAT: + return 4 * 8; + + case PIPE_FORMAT_R32_SSCALED: + case PIPE_FORMAT_R32_SNORM: + case PIPE_FORMAT_R32_USCALED: + case PIPE_FORMAT_R32_UNORM: + case PIPE_FORMAT_R32_FLOAT: + return 4; + case PIPE_FORMAT_R32G32_SSCALED: + case PIPE_FORMAT_R32G32_SNORM: + case PIPE_FORMAT_R32G32_USCALED: + case PIPE_FORMAT_R32G32_UNORM: + case PIPE_FORMAT_R32G32_FLOAT: + return 2 * 4; + case PIPE_FORMAT_R32G32B32_SSCALED: + case PIPE_FORMAT_R32G32B32_SNORM: + case PIPE_FORMAT_R32G32B32_USCALED: + case PIPE_FORMAT_R32G32B32_UNORM: + case PIPE_FORMAT_R32G32B32_FLOAT: + return 3 * 4; + case PIPE_FORMAT_R32G32B32A32_SSCALED: + case PIPE_FORMAT_R32G32B32A32_SNORM: + case PIPE_FORMAT_R32G32B32A32_USCALED: + case PIPE_FORMAT_R32G32B32A32_UNORM: + case PIPE_FORMAT_R32G32B32A32_FLOAT: + return 4 * 4; + + case PIPE_FORMAT_R16_SSCALED: + case PIPE_FORMAT_R16_SNORM: + case PIPE_FORMAT_R16_UNORM: + case PIPE_FORMAT_R16_USCALED: + return 2; + case PIPE_FORMAT_R16G16_SSCALED: + case PIPE_FORMAT_R16G16_SNORM: + case PIPE_FORMAT_R16G16_USCALED: + case PIPE_FORMAT_R16G16_UNORM: + return 2 * 2; + case PIPE_FORMAT_R16G16B16_SSCALED: + case PIPE_FORMAT_R16G16B16_SNORM: + case PIPE_FORMAT_R16G16B16_USCALED: + case PIPE_FORMAT_R16G16B16_UNORM: + return 3 * 2; + case PIPE_FORMAT_R16G16B16A16_SSCALED: + case PIPE_FORMAT_R16G16B16A16_SNORM: + case PIPE_FORMAT_R16G16B16A16_USCALED: + case PIPE_FORMAT_R16G16B16A16_UNORM: + return 4 * 2; + + case PIPE_FORMAT_R8_SSCALED: + case PIPE_FORMAT_R8_SNORM: + case PIPE_FORMAT_R8_USCALED: + case PIPE_FORMAT_R8_UNORM: + return 1; + case PIPE_FORMAT_R8G8_SSCALED: + case PIPE_FORMAT_R8G8_SNORM: + case PIPE_FORMAT_R8G8_USCALED: + case PIPE_FORMAT_R8G8_UNORM: + return 2 * 1; + case PIPE_FORMAT_R8G8B8_SSCALED: + case PIPE_FORMAT_R8G8B8_SNORM: + case PIPE_FORMAT_R8G8B8_USCALED: + case PIPE_FORMAT_R8G8B8_UNORM: + return 3 * 1; + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_R8G8B8A8_SSCALED: + case PIPE_FORMAT_R8G8B8A8_SNORM: + case PIPE_FORMAT_R8G8B8A8_USCALED: + case PIPE_FORMAT_R8G8B8A8_UNORM: + return 4 * 1; + + case 0: + return 0; /* not sure why this is needed */ + + default: + assert(0); + return 0; + } +} + + +/** + * Fetch vertex attributes for 'count' vertices. + */ +static void generic_vertex_fetch(struct spu_vs_context *draw, + struct spu_exec_machine *machine, + const unsigned *elts, + unsigned count) +{ + unsigned nr_attrs = draw->vertex_fetch.nr_attrs; + unsigned attr; + + assert(count <= 4); + +#if DRAW_DBG + printf("SPU: %s count = %u, nr_attrs = %u\n", + __FUNCTION__, count, nr_attrs); +#endif + + /* loop over vertex attributes (vertex shader inputs) + */ + for (attr = 0; attr < nr_attrs; attr++) { + const qword default_values = (qword)(vec_float4){ 0.0, 0.0, 0.0, 1.0 }; + const unsigned pitch = draw->vertex_fetch.pitch[attr]; + const uint64_t src = draw->vertex_fetch.src_ptr[attr]; + const spu_fetch_func fetch = draw->vertex_fetch.fetch[attr]; + unsigned i; + unsigned idx; + const unsigned bytes_per_entry = draw->vertex_fetch.size[attr]; + const unsigned quads_per_entry = (bytes_per_entry + 15) / 16; + qword in[2 * 4]; + + + /* Fetch four attributes for four vertices. + */ + idx = 0; + for (i = 0; i < count; i++) { + const uint64_t addr = src + (elts[i] * pitch); + +#if DRAW_DBG + printf("SPU: fetching = 0x%llx\n", addr); +#endif + + fetch_unaligned(& in[idx], addr, bytes_per_entry); + idx += quads_per_entry; + } + + /* Be nice and zero out any missing vertices. + */ + (void) memset(& in[idx], 0, (8 - idx) * sizeof(qword)); + + + /* Convert all 4 vertices to vectors of float. + */ + (*fetch)(&machine->Inputs[attr].xyzw[0].q, in, default_values, + fetch_shuffle_data); + } +} + + +void spu_update_vertex_fetch( struct spu_vs_context *draw ) +{ + unsigned i; + + + /* Invalidate the vertex cache. + */ + for (i = 0; i < (CACHE_NWAY * CACHE_NSETS); i++) { + CACHELINE_CLEARVALID(i); + } + + + for (i = 0; i < draw->vertex_fetch.nr_attrs; i++) { + draw->vertex_fetch.fetch[i] = + get_fetch_func(draw->vertex_fetch.format[i]); + draw->vertex_fetch.size[i] = + get_vertex_size(draw->vertex_fetch.format[i]); + } + + draw->vertex_fetch.fetch_func = generic_vertex_fetch; +} diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c new file mode 100644 index 0000000000..c1cbbb6d1e --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c @@ -0,0 +1,231 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + * Brian Paul + * Ian Romanick <idr@us.ibm.com> + */ + +#include <spu_mfcio.h> + +#include "pipe/p_util.h" +#include "pipe/p_state.h" +#include "pipe/p_shader_tokens.h" +#include "spu_vertex_shader.h" +#include "spu_exec.h" +#include "pipe/draw/draw_private.h" +#include "pipe/draw/draw_context.h" +#include "pipe/cell/common.h" +#include "spu_main.h" + +static INLINE unsigned +compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) +{ + unsigned mask = 0; + unsigned i; + + /* Do the hardwired planes first: + */ + if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT; + if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT; + if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT; + if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT; + if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT; + if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT; + + /* Followed by any remaining ones: + */ + for (i = 6; i < nr; i++) { + if (dot4(clip, plane[i]) < 0) + mask |= (1<<i); + } + + return mask; +} + + +/** + * Transform vertices with the current vertex program/shader + * Up to four vertices can be shaded at a time. + * \param vbuffer the input vertex data + * \param elts indexes of four input vertices + * \param count number of vertices to shade [1..4] + * \param vOut array of pointers to four output vertices + */ +static void +run_vertex_program(struct spu_vs_context *draw, + unsigned elts[4], unsigned count, + const uint64_t *vOut) +{ + struct spu_exec_machine *machine = &draw->machine; + unsigned int j; + + ALIGN16_DECL(struct spu_exec_vector, inputs, PIPE_ATTRIB_MAX); + ALIGN16_DECL(struct spu_exec_vector, outputs, PIPE_ATTRIB_MAX); + const float *scale = draw->viewport.scale; + const float *trans = draw->viewport.translate; + + assert(count <= 4); + + machine->Processor = TGSI_PROCESSOR_VERTEX; + + ASSERT_ALIGN16(draw->constants); + machine->Consts = (float (*)[4]) draw->constants; + + machine->Inputs = ALIGN16_ASSIGN(inputs); + machine->Outputs = ALIGN16_ASSIGN(outputs); + + spu_vertex_fetch( draw, machine, elts, count ); + + /* run shader */ + spu_exec_machine_run( machine ); + + + /* store machine results */ + for (j = 0; j < count; j++) { + unsigned slot; + float x, y, z, w; + unsigned char buffer[sizeof(struct vertex_header) + + MAX_VERTEX_SIZE] ALIGN16_ATTRIB; + struct vertex_header *const tmpOut = + (struct vertex_header *) buffer; + const unsigned vert_size = ROUNDUP16(sizeof(struct vertex_header) + + (sizeof(float) * 4 + * draw->num_vs_outputs)); + + mfc_get(tmpOut, vOut[j], vert_size, TAG_VERTEX_BUFFER, 0, 0); + wait_on_mask(1 << TAG_VERTEX_BUFFER); + + + /* Handle attr[0] (position) specially: + * + * XXX: Computing the clipmask should be done in the vertex + * program as a set of DP4 instructions appended to the + * user-provided code. + */ + x = tmpOut->clip[0] = machine->Outputs[0].xyzw[0].f[j]; + y = tmpOut->clip[1] = machine->Outputs[0].xyzw[1].f[j]; + z = tmpOut->clip[2] = machine->Outputs[0].xyzw[2].f[j]; + w = tmpOut->clip[3] = machine->Outputs[0].xyzw[3].f[j]; + + tmpOut->clipmask = compute_clipmask(tmpOut->clip, draw->plane, + draw->nr_planes); + tmpOut->edgeflag = 1; + + /* divide by w */ + w = 1.0f / w; + x *= w; + y *= w; + z *= w; + + /* Viewport mapping */ + tmpOut->data[0][0] = x * scale[0] + trans[0]; + tmpOut->data[0][1] = y * scale[1] + trans[1]; + tmpOut->data[0][2] = z * scale[2] + trans[2]; + tmpOut->data[0][3] = w; + + /* Remaining attributes are packed into sequential post-transform + * vertex attrib slots. + */ + for (slot = 1; slot < draw->num_vs_outputs; slot++) { + tmpOut->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + tmpOut->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + tmpOut->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + tmpOut->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } + + mfc_put(tmpOut, vOut[j], vert_size, TAG_VERTEX_BUFFER, 0, 0); + } /* loop over vertices */ +} + + +static void +spu_bind_vertex_shader(struct spu_vs_context *draw, + void *uniforms, + void *planes, + unsigned nr_planes, + unsigned num_outputs + ) +{ + draw->constants = (float (*)[4]) uniforms; + + (void) memcpy(draw->plane, planes, sizeof(float) * 4 * nr_planes); + draw->nr_planes = nr_planes; + draw->num_vs_outputs = num_outputs; + + /* specify the shader to interpret/execute */ + spu_exec_machine_init(&draw->machine, + PIPE_MAX_SAMPLERS, + NULL /*samplers*/, + PIPE_SHADER_VERTEX); +} + + +unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32] + ALIGN16_ATTRIB; + +void +spu_execute_vertex_shader(struct spu_vs_context *draw, + const struct cell_command_vs *vs) +{ + unsigned i; + + const uint64_t immediate_addr = vs->shader.immediates; + const unsigned immediate_size = + ROUNDUP16((sizeof(float) * 4 * vs->shader.num_immediates) + + (immediate_addr & 0x0f)); + + mfc_get(immediates, immediate_addr & ~0x0f, immediate_size, + TAG_VERTEX_BUFFER, 0, 0); + + draw->machine.Instructions = (struct tgsi_full_instruction *) + vs->shader.instructions; + draw->machine.NumInstructions = vs->shader.num_instructions; + + draw->machine.Declarations = (struct tgsi_full_declaration *) + vs->shader.declarations; + draw->machine.NumDeclarations = vs->shader.num_declarations; + + draw->vertex_fetch.nr_attrs = vs->nr_attrs; + + wait_on_mask(1 << TAG_VERTEX_BUFFER); + + (void) memcpy(& draw->machine.Imms, &immediates[immediate_addr & 0x0f], + sizeof(float) * 4 * vs->shader.num_immediates); + + spu_bind_vertex_shader(draw, vs->shader.uniforms, + vs->plane, vs->nr_planes, + vs->shader.num_outputs); + + for (i = 0; i < vs->num_elts; i += 4) { + const unsigned batch_size = MIN2(vs->num_elts - i, 4); + + run_vertex_program(draw, & vs->elts[i], batch_size, &vs->vOut[i]); + } +} diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.h b/src/gallium/drivers/cell/spu/spu_vertex_shader.h new file mode 100644 index 0000000000..b5bf31e67d --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.h @@ -0,0 +1,63 @@ +#ifndef SPU_VERTEX_SHADER_H +#define SPU_VERTEX_SHADER_H + +#include "pipe/p_format.h" +#include "spu_exec.h" + +struct spu_vs_context; + +typedef void (*spu_fetch_func)(qword *out, const qword *in, qword defaults, + const qword *shuffle_data); +typedef void (*spu_full_fetch_func)( struct spu_vs_context *draw, + struct spu_exec_machine *machine, + const unsigned *elts, + unsigned count ); + +struct spu_vs_context { + struct pipe_viewport_state viewport; + + struct { + uint64_t src_ptr[PIPE_ATTRIB_MAX]; + unsigned pitch[PIPE_ATTRIB_MAX]; + unsigned size[PIPE_ATTRIB_MAX]; + enum pipe_format format[PIPE_ATTRIB_MAX]; + unsigned nr_attrs; + boolean dirty; + + spu_fetch_func fetch[PIPE_ATTRIB_MAX]; + spu_full_fetch_func fetch_func; + } vertex_fetch; + + /* Clip derived state: + */ + float plane[12][4]; + unsigned nr_planes; + + struct spu_exec_machine machine; + const float (*constants)[4]; + + unsigned num_vs_outputs; +}; + +extern void spu_update_vertex_fetch(struct spu_vs_context *draw); + +static INLINE void spu_vertex_fetch(struct spu_vs_context *draw, + struct spu_exec_machine *machine, + const unsigned *elts, + unsigned count) +{ + if (draw->vertex_fetch.dirty) { + spu_update_vertex_fetch(draw); + draw->vertex_fetch.dirty = 0; + } + + (*draw->vertex_fetch.fetch_func)(draw, machine, elts, count); +} + +struct cell_command_vs; + +extern void +spu_execute_vertex_shader(struct spu_vs_context *draw, + const struct cell_command_vs *vs); + +#endif /* SPU_VERTEX_SHADER_H */ diff --git a/src/gallium/drivers/cell/spu/spu_ztest.h b/src/gallium/drivers/cell/spu/spu_ztest.h new file mode 100644 index 0000000000..ce8ad00339 --- /dev/null +++ b/src/gallium/drivers/cell/spu/spu_ztest.h @@ -0,0 +1,135 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * Zbuffer/depth test code. + */ + + +#ifndef SPU_ZTEST_H +#define SPU_ZTEST_H + + +#ifdef __SPU__ +#include <spu_intrinsics.h> +#endif + + + +/** + * Perform Z testing for a 16-bit/value Z buffer. + * + * \param zvals vector of four fragment zvalues as floats + * \param zbuf ptr to vector of ushort[8] zbuffer values. Note that this + * contains the Z values for 2 quads, 8 pixels. + * \param x x coordinate of quad (only lsbit is significant) + * \param inMask indicates which fragments in the quad are alive + * \return new mask indicating which fragments are alive after ztest + */ +static INLINE vector unsigned int +spu_z16_test_less(vector float zvals, vector unsigned short *zbuf, + uint x, vector unsigned int inMask) +{ +#define ZERO 0x80 + vector unsigned int zvals_ui4, zbuf_ui4, mask; + + /* convert floats to uints in [0, 65535] */ + zvals_ui4 = spu_convtu(zvals, 32); /* convert to [0, 2^32] */ + zvals_ui4 = spu_rlmask(zvals_ui4, -16); /* right shift 16 */ + + /* XXX this conditional could be removed with a bit of work */ + if (x & 1) { + /* convert zbuffer values from ushorts to uints */ + /* gather lower four ushorts */ + zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf, + (vector unsigned int) *zbuf, + ((vector unsigned char) { + ZERO, ZERO, 8, 9, ZERO, ZERO, 10, 11, + ZERO, ZERO, 12, 13, ZERO, ZERO, 14, 15})); + /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */ + mask = spu_cmpgt(zbuf_ui4, zvals_ui4); + /* mask &= inMask */ + mask = spu_and(mask, inMask); + /* zbuf = mask ? zval : zbuf */ + zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask); + /* convert zbuffer values from uints back to ushorts, preserve lower 4 */ + *zbuf = (vector unsigned short) + spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf, + ((vector unsigned char) { + 16, 17, 18, 19, 20, 21, 22, 23, + 2, 3, 6, 7, 10, 11, 14, 15})); + } + else { + /* convert zbuffer values from ushorts to uints */ + /* gather upper four ushorts */ + zbuf_ui4 = spu_shuffle((vector unsigned int) *zbuf, + (vector unsigned int) *zbuf, + ((vector unsigned char) { + ZERO, ZERO, 0, 1, ZERO, ZERO, 2, 3, + ZERO, ZERO, 4, 5, ZERO, ZERO, 6, 7})); + /* mask = (zbuf_ui4 < zvals_ui4) ? ~0 : 0 */ + mask = spu_cmpgt(zbuf_ui4, zvals_ui4); + /* mask &= inMask */ + mask = spu_and(mask, inMask); + /* zbuf = mask ? zval : zbuf */ + zbuf_ui4 = spu_sel(zbuf_ui4, zvals_ui4, mask); + /* convert zbuffer values from uints back to ushorts, preserve upper 4 */ + *zbuf = (vector unsigned short) + spu_shuffle(zbuf_ui4, (vector unsigned int) *zbuf, + ((vector unsigned char) { + 2, 3, 6, 7, 10, 11, 14, 15, + 24, 25, 26, 27, 28, 29, 30, 31})); + } + return mask; +#undef ZERO +} + + +/** + * As above, but Zbuffer values as 32-bit uints + */ +static INLINE vector unsigned int +spu_z32_test_less(vector float zvals, vector unsigned int *zbuf_ptr, + vector unsigned int inMask) +{ + vector unsigned int zvals_ui4, mask, zbuf = *zbuf_ptr; + + /* convert floats to uints in [0, 0xffffffff] */ + zvals_ui4 = spu_convtu(zvals, 32); + /* mask = (zbuf < zvals_ui4) ? ~0 : 0 */ + mask = spu_cmpgt(zbuf, zvals_ui4); + /* mask &= inMask */ + mask = spu_and(mask, inMask); + /* zbuf = mask ? zval : zbuf */ + *zbuf_ptr = spu_sel(zbuf, zvals_ui4, mask); + + return mask; +} + + +#endif /* SPU_ZTEST_H */ diff --git a/src/gallium/drivers/failover/Makefile b/src/gallium/drivers/failover/Makefile new file mode 100644 index 0000000000..72d0895c74 --- /dev/null +++ b/src/gallium/drivers/failover/Makefile @@ -0,0 +1,21 @@ + +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = failover + +DRIVER_SOURCES = \ + fo_state.c \ + fo_state_emit.c \ + fo_context.c + +C_SOURCES = \ + $(COMMON_SOURCES) \ + $(DRIVER_SOURCES) + +ASM_SOURCES = + +include ../Makefile.template + +symlinks: + diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c new file mode 100644 index 0000000000..7ce4a7df17 --- /dev/null +++ b/src/gallium/drivers/failover/fo_context.c @@ -0,0 +1,155 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_defines.h" +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" +#include "pipe/p_context.h" + +#include "fo_context.h" +#include "fo_winsys.h" + + + +static void failover_destroy( struct pipe_context *pipe ) +{ + struct failover_context *failover = failover_context( pipe ); + + free( failover ); +} + + + +static boolean failover_draw_elements( struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned prim, unsigned start, unsigned count) +{ + struct failover_context *failover = failover_context( pipe ); + + /* If there has been any statechange since last time, try hardware + * rendering again: + */ + if (failover->dirty) { + failover->mode = FO_HW; + } + + /* Try hardware: + */ + if (failover->mode == FO_HW) { + if (!failover->hw->draw_elements( failover->hw, + indexBuffer, + indexSize, + prim, + start, + count )) { + + failover->hw->flush( failover->hw, ~0 ); + failover->mode = FO_SW; + } + } + + /* Possibly try software: + */ + if (failover->mode == FO_SW) { + + if (failover->dirty) + failover_state_emit( failover ); + + failover->sw->draw_elements( failover->sw, + indexBuffer, + indexSize, + prim, + start, + count ); + + /* Be ready to switch back to hardware rendering without an + * intervening flush. Unlikely to be much performance impact to + * this: + */ + failover->sw->flush( failover->sw, ~0 ); + } + + return TRUE; +} + + +static boolean failover_draw_arrays( struct pipe_context *pipe, + unsigned prim, unsigned start, unsigned count) +{ + return failover_draw_elements(pipe, NULL, 0, prim, start, count); +} + + + +struct pipe_context *failover_create( struct pipe_context *hw, + struct pipe_context *sw ) +{ + struct failover_context *failover = CALLOC_STRUCT(failover_context); + if (failover == NULL) + return NULL; + + failover->hw = hw; + failover->sw = sw; + failover->pipe.winsys = hw->winsys; + failover->pipe.destroy = failover_destroy; + failover->pipe.is_format_supported = hw->is_format_supported; + failover->pipe.get_name = hw->get_name; + failover->pipe.get_vendor = hw->get_vendor; + failover->pipe.get_param = hw->get_param; + failover->pipe.get_paramf = hw->get_paramf; + + failover->pipe.draw_arrays = failover_draw_arrays; + failover->pipe.draw_elements = failover_draw_elements; + failover->pipe.clear = hw->clear; + + /* No software occlusion fallback (or other optional functionality) + * at this point - if the hardware doesn't support it, don't + * advertise it to the application. + */ + failover->pipe.begin_query = hw->begin_query; + failover->pipe.end_query = hw->end_query; + + failover_init_state_functions( failover ); + +#if 0 + failover->pipe.surface_alloc = hw->surface_alloc; +#endif + failover->pipe.get_tex_surface = hw->get_tex_surface; + + failover->pipe.surface_copy = hw->surface_copy; + failover->pipe.surface_fill = hw->surface_fill; + failover->pipe.texture_create = hw->texture_create; + failover->pipe.texture_release = hw->texture_release; + failover->pipe.flush = hw->flush; + + failover->dirty = 0; + + return &failover->pipe; +} + diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h new file mode 100644 index 0000000000..1dc87291c9 --- /dev/null +++ b/src/gallium/drivers/failover/fo_context.h @@ -0,0 +1,114 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef FO_CONTEXT_H +#define FO_CONTEXT_H + +#include "pipe/p_state.h" +#include "pipe/p_context.h" + + + +#define FO_NEW_VIEWPORT 0x1 +#define FO_NEW_RASTERIZER 0x2 +#define FO_NEW_FRAGMENT_SHADER 0x4 +#define FO_NEW_BLEND 0x8 +#define FO_NEW_CLIP 0x10 +#define FO_NEW_SCISSOR 0x20 +#define FO_NEW_STIPPLE 0x40 +#define FO_NEW_FRAMEBUFFER 0x80 +#define FO_NEW_ALPHA_TEST 0x100 +#define FO_NEW_DEPTH_STENCIL 0x200 +#define FO_NEW_SAMPLER 0x400 +#define FO_NEW_TEXTURE 0x800 +#define FO_NEW_VERTEX 0x2000 +#define FO_NEW_VERTEX_SHADER 0x4000 +#define FO_NEW_BLEND_COLOR 0x8000 +#define FO_NEW_CLEAR_COLOR 0x10000 +#define FO_NEW_VERTEX_BUFFER 0x20000 +#define FO_NEW_VERTEX_ELEMENT 0x40000 + + + +#define FO_HW 0 +#define FO_SW 1 + +struct fo_state { + void *sw_state; + void *hw_state; +}; +struct failover_context { + struct pipe_context pipe; /**< base class */ + + + /* The most recent drawing state as set by the driver: + */ + const struct fo_state *blend; + const struct fo_state *sampler[PIPE_MAX_SAMPLERS]; + const struct fo_state *depth_stencil; + const struct fo_state *rasterizer; + const struct fo_state *fragment_shader; + const struct fo_state *vertex_shader; + + struct pipe_blend_color blend_color; + struct pipe_clip_state clip; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct pipe_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX]; + struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX]; + + unsigned dirty; + unsigned dirty_sampler; + unsigned dirty_texture; + unsigned dirty_vertex_buffer; + unsigned dirty_vertex_element; + + + unsigned mode; + struct pipe_context *hw; + struct pipe_context *sw; +}; + + + +void failover_init_state_functions( struct failover_context *failover ); +void failover_state_emit( struct failover_context *failover ); + +static INLINE struct failover_context * +failover_context( struct pipe_context *pipe ) +{ + return (struct failover_context *)pipe; +} + + +#endif /* FO_CONTEXT_H */ diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c new file mode 100644 index 0000000000..0fc5568da1 --- /dev/null +++ b/src/gallium/drivers/failover/fo_state.c @@ -0,0 +1,457 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "fo_context.h" + + +/* This looks like a lot of work at the moment - we're keeping a + * duplicate copy of the state up-to-date. + * + * This can change in two ways: + * - With constant state objects we would only need to save a pointer, + * not the whole object. + * - By adding a callback in the state tracker to re-emit state. The + * state tracker knows the current state already and can re-emit it + * without additional complexity. + * + * This works as a proof-of-concept, but a final version will have + * lower overheads. + */ + + + +static void * +failover_create_blend_state( struct pipe_context *pipe, + const struct pipe_blend_state *blend ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_blend_state(failover->sw, blend); + state->hw_state = failover->hw->create_blend_state(failover->hw, blend); + + return state; +} + +static void +failover_bind_blend_state( struct pipe_context *pipe, + void *blend ) +{ + struct failover_context *failover = failover_context(pipe); + struct fo_state *state = (struct fo_state *)blend; + failover->blend = state; + failover->dirty |= FO_NEW_BLEND; + failover->sw->bind_blend_state( failover->sw, state->sw_state ); + failover->hw->bind_blend_state( failover->hw, state->hw_state ); +} + +static void +failover_delete_blend_state( struct pipe_context *pipe, + void *blend ) +{ + struct fo_state *state = (struct fo_state*)blend; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_blend_state(failover->sw, state->sw_state); + failover->hw->delete_blend_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + +static void +failover_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct failover_context *failover = failover_context(pipe); + + failover->blend_color = *blend_color; + failover->dirty |= FO_NEW_BLEND_COLOR; + failover->sw->set_blend_color( failover->sw, blend_color ); + failover->hw->set_blend_color( failover->hw, blend_color ); +} + +static void +failover_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct failover_context *failover = failover_context(pipe); + + failover->clip = *clip; + failover->dirty |= FO_NEW_CLIP; + failover->sw->set_clip_state( failover->sw, clip ); + failover->hw->set_clip_state( failover->hw, clip ); +} + + +static void * +failover_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *templ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_depth_stencil_alpha_state(failover->sw, templ); + state->hw_state = failover->hw->create_depth_stencil_alpha_state(failover->hw, templ); + + return state; +} + +static void +failover_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct failover_context *failover = failover_context(pipe); + struct fo_state *state = (struct fo_state *)depth_stencil; + failover->depth_stencil = state; + failover->dirty |= FO_NEW_DEPTH_STENCIL; + failover->sw->bind_depth_stencil_alpha_state(failover->sw, state->sw_state); + failover->hw->bind_depth_stencil_alpha_state(failover->hw, state->hw_state); +} + +static void +failover_delete_depth_stencil_state(struct pipe_context *pipe, + void *ds) +{ + struct fo_state *state = (struct fo_state*)ds; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_depth_stencil_alpha_state(failover->sw, state->sw_state); + failover->hw->delete_depth_stencil_alpha_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + +static void +failover_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *framebuffer) +{ + struct failover_context *failover = failover_context(pipe); + + failover->framebuffer = *framebuffer; + failover->dirty |= FO_NEW_FRAMEBUFFER; + failover->sw->set_framebuffer_state( failover->sw, framebuffer ); + failover->hw->set_framebuffer_state( failover->hw, framebuffer ); +} + + +static void * +failover_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_fs_state(failover->sw, templ); + state->hw_state = failover->hw->create_fs_state(failover->hw, templ); + + return state; +} + +static void +failover_bind_fs_state(struct pipe_context *pipe, void *fs) +{ + struct failover_context *failover = failover_context(pipe); + struct fo_state *state = (struct fo_state*)fs; + failover->fragment_shader = state; + failover->dirty |= FO_NEW_FRAGMENT_SHADER; + failover->sw->bind_fs_state(failover->sw, state->sw_state); + failover->hw->bind_fs_state(failover->hw, state->hw_state); +} + +static void +failover_delete_fs_state(struct pipe_context *pipe, + void *fs) +{ + struct fo_state *state = (struct fo_state*)fs; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_fs_state(failover->sw, state->sw_state); + failover->hw->delete_fs_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + +static void * +failover_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_vs_state(failover->sw, templ); + state->hw_state = failover->hw->create_vs_state(failover->hw, templ); + + return state; +} + +static void +failover_bind_vs_state(struct pipe_context *pipe, + void *vs) +{ + struct failover_context *failover = failover_context(pipe); + + struct fo_state *state = (struct fo_state*)vs; + failover->vertex_shader = state; + failover->dirty |= FO_NEW_VERTEX_SHADER; + failover->sw->bind_vs_state(failover->sw, state->sw_state); + failover->hw->bind_vs_state(failover->hw, state->hw_state); +} + +static void +failover_delete_vs_state(struct pipe_context *pipe, + void *vs) +{ + struct fo_state *state = (struct fo_state*)vs; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_vs_state(failover->sw, state->sw_state); + failover->hw->delete_vs_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + +static void +failover_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ + struct failover_context *failover = failover_context(pipe); + + failover->poly_stipple = *stipple; + failover->dirty |= FO_NEW_STIPPLE; + failover->sw->set_polygon_stipple( failover->sw, stipple ); + failover->hw->set_polygon_stipple( failover->hw, stipple ); +} + + +static void * +failover_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *templ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_rasterizer_state(failover->sw, templ); + state->hw_state = failover->hw->create_rasterizer_state(failover->hw, templ); + + return state; +} + +static void +failover_bind_rasterizer_state(struct pipe_context *pipe, + void *raster) +{ + struct failover_context *failover = failover_context(pipe); + + struct fo_state *state = (struct fo_state*)raster; + failover->rasterizer = state; + failover->dirty |= FO_NEW_RASTERIZER; + failover->sw->bind_rasterizer_state(failover->sw, state->sw_state); + failover->hw->bind_rasterizer_state(failover->hw, state->hw_state); +} + +static void +failover_delete_rasterizer_state(struct pipe_context *pipe, + void *raster) +{ + struct fo_state *state = (struct fo_state*)raster; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_rasterizer_state(failover->sw, state->sw_state); + failover->hw->delete_rasterizer_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + + +static void +failover_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct failover_context *failover = failover_context(pipe); + + failover->scissor = *scissor; + failover->dirty |= FO_NEW_SCISSOR; + failover->sw->set_scissor_state( failover->sw, scissor ); + failover->hw->set_scissor_state( failover->hw, scissor ); +} + + +static void * +failover_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *templ) +{ + struct fo_state *state = malloc(sizeof(struct fo_state)); + struct failover_context *failover = failover_context(pipe); + + state->sw_state = failover->sw->create_sampler_state(failover->sw, templ); + state->hw_state = failover->hw->create_sampler_state(failover->hw, templ); + + return state; +} + +static void +failover_bind_sampler_state(struct pipe_context *pipe, + unsigned unit, void *sampler) +{ + struct failover_context *failover = failover_context(pipe); + struct fo_state *state = (struct fo_state*)sampler; + failover->sampler[unit] = state; + failover->dirty |= FO_NEW_SAMPLER; + failover->dirty_sampler |= (1<<unit); + failover->sw->bind_sampler_state(failover->sw, unit, + state->sw_state); + failover->hw->bind_sampler_state(failover->hw, unit, + state->hw_state); +} + +static void +failover_delete_sampler_state(struct pipe_context *pipe, void *sampler) +{ + struct fo_state *state = (struct fo_state*)sampler; + struct failover_context *failover = failover_context(pipe); + + failover->sw->delete_sampler_state(failover->sw, state->sw_state); + failover->hw->delete_sampler_state(failover->hw, state->hw_state); + state->sw_state = 0; + state->hw_state = 0; + free(state); +} + + +static void +failover_set_sampler_texture(struct pipe_context *pipe, + unsigned unit, + struct pipe_texture *texture) +{ + struct failover_context *failover = failover_context(pipe); + + failover->texture[unit] = texture; + failover->dirty |= FO_NEW_TEXTURE; + failover->dirty_texture |= (1<<unit); + failover->sw->set_sampler_texture( failover->sw, unit, texture ); + failover->hw->set_sampler_texture( failover->hw, unit, texture ); +} + + +static void +failover_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct failover_context *failover = failover_context(pipe); + + failover->viewport = *viewport; + failover->dirty |= FO_NEW_VIEWPORT; + failover->sw->set_viewport_state( failover->sw, viewport ); + failover->hw->set_viewport_state( failover->hw, viewport ); +} + + +static void +failover_set_vertex_buffer(struct pipe_context *pipe, + unsigned unit, + const struct pipe_vertex_buffer *vertex_buffer) +{ + struct failover_context *failover = failover_context(pipe); + + failover->vertex_buffer[unit] = *vertex_buffer; + failover->dirty |= FO_NEW_VERTEX_BUFFER; + failover->dirty_vertex_buffer |= (1<<unit); + failover->sw->set_vertex_buffer( failover->sw, unit, vertex_buffer ); + failover->hw->set_vertex_buffer( failover->hw, unit, vertex_buffer ); +} + + +static void +failover_set_vertex_element(struct pipe_context *pipe, + unsigned unit, + const struct pipe_vertex_element *vertex_element) +{ + struct failover_context *failover = failover_context(pipe); + + failover->vertex_element[unit] = *vertex_element; + failover->dirty |= FO_NEW_VERTEX_ELEMENT; + failover->dirty_vertex_element |= (1<<unit); + failover->sw->set_vertex_element( failover->sw, unit, vertex_element ); + failover->hw->set_vertex_element( failover->hw, unit, vertex_element ); +} + +void +failover_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct failover_context *failover = failover_context(pipe); + + assert(shader < PIPE_SHADER_TYPES); + assert(index == 0); + + failover->sw->set_constant_buffer(failover->sw, shader, index, buf); + failover->hw->set_constant_buffer(failover->hw, shader, index, buf); +} + + +void +failover_init_state_functions( struct failover_context *failover ) +{ + failover->pipe.create_blend_state = failover_create_blend_state; + failover->pipe.bind_blend_state = failover_bind_blend_state; + failover->pipe.delete_blend_state = failover_delete_blend_state; + failover->pipe.create_sampler_state = failover_create_sampler_state; + failover->pipe.bind_sampler_state = failover_bind_sampler_state; + failover->pipe.delete_sampler_state = failover_delete_sampler_state; + failover->pipe.create_depth_stencil_alpha_state = failover_create_depth_stencil_state; + failover->pipe.bind_depth_stencil_alpha_state = failover_bind_depth_stencil_state; + failover->pipe.delete_depth_stencil_alpha_state = failover_delete_depth_stencil_state; + failover->pipe.create_rasterizer_state = failover_create_rasterizer_state; + failover->pipe.bind_rasterizer_state = failover_bind_rasterizer_state; + failover->pipe.delete_rasterizer_state = failover_delete_rasterizer_state; + failover->pipe.create_fs_state = failover_create_fs_state; + failover->pipe.bind_fs_state = failover_bind_fs_state; + failover->pipe.delete_fs_state = failover_delete_fs_state; + failover->pipe.create_vs_state = failover_create_vs_state; + failover->pipe.bind_vs_state = failover_bind_vs_state; + failover->pipe.delete_vs_state = failover_delete_vs_state; + + failover->pipe.set_blend_color = failover_set_blend_color; + failover->pipe.set_clip_state = failover_set_clip_state; + failover->pipe.set_framebuffer_state = failover_set_framebuffer_state; + failover->pipe.set_polygon_stipple = failover_set_polygon_stipple; + failover->pipe.set_scissor_state = failover_set_scissor_state; + failover->pipe.set_sampler_texture = failover_set_sampler_texture; + failover->pipe.set_viewport_state = failover_set_viewport_state; + failover->pipe.set_vertex_buffer = failover_set_vertex_buffer; + failover->pipe.set_vertex_element = failover_set_vertex_element; + failover->pipe.set_constant_buffer = failover_set_constant_buffer; +} diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c new file mode 100644 index 0000000000..c663dd4947 --- /dev/null +++ b/src/gallium/drivers/failover/fo_state_emit.c @@ -0,0 +1,137 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "fo_context.h" + +/* This looks like a lot of work at the moment - we're keeping a + * duplicate copy of the state up-to-date. + * + * This can change in two ways: + * - With constant state objects we would only need to save a pointer, + * not the whole object. + * - By adding a callback in the state tracker to re-emit state. The + * state tracker knows the current state already and can re-emit it + * without additional complexity. + * + * This works as a proof-of-concept, but a final version will have + * lower overheads. + */ + + +/* Bring the software pipe uptodate with current state. + * + * With constant state objects we would probably just send all state + * to both rasterizers all the time??? + */ +void +failover_state_emit( struct failover_context *failover ) +{ + unsigned i; + + if (failover->dirty & FO_NEW_BLEND) + failover->sw->bind_blend_state( failover->sw, + failover->blend->sw_state ); + + if (failover->dirty & FO_NEW_BLEND_COLOR) + failover->sw->set_blend_color( failover->sw, &failover->blend_color ); + + if (failover->dirty & FO_NEW_CLIP) + failover->sw->set_clip_state( failover->sw, &failover->clip ); + + if (failover->dirty & FO_NEW_DEPTH_STENCIL) + failover->sw->bind_depth_stencil_alpha_state( failover->sw, + failover->depth_stencil->sw_state ); + + if (failover->dirty & FO_NEW_FRAMEBUFFER) + failover->sw->set_framebuffer_state( failover->sw, &failover->framebuffer ); + + if (failover->dirty & FO_NEW_FRAGMENT_SHADER) + failover->sw->bind_fs_state( failover->sw, + failover->fragment_shader->sw_state ); + + if (failover->dirty & FO_NEW_VERTEX_SHADER) + failover->sw->bind_vs_state( failover->sw, + failover->vertex_shader->sw_state ); + + if (failover->dirty & FO_NEW_STIPPLE) + failover->sw->set_polygon_stipple( failover->sw, &failover->poly_stipple ); + + if (failover->dirty & FO_NEW_RASTERIZER) + failover->sw->bind_rasterizer_state( failover->sw, + failover->rasterizer->sw_state ); + + if (failover->dirty & FO_NEW_SCISSOR) + failover->sw->set_scissor_state( failover->sw, &failover->scissor ); + + if (failover->dirty & FO_NEW_VIEWPORT) + failover->sw->set_viewport_state( failover->sw, &failover->viewport ); + + if (failover->dirty & FO_NEW_SAMPLER) { + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + if (failover->dirty_sampler & (1<<i)) { + failover->sw->bind_sampler_state( failover->sw, i, + failover->sampler[i]->sw_state ); + } + } + } + + if (failover->dirty & FO_NEW_TEXTURE) { + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + if (failover->dirty_texture & (1<<i)) { + failover->sw->set_sampler_texture( failover->sw, i, + failover->texture[i] ); + } + } + } + + if (failover->dirty & FO_NEW_VERTEX_BUFFER) { + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (failover->dirty_vertex_buffer & (1<<i)) { + failover->sw->set_vertex_buffer( failover->sw, i, + &failover->vertex_buffer[i] ); + } + } + } + + if (failover->dirty & FO_NEW_VERTEX_ELEMENT) { + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (failover->dirty_vertex_element & (1<<i)) { + failover->sw->set_vertex_element( failover->sw, i, + &failover->vertex_element[i] ); + } + } + } + + failover->dirty = 0; + failover->dirty_vertex_element = 0; + failover->dirty_vertex_buffer = 0; + failover->dirty_texture = 0; + failover->dirty_sampler = 0; +} diff --git a/src/gallium/drivers/failover/fo_winsys.h b/src/gallium/drivers/failover/fo_winsys.h new file mode 100644 index 0000000000..a8ce997a1f --- /dev/null +++ b/src/gallium/drivers/failover/fo_winsys.h @@ -0,0 +1,45 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef FO_WINSYS_H +#define FO_WINSYS_H + + +/* This is the interface that failover requires any window system + * hosting it to implement. This is the only include file in failover + * which is public. + */ + + +struct pipe_context; + + +struct pipe_context *failover_create( struct pipe_context *hw, + struct pipe_context *sw ); + + +#endif /* FO_WINSYS_H */ diff --git a/src/gallium/drivers/i915simple/Makefile b/src/gallium/drivers/i915simple/Makefile new file mode 100644 index 0000000000..2f91de3afc --- /dev/null +++ b/src/gallium/drivers/i915simple/Makefile @@ -0,0 +1,38 @@ + +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = i915simple + +DRIVER_SOURCES = \ + i915_blit.c \ + i915_clear.c \ + i915_flush.c \ + i915_context.c \ + i915_context.c \ + i915_debug.c \ + i915_debug_fp.c \ + i915_state.c \ + i915_state_immediate.c \ + i915_state_dynamic.c \ + i915_state_derived.c \ + i915_state_emit.c \ + i915_state_sampler.c \ + i915_strings.c \ + i915_prim_emit.c \ + i915_prim_vbuf.c \ + i915_texture.c \ + i915_fpc_emit.c \ + i915_fpc_translate.c \ + i915_surface.c + +C_SOURCES = \ + $(COMMON_SOURCES) \ + $(DRIVER_SOURCES) + +ASM_SOURCES = + +include ../Makefile.template + +symlinks: + diff --git a/src/gallium/drivers/i915simple/SConscript b/src/gallium/drivers/i915simple/SConscript new file mode 100644 index 0000000000..f5fb96b995 --- /dev/null +++ b/src/gallium/drivers/i915simple/SConscript @@ -0,0 +1,29 @@ +Import('*') + +env = env.Clone() + +i915simple = env.ConvenienceLibrary( + target = 'i915simple', + source = [ + 'i915_blit.c', + 'i915_clear.c', + 'i915_context.c', + 'i915_debug.c', + 'i915_debug_fp.c', + 'i915_flush.c', + 'i915_fpc_emit.c', + 'i915_fpc_translate.c', + 'i915_prim_emit.c', + 'i915_prim_vbuf.c', + 'i915_state.c', + 'i915_state_derived.c', + 'i915_state_dynamic.c', + 'i915_state_emit.c', + 'i915_state_immediate.c', + 'i915_state_sampler.c', + 'i915_strings.c', + 'i915_surface.c', + 'i915_texture.c', + ]) + +Export('i915simple') diff --git a/src/gallium/drivers/i915simple/i915_batch.h b/src/gallium/drivers/i915simple/i915_batch.h new file mode 100644 index 0000000000..fb88cd6db0 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_batch.h @@ -0,0 +1,54 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_BATCH_H +#define I915_BATCH_H + +#include "i915_winsys.h" +#include "i915_debug.h" + +#define BATCH_LOCALS + +#define BEGIN_BATCH( dwords, relocs ) \ + (i915->batch_start = i915->winsys->batch_start( i915->winsys, dwords, relocs )) + +#define OUT_BATCH( dword ) \ + i915->winsys->batch_dword( i915->winsys, dword ) + +#define OUT_RELOC( buf, flags, delta ) \ + i915->winsys->batch_reloc( i915->winsys, buf, flags, delta ) + +#define ADVANCE_BATCH() + +#define FLUSH_BATCH() do { \ + if (0) i915_dump_batchbuffer( i915 ); \ + i915->winsys->batch_flush( i915->winsys ); \ + i915->batch_start = NULL; \ + i915->hardware_dirty = ~0; \ +} while (0) + +#endif diff --git a/src/gallium/drivers/i915simple/i915_blit.c b/src/gallium/drivers/i915simple/i915_blit.c new file mode 100644 index 0000000000..db4671ff55 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_blit.c @@ -0,0 +1,162 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "i915_context.h" +#include "i915_winsys.h" +#include "i915_blit.h" +#include "i915_reg.h" +#include "i915_batch.h" + +#define FILE_DEBUG_FLAG DEBUG_BLIT + +void +i915_fill_blit(struct i915_context *i915, + unsigned cpp, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + short x, short y, + short w, short h, + unsigned color) +{ + unsigned BR13, CMD; + BATCH_LOCALS; + + dst_pitch *= (short) cpp; + + switch (cpp) { + case 1: + case 2: + case 3: + BR13 = dst_pitch | (0xF0 << 16) | (1 << 24); + CMD = XY_COLOR_BLT_CMD; + break; + case 4: + BR13 = dst_pitch | (0xF0 << 16) | (1 << 24) | (1 << 25); + CMD = (XY_COLOR_BLT_CMD | XY_COLOR_BLT_WRITE_ALPHA | + XY_COLOR_BLT_WRITE_RGB); + break; + default: + return; + } + +// DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", +// __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h); + + + if (!BEGIN_BATCH(6, 1)) { + FLUSH_BATCH(); + assert(BEGIN_BATCH(6, 1)); + } + OUT_BATCH(CMD); + OUT_BATCH(BR13); + OUT_BATCH((y << 16) | x); + OUT_BATCH(((y + h) << 16) | (x + w)); + OUT_RELOC( dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset); + OUT_BATCH(color); + ADVANCE_BATCH(); +} + + +void +i915_copy_blit( struct i915_context *i915, + unsigned do_flip, + unsigned cpp, + short src_pitch, + struct pipe_buffer *src_buffer, + unsigned src_offset, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + short src_x, short src_y, + short dst_x, short dst_y, + short w, short h ) +{ + unsigned CMD, BR13; + int dst_y2 = dst_y + h; + int dst_x2 = dst_x + w; + BATCH_LOCALS; + + + I915_DBG(i915, + "%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n", + __FUNCTION__, + src_buffer, src_pitch, src_offset, src_x, src_y, + dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h); + + src_pitch *= (short) cpp; + dst_pitch *= (short) cpp; + + switch (cpp) { + case 1: + case 2: + case 3: + BR13 = (((int) dst_pitch) & 0xffff) | + (0xCC << 16) | (1 << 24); + CMD = XY_SRC_COPY_BLT_CMD; + break; + case 4: + BR13 = + (((int) dst_pitch) & 0xffff) | + (0xCC << 16) | (1 << 24) | (1 << 25); + CMD = + (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA | + XY_SRC_COPY_BLT_WRITE_RGB); + break; + default: + return; + } + + if (dst_y2 < dst_y || + dst_x2 < dst_x) { + return; + } + + /* Hardware can handle negative pitches but loses the ability to do + * proper overlapping blits in that case. We don't really have a + * need for either at this stage. + */ + assert (dst_pitch > 0 && src_pitch > 0); + + + if (!BEGIN_BATCH(8, 2)) { + FLUSH_BATCH(); + assert(BEGIN_BATCH(8, 2)); + } + OUT_BATCH(CMD); + OUT_BATCH(BR13); + OUT_BATCH((dst_y << 16) | dst_x); + OUT_BATCH((dst_y2 << 16) | dst_x2); + OUT_RELOC(dst_buffer, I915_BUFFER_ACCESS_WRITE, dst_offset); + OUT_BATCH((src_y << 16) | src_x); + OUT_BATCH(((int) src_pitch & 0xffff)); + OUT_RELOC(src_buffer, I915_BUFFER_ACCESS_READ, src_offset); + ADVANCE_BATCH(); +} + + diff --git a/src/gallium/drivers/i915simple/i915_blit.h b/src/gallium/drivers/i915simple/i915_blit.h new file mode 100644 index 0000000000..6e5b44e124 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_blit.h @@ -0,0 +1,55 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_BLIT_H +#define I915_BLIT_H + +#include "i915_context.h" + +extern void i915_copy_blit(struct i915_context *i915, + unsigned do_flip, + unsigned cpp, + short src_pitch, + struct pipe_buffer *src_buffer, + unsigned src_offset, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + short srcx, short srcy, + short dstx, short dsty, + short w, short h ); + +extern void i915_fill_blit(struct i915_context *i915, + unsigned cpp, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + short x, short y, + short w, short h, unsigned color); + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_clear.c b/src/gallium/drivers/i915simple/i915_clear.c new file mode 100644 index 0000000000..cde69daacc --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_clear.c @@ -0,0 +1,47 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + */ + + +#include "pipe/p_defines.h" +#include "i915_context.h" +#include "i915_state.h" + + +/** + * Clear the given surface to the specified value. + * No masking, no scissor (clear entire buffer). + */ +void +i915_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue) +{ + pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, clearValue); +} diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c new file mode 100644 index 0000000000..497623a700 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_context.c @@ -0,0 +1,320 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "i915_context.h" +#include "i915_winsys.h" +#include "i915_state.h" +#include "i915_batch.h" +#include "i915_texture.h" +#include "i915_reg.h" + +#include "pipe/draw/draw_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" + + +/** + * Query format support for creating a texture, drawing surface, etc. + * \param format the format to test + * \param type one of PIPE_TEXTURE, PIPE_SURFACE + */ +static boolean +i915_is_format_supported( struct pipe_context *pipe, + enum pipe_format format, uint type ) +{ + static const enum pipe_format tex_supported[] = { + PIPE_FORMAT_R8G8B8A8_UNORM, + PIPE_FORMAT_A8R8G8B8_UNORM, + PIPE_FORMAT_R5G6B5_UNORM, + PIPE_FORMAT_U_L8, + PIPE_FORMAT_U_A8, + PIPE_FORMAT_U_I8, + PIPE_FORMAT_U_A8_L8, + PIPE_FORMAT_YCBCR, + PIPE_FORMAT_YCBCR_REV, + PIPE_FORMAT_S8Z24_UNORM, + PIPE_FORMAT_NONE /* list terminator */ + }; + static const enum pipe_format surface_supported[] = { + PIPE_FORMAT_A8R8G8B8_UNORM, + PIPE_FORMAT_R5G6B5_UNORM, + PIPE_FORMAT_S8Z24_UNORM, + /*PIPE_FORMAT_R16G16B16A16_SNORM,*/ + PIPE_FORMAT_NONE /* list terminator */ + }; + const enum pipe_format *list; + uint i; + + switch (type) { + case PIPE_TEXTURE: + list = tex_supported; + break; + case PIPE_SURFACE: + list = surface_supported; + break; + default: + assert(0); + } + + for (i = 0; list[i] != PIPE_FORMAT_NONE; i++) { + if (list[i] == format) + return TRUE; + } + + return FALSE; +} + + +static int +i915_get_param(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return 8; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 0; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 0; + case PIPE_CAP_POINT_SPRITE: + return 0; + case PIPE_CAP_MAX_RENDER_TARGETS: + return 1; + case PIPE_CAP_OCCLUSION_QUERY: + return 0; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 11; /* max 1024x1024 */ + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 8; /* max 128x128x128 */ + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 11; /* max 1024x1024 */ + default: + return 0; + } +} + + +static float +i915_get_paramf(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 7.5; + + case PIPE_CAP_MAX_POINT_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 255.0; + + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 4.0; + + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; + + default: + return 0; + } +} + + +static void i915_destroy( struct pipe_context *pipe ) +{ + struct i915_context *i915 = i915_context( pipe ); + + draw_destroy( i915->draw ); + + FREE( i915 ); +} + + + + +static boolean +i915_draw_elements( struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned prim, unsigned start, unsigned count) +{ + struct i915_context *i915 = i915_context( pipe ); + struct draw_context *draw = i915->draw; + unsigned i; + + if (i915->dirty) + i915_update_derived( i915 ); + + /* + * Map vertex buffers + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (i915->vertex_buffer[i].buffer) { + void *buf + = pipe->winsys->buffer_map(pipe->winsys, + i915->vertex_buffer[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_vertex_buffer(draw, i, buf); + } + } + /* Map index buffer, if present */ + if (indexBuffer) { + void *mapped_indexes + = pipe->winsys->buffer_map(pipe->winsys, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); + } + else { + /* no index/element buffer */ + draw_set_mapped_element_buffer(draw, 0, NULL); + } + + + draw_set_mapped_constant_buffer(draw, + i915->current.constants[PIPE_SHADER_VERTEX]); + + /* draw! */ + draw_arrays(i915->draw, prim, start, count); + + /* + * unmap vertex/index buffers + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (i915->vertex_buffer[i].buffer) { + pipe->winsys->buffer_unmap(pipe->winsys, i915->vertex_buffer[i].buffer); + draw_set_mapped_vertex_buffer(draw, i, NULL); + } + } + if (indexBuffer) { + pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + draw_set_mapped_element_buffer(draw, 0, NULL); + } + + return TRUE; +} + + +static boolean i915_draw_arrays( struct pipe_context *pipe, + unsigned prim, unsigned start, unsigned count) +{ + return i915_draw_elements(pipe, NULL, 0, prim, start, count); +} + + + +struct pipe_context *i915_create( struct pipe_winsys *pipe_winsys, + struct i915_winsys *i915_winsys, + unsigned pci_id ) +{ + struct i915_context *i915; + unsigned is_i945 = 0; + + switch (pci_id) { + case PCI_CHIP_I915_G: + case PCI_CHIP_I915_GM: + break; + + case PCI_CHIP_I945_G: + case PCI_CHIP_I945_GM: + case PCI_CHIP_I945_GME: + case PCI_CHIP_G33_G: + case PCI_CHIP_Q33_G: + case PCI_CHIP_Q35_G: + is_i945 = 1; + break; + + default: + pipe_winsys->printf(pipe_winsys, + "%s: unknown pci id 0x%x, cannot create context\n", + __FUNCTION__, pci_id); + return NULL; + } + + i915 = CALLOC_STRUCT(i915_context); + if (i915 == NULL) + return NULL; + + i915->winsys = i915_winsys; + i915->pipe.winsys = pipe_winsys; + + i915->pipe.destroy = i915_destroy; + i915->pipe.is_format_supported = i915_is_format_supported; + i915->pipe.get_param = i915_get_param; + i915->pipe.get_paramf = i915_get_paramf; + + i915->pipe.clear = i915_clear; + + + i915->pipe.draw_arrays = i915_draw_arrays; + i915->pipe.draw_elements = i915_draw_elements; + + /* + * Create drawing context and plug our rendering stage into it. + */ + i915->draw = draw_create(); + assert(i915->draw); + if (GETENV("I915_VBUF")) { + draw_set_rasterize_stage(i915->draw, i915_draw_vbuf_stage(i915)); + } + else { + draw_set_rasterize_stage(i915->draw, i915_draw_render_stage(i915)); + } + + i915_init_surface_functions(i915); + i915_init_state_functions(i915); + i915_init_flush_functions(i915); + i915_init_string_functions(i915); + + i915->pci_id = pci_id; + i915->flags.is_i945 = is_i945; + + i915->pipe.texture_create = i915_texture_create; + i915->pipe.texture_release = i915_texture_release; + + i915->dirty = ~0; + i915->hardware_dirty = ~0; + + /* Batch stream debugging is a bit hacked up at the moment: + */ + i915->batch_start = NULL; + + /* + * XXX we could plug GL selection/feedback into the drawing pipeline + * by specifying a different setup/render stage. + */ + + return &i915->pipe; +} + diff --git a/src/gallium/drivers/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h new file mode 100644 index 0000000000..b4ea63c3e7 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_context.h @@ -0,0 +1,304 @@ + /************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_CONTEXT_H +#define I915_CONTEXT_H + + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" + +#include "pipe/draw/draw_vertex.h" + + +#define I915_TEX_UNITS 8 + +#define I915_DYNAMIC_MODES4 0 +#define I915_DYNAMIC_DEPTHSCALE_0 1 /* just the header */ +#define I915_DYNAMIC_DEPTHSCALE_1 2 +#define I915_DYNAMIC_IAB 3 +#define I915_DYNAMIC_BC_0 4 /* just the header */ +#define I915_DYNAMIC_BC_1 5 +#define I915_DYNAMIC_BFO_0 6 +#define I915_DYNAMIC_BFO_1 7 +#define I915_DYNAMIC_STP_0 8 +#define I915_DYNAMIC_STP_1 9 +#define I915_DYNAMIC_SC_ENA_0 10 +#define I915_DYNAMIC_SC_RECT_0 11 +#define I915_DYNAMIC_SC_RECT_1 12 +#define I915_DYNAMIC_SC_RECT_2 13 +#define I915_MAX_DYNAMIC 14 + + +#define I915_IMMEDIATE_S0 0 +#define I915_IMMEDIATE_S1 1 +#define I915_IMMEDIATE_S2 2 +#define I915_IMMEDIATE_S3 3 +#define I915_IMMEDIATE_S4 4 +#define I915_IMMEDIATE_S5 5 +#define I915_IMMEDIATE_S6 6 +#define I915_IMMEDIATE_S7 7 +#define I915_MAX_IMMEDIATE 8 + +/* These must mach the order of LI0_STATE_* bits, as they will be used + * to generate hardware packets: + */ +#define I915_CACHE_STATIC 0 +#define I915_CACHE_DYNAMIC 1 /* handled specially */ +#define I915_CACHE_SAMPLER 2 +#define I915_CACHE_MAP 3 +#define I915_CACHE_PROGRAM 4 +#define I915_CACHE_CONSTANTS 5 +#define I915_MAX_CACHE 6 + +#define I915_MAX_CONSTANT 32 + + + +struct i915_cache_context; + +/* Use to calculate differences between state emitted to hardware and + * current driver-calculated state. + */ +struct i915_state +{ + unsigned immediate[I915_MAX_IMMEDIATE]; + unsigned dynamic[I915_MAX_DYNAMIC]; + + float constants[PIPE_SHADER_TYPES][I915_MAX_CONSTANT][4]; + /** number of constants passed in through a constant buffer */ + uint num_user_constants[PIPE_SHADER_TYPES]; + /** user constants, plus extra constants from shader translation */ + uint num_constants[PIPE_SHADER_TYPES]; + + uint *program; + uint program_len; + + /* texture sampler state */ + unsigned sampler[I915_TEX_UNITS][3]; + unsigned sampler_enable_flags; + unsigned sampler_enable_nr; + + /* texture image buffers */ + unsigned texbuffer[I915_TEX_UNITS][2]; + + /** Describes the current hardware vertex layout */ + struct vertex_info vertex_info; + + unsigned id; /* track lost context events */ +}; + +struct i915_blend_state { + unsigned iab; + unsigned modes4; + unsigned LIS5; + unsigned LIS6; +}; + +struct i915_depth_stencil_state { + unsigned stencil_modes4; + unsigned bfo[2]; + unsigned stencil_LIS5; + unsigned depth_LIS6; +}; + +struct i915_rasterizer_state { + int light_twoside : 1; + unsigned st; + enum interp_mode color_interp; + + unsigned LIS4; + unsigned LIS7; + unsigned sc[1]; + + const struct pipe_rasterizer_state *templ; + + union { float f; unsigned u; } ds[2]; +}; + +struct i915_sampler_state { + unsigned state[3]; + const struct pipe_sampler_state *templ; +}; + + +struct i915_texture { + struct pipe_texture base; + + /* Derived from the above: + */ + unsigned pitch; + unsigned depth_pitch; /* per-image on i945? */ + unsigned total_height; + + unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS]; + + /* Explicitly store the offset of each image for each cube face or + * depth value. Pretty much have to accept that hardware formats + * are going to be so diverse that there is no unified way to + * compute the offsets of depth/cube images within a mipmap level, + * so have to store them as a lookup table: + */ + unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */ + + /* Includes image offset tables: + */ + unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; +}; + +struct i915_context +{ + struct pipe_context pipe; + struct i915_winsys *winsys; + struct draw_context *draw; + + /* The most recent drawing state as set by the driver: + */ + const struct i915_blend_state *blend; + const struct i915_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + const struct i915_depth_stencil_state *depth_stencil; + const struct i915_rasterizer_state *rasterizer; + const struct pipe_shader_state *fs; + + struct pipe_blend_color blend_color; + struct pipe_clip_state clip; + struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct i915_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX]; + + unsigned dirty; + + unsigned *batch_start; + + /** Vertex buffer */ + struct pipe_buffer *vbo; + + struct i915_state current; + unsigned hardware_dirty; + + unsigned debug; + unsigned pci_id; + + struct { + unsigned is_i945:1; + } flags; +}; + +/* A flag for each state_tracker state object: + */ +#define I915_NEW_VIEWPORT 0x1 +#define I915_NEW_RASTERIZER 0x2 +#define I915_NEW_FS 0x4 +#define I915_NEW_BLEND 0x8 +#define I915_NEW_CLIP 0x10 +#define I915_NEW_SCISSOR 0x20 +#define I915_NEW_STIPPLE 0x40 +#define I915_NEW_FRAMEBUFFER 0x80 +#define I915_NEW_ALPHA_TEST 0x100 +#define I915_NEW_DEPTH_STENCIL 0x200 +#define I915_NEW_SAMPLER 0x400 +#define I915_NEW_TEXTURE 0x800 +#define I915_NEW_CONSTANTS 0x1000 +#define I915_NEW_VBO 0x2000 + + +/* Driver's internally generated state flags: + */ +#define I915_NEW_VERTEX_FORMAT 0x10000 + + +/* Dirty flags for hardware emit + */ +#define I915_HW_STATIC (1<<I915_CACHE_STATIC) +#define I915_HW_DYNAMIC (1<<I915_CACHE_DYNAMIC) +#define I915_HW_SAMPLER (1<<I915_CACHE_SAMPLER) +#define I915_HW_MAP (1<<I915_CACHE_MAP) +#define I915_HW_PROGRAM (1<<I915_CACHE_PROGRAM) +#define I915_HW_CONSTANTS (1<<I915_CACHE_CONSTANTS) +#define I915_HW_IMMEDIATE (1<<(I915_MAX_CACHE+0)) +#define I915_HW_INVARIENT (1<<(I915_MAX_CACHE+1)) + + +/*********************************************************************** + * i915_prim_emit.c: + */ +struct draw_stage *i915_draw_render_stage( struct i915_context *i915 ); + + +/*********************************************************************** + * i915_prim_vbuf.c: + */ +struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 ); + + +/*********************************************************************** + * i915_state_emit.c: + */ +void i915_emit_hardware_state(struct i915_context *i915 ); + + + +/*********************************************************************** + * i915_clear.c: + */ +void i915_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue); + + +/*********************************************************************** + * i915_surface.c: + */ +void i915_init_surface_functions( struct i915_context *i915 ); + +void i915_init_state_functions( struct i915_context *i915 ); +void i915_init_flush_functions( struct i915_context *i915 ); +void i915_init_string_functions( struct i915_context *i915 ); + + + +/*********************************************************************** + * Inline conversion functions. These are better-typed than the + * macros used previously: + */ +static INLINE struct i915_context * +i915_context( struct pipe_context *pipe ) +{ + return (struct i915_context *)pipe; +} + + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_debug.c b/src/gallium/drivers/i915simple/i915_debug.c new file mode 100644 index 0000000000..94db44e1aa --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_debug.c @@ -0,0 +1,901 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +//#include "imports.h" + +#include "i915_reg.h" +#include "i915_context.h" +#include "i915_winsys.h" +#include "i915_debug.h" +#include "pipe/p_winsys.h" + + +static void +PRINTF( + struct debug_stream *stream, + const char *fmt, + ... ) +{ + va_list args; + char buffer[256]; + + va_start( args, fmt ); + vsprintf( buffer, fmt, args ); + stream->winsys->printf( stream->winsys, buffer ); + va_end( args ); +} + + +static boolean debug( struct debug_stream *stream, const char *name, unsigned len ) +{ + unsigned i; + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + + if (len == 0) { + PRINTF(stream, "Error - zero length packet (0x%08x)\n", stream->ptr[0]); + assert(0); + return FALSE; + } + + if (stream->print_addresses) + PRINTF(stream, "%08x: ", stream->offset); + + + PRINTF(stream, "%s (%d dwords):\n", name, len); + for (i = 0; i < len; i++) + PRINTF(stream, "\t0x%08x\n", ptr[i]); + PRINTF(stream, "\n"); + + stream->offset += len * sizeof(unsigned); + + return TRUE; +} + + +static const char *get_prim_name( unsigned val ) +{ + switch (val & PRIM3D_MASK) { + case PRIM3D_TRILIST: return "TRILIST"; break; + case PRIM3D_TRISTRIP: return "TRISTRIP"; break; + case PRIM3D_TRISTRIP_RVRSE: return "TRISTRIP_RVRSE"; break; + case PRIM3D_TRIFAN: return "TRIFAN"; break; + case PRIM3D_POLY: return "POLY"; break; + case PRIM3D_LINELIST: return "LINELIST"; break; + case PRIM3D_LINESTRIP: return "LINESTRIP"; break; + case PRIM3D_RECTLIST: return "RECTLIST"; break; + case PRIM3D_POINTLIST: return "POINTLIST"; break; + case PRIM3D_DIB: return "DIB"; break; + case PRIM3D_CLEAR_RECT: return "CLEAR_RECT"; break; + case PRIM3D_ZONE_INIT: return "ZONE_INIT"; break; + default: return "????"; break; + } +} + +static boolean debug_prim( struct debug_stream *stream, const char *name, + boolean dump_floats, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + const char *prim = get_prim_name( ptr[0] ); + unsigned i; + + + + PRINTF(stream, "%s %s (%d dwords):\n", name, prim, len); + PRINTF(stream, "\t0x%08x\n", ptr[0]); + for (i = 1; i < len; i++) { + if (dump_floats) + PRINTF(stream, "\t0x%08x // %f\n", ptr[i], *(float *)&ptr[i]); + else + PRINTF(stream, "\t0x%08x\n", ptr[i]); + } + + + PRINTF(stream, "\n"); + + stream->offset += len * sizeof(unsigned); + + return TRUE; +} + + + + +static boolean debug_program( struct debug_stream *stream, const char *name, unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + + if (len == 0) { + PRINTF(stream, "Error - zero length packet (0x%08x)\n", stream->ptr[0]); + assert(0); + return FALSE; + } + + if (stream->print_addresses) + PRINTF(stream, "%08x: ", stream->offset); + + PRINTF(stream, "%s (%d dwords):\n", name, len); + i915_disassemble_program( stream, ptr, len ); + + stream->offset += len * sizeof(unsigned); + return TRUE; +} + + +static boolean debug_chain( struct debug_stream *stream, const char *name, unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned old_offset = stream->offset + len * sizeof(unsigned); + unsigned i; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + for (i = 0; i < len; i++) + PRINTF(stream, "\t0x%08x\n", ptr[i]); + + stream->offset = ptr[1] & ~0x3; + + if (stream->offset < old_offset) + PRINTF(stream, "\n... skipping backwards from 0x%x --> 0x%x ...\n\n", + old_offset, stream->offset ); + else + PRINTF(stream, "\n... skipping from 0x%x --> 0x%x ...\n\n", + old_offset, stream->offset ); + + + return TRUE; +} + + +static boolean debug_variable_length_prim( struct debug_stream *stream ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + const char *prim = get_prim_name( ptr[0] ); + unsigned i, len; + + ushort *idx = (ushort *)(ptr+1); + for (i = 0; idx[i] != 0xffff; i++) + ; + + len = 1+(i+2)/2; + + PRINTF(stream, "3DPRIM, %s variable length %d indicies (%d dwords):\n", prim, i, len); + for (i = 0; i < len; i++) + PRINTF(stream, "\t0x%08x\n", ptr[i]); + PRINTF(stream, "\n"); + + stream->offset += len * sizeof(unsigned); + return TRUE; +} + + +static void +BITS( + struct debug_stream *stream, + unsigned dw, + unsigned hi, + unsigned lo, + const char *fmt, + ... ) +{ + va_list args; + char buffer[256]; + unsigned himask = ~0UL >> (31 - (hi)); + + PRINTF(stream, "\t\t "); + + va_start( args, fmt ); + vsprintf( buffer, fmt, args ); + stream->winsys->printf( stream->winsys, buffer ); + va_end( args ); + + PRINTF(stream, ": 0x%x\n", ((dw) & himask) >> (lo)); +} + +#define MBZ( dw, hi, lo) do { \ + unsigned x = (dw) >> (lo); \ + unsigned lomask = (1 << (lo)) - 1; \ + unsigned himask; \ + himask = (1UL << (hi)) - 1; \ + assert ((x & himask & ~lomask) == 0); \ +} while (0) + +static void +FLAG( + struct debug_stream *stream, + unsigned dw, + unsigned bit, + const char *fmt, + ... ) +{ + if (((dw) >> (bit)) & 1) { + va_list args; + char buffer[256]; + + PRINTF(stream, "\t\t "); + + va_start( args, fmt ); + vsprintf( buffer, fmt, args ); + stream->winsys->printf( stream->winsys, buffer ); + va_end( args ); + + PRINTF(stream, "\n"); + } +} + +static boolean debug_load_immediate( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned bits = (ptr[0] >> 4) & 0xff; + unsigned j = 0; + + PRINTF(stream, "%s (%d dwords, flags: %x):\n", name, len, bits); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + if (bits & (1<<0)) { + PRINTF(stream, "\t LIS0: 0x%08x\n", ptr[j]); + PRINTF(stream, "\t vb address: 0x%08x\n", (ptr[j] & ~0x3)); + BITS(stream, ptr[j], 0, 0, "vb invalidate disable"); + j++; + } + if (bits & (1<<1)) { + PRINTF(stream, "\t LIS1: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 29, 24, "vb dword width"); + BITS(stream, ptr[j], 21, 16, "vb dword pitch"); + BITS(stream, ptr[j], 15, 0, "vb max index"); + j++; + } + if (bits & (1<<2)) { + int i; + PRINTF(stream, "\t LIS2: 0x%08x\n", ptr[j]); + for (i = 0; i < 8; i++) { + unsigned tc = (ptr[j] >> (i * 4)) & 0xf; + if (tc != 0xf) + BITS(stream, tc, 3, 0, "tex coord %d", i); + } + j++; + } + if (bits & (1<<3)) { + PRINTF(stream, "\t LIS3: 0x%08x\n", ptr[j]); + j++; + } + if (bits & (1<<4)) { + PRINTF(stream, "\t LIS4: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 31, 23, "point width"); + BITS(stream, ptr[j], 22, 19, "line width"); + FLAG(stream, ptr[j], 18, "alpha flatshade"); + FLAG(stream, ptr[j], 17, "fog flatshade"); + FLAG(stream, ptr[j], 16, "spec flatshade"); + FLAG(stream, ptr[j], 15, "rgb flatshade"); + BITS(stream, ptr[j], 14, 13, "cull mode"); + FLAG(stream, ptr[j], 12, "vfmt: point width"); + FLAG(stream, ptr[j], 11, "vfmt: specular/fog"); + FLAG(stream, ptr[j], 10, "vfmt: rgba"); + FLAG(stream, ptr[j], 9, "vfmt: depth offset"); + BITS(stream, ptr[j], 8, 6, "vfmt: position (2==xyzw)"); + FLAG(stream, ptr[j], 5, "force dflt diffuse"); + FLAG(stream, ptr[j], 4, "force dflt specular"); + FLAG(stream, ptr[j], 3, "local depth offset enable"); + FLAG(stream, ptr[j], 2, "vfmt: fp32 fog coord"); + FLAG(stream, ptr[j], 1, "sprite point"); + FLAG(stream, ptr[j], 0, "antialiasing"); + j++; + } + if (bits & (1<<5)) { + PRINTF(stream, "\t LIS5: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 31, 28, "rgba write disables"); + FLAG(stream, ptr[j], 27, "force dflt point width"); + FLAG(stream, ptr[j], 26, "last pixel enable"); + FLAG(stream, ptr[j], 25, "global z offset enable"); + FLAG(stream, ptr[j], 24, "fog enable"); + BITS(stream, ptr[j], 23, 16, "stencil ref"); + BITS(stream, ptr[j], 15, 13, "stencil test"); + BITS(stream, ptr[j], 12, 10, "stencil fail op"); + BITS(stream, ptr[j], 9, 7, "stencil pass z fail op"); + BITS(stream, ptr[j], 6, 4, "stencil pass z pass op"); + FLAG(stream, ptr[j], 3, "stencil write enable"); + FLAG(stream, ptr[j], 2, "stencil test enable"); + FLAG(stream, ptr[j], 1, "color dither enable"); + FLAG(stream, ptr[j], 0, "logiop enable"); + j++; + } + if (bits & (1<<6)) { + PRINTF(stream, "\t LIS6: 0x%08x\n", ptr[j]); + FLAG(stream, ptr[j], 31, "alpha test enable"); + BITS(stream, ptr[j], 30, 28, "alpha func"); + BITS(stream, ptr[j], 27, 20, "alpha ref"); + FLAG(stream, ptr[j], 19, "depth test enable"); + BITS(stream, ptr[j], 18, 16, "depth func"); + FLAG(stream, ptr[j], 15, "blend enable"); + BITS(stream, ptr[j], 14, 12, "blend func"); + BITS(stream, ptr[j], 11, 8, "blend src factor"); + BITS(stream, ptr[j], 7, 4, "blend dst factor"); + FLAG(stream, ptr[j], 3, "depth write enable"); + FLAG(stream, ptr[j], 2, "color write enable"); + BITS(stream, ptr[j], 1, 0, "provoking vertex"); + j++; + } + + + PRINTF(stream, "\n"); + + assert(j == len); + + stream->offset += len * sizeof(unsigned); + + return TRUE; +} + + + +static boolean debug_load_indirect( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned bits = (ptr[0] >> 8) & 0x3f; + unsigned i, j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + for (i = 0; i < 6; i++) { + if (bits & (1<<i)) { + switch (1<<(8+i)) { + case LI0_STATE_STATIC_INDIRECT: + PRINTF(stream, " STATIC: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + PRINTF(stream, " 0x%08x\n", ptr[j++]); + break; + case LI0_STATE_DYNAMIC_INDIRECT: + PRINTF(stream, " DYNAMIC: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + break; + case LI0_STATE_SAMPLER: + PRINTF(stream, " SAMPLER: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + PRINTF(stream, " 0x%08x\n", ptr[j++]); + break; + case LI0_STATE_MAP: + PRINTF(stream, " MAP: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + PRINTF(stream, " 0x%08x\n", ptr[j++]); + break; + case LI0_STATE_PROGRAM: + PRINTF(stream, " PROGRAM: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + PRINTF(stream, " 0x%08x\n", ptr[j++]); + break; + case LI0_STATE_CONSTANTS: + PRINTF(stream, " CONSTANTS: 0x%08x | %x\n", ptr[j]&~3, ptr[j]&3); j++; + PRINTF(stream, " 0x%08x\n", ptr[j++]); + break; + default: + assert(0); + break; + } + } + } + + if (bits == 0) { + PRINTF(stream, "\t DUMMY: 0x%08x\n", ptr[j++]); + } + + PRINTF(stream, "\n"); + + + assert(j == len); + + stream->offset += len * sizeof(unsigned); + + return TRUE; +} + +static void BR13( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x\n", val); + FLAG(stream, val, 30, "clipping enable"); + BITS(stream, val, 25, 24, "color depth (3==32bpp)"); + BITS(stream, val, 23, 16, "raster op"); + BITS(stream, val, 15, 0, "dest pitch"); +} + + +static void BR22( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x\n", val); + BITS(stream, val, 31, 16, "dest y1"); + BITS(stream, val, 15, 0, "dest x1"); +} + +static void BR23( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x\n", val); + BITS(stream, val, 31, 16, "dest y2"); + BITS(stream, val, 15, 0, "dest x2"); +} + +static void BR09( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x -- dest address\n", val); +} + +static void BR26( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x\n", val); + BITS(stream, val, 31, 16, "src y1"); + BITS(stream, val, 15, 0, "src x1"); +} + +static void BR11( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x\n", val); + BITS(stream, val, 15, 0, "src pitch"); +} + +static void BR12( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x -- src address\n", val); +} + +static void BR16( struct debug_stream *stream, + unsigned val ) +{ + PRINTF(stream, "\t0x%08x -- color\n", val); +} + +static boolean debug_copy_blit( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + int j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + BR13(stream, ptr[j++]); + BR22(stream, ptr[j++]); + BR23(stream, ptr[j++]); + BR09(stream, ptr[j++]); + BR26(stream, ptr[j++]); + BR11(stream, ptr[j++]); + BR12(stream, ptr[j++]); + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_color_blit( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + int j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + BR13(stream, ptr[j++]); + BR22(stream, ptr[j++]); + BR23(stream, ptr[j++]); + BR09(stream, ptr[j++]); + BR16(stream, ptr[j++]); + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_modes4( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + int j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 21, 18, "logicop func"); + FLAG(stream, ptr[j], 17, "stencil test mask modify-enable"); + FLAG(stream, ptr[j], 16, "stencil write mask modify-enable"); + BITS(stream, ptr[j], 15, 8, "stencil test mask"); + BITS(stream, ptr[j], 7, 0, "stencil write mask"); + j++; + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_map_state( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + { + PRINTF(stream, "\t0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 15, 0, "map mask"); + j++; + } + + while (j < len) { + { + PRINTF(stream, "\t TMn.0: 0x%08x\n", ptr[j]); + PRINTF(stream, "\t map address: 0x%08x\n", (ptr[j] & ~0x3)); + FLAG(stream, ptr[j], 1, "vertical line stride"); + FLAG(stream, ptr[j], 0, "vertical line stride offset"); + j++; + } + + { + PRINTF(stream, "\t TMn.1: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 31, 21, "height"); + BITS(stream, ptr[j], 20, 10, "width"); + BITS(stream, ptr[j], 9, 7, "surface format"); + BITS(stream, ptr[j], 6, 3, "texel format"); + FLAG(stream, ptr[j], 2, "use fence regs"); + FLAG(stream, ptr[j], 1, "tiled surface"); + FLAG(stream, ptr[j], 0, "tile walk ymajor"); + j++; + } + { + PRINTF(stream, "\t TMn.2: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 31, 21, "dword pitch"); + BITS(stream, ptr[j], 20, 15, "cube face enables"); + BITS(stream, ptr[j], 14, 9, "max lod"); + FLAG(stream, ptr[j], 8, "mip layout right"); + BITS(stream, ptr[j], 7, 0, "depth"); + j++; + } + } + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_sampler_state( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + { + PRINTF(stream, "\t0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 15, 0, "sampler mask"); + j++; + } + + while (j < len) { + { + PRINTF(stream, "\t TSn.0: 0x%08x\n", ptr[j]); + FLAG(stream, ptr[j], 31, "reverse gamma"); + FLAG(stream, ptr[j], 30, "planar to packed"); + FLAG(stream, ptr[j], 29, "yuv->rgb"); + BITS(stream, ptr[j], 28, 27, "chromakey index"); + BITS(stream, ptr[j], 26, 22, "base mip level"); + BITS(stream, ptr[j], 21, 20, "mip mode filter"); + BITS(stream, ptr[j], 19, 17, "mag mode filter"); + BITS(stream, ptr[j], 16, 14, "min mode filter"); + BITS(stream, ptr[j], 13, 5, "lod bias (s4.4)"); + FLAG(stream, ptr[j], 4, "shadow enable"); + FLAG(stream, ptr[j], 3, "max-aniso-4"); + BITS(stream, ptr[j], 2, 0, "shadow func"); + j++; + } + + { + PRINTF(stream, "\t TSn.1: 0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 31, 24, "min lod"); + MBZ( ptr[j], 23, 18 ); + FLAG(stream, ptr[j], 17, "kill pixel enable"); + FLAG(stream, ptr[j], 16, "keyed tex filter mode"); + FLAG(stream, ptr[j], 15, "chromakey enable"); + BITS(stream, ptr[j], 14, 12, "tcx wrap mode"); + BITS(stream, ptr[j], 11, 9, "tcy wrap mode"); + BITS(stream, ptr[j], 8, 6, "tcz wrap mode"); + FLAG(stream, ptr[j], 5, "normalized coords"); + BITS(stream, ptr[j], 4, 1, "map (surface) index"); + FLAG(stream, ptr[j], 0, "EAST deinterlacer enable"); + j++; + } + { + PRINTF(stream, "\t TSn.2: 0x%08x (default color)\n", ptr[j]); + j++; + } + } + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_dest_vars( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + int j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + { + PRINTF(stream, "\t0x%08x\n", ptr[j]); + FLAG(stream, ptr[j], 31, "early classic ztest"); + FLAG(stream, ptr[j], 30, "opengl tex default color"); + FLAG(stream, ptr[j], 29, "bypass iz"); + FLAG(stream, ptr[j], 28, "lod preclamp"); + BITS(stream, ptr[j], 27, 26, "dither pattern"); + FLAG(stream, ptr[j], 25, "linear gamma blend"); + FLAG(stream, ptr[j], 24, "debug dither"); + BITS(stream, ptr[j], 23, 20, "dstorg x"); + BITS(stream, ptr[j], 19, 16, "dstorg y"); + MBZ (ptr[j], 15, 15 ); + BITS(stream, ptr[j], 14, 12, "422 write select"); + BITS(stream, ptr[j], 11, 8, "cbuf format"); + BITS(stream, ptr[j], 3, 2, "zbuf format"); + FLAG(stream, ptr[j], 1, "vert line stride"); + FLAG(stream, ptr[j], 1, "vert line stride offset"); + j++; + } + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean debug_buf_info( struct debug_stream *stream, + const char *name, + unsigned len ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + int j = 0; + + PRINTF(stream, "%s (%d dwords):\n", name, len); + PRINTF(stream, "\t0x%08x\n", ptr[j++]); + + { + PRINTF(stream, "\t0x%08x\n", ptr[j]); + BITS(stream, ptr[j], 28, 28, "aux buffer id"); + BITS(stream, ptr[j], 27, 24, "buffer id (7=depth, 3=back)"); + FLAG(stream, ptr[j], 23, "use fence regs"); + FLAG(stream, ptr[j], 22, "tiled surface"); + FLAG(stream, ptr[j], 21, "tile walk ymajor"); + MBZ (ptr[j], 20, 14); + BITS(stream, ptr[j], 13, 2, "dword pitch"); + MBZ (ptr[j], 2, 0); + j++; + } + + PRINTF(stream, "\t0x%08x -- buffer base address\n", ptr[j++]); + + stream->offset += len * sizeof(unsigned); + assert(j == len); + return TRUE; +} + +static boolean i915_debug_packet( struct debug_stream *stream ) +{ + unsigned *ptr = (unsigned *)(stream->ptr + stream->offset); + unsigned cmd = *ptr; + + switch (((cmd >> 29) & 0x7)) { + case 0x0: + switch ((cmd >> 23) & 0x3f) { + case 0x0: + return debug(stream, "MI_NOOP", 1); + case 0x3: + return debug(stream, "MI_WAIT_FOR_EVENT", 1); + case 0x4: + return debug(stream, "MI_FLUSH", 1); + case 0xA: + debug(stream, "MI_BATCH_BUFFER_END", 1); + return FALSE; + case 0x22: + return debug(stream, "MI_LOAD_REGISTER_IMM", 3); + case 0x31: + return debug_chain(stream, "MI_BATCH_BUFFER_START", 2); + default: + (void)debug(stream, "UNKNOWN 0x0 case!", 1); + assert(0); + break; + } + break; + case 0x1: + (void) debug(stream, "UNKNOWN 0x1 case!", 1); + assert(0); + break; + case 0x2: + switch ((cmd >> 22) & 0xff) { + case 0x50: + return debug_color_blit(stream, "XY_COLOR_BLT", (cmd & 0xff) + 2); + case 0x53: + return debug_copy_blit(stream, "XY_SRC_COPY_BLT", (cmd & 0xff) + 2); + default: + return debug(stream, "blit command", (cmd & 0xff) + 2); + } + break; + case 0x3: + switch ((cmd >> 24) & 0x1f) { + case 0x6: + return debug(stream, "3DSTATE_ANTI_ALIASING", 1); + case 0x7: + return debug(stream, "3DSTATE_RASTERIZATION_RULES", 1); + case 0x8: + return debug(stream, "3DSTATE_BACKFACE_STENCIL_OPS", 2); + case 0x9: + return debug(stream, "3DSTATE_BACKFACE_STENCIL_MASKS", 1); + case 0xb: + return debug(stream, "3DSTATE_INDEPENDENT_ALPHA_BLEND", 1); + case 0xc: + return debug(stream, "3DSTATE_MODES5", 1); + case 0xd: + return debug_modes4(stream, "3DSTATE_MODES4", 1); + case 0x15: + return debug(stream, "3DSTATE_FOG_COLOR", 1); + case 0x16: + return debug(stream, "3DSTATE_COORD_SET_BINDINGS", 1); + case 0x1c: + /* 3DState16NP */ + switch((cmd >> 19) & 0x1f) { + case 0x10: + return debug(stream, "3DSTATE_SCISSOR_ENABLE", 1); + case 0x11: + return debug(stream, "3DSTATE_DEPTH_SUBRECTANGLE_DISABLE", 1); + default: + (void) debug(stream, "UNKNOWN 0x1c case!", 1); + assert(0); + break; + } + break; + case 0x1d: + /* 3DStateMW */ + switch ((cmd >> 16) & 0xff) { + case 0x0: + return debug_map_state(stream, "3DSTATE_MAP_STATE", (cmd & 0x1f) + 2); + case 0x1: + return debug_sampler_state(stream, "3DSTATE_SAMPLER_STATE", (cmd & 0x1f) + 2); + case 0x4: + return debug_load_immediate(stream, "3DSTATE_LOAD_STATE_IMMEDIATE", (cmd & 0xf) + 2); + case 0x5: + return debug_program(stream, "3DSTATE_PIXEL_SHADER_PROGRAM", (cmd & 0x1ff) + 2); + case 0x6: + return debug(stream, "3DSTATE_PIXEL_SHADER_CONSTANTS", (cmd & 0xff) + 2); + case 0x7: + return debug_load_indirect(stream, "3DSTATE_LOAD_INDIRECT", (cmd & 0xff) + 2); + case 0x80: + return debug(stream, "3DSTATE_DRAWING_RECTANGLE", (cmd & 0xffff) + 2); + case 0x81: + return debug(stream, "3DSTATE_SCISSOR_RECTANGLE", (cmd & 0xffff) + 2); + case 0x83: + return debug(stream, "3DSTATE_SPAN_STIPPLE", (cmd & 0xffff) + 2); + case 0x85: + return debug_dest_vars(stream, "3DSTATE_DEST_BUFFER_VARS", (cmd & 0xffff) + 2); + case 0x88: + return debug(stream, "3DSTATE_CONSTANT_BLEND_COLOR", (cmd & 0xffff) + 2); + case 0x89: + return debug(stream, "3DSTATE_FOG_MODE", (cmd & 0xffff) + 2); + case 0x8e: + return debug_buf_info(stream, "3DSTATE_BUFFER_INFO", (cmd & 0xffff) + 2); + case 0x97: + return debug(stream, "3DSTATE_DEPTH_OFFSET_SCALE", (cmd & 0xffff) + 2); + case 0x98: + return debug(stream, "3DSTATE_DEFAULT_Z", (cmd & 0xffff) + 2); + case 0x99: + return debug(stream, "3DSTATE_DEFAULT_DIFFUSE", (cmd & 0xffff) + 2); + case 0x9a: + return debug(stream, "3DSTATE_DEFAULT_SPECULAR", (cmd & 0xffff) + 2); + case 0x9c: + return debug(stream, "3DSTATE_CLEAR_PARAMETERS", (cmd & 0xffff) + 2); + default: + assert(0); + return 0; + } + break; + case 0x1e: + if (cmd & (1 << 23)) + return debug(stream, "???", (cmd & 0xffff) + 1); + else + return debug(stream, "", 1); + break; + case 0x1f: + if ((cmd & (1 << 23)) == 0) + return debug_prim(stream, "3DPRIM (inline)", 1, (cmd & 0x1ffff) + 2); + else if (cmd & (1 << 17)) + { + if ((cmd & 0xffff) == 0) + return debug_variable_length_prim(stream); + else + return debug_prim(stream, "3DPRIM (indexed)", 0, (((cmd & 0xffff) + 1) / 2) + 1); + } + else + return debug_prim(stream, "3DPRIM (indirect sequential)", 0, 2); + break; + default: + return debug(stream, "", 0); + } + default: + assert(0); + return 0; + } + + assert(0); + return 0; +} + + + +void +i915_dump_batchbuffer( struct i915_context *i915 ) +{ + struct debug_stream stream; + unsigned *start = i915->batch_start; + unsigned *end = i915->winsys->batch_start( i915->winsys, 0, 0 ); + unsigned long bytes = (unsigned long) (end - start) * 4; + boolean done = FALSE; + + stream.offset = 0; + stream.ptr = (char *)start; + stream.print_addresses = 0; + stream.winsys = i915->pipe.winsys; + + if (!start || !end) { + stream.winsys->printf( stream.winsys, "\n\nBATCH: ???\n"); + return; + } + + stream.winsys->printf( stream.winsys, "\n\nBATCH: (%d)\n", bytes / 4); + + while (!done && + stream.offset < bytes) + { + if (!i915_debug_packet( &stream )) + break; + + assert(stream.offset <= bytes && + stream.offset >= 0); + } + + stream.winsys->printf( stream.winsys, "END-BATCH\n\n\n"); +} + + diff --git a/src/gallium/drivers/i915simple/i915_debug.h b/src/gallium/drivers/i915simple/i915_debug.h new file mode 100644 index 0000000000..0bcd094233 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_debug.h @@ -0,0 +1,117 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef I915_DEBUG_H +#define I915_DEBUG_H + +#include <stdarg.h> + +struct i915_context; + +struct debug_stream +{ + unsigned offset; /* current gtt offset */ + char *ptr; /* pointer to gtt offset zero */ + char *end; /* pointer to gtt offset zero */ + unsigned print_addresses; + struct pipe_winsys *winsys; +}; + + +/* Internal functions + */ +void i915_disassemble_program(struct debug_stream *stream, + const unsigned *program, unsigned sz); + +void i915_print_ureg(const char *msg, unsigned ureg); + + +#define DEBUG_BATCH 0x1 +#define DEBUG_BLIT 0x2 +#define DEBUG_BUFFER 0x4 +#define DEBUG_CONSTANTS 0x8 +#define DEBUG_CONTEXT 0x10 +#define DEBUG_DRAW 0x20 +#define DEBUG_DYNAMIC 0x40 +#define DEBUG_FLUSH 0x80 +#define DEBUG_MAP 0x100 +#define DEBUG_PROGRAM 0x200 +#define DEBUG_REGIONS 0x400 +#define DEBUG_SAMPLER 0x800 +#define DEBUG_STATIC 0x1000 +#define DEBUG_SURFACE 0x2000 +#define DEBUG_WINSYS 0x4000 + +#include "pipe/p_compiler.h" + +#if defined(DEBUG) && defined(FILE_DEBUG_FLAG) + +#include "pipe/p_winsys.h" + +static INLINE void +I915_DBG( + struct i915_context *i915, + const char *fmt, + ... ) +{ + if ((i915)->debug & FILE_DEBUG_FLAG) { + va_list args; + char buffer[256]; + + va_start( args, fmt ); + vsprintf( buffer, fmt, args ); + i915->pipe.winsys->printf( i915->pipe.winsys, buffer ); + va_end( args ); + } +} + +#else + +static INLINE void +I915_DBG( + struct i915_context *i915, + const char *fmt, + ... ) +{ + (void) i915; + (void) fmt; +} + +#endif + + +void i915_dump_batchbuffer( struct i915_context *i915 ); + + + +void i915_debug_init( struct i915_context *i915 ); + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_debug_fp.c b/src/gallium/drivers/i915simple/i915_debug_fp.c new file mode 100644 index 0000000000..ebfdb3d93c --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_debug_fp.c @@ -0,0 +1,366 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "i915_reg.h" +#include "i915_debug.h" +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" + + +static void +PRINTF( + struct debug_stream *stream, + const char *fmt, + ... ) +{ + va_list args; + char buffer[256]; + + va_start( args, fmt ); + vsprintf( buffer, fmt, args ); + stream->winsys->printf( stream->winsys, buffer ); + va_end( args ); +} + + +static const char *opcodes[0x20] = { + "NOP", + "ADD", + "MOV", + "MUL", + "MAD", + "DP2ADD", + "DP3", + "DP4", + "FRC", + "RCP", + "RSQ", + "EXP", + "LOG", + "CMP", + "MIN", + "MAX", + "FLR", + "MOD", + "TRC", + "SGE", + "SLT", + "TEXLD", + "TEXLDP", + "TEXLDB", + "TEXKILL", + "DCL", + "0x1a", + "0x1b", + "0x1c", + "0x1d", + "0x1e", + "0x1f", +}; + + +static const int args[0x20] = { + 0, /* 0 nop */ + 2, /* 1 add */ + 1, /* 2 mov */ + 2, /* 3 m ul */ + 3, /* 4 mad */ + 3, /* 5 dp2add */ + 2, /* 6 dp3 */ + 2, /* 7 dp4 */ + 1, /* 8 frc */ + 1, /* 9 rcp */ + 1, /* a rsq */ + 1, /* b exp */ + 1, /* c log */ + 3, /* d cmp */ + 2, /* e min */ + 2, /* f max */ + 1, /* 10 flr */ + 1, /* 11 mod */ + 1, /* 12 trc */ + 2, /* 13 sge */ + 2, /* 14 slt */ + 1, + 1, + 1, + 1, + 0, + 0, + 0, + 0, + 0, + 0, + 0, +}; + + +static const char *regname[0x8] = { + "R", + "T", + "CONST", + "S", + "OC", + "OD", + "U", + "UNKNOWN", +}; + +static void +print_reg_type_nr(struct debug_stream *stream, unsigned type, unsigned nr) +{ + switch (type) { + case REG_TYPE_T: + switch (nr) { + case T_DIFFUSE: + PRINTF(stream, "T_DIFFUSE"); + return; + case T_SPECULAR: + PRINTF(stream, "T_SPECULAR"); + return; + case T_FOG_W: + PRINTF(stream, "T_FOG_W"); + return; + default: + PRINTF(stream, "T_TEX%d", nr); + return; + } + case REG_TYPE_OC: + if (nr == 0) { + PRINTF(stream, "oC"); + return; + } + break; + case REG_TYPE_OD: + if (nr == 0) { + PRINTF(stream, "oD"); + return; + } + break; + default: + break; + } + + PRINTF(stream, "%s[%d]", regname[type], nr); +} + +#define REG_SWIZZLE_MASK 0x7777 +#define REG_NEGATE_MASK 0x8888 + +#define REG_SWIZZLE_XYZW ((SRC_X << A2_SRC2_CHANNEL_X_SHIFT) | \ + (SRC_Y << A2_SRC2_CHANNEL_Y_SHIFT) | \ + (SRC_Z << A2_SRC2_CHANNEL_Z_SHIFT) | \ + (SRC_W << A2_SRC2_CHANNEL_W_SHIFT)) + + +static void +print_reg_neg_swizzle(struct debug_stream *stream, unsigned reg) +{ + int i; + + if ((reg & REG_SWIZZLE_MASK) == REG_SWIZZLE_XYZW && + (reg & REG_NEGATE_MASK) == 0) + return; + + PRINTF(stream, "."); + + for (i = 3; i >= 0; i--) { + if (reg & (1 << ((i * 4) + 3))) + PRINTF(stream, "-"); + + switch ((reg >> (i * 4)) & 0x7) { + case 0: + PRINTF(stream, "x"); + break; + case 1: + PRINTF(stream, "y"); + break; + case 2: + PRINTF(stream, "z"); + break; + case 3: + PRINTF(stream, "w"); + break; + case 4: + PRINTF(stream, "0"); + break; + case 5: + PRINTF(stream, "1"); + break; + default: + PRINTF(stream, "?"); + break; + } + } +} + + +static void +print_src_reg(struct debug_stream *stream, unsigned dword) +{ + unsigned nr = (dword >> A2_SRC2_NR_SHIFT) & REG_NR_MASK; + unsigned type = (dword >> A2_SRC2_TYPE_SHIFT) & REG_TYPE_MASK; + print_reg_type_nr(stream, type, nr); + print_reg_neg_swizzle(stream, dword); +} + + +static void +print_dest_reg(struct debug_stream *stream, unsigned dword) +{ + unsigned nr = (dword >> A0_DEST_NR_SHIFT) & REG_NR_MASK; + unsigned type = (dword >> A0_DEST_TYPE_SHIFT) & REG_TYPE_MASK; + print_reg_type_nr(stream, type, nr); + if ((dword & A0_DEST_CHANNEL_ALL) == A0_DEST_CHANNEL_ALL) + return; + PRINTF(stream, "."); + if (dword & A0_DEST_CHANNEL_X) + PRINTF(stream, "x"); + if (dword & A0_DEST_CHANNEL_Y) + PRINTF(stream, "y"); + if (dword & A0_DEST_CHANNEL_Z) + PRINTF(stream, "z"); + if (dword & A0_DEST_CHANNEL_W) + PRINTF(stream, "w"); +} + + +#define GET_SRC0_REG(r0, r1) ((r0<<14)|(r1>>A1_SRC0_CHANNEL_W_SHIFT)) +#define GET_SRC1_REG(r0, r1) ((r0<<8)|(r1>>A2_SRC1_CHANNEL_W_SHIFT)) +#define GET_SRC2_REG(r) (r) + + +static void +print_arith_op(struct debug_stream *stream, + unsigned opcode, const unsigned * program) +{ + if (opcode != A0_NOP) { + print_dest_reg(stream, program[0]); + if (program[0] & A0_DEST_SATURATE) + PRINTF(stream, " = SATURATE "); + else + PRINTF(stream, " = "); + } + + PRINTF(stream, "%s ", opcodes[opcode]); + + print_src_reg(stream, GET_SRC0_REG(program[0], program[1])); + if (args[opcode] == 1) { + PRINTF(stream, "\n"); + return; + } + + PRINTF(stream, ", "); + print_src_reg(stream, GET_SRC1_REG(program[1], program[2])); + if (args[opcode] == 2) { + PRINTF(stream, "\n"); + return; + } + + PRINTF(stream, ", "); + print_src_reg(stream, GET_SRC2_REG(program[2])); + PRINTF(stream, "\n"); + return; +} + + +static void +print_tex_op(struct debug_stream *stream, + unsigned opcode, const unsigned * program) +{ + print_dest_reg(stream, program[0] | A0_DEST_CHANNEL_ALL); + PRINTF(stream, " = "); + + PRINTF(stream, "%s ", opcodes[opcode]); + + PRINTF(stream, "S[%d],", program[0] & T0_SAMPLER_NR_MASK); + + print_reg_type_nr(stream, + (program[1] >> T1_ADDRESS_REG_TYPE_SHIFT) & + REG_TYPE_MASK, + (program[1] >> T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK); + PRINTF(stream, "\n"); +} + +static void +print_texkil_op(struct debug_stream *stream, + unsigned opcode, const unsigned * program) +{ + PRINTF(stream, "TEXKIL "); + + print_reg_type_nr(stream, + (program[1] >> T1_ADDRESS_REG_TYPE_SHIFT) & + REG_TYPE_MASK, + (program[1] >> T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK); + PRINTF(stream, "\n"); +} + +static void +print_dcl_op(struct debug_stream *stream, + unsigned opcode, const unsigned * program) +{ + PRINTF(stream, "%s ", opcodes[opcode]); + print_dest_reg(stream, + program[0] | A0_DEST_CHANNEL_ALL); + PRINTF(stream, "\n"); +} + + +void +i915_disassemble_program(struct debug_stream *stream, + const unsigned * program, unsigned sz) +{ + unsigned size = program[0] & 0x1ff; + unsigned i; + + PRINTF(stream, "\t\tBEGIN\n"); + + assert(size + 2 == sz); + + program++; + for (i = 1; i < sz; i += 3, program += 3) { + unsigned opcode = program[0] & (0x1f << 24); + + PRINTF(stream, "\t\t"); + + if ((int) opcode >= A0_NOP && opcode <= A0_SLT) + print_arith_op(stream, opcode >> 24, program); + else if (opcode >= T0_TEXLD && opcode < T0_TEXKILL) + print_tex_op(stream, opcode >> 24, program); + else if (opcode == T0_TEXKILL) + print_texkil_op(stream, opcode >> 24, program); + else if (opcode == D0_DCL) + print_dcl_op(stream, opcode >> 24, program); + else + PRINTF(stream, "Unknown opcode 0x%x\n", opcode); + } + + PRINTF(stream, "\t\tEND\n\n"); +} + + diff --git a/src/gallium/drivers/i915simple/i915_flush.c b/src/gallium/drivers/i915simple/i915_flush.c new file mode 100644 index 0000000000..3c2069b827 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_flush.c @@ -0,0 +1,81 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_defines.h" +#include "i915_context.h" +#include "i915_reg.h" +#include "i915_batch.h" + + +/** + * In future we may want a fence-like interface instead of finish. + */ +static void i915_flush( struct pipe_context *pipe, + unsigned flags ) +{ + struct i915_context *i915 = i915_context(pipe); + + /* Do we need to emit an MI_FLUSH command to flush the hardware + * caches? + */ + if (flags & (PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE)) { + unsigned flush = MI_FLUSH; + + if (!(flags & PIPE_FLUSH_RENDER_CACHE)) + flush |= INHIBIT_FLUSH_RENDER_CACHE; + + if (flags & PIPE_FLUSH_TEXTURE_CACHE) + flush |= FLUSH_MAP_CACHE; + + if (!BEGIN_BATCH(1, 0)) { + FLUSH_BATCH(); + assert(BEGIN_BATCH(1, 0)); + } + OUT_BATCH( flush ); + ADVANCE_BATCH(); + } + + /* If there are no flags, just flush pending commands to hardware: + */ + FLUSH_BATCH(); + + if (flags & PIPE_FLUSH_WAIT) { + i915->winsys->batch_finish(i915->winsys); + } +} + + + +void i915_init_flush_functions( struct i915_context *i915 ) +{ + i915->pipe.flush = i915_flush; +} diff --git a/src/gallium/drivers/i915simple/i915_fpc.h b/src/gallium/drivers/i915simple/i915_fpc.h new file mode 100644 index 0000000000..8c7b68aefb --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_fpc.h @@ -0,0 +1,213 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef I915_FPC_H +#define I915_FPC_H + +#include "pipe/p_util.h" + +#include "i915_context.h" +#include "i915_reg.h" + + + +#define I915_PROGRAM_SIZE 192 + + + +/** + * Program translation state + */ +struct i915_fp_compile { + const struct pipe_shader_state *shader; + + struct vertex_info *vertex_info; + + uint declarations[I915_PROGRAM_SIZE]; + uint program[I915_PROGRAM_SIZE]; + + uint input_semantic_name[PIPE_MAX_SHADER_INPUTS]; + uint input_semantic_index[PIPE_MAX_SHADER_INPUTS]; + + uint output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; + uint output_semantic_index[PIPE_MAX_SHADER_OUTPUTS]; + + /** points into the i915->current.constants array: */ + float (*constants)[4]; + uint num_constants; + uint constant_flags[I915_MAX_CONSTANT]; /**< status of each constant */ + + uint *csr; /**< Cursor, points into program. */ + + uint *decl; /**< Cursor, points into declarations. */ + + uint decl_s; /**< flags for which s regs need to be decl'd */ + uint decl_t; /**< flags for which t regs need to be decl'd */ + + uint temp_flag; /**< Tracks temporary regs which are in use */ + uint utemp_flag; /**< Tracks TYPE_U temporary regs which are in use */ + + uint nr_tex_indirect; + uint nr_tex_insn; + uint nr_alu_insn; + uint nr_decl_insn; + + boolean error; /**< Set if i915_program_error() is called */ + uint wpos_tex; + uint NumNativeInstructions; + uint NumNativeAluInstructions; + uint NumNativeTexInstructions; + uint NumNativeTexIndirections; +}; + + +/* Having zero and one in here makes the definition of swizzle a lot + * easier. + */ +#define UREG_TYPE_SHIFT 29 +#define UREG_NR_SHIFT 24 +#define UREG_CHANNEL_X_NEGATE_SHIFT 23 +#define UREG_CHANNEL_X_SHIFT 20 +#define UREG_CHANNEL_Y_NEGATE_SHIFT 19 +#define UREG_CHANNEL_Y_SHIFT 16 +#define UREG_CHANNEL_Z_NEGATE_SHIFT 15 +#define UREG_CHANNEL_Z_SHIFT 12 +#define UREG_CHANNEL_W_NEGATE_SHIFT 11 +#define UREG_CHANNEL_W_SHIFT 8 +#define UREG_CHANNEL_ZERO_NEGATE_MBZ 5 +#define UREG_CHANNEL_ZERO_SHIFT 4 +#define UREG_CHANNEL_ONE_NEGATE_MBZ 1 +#define UREG_CHANNEL_ONE_SHIFT 0 + +#define UREG_BAD 0xffffffff /* not a valid ureg */ + +#define X SRC_X +#define Y SRC_Y +#define Z SRC_Z +#define W SRC_W +#define ZERO SRC_ZERO +#define ONE SRC_ONE + +/* Construct a ureg: + */ +#define UREG( type, nr ) (((type)<< UREG_TYPE_SHIFT) | \ + ((nr) << UREG_NR_SHIFT) | \ + (X << UREG_CHANNEL_X_SHIFT) | \ + (Y << UREG_CHANNEL_Y_SHIFT) | \ + (Z << UREG_CHANNEL_Z_SHIFT) | \ + (W << UREG_CHANNEL_W_SHIFT) | \ + (ZERO << UREG_CHANNEL_ZERO_SHIFT) | \ + (ONE << UREG_CHANNEL_ONE_SHIFT)) + +#define GET_CHANNEL_SRC( reg, channel ) ((reg<<(channel*4)) & (0xf<<20)) +#define CHANNEL_SRC( src, channel ) (src>>(channel*4)) + +#define GET_UREG_TYPE(reg) (((reg)>>UREG_TYPE_SHIFT)®_TYPE_MASK) +#define GET_UREG_NR(reg) (((reg)>>UREG_NR_SHIFT)®_NR_MASK) + + + +#define UREG_XYZW_CHANNEL_MASK 0x00ffff00 + +/* One neat thing about the UREG representation: + */ +static INLINE int +swizzle(int reg, uint x, uint y, uint z, uint w) +{ + assert(x <= SRC_ONE); + assert(y <= SRC_ONE); + assert(z <= SRC_ONE); + assert(w <= SRC_ONE); + return ((reg & ~UREG_XYZW_CHANNEL_MASK) | + CHANNEL_SRC(GET_CHANNEL_SRC(reg, x), 0) | + CHANNEL_SRC(GET_CHANNEL_SRC(reg, y), 1) | + CHANNEL_SRC(GET_CHANNEL_SRC(reg, z), 2) | + CHANNEL_SRC(GET_CHANNEL_SRC(reg, w), 3)); +} + + + +/*********************************************************************** + * Public interface for the compiler + */ +extern void i915_translate_fragment_program( struct i915_context *i915 ); + + + +extern uint i915_get_temp(struct i915_fp_compile *p); +extern uint i915_get_utemp(struct i915_fp_compile *p); +extern void i915_release_utemps(struct i915_fp_compile *p); + + +extern uint i915_emit_texld(struct i915_fp_compile *p, + uint dest, + uint destmask, + uint sampler, uint coord, uint op); + +extern uint i915_emit_arith(struct i915_fp_compile *p, + uint op, + uint dest, + uint mask, + uint saturate, + uint src0, uint src1, uint src2); + +extern uint i915_emit_decl(struct i915_fp_compile *p, + uint type, uint nr, uint d0_flags); + + +extern uint i915_emit_const1f(struct i915_fp_compile *p, float c0); + +extern uint i915_emit_const2f(struct i915_fp_compile *p, + float c0, float c1); + +extern uint i915_emit_const4fv(struct i915_fp_compile *p, + const float * c); + +extern uint i915_emit_const4f(struct i915_fp_compile *p, + float c0, float c1, + float c2, float c3); + + +/*====================================================================== + * i915_fpc_debug.c + */ +extern void i915_disassemble_program(const uint * program, uint sz); + + +/*====================================================================== + * i915_fpc_translate.c + */ + +extern void +i915_program_error(struct i915_fp_compile *p, const char *msg, ...); + +extern void +i915_translate_fragment_program(struct i915_context *i915); + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_fpc_emit.c b/src/gallium/drivers/i915simple/i915_fpc_emit.c new file mode 100644 index 0000000000..74924ff0a1 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_fpc_emit.c @@ -0,0 +1,375 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "i915_reg.h" +#include "i915_context.h" +#include "i915_fpc.h" + + +#define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) +#define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) +#define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT) +#define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT) +#define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT) +#define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT) +#define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT) +#define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT) + +/* These are special, and don't have swizzle/negate bits. + */ +#define T0_SAMPLER( reg ) (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT) +#define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \ + (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT)) + + +/* Macros for translating UREG's into the various register fields used + * by the I915 programmable unit. + */ +#define UREG_A0_DEST_SHIFT_LEFT (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT) +#define UREG_A0_SRC0_SHIFT_LEFT (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT) +#define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT) +#define UREG_A1_SRC1_SHIFT_LEFT (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT) +#define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT) +#define UREG_A2_SRC2_SHIFT_LEFT (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT) + +#define UREG_MASK 0xffffff00 +#define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \ + (REG_NR_MASK << UREG_NR_SHIFT)) + + +#define I915_CONSTFLAG_PARAM 0x1f + +uint +i915_get_temp(struct i915_fp_compile *p) +{ + int bit = ffs(~p->temp_flag); + if (!bit) { + i915_program_error(p, "i915_get_temp: out of temporaries\n"); + return 0; + } + + p->temp_flag |= 1 << (bit - 1); + return UREG(REG_TYPE_R, (bit - 1)); +} + + +uint +i915_get_utemp(struct i915_fp_compile * p) +{ + int bit = ffs(~p->utemp_flag); + if (!bit) { + i915_program_error(p, "i915_get_utemp: out of temporaries\n"); + return 0; + } + + p->utemp_flag |= 1 << (bit - 1); + return UREG(REG_TYPE_U, (bit - 1)); +} + +void +i915_release_utemps(struct i915_fp_compile *p) +{ + p->utemp_flag = ~0x7; +} + + +uint +i915_emit_decl(struct i915_fp_compile *p, + uint type, uint nr, uint d0_flags) +{ + uint reg = UREG(type, nr); + + if (type == REG_TYPE_T) { + if (p->decl_t & (1 << nr)) + return reg; + + p->decl_t |= (1 << nr); + } + else if (type == REG_TYPE_S) { + if (p->decl_s & (1 << nr)) + return reg; + + p->decl_s |= (1 << nr); + } + else + return reg; + + *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags); + *(p->decl++) = D1_MBZ; + *(p->decl++) = D2_MBZ; + + p->nr_decl_insn++; + return reg; +} + +uint +i915_emit_arith(struct i915_fp_compile * p, + uint op, + uint dest, + uint mask, + uint saturate, uint src0, uint src1, uint src2) +{ + uint c[3]; + uint nr_const = 0; + + assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); + dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)); + assert(dest); + + if (GET_UREG_TYPE(src0) == REG_TYPE_CONST) + c[nr_const++] = 0; + if (GET_UREG_TYPE(src1) == REG_TYPE_CONST) + c[nr_const++] = 1; + if (GET_UREG_TYPE(src2) == REG_TYPE_CONST) + c[nr_const++] = 2; + + /* Recursively call this function to MOV additional const values + * into temporary registers. Use utemp registers for this - + * currently shouldn't be possible to run out, but keep an eye on + * this. + */ + if (nr_const > 1) { + uint s[3], first, i, old_utemp_flag; + + s[0] = src0; + s[1] = src1; + s[2] = src2; + old_utemp_flag = p->utemp_flag; + + first = GET_UREG_NR(s[c[0]]); + for (i = 1; i < nr_const; i++) { + if (GET_UREG_NR(s[c[i]]) != first) { + uint tmp = i915_get_utemp(p); + + i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0, + s[c[i]], 0, 0); + s[c[i]] = tmp; + } + } + + src0 = s[0]; + src1 = s[1]; + src2 = s[2]; + p->utemp_flag = old_utemp_flag; /* restore */ + } + + *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0)); + *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1)); + *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2)); + + p->nr_alu_insn++; + return dest; +} + +uint i915_emit_texld( struct i915_fp_compile *p, + uint dest, + uint destmask, + uint sampler, + uint coord, + uint op ) +{ + uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord)); + if (coord != k) { + /* No real way to work around this in the general case - need to + * allocate and declare a new temporary register (a utemp won't + * do). Will fallback for now. + */ + i915_program_error(p, "Can't (yet) swizzle TEX arguments"); + assert(0); + return 0; + } + + /* Don't worry about saturate as we only support + */ + if (destmask != A0_DEST_CHANNEL_ALL) { + uint tmp = i915_get_utemp(p); + i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op ); + i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 ); + return dest; + } + else { + assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); + assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest))); + + if (GET_UREG_TYPE(coord) != REG_TYPE_T) { + p->nr_tex_indirect++; + } + + *(p->csr++) = (op | + T0_DEST( dest ) | + T0_SAMPLER( sampler )); + + *(p->csr++) = T1_ADDRESS_REG( coord ); + *(p->csr++) = T2_MBZ; + + p->nr_tex_insn++; + return dest; + } +} + + +uint +i915_emit_const1f(struct i915_fp_compile * p, float c0) +{ + unsigned reg, idx; + + if (c0 == 0.0) + return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO); + if (c0 == 1.0) + return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE); + + for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { + if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM) + continue; + for (idx = 0; idx < 4; idx++) { + if (!(p->constant_flags[reg] & (1 << idx)) || + p->constants[reg][idx] == c0) { + p->constants[reg][idx] = c0; + p->constant_flags[reg] |= 1 << idx; + if (reg + 1 > p->num_constants) + p->num_constants = reg + 1; + return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE); + } + } + } + + i915_program_error(p, "i915_emit_const1f: out of constants\n"); + return 0; +} + +uint +i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1) +{ + unsigned reg, idx; + + if (c0 == 0.0) + return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W); + if (c0 == 1.0) + return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W); + + if (c1 == 0.0) + return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W); + if (c1 == 1.0) + return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W); + + for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { + if (p->constant_flags[reg] == 0xf || + p->constant_flags[reg] == I915_CONSTFLAG_PARAM) + continue; + for (idx = 0; idx < 3; idx++) { + if (!(p->constant_flags[reg] & (3 << idx))) { + p->constants[reg][idx + 0] = c0; + p->constants[reg][idx + 1] = c1; + p->constant_flags[reg] |= 3 << idx; + if (reg + 1 > p->num_constants) + p->num_constants = reg + 1; + return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE); + } + } + } + + i915_program_error(p, "i915_emit_const2f: out of constants\n"); + return 0; +} + + + +uint +i915_emit_const4f(struct i915_fp_compile * p, + float c0, float c1, float c2, float c3) +{ + unsigned reg; + + for (reg = 0; reg < I915_MAX_CONSTANT; reg++) { + if (p->constant_flags[reg] == 0xf && + p->constants[reg][0] == c0 && + p->constants[reg][1] == c1 && + p->constants[reg][2] == c2 && + p->constants[reg][3] == c3) { + return UREG(REG_TYPE_CONST, reg); + } + else if (p->constant_flags[reg] == 0) { + + p->constants[reg][0] = c0; + p->constants[reg][1] = c1; + p->constants[reg][2] = c2; + p->constants[reg][3] = c3; + p->constant_flags[reg] = 0xf; + if (reg + 1 > p->num_constants) + p->num_constants = reg + 1; + return UREG(REG_TYPE_CONST, reg); + } + } + + i915_program_error(p, "i915_emit_const4f: out of constants\n"); + return 0; +} + + +uint +i915_emit_const4fv(struct i915_fp_compile * p, const float * c) +{ + return i915_emit_const4f(p, c[0], c[1], c[2], c[3]); +} + + +#if 00000/*UNUSED*/ +/* Reserve a slot in the constant file for a Mesa state parameter. + * These will later need to be tracked on statechanges, but that is + * done elsewhere. + */ +uint +i915_emit_param4fv(struct i915_fp_compile * p, const float * values) +{ + struct i915_fragment_program *fp = p->fp; + int i; + + for (i = 0; i < fp->nr_params; i++) { + if (fp->param[i].values == values) + return UREG(REG_TYPE_CONST, fp->param[i].reg); + } + + if (p->constants->nr_constants == I915_MAX_CONSTANT || + fp->nr_params == I915_MAX_CONSTANT) { + i915_program_error(p, "i915_emit_param4fv: out of constants\n"); + return 0; + } + + { + int reg = p->constants->nr_constants++; + int i = fp->nr_params++; + + assert (p->constant_flags[reg] == 0); + p->constant_flags[reg] = I915_CONSTFLAG_PARAM; + + fp->param[i].values = values; + fp->param[i].reg = reg; + + return UREG(REG_TYPE_CONST, reg); + } +} +#endif diff --git a/src/gallium/drivers/i915simple/i915_fpc_translate.c b/src/gallium/drivers/i915simple/i915_fpc_translate.c new file mode 100644 index 0000000000..868f0c7e04 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_fpc_translate.c @@ -0,0 +1,1135 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <stdarg.h> + +#include "i915_reg.h" +#include "i915_context.h" +#include "i915_fpc.h" + +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" + +#include "pipe/draw/draw_vertex.h" + + +/** + * Simple pass-through fragment shader to use when we don't have + * a real shader (or it fails to compile for some reason). + */ +static unsigned passthrough[] = +{ + _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1), + + /* declare input color: + */ + (D0_DCL | + (REG_TYPE_T << D0_TYPE_SHIFT) | + (T_DIFFUSE << D0_NR_SHIFT) | + D0_CHANNEL_ALL), + 0, + 0, + + /* move to output color: + */ + (A0_MOV | + (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) | + A0_DEST_CHANNEL_ALL | + (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) | + (T_DIFFUSE << A0_SRC0_NR_SHIFT)), + 0x01230000, /* .xyzw */ + 0 +}; + + +/* 1, -1/3!, 1/5!, -1/7! */ +static const float sin_constants[4] = { 1.0, + -1.0f / (3 * 2 * 1), + 1.0f / (5 * 4 * 3 * 2 * 1), + -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1) +}; + +/* 1, -1/2!, 1/4!, -1/6! */ +static const float cos_constants[4] = { 1.0, + -1.0f / (2 * 1), + 1.0f / (4 * 3 * 2 * 1), + -1.0f / (6 * 5 * 4 * 3 * 2 * 1) +}; + + + +/** + * component-wise negation of ureg + */ +static INLINE int +negate(int reg, int x, int y, int z, int w) +{ + /* Another neat thing about the UREG representation */ + return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) | + ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) | + ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) | + ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT)); +} + + +static void +i915_use_passthrough_shader(struct i915_context *i915) +{ + debug_printf("**** Using i915 pass-through fragment shader\n"); + + i915->current.program = (uint *) MALLOC(sizeof(passthrough)); + if (i915->current.program) { + memcpy(i915->current.program, passthrough, sizeof(passthrough)); + i915->current.program_len = Elements(passthrough); + } + + i915->current.num_constants[PIPE_SHADER_FRAGMENT] = 0; + i915->current.num_user_constants[PIPE_SHADER_FRAGMENT] = 0; +} + + +void +i915_program_error(struct i915_fp_compile *p, const char *msg, ...) +{ + va_list args; + char buffer[1024]; + + debug_printf("i915_program_error: "); + va_start( args, msg ); + vsprintf( buffer, msg, args ); + va_end( args ); + debug_printf(buffer); + debug_printf("\n"); + + p->error = 1; +} + + + +/** + * Construct a ureg for the given source register. Will emit + * constants, apply swizzling and negation as needed. + */ +static uint +src_vector(struct i915_fp_compile *p, + const struct tgsi_full_src_register *source) +{ + uint index = source->SrcRegister.Index; + uint src, sem_name, sem_ind; + + switch (source->SrcRegister.File) { + case TGSI_FILE_TEMPORARY: + if (source->SrcRegister.Index >= I915_MAX_TEMPORARY) { + i915_program_error(p, "Exceeded max temporary reg"); + return 0; + } + src = UREG(REG_TYPE_R, index); + break; + case TGSI_FILE_INPUT: + /* XXX: Packing COL1, FOGC into a single attribute works for + * texenv programs, but will fail for real fragment programs + * that use these attributes and expect them to be a full 4 + * components wide. Could use a texcoord to pass these + * attributes if necessary, but that won't work in the general + * case. + * + * We also use a texture coordinate to pass wpos when possible. + */ + + /* use vertex format info to map a slot number to a VF attrib */ + assert(index < p->vertex_info->num_attribs); + + sem_name = p->input_semantic_name[index]; + sem_ind = p->input_semantic_index[index]; + + switch (sem_name) { + case TGSI_SEMANTIC_POSITION: + debug_printf("SKIP SEM POS\n"); + /* + assert(p->wpos_tex != -1); + src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL); + */ + break; + case TGSI_SEMANTIC_COLOR: + if (sem_ind == 0) { + src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL); + } + else { + /* secondary color */ + assert(sem_ind == 1); + src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ); + src = swizzle(src, X, Y, Z, ONE); + } + break; + case TGSI_SEMANTIC_FOG: + src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W); + src = swizzle(src, W, W, W, W); + break; + case TGSI_SEMANTIC_GENERIC: + /* usually a texcoord */ + src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + sem_ind, D0_CHANNEL_ALL); + break; + default: + i915_program_error(p, "Bad source->Index"); + return 0; + } + break; + + case TGSI_FILE_IMMEDIATE: + /* XXX unfinished - need to append immediates onto const buffer */ + /* fall-through */ + case TGSI_FILE_CONSTANT: + src = UREG(REG_TYPE_CONST, index); + break; + + default: + i915_program_error(p, "Bad source->File"); + return 0; + } + + if (source->SrcRegister.Extended) { + src = swizzle(src, + source->SrcRegisterExtSwz.ExtSwizzleX, + source->SrcRegisterExtSwz.ExtSwizzleY, + source->SrcRegisterExtSwz.ExtSwizzleZ, + source->SrcRegisterExtSwz.ExtSwizzleW); + } + else { + src = swizzle(src, + source->SrcRegister.SwizzleX, + source->SrcRegister.SwizzleY, + source->SrcRegister.SwizzleZ, + source->SrcRegister.SwizzleW); + } + + + /* There's both negate-all-components and per-component negation. + * Try to handle both here. + */ + { + int nx = source->SrcRegisterExtSwz.NegateX; + int ny = source->SrcRegisterExtSwz.NegateY; + int nz = source->SrcRegisterExtSwz.NegateZ; + int nw = source->SrcRegisterExtSwz.NegateW; + if (source->SrcRegister.Negate) { + nx = !nx; + ny = !ny; + nz = !nz; + nw = !nw; + } + src = negate(src, nx, ny, nz, nw); + } + + /* no abs() or post-abs negation */ +#if 0 + /* XXX assertions disabled to allow arbfplight.c to run */ + /* XXX enable these assertions, or fix things */ + assert(!source->SrcRegisterExtMod.Absolute); + assert(!source->SrcRegisterExtMod.Negate); +#endif + return src; +} + + +/** + * Construct a ureg for a destination register. + */ +static uint +get_result_vector(struct i915_fp_compile *p, + const struct tgsi_full_dst_register *dest) +{ + switch (dest->DstRegister.File) { + case TGSI_FILE_OUTPUT: + { + uint sem_name = p->output_semantic_name[dest->DstRegister.Index]; + switch (sem_name) { + case TGSI_SEMANTIC_POSITION: + return UREG(REG_TYPE_OD, 0); + case TGSI_SEMANTIC_COLOR: + return UREG(REG_TYPE_OC, 0); + default: + i915_program_error(p, "Bad inst->DstReg.Index/semantics"); + return 0; + } + } + case TGSI_FILE_TEMPORARY: + return UREG(REG_TYPE_R, dest->DstRegister.Index); + default: + i915_program_error(p, "Bad inst->DstReg.File"); + return 0; + } +} + + +/** + * Compute flags for saturation and writemask. + */ +static uint +get_result_flags(const struct tgsi_full_instruction *inst) +{ + const uint writeMask + = inst->FullDstRegisters[0].DstRegister.WriteMask; + uint flags = 0x0; + + if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE) + flags |= A0_DEST_SATURATE; + + if (writeMask & TGSI_WRITEMASK_X) + flags |= A0_DEST_CHANNEL_X; + if (writeMask & TGSI_WRITEMASK_Y) + flags |= A0_DEST_CHANNEL_Y; + if (writeMask & TGSI_WRITEMASK_Z) + flags |= A0_DEST_CHANNEL_Z; + if (writeMask & TGSI_WRITEMASK_W) + flags |= A0_DEST_CHANNEL_W; + + return flags; +} + + +/** + * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token + */ +static uint +translate_tex_src_target(struct i915_fp_compile *p, uint tex) +{ + switch (tex) { + case TGSI_TEXTURE_1D: + return D0_SAMPLE_TYPE_2D; + case TGSI_TEXTURE_2D: + return D0_SAMPLE_TYPE_2D; + case TGSI_TEXTURE_RECT: + return D0_SAMPLE_TYPE_2D; + case TGSI_TEXTURE_3D: + return D0_SAMPLE_TYPE_VOLUME; + case TGSI_TEXTURE_CUBE: + return D0_SAMPLE_TYPE_CUBE; + default: + i915_program_error(p, "TexSrc type"); + return 0; + } +} + + +/** + * Generate texel lookup instruction. + */ +static void +emit_tex(struct i915_fp_compile *p, + const struct tgsi_full_instruction *inst, + uint opcode) +{ + uint texture = inst->InstructionExtTexture.Texture; + uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; + uint tex = translate_tex_src_target( p, texture ); + uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex); + uint coord = src_vector( p, &inst->FullSrcRegisters[0]); + + i915_emit_texld( p, + get_result_vector( p, &inst->FullDstRegisters[0] ), + get_result_flags( inst ), + sampler, + coord, + opcode); +} + + +/** + * Generate a simple arithmetic instruction + * \param opcode the i915 opcode + * \param numArgs the number of input/src arguments + */ +static void +emit_simple_arith(struct i915_fp_compile *p, + const struct tgsi_full_instruction *inst, + uint opcode, uint numArgs) +{ + uint arg1, arg2, arg3; + + assert(numArgs <= 3); + + arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->FullSrcRegisters[0] ); + arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->FullSrcRegisters[1] ); + arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->FullSrcRegisters[2] ); + + i915_emit_arith( p, + opcode, + get_result_vector( p, &inst->FullDstRegisters[0]), + get_result_flags( inst ), 0, + arg1, + arg2, + arg3 ); +} + +#ifndef M_PI +#define M_PI 3.14159265358979323846 +#endif + +/* + * Translate TGSI instruction to i915 instruction. + * + * Possible concerns: + * + * SIN, COS -- could use another taylor step? + * LIT -- results seem a little different to sw mesa + * LOG -- different to mesa on negative numbers, but this is conformant. + */ +static void +i915_translate_instruction(struct i915_fp_compile *p, + const struct tgsi_full_instruction *inst) +{ + uint writemask; + uint src0, src1, src2, flags; + uint tmp = 0; + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_ABS: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + i915_emit_arith(p, + A0_MAX, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + src0, negate(src0, 1, 1, 1, 1), 0); + break; + + case TGSI_OPCODE_ADD: + emit_simple_arith(p, inst, A0_ADD, 2); + break; + + case TGSI_OPCODE_CMP: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + src2 = src_vector(p, &inst->FullSrcRegisters[2]); + i915_emit_arith(p, A0_CMP, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), + 0, src0, src2, src1); /* NOTE: order of src2, src1 */ + break; + + case TGSI_OPCODE_COS: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_X, 0, + src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0); + + i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0); + + /* By choosing different taylor constants, could get rid of this mul: + */ + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_X, 0, + tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0); + + /* + * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1 + * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1 + * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1 + * result = DP4 t0, cos_constants + */ + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XY, 0, + swizzle(tmp, X, X, ONE, ONE), + swizzle(tmp, X, ONE, ONE, ONE), 0); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XYZ, 0, + swizzle(tmp, X, Y, X, ONE), + swizzle(tmp, X, X, ONE, ONE), 0); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XYZ, 0, + swizzle(tmp, X, X, Z, ONE), + swizzle(tmp, Z, ONE, ONE, ONE), 0); + + i915_emit_arith(p, + A0_DP4, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(tmp, ONE, Z, Y, X), + i915_emit_const4fv(p, cos_constants), 0); + break; + + case TGSI_OPCODE_DP3: + emit_simple_arith(p, inst, A0_DP3, 2); + break; + + case TGSI_OPCODE_DP4: + emit_simple_arith(p, inst, A0_DP4, 2); + break; + + case TGSI_OPCODE_DPH: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + + i915_emit_arith(p, + A0_DP4, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, X, Y, Z, ONE), src1, 0); + break; + + case TGSI_OPCODE_DST: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + + /* result[0] = 1 * 1; + * result[1] = a[1] * b[1]; + * result[2] = a[2] * 1; + * result[3] = 1 * b[3]; + */ + i915_emit_arith(p, + A0_MUL, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, ONE, Y, Z, ONE), + swizzle(src1, ONE, Y, ONE, W), 0); + break; + + case TGSI_OPCODE_END: + /* no-op */ + break; + + case TGSI_OPCODE_EX2: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + + i915_emit_arith(p, + A0_EXP, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, X, X, X, X), 0, 0); + break; + + case TGSI_OPCODE_FLR: + emit_simple_arith(p, inst, A0_FLR, 1); + break; + + case TGSI_OPCODE_FRC: + emit_simple_arith(p, inst, A0_FRC, 1); + break; + + case TGSI_OPCODE_KIL: + /* unconditional kill */ + assert(0); /* not tested yet */ +#if 0 + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */ + 0, src0, T0_TEXKILL); +#endif + break; + + case TGSI_OPCODE_KILP: + /* kill if src[0].x < 0 || src[0].y < 0 ... */ + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */ + 0, src0, T0_TEXKILL); + break; + + case TGSI_OPCODE_LG2: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + + i915_emit_arith(p, + A0_LOG, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, X, X, X, X), 0, 0); + break; + + case TGSI_OPCODE_LIT: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + /* tmp = max( a.xyzw, a.00zw ) + * XXX: Clamp tmp.w to -128..128 + * tmp.y = log(tmp.y) + * tmp.y = tmp.w * tmp.y + * tmp.y = exp(tmp.y) + * result = cmp (a.11-x1, a.1x01, a.1xy1 ) + */ + i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0, + src0, swizzle(src0, ZERO, ZERO, Z, W), 0); + + i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0, + swizzle(tmp, Y, Y, Y, Y), 0, 0); + + i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0, + swizzle(tmp, ZERO, Y, ZERO, ZERO), + swizzle(tmp, ZERO, W, ZERO, ZERO), 0); + + i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0, + swizzle(tmp, Y, Y, Y, Y), 0, 0); + + i915_emit_arith(p, A0_CMP, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0), + swizzle(tmp, ONE, X, ZERO, ONE), + swizzle(tmp, ONE, X, Y, ONE)); + + break; + + case TGSI_OPCODE_LRP: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + src2 = src_vector(p, &inst->FullSrcRegisters[2]); + flags = get_result_flags(inst); + tmp = i915_get_utemp(p); + + /* b*a + c*(1-a) + * + * b*a + c - ca + * + * tmp = b*a + c, + * result = (-c)*a + tmp + */ + i915_emit_arith(p, A0_MAD, tmp, + flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2); + + i915_emit_arith(p, A0_MAD, + get_result_vector(p, &inst->FullDstRegisters[0]), + flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp); + break; + + case TGSI_OPCODE_MAD: + emit_simple_arith(p, inst, A0_MAD, 3); + break; + + case TGSI_OPCODE_MAX: + emit_simple_arith(p, inst, A0_MAX, 2); + break; + + case TGSI_OPCODE_MIN: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + tmp = i915_get_utemp(p); + flags = get_result_flags(inst); + + i915_emit_arith(p, + A0_MAX, + tmp, flags & A0_DEST_CHANNEL_ALL, 0, + negate(src0, 1, 1, 1, 1), + negate(src1, 1, 1, 1, 1), 0); + + i915_emit_arith(p, + A0_MOV, + get_result_vector(p, &inst->FullDstRegisters[0]), + flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0); + break; + + case TGSI_OPCODE_MOV: + /* aka TGSI_OPCODE_SWZ */ + emit_simple_arith(p, inst, A0_MOV, 1); + break; + + case TGSI_OPCODE_MUL: + emit_simple_arith(p, inst, A0_MUL, 2); + break; + + case TGSI_OPCODE_POW: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + tmp = i915_get_utemp(p); + flags = get_result_flags(inst); + + /* XXX: masking on intermediate values, here and elsewhere. + */ + i915_emit_arith(p, + A0_LOG, + tmp, A0_DEST_CHANNEL_X, 0, + swizzle(src0, X, X, X, X), 0, 0); + + i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0); + + i915_emit_arith(p, + A0_EXP, + get_result_vector(p, &inst->FullDstRegisters[0]), + flags, 0, swizzle(tmp, X, X, X, X), 0, 0); + break; + + case TGSI_OPCODE_RET: + /* XXX: no-op? */ + break; + + case TGSI_OPCODE_RCP: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + + i915_emit_arith(p, + A0_RCP, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, X, X, X, X), 0, 0); + break; + + case TGSI_OPCODE_RSQ: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + + i915_emit_arith(p, + A0_RSQ, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, X, X, X, X), 0, 0); + break; + + case TGSI_OPCODE_SCS: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + /* + * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1 + * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x + * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x + * scs.x = DP4 t1, sin_constants + * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1 + * scs.y = DP4 t1, cos_constants + */ + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XY, 0, + swizzle(src0, X, X, ONE, ONE), + swizzle(src0, X, ONE, ONE, ONE), 0); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_ALL, 0, + swizzle(tmp, X, Y, X, Y), + swizzle(tmp, X, X, ONE, ONE), 0); + + writemask = inst->FullDstRegisters[0].DstRegister.WriteMask; + + if (writemask & TGSI_WRITEMASK_Y) { + uint tmp1; + + if (writemask & TGSI_WRITEMASK_X) + tmp1 = i915_get_utemp(p); + else + tmp1 = tmp; + + i915_emit_arith(p, + A0_MUL, + tmp1, A0_DEST_CHANNEL_ALL, 0, + swizzle(tmp, X, Y, Y, W), + swizzle(tmp, X, Z, ONE, ONE), 0); + + i915_emit_arith(p, + A0_DP4, + get_result_vector(p, &inst->FullDstRegisters[0]), + A0_DEST_CHANNEL_Y, 0, + swizzle(tmp1, W, Z, Y, X), + i915_emit_const4fv(p, sin_constants), 0); + } + + if (writemask & TGSI_WRITEMASK_X) { + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XYZ, 0, + swizzle(tmp, X, X, Z, ONE), + swizzle(tmp, Z, ONE, ONE, ONE), 0); + + i915_emit_arith(p, + A0_DP4, + get_result_vector(p, &inst->FullDstRegisters[0]), + A0_DEST_CHANNEL_X, 0, + swizzle(tmp, ONE, Z, Y, X), + i915_emit_const4fv(p, cos_constants), 0); + } + break; + + case TGSI_OPCODE_SGE: + emit_simple_arith(p, inst, A0_SGE, 2); + break; + + case TGSI_OPCODE_SIN: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + tmp = i915_get_utemp(p); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_X, 0, + src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0); + + i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0); + + /* By choosing different taylor constants, could get rid of this mul: + */ + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_X, 0, + tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0); + + /* + * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1 + * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x + * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x + * result = DP4 t1.wzyx, sin_constants + */ + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_XY, 0, + swizzle(tmp, X, X, ONE, ONE), + swizzle(tmp, X, ONE, ONE, ONE), 0); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_ALL, 0, + swizzle(tmp, X, Y, X, Y), + swizzle(tmp, X, X, ONE, ONE), 0); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_ALL, 0, + swizzle(tmp, X, Y, Y, W), + swizzle(tmp, X, Z, ONE, ONE), 0); + + i915_emit_arith(p, + A0_DP4, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(tmp, W, Z, Y, X), + i915_emit_const4fv(p, sin_constants), 0); + break; + + case TGSI_OPCODE_SLT: + emit_simple_arith(p, inst, A0_SLT, 2); + break; + + case TGSI_OPCODE_SUB: + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + + i915_emit_arith(p, + A0_ADD, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + src0, negate(src1, 1, 1, 1, 1), 0); + break; + + case TGSI_OPCODE_TEX: + if (inst->FullSrcRegisters[0].SrcRegisterExtSwz.ExtDivide + == TGSI_EXTSWIZZLE_W) { + emit_tex(p, inst, T0_TEXLDP); + } + else { + emit_tex(p, inst, T0_TEXLD); + } + break; + + case TGSI_OPCODE_TXB: + emit_tex(p, inst, T0_TEXLDB); + break; + + case TGSI_OPCODE_XPD: + /* Cross product: + * result.x = src0.y * src1.z - src0.z * src1.y; + * result.y = src0.z * src1.x - src0.x * src1.z; + * result.z = src0.x * src1.y - src0.y * src1.x; + * result.w = undef; + */ + src0 = src_vector(p, &inst->FullSrcRegisters[0]); + src1 = src_vector(p, &inst->FullSrcRegisters[1]); + tmp = i915_get_utemp(p); + + i915_emit_arith(p, + A0_MUL, + tmp, A0_DEST_CHANNEL_ALL, 0, + swizzle(src0, Z, X, Y, ONE), + swizzle(src1, Y, Z, X, ONE), 0); + + i915_emit_arith(p, + A0_MAD, + get_result_vector(p, &inst->FullDstRegisters[0]), + get_result_flags(inst), 0, + swizzle(src0, Y, Z, X, ONE), + swizzle(src1, Z, X, Y, ONE), + negate(tmp, 1, 1, 1, 0)); + break; + + default: + i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode); + return; + } + + i915_release_utemps(p); +} + + +/** + * Translate TGSI fragment shader into i915 hardware instructions. + * \param p the translation state + * \param tokens the TGSI token array + */ +static void +i915_translate_instructions(struct i915_fp_compile *p, + const struct tgsi_token *tokens) +{ + struct tgsi_parse_context parse; + + tgsi_parse_init( &parse, tokens ); + + while( !tgsi_parse_end_of_tokens( &parse ) ) { + + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + if (parse.FullToken.FullDeclaration.Declaration.File + == TGSI_FILE_INPUT) { + /* save input register info for use in src_vector() */ + uint ind, sem, semi; + ind = parse.FullToken.FullDeclaration.u.DeclarationRange.First; + sem = parse.FullToken.FullDeclaration.Semantic.SemanticName; + semi = parse.FullToken.FullDeclaration.Semantic.SemanticIndex; + /*debug_printf("FS Input DECL [%u] sem %u\n", ind, sem);*/ + p->input_semantic_name[ind] = sem; + p->input_semantic_index[ind] = semi; + } + else if (parse.FullToken.FullDeclaration.Declaration.File + == TGSI_FILE_OUTPUT) { + /* save output register info for use in get_result_vector() */ + uint ind, sem, semi; + ind = parse.FullToken.FullDeclaration.u.DeclarationRange.First; + sem = parse.FullToken.FullDeclaration.Semantic.SemanticName; + semi = parse.FullToken.FullDeclaration.Semantic.SemanticIndex; + /*debug_printf("FS Output DECL [%u] sem %u\n", ind, sem);*/ + p->output_semantic_name[ind] = sem; + p->output_semantic_index[ind] = semi; + } + break; + + case TGSI_TOKEN_TYPE_IMMEDIATE: + /* XXX append the immediate to the const buffer... */ + break; + + case TGSI_TOKEN_TYPE_INSTRUCTION: + i915_translate_instruction(p, &parse.FullToken.FullInstruction); + break; + + default: + assert( 0 ); + } + + } /* while */ + + tgsi_parse_free (&parse); +} + + +static struct i915_fp_compile * +i915_init_compile(struct i915_context *i915, + const struct pipe_shader_state *fs) +{ + struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile); + + p->shader = i915->fs; + + p->vertex_info = &i915->current.vertex_info; + + /* new constants found during translation get appended after the + * user-provided constants. + */ + p->constants = i915->current.constants[PIPE_SHADER_FRAGMENT]; + p->num_constants = i915->current.num_user_constants[PIPE_SHADER_FRAGMENT]; + + p->nr_tex_indirect = 1; /* correct? */ + p->nr_tex_insn = 0; + p->nr_alu_insn = 0; + p->nr_decl_insn = 0; + + memset(p->constant_flags, 0, sizeof(p->constant_flags)); + + p->csr = p->program; + p->decl = p->declarations; + p->decl_s = 0; + p->decl_t = 0; + p->temp_flag = 0xffff000; + p->utemp_flag = ~0x7; + + p->wpos_tex = -1; + + /* initialize the first program word */ + *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM; + + return p; +} + + +/* Copy compile results to the fragment program struct and destroy the + * compilation context. + */ +static void +i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p) +{ + unsigned long program_size = (unsigned long) (p->csr - p->program); + unsigned long decl_size = (unsigned long) (p->decl - p->declarations); + + if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT) + i915_program_error(p, "Exceeded max nr indirect texture lookups"); + + if (p->nr_tex_insn > I915_MAX_TEX_INSN) + i915_program_error(p, "Exceeded max TEX instructions"); + + if (p->nr_alu_insn > I915_MAX_ALU_INSN) + i915_program_error(p, "Exceeded max ALU instructions"); + + if (p->nr_decl_insn > I915_MAX_DECL_INSN) + i915_program_error(p, "Exceeded max DECL instructions"); + + /* free old program, if present */ + if (i915->current.program) { + FREE(i915->current.program); + i915->current.program_len = 0; + } + + if (p->error) { + p->NumNativeInstructions = 0; + p->NumNativeAluInstructions = 0; + p->NumNativeTexInstructions = 0; + p->NumNativeTexIndirections = 0; + + i915_use_passthrough_shader(i915); + } + else { + p->NumNativeInstructions + = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn; + p->NumNativeAluInstructions = p->nr_alu_insn; + p->NumNativeTexInstructions = p->nr_tex_insn; + p->NumNativeTexIndirections = p->nr_tex_indirect; + + /* patch in the program length */ + p->declarations[0] |= program_size + decl_size - 2; + + /* Copy compilation results to fragment program struct: + */ + i915->current.program + = (uint *) MALLOC((program_size + decl_size) * sizeof(uint)); + if (i915->current.program) { + i915->current.program_len = program_size + decl_size; + + memcpy(i915->current.program, + p->declarations, + decl_size * sizeof(uint)); + + memcpy(i915->current.program + decl_size, + p->program, + program_size * sizeof(uint)); + } + + /* update number of constants */ + i915->current.num_constants[PIPE_SHADER_FRAGMENT] = p->num_constants; + assert(i915->current.num_constants[PIPE_SHADER_FRAGMENT] + >= i915->current.num_user_constants[PIPE_SHADER_FRAGMENT]); + } + + /* Release the compilation struct: + */ + FREE(p); +} + + +/** + * Find an unused texture coordinate slot to use for fragment WPOS. + * Update p->fp->wpos_tex with the result (-1 if no used texcoord slot is found). + */ +static void +i915_find_wpos_space(struct i915_fp_compile *p) +{ +#if 0 + const uint inputs + = p->shader->inputs_read | (1 << TGSI_ATTRIB_POS); /*XXX hack*/ + uint i; + + p->wpos_tex = -1; + + if (inputs & (1 << TGSI_ATTRIB_POS)) { + for (i = 0; i < I915_TEX_UNITS; i++) { + if ((inputs & (1 << (TGSI_ATTRIB_TEX0 + i))) == 0) { + p->wpos_tex = i; + return; + } + } + + i915_program_error(p, "No free texcoord for wpos value"); + } +#else + if (p->shader->input_semantic_name[0] == TGSI_SEMANTIC_POSITION) { + /* frag shader using the fragment position input */ +#if 0 + assert(0); +#endif + } +#endif +} + + + + +/** + * Rather than trying to intercept and jiggle depth writes during + * emit, just move the value into its correct position at the end of + * the program: + */ +static void +i915_fixup_depth_write(struct i915_fp_compile *p) +{ + /* XXX assuming pos/depth is always in output[0] */ + if (p->shader->output_semantic_name[0] == TGSI_SEMANTIC_POSITION) { + const uint depth = UREG(REG_TYPE_OD, 0); + + i915_emit_arith(p, + A0_MOV, /* opcode */ + depth, /* dest reg */ + A0_DEST_CHANNEL_W, /* write mask */ + 0, /* saturate? */ + swizzle(depth, X, Y, Z, Z), /* src0 */ + 0, 0 /* src1, src2 */); + } +} + + +void +i915_translate_fragment_program( struct i915_context *i915 ) +{ + struct i915_fp_compile *p = i915_init_compile(i915, i915->fs); + const struct tgsi_token *tokens = i915->fs->tokens; + + i915_find_wpos_space(p); + + i915_translate_instructions(p, tokens); + i915_fixup_depth_write(p); + + i915_fini_compile(i915, p); +} diff --git a/src/gallium/drivers/i915simple/i915_prim_emit.c b/src/gallium/drivers/i915simple/i915_prim_emit.c new file mode 100644 index 0000000000..c4a706c37d --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_prim_emit.c @@ -0,0 +1,215 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/draw/draw_private.h" +#include "pipe/p_util.h" + +#include "i915_context.h" +#include "i915_winsys.h" +#include "i915_reg.h" +#include "i915_state.h" +#include "i915_batch.h" + + + +/** + * Primitive emit to hardware. No support for vertex buffers or any + * nice fast paths. + */ +struct setup_stage { + struct draw_stage stage; /**< This must be first (base class) */ + + struct i915_context *i915; +}; + + + +/** + * Basically a cast wrapper. + */ +static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) +{ + return (struct setup_stage *)stage; +} + + +/** + * Extract the needed fields from vertex_header and emit i915 dwords. + * Recall that the vertices are constructed by the 'draw' module and + * have a couple of slots at the beginning (1-dword header, 4-dword + * clip pos) that we ignore here. + */ +static INLINE void +emit_hw_vertex( struct i915_context *i915, + const struct vertex_header *vertex) +{ + const struct vertex_info *vinfo = &i915->current.vertex_info; + uint i; + uint count = 0; /* for debug/sanity */ + + for (i = 0; i < vinfo->num_attribs; i++) { + switch (vinfo->emit[i]) { + case EMIT_OMIT: + /* no-op */ + break; + case EMIT_1F: + OUT_BATCH( fui(vertex->data[i][0]) ); + count++; + break; + case EMIT_2F: + OUT_BATCH( fui(vertex->data[i][0]) ); + OUT_BATCH( fui(vertex->data[i][1]) ); + count += 2; + break; + case EMIT_3F: + OUT_BATCH( fui(vertex->data[i][0]) ); + OUT_BATCH( fui(vertex->data[i][1]) ); + OUT_BATCH( fui(vertex->data[i][2]) ); + count += 3; + break; + case EMIT_4F: + OUT_BATCH( fui(vertex->data[i][0]) ); + OUT_BATCH( fui(vertex->data[i][1]) ); + OUT_BATCH( fui(vertex->data[i][2]) ); + OUT_BATCH( fui(vertex->data[i][3]) ); + count += 4; + break; + case EMIT_4UB: + OUT_BATCH( pack_ub4(float_to_ubyte( vertex->data[i][2] ), + float_to_ubyte( vertex->data[i][1] ), + float_to_ubyte( vertex->data[i][0] ), + float_to_ubyte( vertex->data[i][3] )) ); + count += 1; + break; + default: + assert(0); + } + } + assert(count == vinfo->size); +} + + + +static INLINE void +emit_prim( struct draw_stage *stage, + struct prim_header *prim, + unsigned hwprim, + unsigned nr ) +{ + struct i915_context *i915 = setup_stage(stage)->i915; + unsigned vertex_size = i915->current.vertex_info.size * 4; /* in bytes */ + unsigned i; + + assert(vertex_size >= 12); /* never smaller than 12 bytes */ + + if (i915->dirty) + i915_update_derived( i915 ); + + if (i915->hardware_dirty) + i915_emit_hardware_state( i915 ); + + if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) { + FLUSH_BATCH(); + + /* Make sure state is re-emitted after a flush: + */ + i915_update_derived( i915 ); + i915_emit_hardware_state( i915 ); + + if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) { + assert(0); + return; + } + } + + /* Emit each triangle as a single primitive. I told you this was + * simple. + */ + OUT_BATCH(_3DPRIMITIVE | + hwprim | + ((4 + vertex_size * nr)/4 - 2)); + + for (i = 0; i < nr; i++) + emit_hw_vertex(i915, prim->v[i]); +} + + +static void +setup_tri( struct draw_stage *stage, struct prim_header *prim ) +{ + emit_prim( stage, prim, PRIM3D_TRILIST, 3 ); +} + + +static void +setup_line(struct draw_stage *stage, struct prim_header *prim) +{ + emit_prim( stage, prim, PRIM3D_LINELIST, 2 ); +} + + +static void +setup_point(struct draw_stage *stage, struct prim_header *prim) +{ + emit_prim( stage, prim, PRIM3D_POINTLIST, 1 ); +} + + +static void setup_flush( struct draw_stage *stage, unsigned flags ) +{ +} + +static void reset_stipple_counter( struct draw_stage *stage ) +{ +} + +static void render_destroy( struct draw_stage *stage ) +{ + FREE( stage ); +} + + +/** + * Create a new primitive setup/render stage. This gets plugged into + * the 'draw' module's pipeline. + */ +struct draw_stage *i915_draw_render_stage( struct i915_context *i915 ) +{ + struct setup_stage *setup = CALLOC_STRUCT(setup_stage); + + setup->i915 = i915; + setup->stage.draw = i915->draw; + setup->stage.point = setup_point; + setup->stage.line = setup_line; + setup->stage.tri = setup_tri; + setup->stage.flush = setup_flush; + setup->stage.reset_stipple_counter = reset_stipple_counter; + setup->stage.destroy = render_destroy; + + return &setup->stage; +} diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c new file mode 100644 index 0000000000..e069773fd4 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c @@ -0,0 +1,254 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \file + * Build post-transformation, post-clipping vertex buffers and element + * lists by hooking into the end of the primitive pipeline and + * manipulating the vertex_id field in the vertex headers. + * + * XXX: work in progress + * + * \author José Fonseca <jrfonseca@tungstengraphics.com> + * \author Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/draw/draw_vbuf.h" +#include "pipe/p_debug.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" + +#include "i915_context.h" +#include "i915_reg.h" +#include "i915_winsys.h" +#include "i915_batch.h" +#include "i915_state.h" + + +/** + * Primitive renderer for i915. + */ +struct i915_vbuf_render { + struct vbuf_render base; + + struct i915_context *i915; + + /** Vertex size in bytes */ + unsigned vertex_size; + + /** Hardware primitive */ + unsigned hwprim; +}; + + +/** + * Basically a cast wrapper. + */ +static INLINE struct i915_vbuf_render * +i915_vbuf_render( struct vbuf_render *render ) +{ + assert(render); + return (struct i915_vbuf_render *)render; +} + + +static const struct vertex_info * +i915_vbuf_render_get_vertex_info( struct vbuf_render *render ) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + return &i915->current.vertex_info; +} + + +static void * +i915_vbuf_render_allocate_vertices( struct vbuf_render *render, + ushort vertex_size, + ushort nr_vertices ) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + struct pipe_winsys *winsys = i915->pipe.winsys; + size_t size = (size_t)vertex_size * (size_t)nr_vertices; + + /* FIXME: handle failure */ + assert(!i915->vbo); + i915->vbo = winsys->buffer_create(winsys, 64, I915_BUFFER_USAGE_LIT_VERTEX, + size); + + i915->dirty |= I915_NEW_VBO; + + return winsys->buffer_map(winsys, + i915->vbo, + PIPE_BUFFER_USAGE_CPU_WRITE); +} + + +static void +i915_vbuf_render_set_primitive( struct vbuf_render *render, + unsigned prim ) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + + switch(prim) { + case PIPE_PRIM_POINTS: + i915_render->hwprim = PRIM3D_POINTLIST; + break; + case PIPE_PRIM_LINES: + i915_render->hwprim = PRIM3D_LINELIST; + break; + case PIPE_PRIM_TRIANGLES: + i915_render->hwprim = PRIM3D_TRILIST; + break; + default: + assert(0); + } +} + + +static void +i915_vbuf_render_draw( struct vbuf_render *render, + const ushort *indices, + uint nr_indices) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + unsigned i; + + assert(nr_indices); + + assert((i915->dirty & ~I915_NEW_VBO) == 0); + + if (i915->dirty) + i915_update_derived( i915 ); + + if (i915->hardware_dirty) + i915_emit_hardware_state( i915 ); + + if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { + FLUSH_BATCH(); + + /* Make sure state is re-emitted after a flush: + */ + i915_update_derived( i915 ); + i915_emit_hardware_state( i915 ); + + if (!BEGIN_BATCH( 1 + (nr_indices + 1)/2, 1 )) { + assert(0); + return; + } + } + + OUT_BATCH( _3DPRIMITIVE | + PRIM_INDIRECT | + i915_render->hwprim | + PRIM_INDIRECT_ELTS | + nr_indices ); + for (i = 0; i + 1 < nr_indices; i += 2) { + OUT_BATCH( indices[i] | + (indices[i + 1] << 16) ); + } + if (i < nr_indices) { + OUT_BATCH( indices[i] ); + } +} + + +static void +i915_vbuf_render_release_vertices( struct vbuf_render *render, + void *vertices, + unsigned vertex_size, + unsigned vertices_used ) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + struct i915_context *i915 = i915_render->i915; + struct pipe_winsys *winsys = i915->pipe.winsys; + + assert(i915->vbo); + winsys->buffer_unmap(winsys, i915->vbo); + pipe_buffer_reference(winsys, &i915->vbo, NULL); +} + + +static void +i915_vbuf_render_destroy( struct vbuf_render *render ) +{ + struct i915_vbuf_render *i915_render = i915_vbuf_render(render); + FREE(i915_render); +} + + +/** + * Create a new primitive render. + */ +static struct vbuf_render * +i915_vbuf_render_create( struct i915_context *i915 ) +{ + struct i915_vbuf_render *i915_render = CALLOC_STRUCT(i915_vbuf_render); + + i915_render->i915 = i915; + + i915_render->base.max_vertex_buffer_bytes = 128*1024; + + /* NOTE: it must be such that state and vertices indices fit in a single + * batch buffer. + */ + i915_render->base.max_indices = 16*1024; + + i915_render->base.get_vertex_info = i915_vbuf_render_get_vertex_info; + i915_render->base.allocate_vertices = i915_vbuf_render_allocate_vertices; + i915_render->base.set_primitive = i915_vbuf_render_set_primitive; + i915_render->base.draw = i915_vbuf_render_draw; + i915_render->base.release_vertices = i915_vbuf_render_release_vertices; + i915_render->base.destroy = i915_vbuf_render_destroy; + + return &i915_render->base; +} + + +/** + * Create a new primitive vbuf/render stage. + */ +struct draw_stage *i915_draw_vbuf_stage( struct i915_context *i915 ) +{ + struct vbuf_render *render; + struct draw_stage *stage; + + render = i915_vbuf_render_create(i915); + if(!render) + return NULL; + + stage = draw_vbuf_stage( i915->draw, render ); + if(!stage) { + render->destroy(render); + return NULL; + } + + return stage; +} diff --git a/src/gallium/drivers/i915simple/i915_reg.h b/src/gallium/drivers/i915simple/i915_reg.h new file mode 100644 index 0000000000..04620fec68 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_reg.h @@ -0,0 +1,978 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef I915_REG_H +#define I915_REG_H + + +#define I915_SET_FIELD( var, mask, value ) (var &= ~(mask), var |= value) + +#define CMD_3D (0x3<<29) + +#define PRIM3D_INLINE (CMD_3D | (0x1f<<24)) +#define PRIM3D_TRILIST (0x0<<18) +#define PRIM3D_TRISTRIP (0x1<<18) +#define PRIM3D_TRISTRIP_RVRSE (0x2<<18) +#define PRIM3D_TRIFAN (0x3<<18) +#define PRIM3D_POLY (0x4<<18) +#define PRIM3D_LINELIST (0x5<<18) +#define PRIM3D_LINESTRIP (0x6<<18) +#define PRIM3D_RECTLIST (0x7<<18) +#define PRIM3D_POINTLIST (0x8<<18) +#define PRIM3D_DIB (0x9<<18) +#define PRIM3D_CLEAR_RECT (0xa<<18) +#define PRIM3D_ZONE_INIT (0xd<<18) +#define PRIM3D_MASK (0x1f<<18) + +/* p137 */ +#define _3DSTATE_AA_CMD (CMD_3D | (0x06<<24)) +#define AA_LINE_ECAAR_WIDTH_ENABLE (1<<16) +#define AA_LINE_ECAAR_WIDTH_0_5 0 +#define AA_LINE_ECAAR_WIDTH_1_0 (1<<14) +#define AA_LINE_ECAAR_WIDTH_2_0 (2<<14) +#define AA_LINE_ECAAR_WIDTH_4_0 (3<<14) +#define AA_LINE_REGION_WIDTH_ENABLE (1<<8) +#define AA_LINE_REGION_WIDTH_0_5 0 +#define AA_LINE_REGION_WIDTH_1_0 (1<<6) +#define AA_LINE_REGION_WIDTH_2_0 (2<<6) +#define AA_LINE_REGION_WIDTH_4_0 (3<<6) + +/* 3DSTATE_BACKFACE_STENCIL_OPS, p138*/ +#define _3DSTATE_BACKFACE_STENCIL_OPS (CMD_3D | (0x8<<24)) +#define BFO_ENABLE_STENCIL_REF (1<<23) +#define BFO_STENCIL_REF_SHIFT 15 +#define BFO_STENCIL_REF_MASK (0xff<<15) +#define BFO_ENABLE_STENCIL_FUNCS (1<<14) +#define BFO_STENCIL_TEST_SHIFT 11 +#define BFO_STENCIL_TEST_MASK (0x7<<11) +#define BFO_STENCIL_FAIL_SHIFT 8 +#define BFO_STENCIL_FAIL_MASK (0x7<<8) +#define BFO_STENCIL_PASS_Z_FAIL_SHIFT 5 +#define BFO_STENCIL_PASS_Z_FAIL_MASK (0x7<<5) +#define BFO_STENCIL_PASS_Z_PASS_SHIFT 2 +#define BFO_STENCIL_PASS_Z_PASS_MASK (0x7<<2) +#define BFO_ENABLE_STENCIL_TWO_SIDE (1<<1) +#define BFO_STENCIL_TWO_SIDE (1<<0) + + +/* 3DSTATE_BACKFACE_STENCIL_MASKS, p140 */ +#define _3DSTATE_BACKFACE_STENCIL_MASKS (CMD_3D | (0x9<<24)) +#define BFM_ENABLE_STENCIL_TEST_MASK (1<<17) +#define BFM_ENABLE_STENCIL_WRITE_MASK (1<<16) +#define BFM_STENCIL_TEST_MASK_SHIFT 8 +#define BFM_STENCIL_TEST_MASK_MASK (0xff<<8) +#define BFM_STENCIL_WRITE_MASK_SHIFT 0 +#define BFM_STENCIL_WRITE_MASK_MASK (0xff<<0) + + + +/* 3DSTATE_BIN_CONTROL p141 */ + +/* p143 */ +#define _3DSTATE_BUF_INFO_CMD (CMD_3D | (0x1d<<24) | (0x8e<<16) | 1) +/* Dword 1 */ +#define BUF_3D_ID_COLOR_BACK (0x3<<24) +#define BUF_3D_ID_DEPTH (0x7<<24) +#define BUF_3D_USE_FENCE (1<<23) +#define BUF_3D_TILED_SURFACE (1<<22) +#define BUF_3D_TILE_WALK_X 0 +#define BUF_3D_TILE_WALK_Y (1<<21) +#define BUF_3D_PITCH(x) (((x)/4)<<2) +/* Dword 2 */ +#define BUF_3D_ADDR(x) ((x) & ~0x3) + + +/* 3DSTATE_CHROMA_KEY */ + +/* 3DSTATE_CLEAR_PARAMETERS, p150 */ +#define _3DSTATE_CLEAR_PARAMETERS (CMD_3D | (0x1d<<24) | (0x9c<<16) | 5) +/* Dword 1 */ +#define CLEARPARAM_CLEAR_RECT (1 << 16) +#define CLEARPARAM_ZONE_INIT (0 << 16) +#define CLEARPARAM_WRITE_COLOR (1 << 2) +#define CLEARPARAM_WRITE_DEPTH (1 << 1) +#define CLEARPARAM_WRITE_STENCIL (1 << 0) + +/* 3DSTATE_CONSTANT_BLEND_COLOR, p153 */ +#define _3DSTATE_CONST_BLEND_COLOR_CMD (CMD_3D | (0x1d<<24) | (0x88<<16)) + + + +/* 3DSTATE_COORD_SET_BINDINGS, p154 */ +#define _3DSTATE_COORD_SET_BINDINGS (CMD_3D | (0x16<<24)) +#define CSB_TCB(iunit, eunit) ((eunit)<<(iunit*3)) + +/* p156 */ +#define _3DSTATE_DFLT_DIFFUSE_CMD (CMD_3D | (0x1d<<24) | (0x99<<16)) + +/* p157 */ +#define _3DSTATE_DFLT_SPEC_CMD (CMD_3D | (0x1d<<24) | (0x9a<<16)) + +/* p158 */ +#define _3DSTATE_DFLT_Z_CMD (CMD_3D | (0x1d<<24) | (0x98<<16)) + + +/* 3DSTATE_DEPTH_OFFSET_SCALE, p159 */ +#define _3DSTATE_DEPTH_OFFSET_SCALE (CMD_3D | (0x1d<<24) | (0x97<<16)) +/* scale in dword 1 */ + + +/* 3DSTATE_DEPTH_SUBRECT_DISABLE, p160 */ +#define _3DSTATE_DEPTH_SUBRECT_DISABLE (CMD_3D | (0x1c<<24) | (0x11<<19) | 0x2) + +/* p161 */ +#define _3DSTATE_DST_BUF_VARS_CMD (CMD_3D | (0x1d<<24) | (0x85<<16)) +/* Dword 1 */ +#define TEX_DEFAULT_COLOR_OGL (0<<30) +#define TEX_DEFAULT_COLOR_D3D (1<<30) +#define ZR_EARLY_DEPTH (1<<29) +#define LOD_PRECLAMP_OGL (1<<28) +#define LOD_PRECLAMP_D3D (0<<28) +#define DITHER_FULL_ALWAYS (0<<26) +#define DITHER_FULL_ON_FB_BLEND (1<<26) +#define DITHER_CLAMPED_ALWAYS (2<<26) +#define LINEAR_GAMMA_BLEND_32BPP (1<<25) +#define DEBUG_DISABLE_ENH_DITHER (1<<24) +#define DSTORG_HORT_BIAS(x) ((x)<<20) +#define DSTORG_VERT_BIAS(x) ((x)<<16) +#define COLOR_4_2_2_CHNL_WRT_ALL 0 +#define COLOR_4_2_2_CHNL_WRT_Y (1<<12) +#define COLOR_4_2_2_CHNL_WRT_CR (2<<12) +#define COLOR_4_2_2_CHNL_WRT_CB (3<<12) +#define COLOR_4_2_2_CHNL_WRT_CRCB (4<<12) +#define COLOR_BUF_8BIT 0 +#define COLOR_BUF_RGB555 (1<<8) +#define COLOR_BUF_RGB565 (2<<8) +#define COLOR_BUF_ARGB8888 (3<<8) +#define DEPTH_FRMT_16_FIXED 0 +#define DEPTH_FRMT_16_FLOAT (1<<2) +#define DEPTH_FRMT_24_FIXED_8_OTHER (2<<2) +#define VERT_LINE_STRIDE_1 (1<<1) +#define VERT_LINE_STRIDE_0 (0<<1) +#define VERT_LINE_STRIDE_OFS_1 1 +#define VERT_LINE_STRIDE_OFS_0 0 + +/* p166 */ +#define _3DSTATE_DRAW_RECT_CMD (CMD_3D|(0x1d<<24)|(0x80<<16)|3) +/* Dword 1 */ +#define DRAW_RECT_DIS_DEPTH_OFS (1<<30) +#define DRAW_DITHER_OFS_X(x) ((x)<<26) +#define DRAW_DITHER_OFS_Y(x) ((x)<<24) +/* Dword 2 */ +#define DRAW_YMIN(x) ((x)<<16) +#define DRAW_XMIN(x) (x) +/* Dword 3 */ +#define DRAW_YMAX(x) ((x)<<16) +#define DRAW_XMAX(x) (x) +/* Dword 4 */ +#define DRAW_YORG(x) ((x)<<16) +#define DRAW_XORG(x) (x) + + +/* 3DSTATE_FILTER_COEFFICIENTS_4X4, p170 */ + +/* 3DSTATE_FILTER_COEFFICIENTS_6X5, p172 */ + + +/* _3DSTATE_FOG_COLOR, p173 */ +#define _3DSTATE_FOG_COLOR_CMD (CMD_3D|(0x15<<24)) +#define FOG_COLOR_RED(x) ((x)<<16) +#define FOG_COLOR_GREEN(x) ((x)<<8) +#define FOG_COLOR_BLUE(x) (x) + +/* _3DSTATE_FOG_MODE, p174 */ +#define _3DSTATE_FOG_MODE_CMD (CMD_3D|(0x1d<<24)|(0x89<<16)|2) +/* Dword 1 */ +#define FMC1_FOGFUNC_MODIFY_ENABLE (1<<31) +#define FMC1_FOGFUNC_VERTEX (0<<28) +#define FMC1_FOGFUNC_PIXEL_EXP (1<<28) +#define FMC1_FOGFUNC_PIXEL_EXP2 (2<<28) +#define FMC1_FOGFUNC_PIXEL_LINEAR (3<<28) +#define FMC1_FOGFUNC_MASK (3<<28) +#define FMC1_FOGINDEX_MODIFY_ENABLE (1<<27) +#define FMC1_FOGINDEX_Z (0<<25) +#define FMC1_FOGINDEX_W (1<<25) +#define FMC1_C1_C2_MODIFY_ENABLE (1<<24) +#define FMC1_DENSITY_MODIFY_ENABLE (1<<23) +#define FMC1_C1_ONE (1<<13) +#define FMC1_C1_MASK (0xffff<<4) +/* Dword 2 */ +#define FMC2_C2_ONE (1<<16) +/* Dword 3 */ +#define FMC3_D_ONE (1<<16) + + + +/* _3DSTATE_INDEPENDENT_ALPHA_BLEND, p177 */ +#define _3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD (CMD_3D|(0x0b<<24)) +#define IAB_MODIFY_ENABLE (1<<23) +#define IAB_ENABLE (1<<22) +#define IAB_MODIFY_FUNC (1<<21) +#define IAB_FUNC_SHIFT 16 +#define IAB_MODIFY_SRC_FACTOR (1<<11) +#define IAB_SRC_FACTOR_SHIFT 6 +#define IAB_SRC_FACTOR_MASK (BLENDFACT_MASK<<6) +#define IAB_MODIFY_DST_FACTOR (1<<5) +#define IAB_DST_FACTOR_SHIFT 0 +#define IAB_DST_FACTOR_MASK (BLENDFACT_MASK<<0) + + +#define BLENDFUNC_ADD 0x0 +#define BLENDFUNC_SUBTRACT 0x1 +#define BLENDFUNC_REVERSE_SUBTRACT 0x2 +#define BLENDFUNC_MIN 0x3 +#define BLENDFUNC_MAX 0x4 +#define BLENDFUNC_MASK 0x7 + +/* 3DSTATE_LOAD_INDIRECT, p180 */ + +#define _3DSTATE_LOAD_INDIRECT (CMD_3D|(0x1d<<24)|(0x7<<16)) +#define LI0_STATE_STATIC_INDIRECT (0x01<<8) +#define LI0_STATE_DYNAMIC_INDIRECT (0x02<<8) +#define LI0_STATE_SAMPLER (0x04<<8) +#define LI0_STATE_MAP (0x08<<8) +#define LI0_STATE_PROGRAM (0x10<<8) +#define LI0_STATE_CONSTANTS (0x20<<8) + +#define SIS0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define SIS0_FORCE_LOAD (1<<1) +#define SIS0_BUFFER_VALID (1<<0) +#define SIS1_BUFFER_LENGTH(x) ((x)&0xff) + +#define DIS0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define DIS0_BUFFER_RESET (1<<1) +#define DIS0_BUFFER_VALID (1<<0) + +#define SSB0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define SSB0_FORCE_LOAD (1<<1) +#define SSB0_BUFFER_VALID (1<<0) +#define SSB1_BUFFER_LENGTH(x) ((x)&0xff) + +#define MSB0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define MSB0_FORCE_LOAD (1<<1) +#define MSB0_BUFFER_VALID (1<<0) +#define MSB1_BUFFER_LENGTH(x) ((x)&0xff) + +#define PSP0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define PSP0_FORCE_LOAD (1<<1) +#define PSP0_BUFFER_VALID (1<<0) +#define PSP1_BUFFER_LENGTH(x) ((x)&0xff) + +#define PSC0_BUFFER_ADDRESS(x) ((x)&~0x3) +#define PSC0_FORCE_LOAD (1<<1) +#define PSC0_BUFFER_VALID (1<<0) +#define PSC1_BUFFER_LENGTH(x) ((x)&0xff) + + + + + +/* _3DSTATE_RASTERIZATION_RULES */ +#define _3DSTATE_RASTER_RULES_CMD (CMD_3D|(0x07<<24)) +#define ENABLE_POINT_RASTER_RULE (1<<15) +#define OGL_POINT_RASTER_RULE (1<<13) +#define ENABLE_TEXKILL_3D_4D (1<<10) +#define TEXKILL_3D (0<<9) +#define TEXKILL_4D (1<<9) +#define ENABLE_LINE_STRIP_PROVOKE_VRTX (1<<8) +#define ENABLE_TRI_FAN_PROVOKE_VRTX (1<<5) +#define LINE_STRIP_PROVOKE_VRTX(x) ((x)<<6) +#define TRI_FAN_PROVOKE_VRTX(x) ((x)<<3) + +/* _3DSTATE_SCISSOR_ENABLE, p256 */ +#define _3DSTATE_SCISSOR_ENABLE_CMD (CMD_3D|(0x1c<<24)|(0x10<<19)) +#define ENABLE_SCISSOR_RECT ((1<<1) | 1) +#define DISABLE_SCISSOR_RECT (1<<1) + +/* _3DSTATE_SCISSOR_RECTANGLE_0, p257 */ +#define _3DSTATE_SCISSOR_RECT_0_CMD (CMD_3D|(0x1d<<24)|(0x81<<16)|1) +/* Dword 1 */ +#define SCISSOR_RECT_0_YMIN(x) ((x)<<16) +#define SCISSOR_RECT_0_XMIN(x) (x) +/* Dword 2 */ +#define SCISSOR_RECT_0_YMAX(x) ((x)<<16) +#define SCISSOR_RECT_0_XMAX(x) (x) + +/* p189 */ +#define _3DSTATE_LOAD_STATE_IMMEDIATE_1 ((0x3<<29)|(0x1d<<24)|(0x04<<16)) +#define I1_LOAD_S(n) (1<<(4+n)) + +#define S0_VB_OFFSET_MASK 0xffffffc +#define S0_AUTO_CACHE_INV_DISABLE (1<<0) + +#define S1_VERTEX_WIDTH_SHIFT 24 +#define S1_VERTEX_WIDTH_MASK (0x3f<<24) +#define S1_VERTEX_PITCH_SHIFT 16 +#define S1_VERTEX_PITCH_MASK (0x3f<<16) + +#define TEXCOORDFMT_2D 0x0 +#define TEXCOORDFMT_3D 0x1 +#define TEXCOORDFMT_4D 0x2 +#define TEXCOORDFMT_1D 0x3 +#define TEXCOORDFMT_2D_16 0x4 +#define TEXCOORDFMT_4D_16 0x5 +#define TEXCOORDFMT_NOT_PRESENT 0xf +#define S2_TEXCOORD_FMT0_MASK 0xf +#define S2_TEXCOORD_FMT1_SHIFT 4 +#define S2_TEXCOORD_FMT(unit, type) ((type)<<(unit*4)) +#define S2_TEXCOORD_NONE (~0) + +/* S3 not interesting */ + +#define S4_POINT_WIDTH_SHIFT 23 +#define S4_POINT_WIDTH_MASK (0x1ff<<23) +#define S4_LINE_WIDTH_SHIFT 19 +#define S4_LINE_WIDTH_ONE (0x2<<19) +#define S4_LINE_WIDTH_MASK (0xf<<19) +#define S4_FLATSHADE_ALPHA (1<<18) +#define S4_FLATSHADE_FOG (1<<17) +#define S4_FLATSHADE_SPECULAR (1<<16) +#define S4_FLATSHADE_COLOR (1<<15) +#define S4_CULLMODE_BOTH (0<<13) +#define S4_CULLMODE_NONE (1<<13) +#define S4_CULLMODE_CW (2<<13) +#define S4_CULLMODE_CCW (3<<13) +#define S4_CULLMODE_MASK (3<<13) +#define S4_VFMT_POINT_WIDTH (1<<12) +#define S4_VFMT_SPEC_FOG (1<<11) +#define S4_VFMT_COLOR (1<<10) +#define S4_VFMT_DEPTH_OFFSET (1<<9) +#define S4_VFMT_XYZ (1<<6) +#define S4_VFMT_XYZW (2<<6) +#define S4_VFMT_XY (3<<6) +#define S4_VFMT_XYW (4<<6) +#define S4_VFMT_XYZW_MASK (7<<6) +#define S4_FORCE_DEFAULT_DIFFUSE (1<<5) +#define S4_FORCE_DEFAULT_SPECULAR (1<<4) +#define S4_LOCAL_DEPTH_OFFSET_ENABLE (1<<3) +#define S4_VFMT_FOG_PARAM (1<<2) +#define S4_SPRITE_POINT_ENABLE (1<<1) +#define S4_LINE_ANTIALIAS_ENABLE (1<<0) + +#define S4_VFMT_MASK (S4_VFMT_POINT_WIDTH | \ + S4_VFMT_SPEC_FOG | \ + S4_VFMT_COLOR | \ + S4_VFMT_DEPTH_OFFSET | \ + S4_VFMT_XYZW_MASK | \ + S4_VFMT_FOG_PARAM) + + +#define S5_WRITEDISABLE_ALPHA (1<<31) +#define S5_WRITEDISABLE_RED (1<<30) +#define S5_WRITEDISABLE_GREEN (1<<29) +#define S5_WRITEDISABLE_BLUE (1<<28) +#define S5_WRITEDISABLE_MASK (0xf<<28) +#define S5_FORCE_DEFAULT_POINT_SIZE (1<<27) +#define S5_LAST_PIXEL_ENABLE (1<<26) +#define S5_GLOBAL_DEPTH_OFFSET_ENABLE (1<<25) +#define S5_FOG_ENABLE (1<<24) +#define S5_STENCIL_REF_SHIFT 16 +#define S5_STENCIL_REF_MASK (0xff<<16) +#define S5_STENCIL_TEST_FUNC_SHIFT 13 +#define S5_STENCIL_TEST_FUNC_MASK (0x7<<13) +#define S5_STENCIL_FAIL_SHIFT 10 +#define S5_STENCIL_FAIL_MASK (0x7<<10) +#define S5_STENCIL_PASS_Z_FAIL_SHIFT 7 +#define S5_STENCIL_PASS_Z_FAIL_MASK (0x7<<7) +#define S5_STENCIL_PASS_Z_PASS_SHIFT 4 +#define S5_STENCIL_PASS_Z_PASS_MASK (0x7<<4) +#define S5_STENCIL_WRITE_ENABLE (1<<3) +#define S5_STENCIL_TEST_ENABLE (1<<2) +#define S5_COLOR_DITHER_ENABLE (1<<1) +#define S5_LOGICOP_ENABLE (1<<0) + + +#define S6_ALPHA_TEST_ENABLE (1<<31) +#define S6_ALPHA_TEST_FUNC_SHIFT 28 +#define S6_ALPHA_TEST_FUNC_MASK (0x7<<28) +#define S6_ALPHA_REF_SHIFT 20 +#define S6_ALPHA_REF_MASK (0xff<<20) +#define S6_DEPTH_TEST_ENABLE (1<<19) +#define S6_DEPTH_TEST_FUNC_SHIFT 16 +#define S6_DEPTH_TEST_FUNC_MASK (0x7<<16) +#define S6_CBUF_BLEND_ENABLE (1<<15) +#define S6_CBUF_BLEND_FUNC_SHIFT 12 +#define S6_CBUF_BLEND_FUNC_MASK (0x7<<12) +#define S6_CBUF_SRC_BLEND_FACT_SHIFT 8 +#define S6_CBUF_SRC_BLEND_FACT_MASK (0xf<<8) +#define S6_CBUF_DST_BLEND_FACT_SHIFT 4 +#define S6_CBUF_DST_BLEND_FACT_MASK (0xf<<4) +#define S6_DEPTH_WRITE_ENABLE (1<<3) +#define S6_COLOR_WRITE_ENABLE (1<<2) +#define S6_TRISTRIP_PV_SHIFT 0 +#define S6_TRISTRIP_PV_MASK (0x3<<0) + +#define S7_DEPTH_OFFSET_CONST_MASK ~0 + + + +#define DST_BLND_FACT(f) ((f)<<S6_CBUF_DST_BLEND_FACT_SHIFT) +#define SRC_BLND_FACT(f) ((f)<<S6_CBUF_SRC_BLEND_FACT_SHIFT) +#define DST_ABLND_FACT(f) ((f)<<IAB_DST_FACTOR_SHIFT) +#define SRC_ABLND_FACT(f) ((f)<<IAB_SRC_FACTOR_SHIFT) + + + + +/* 3DSTATE_MAP_DEINTERLACER_PARAMETERS */ + +/* 3DSTATE_MAP_PALETTE_LOAD_32, p206 */ +#define _3DSTATE_MAP_PALETTE_LOAD_32 (CMD_3D|(0x1d<<24)|(0x8f<<16)) +/* subsequent dwords up to length (max 16) are ARGB8888 color values */ + +/* _3DSTATE_MODES_4, p218 */ +#define _3DSTATE_MODES_4_CMD (CMD_3D|(0x0d<<24)) +#define ENABLE_LOGIC_OP_FUNC (1<<23) +#define LOGIC_OP_FUNC(x) ((x)<<18) +#define LOGICOP_MASK (0xf<<18) +#define MODE4_ENABLE_STENCIL_TEST_MASK ((1<<17)|(0xff00)) +#define ENABLE_STENCIL_TEST_MASK (1<<17) +#define STENCIL_TEST_MASK(x) (((x)&0xff)<<8) +#define MODE4_ENABLE_STENCIL_WRITE_MASK ((1<<16)|(0x00ff)) +#define ENABLE_STENCIL_WRITE_MASK (1<<16) +#define STENCIL_WRITE_MASK(x) ((x)&0xff) + +/* _3DSTATE_MODES_5, p220 */ +#define _3DSTATE_MODES_5_CMD (CMD_3D|(0x0c<<24)) +#define PIPELINE_FLUSH_RENDER_CACHE (1<<18) +#define PIPELINE_FLUSH_TEXTURE_CACHE (1<<16) + + +/* p221 */ +#define _3DSTATE_PIXEL_SHADER_CONSTANTS (CMD_3D|(0x1d<<24)|(0x6<<16)) +#define PS1_REG(n) (1<<(n)) +#define PS2_CONST_X(n) (n) +#define PS3_CONST_Y(n) (n) +#define PS4_CONST_Z(n) (n) +#define PS5_CONST_W(n) (n) + +/* p222 */ + + +#define I915_MAX_TEX_INDIRECT 4 +#define I915_MAX_TEX_INSN 32 +#define I915_MAX_ALU_INSN 64 +#define I915_MAX_DECL_INSN 27 +#define I915_MAX_TEMPORARY 16 + + +/* Each instruction is 3 dwords long, though most don't require all + * this space. Maximum of 123 instructions. Smaller maxes per insn + * type. + */ +#define _3DSTATE_PIXEL_SHADER_PROGRAM (CMD_3D|(0x1d<<24)|(0x5<<16)) + +#define REG_TYPE_R 0 /* temporary regs, no need to + * dcl, must be written before + * read -- Preserved between + * phases. + */ +#define REG_TYPE_T 1 /* Interpolated values, must be + * dcl'ed before use. + * + * 0..7: texture coord, + * 8: diffuse spec, + * 9: specular color, + * 10: fog parameter in w. + */ +#define REG_TYPE_CONST 2 /* Restriction: only one const + * can be referenced per + * instruction, though it may be + * selected for multiple inputs. + * Constants not initialized + * default to zero. + */ +#define REG_TYPE_S 3 /* sampler */ +#define REG_TYPE_OC 4 /* output color (rgba) */ +#define REG_TYPE_OD 5 /* output depth (w), xyz are + * temporaries. If not written, + * interpolated depth is used? + */ +#define REG_TYPE_U 6 /* unpreserved temporaries */ +#define REG_TYPE_MASK 0x7 +#define REG_NR_MASK 0xf + + +/* REG_TYPE_T: + */ +#define T_TEX0 0 +#define T_TEX1 1 +#define T_TEX2 2 +#define T_TEX3 3 +#define T_TEX4 4 +#define T_TEX5 5 +#define T_TEX6 6 +#define T_TEX7 7 +#define T_DIFFUSE 8 +#define T_SPECULAR 9 +#define T_FOG_W 10 /* interpolated fog is in W coord */ + +/* Arithmetic instructions */ + +/* .replicate_swizzle == selection and replication of a particular + * scalar channel, ie., .xxxx, .yyyy, .zzzz or .wwww + */ +#define A0_NOP (0x0<<24) /* no operation */ +#define A0_ADD (0x1<<24) /* dst = src0 + src1 */ +#define A0_MOV (0x2<<24) /* dst = src0 */ +#define A0_MUL (0x3<<24) /* dst = src0 * src1 */ +#define A0_MAD (0x4<<24) /* dst = src0 * src1 + src2 */ +#define A0_DP2ADD (0x5<<24) /* dst.xyzw = src0.xy dot src1.xy + src2.replicate_swizzle */ +#define A0_DP3 (0x6<<24) /* dst.xyzw = src0.xyz dot src1.xyz */ +#define A0_DP4 (0x7<<24) /* dst.xyzw = src0.xyzw dot src1.xyzw */ +#define A0_FRC (0x8<<24) /* dst = src0 - floor(src0) */ +#define A0_RCP (0x9<<24) /* dst.xyzw = 1/(src0.replicate_swizzle) */ +#define A0_RSQ (0xa<<24) /* dst.xyzw = 1/(sqrt(abs(src0.replicate_swizzle))) */ +#define A0_EXP (0xb<<24) /* dst.xyzw = exp2(src0.replicate_swizzle) */ +#define A0_LOG (0xc<<24) /* dst.xyzw = log2(abs(src0.replicate_swizzle)) */ +#define A0_CMP (0xd<<24) /* dst = (src0 >= 0.0) ? src1 : src2 */ +#define A0_MIN (0xe<<24) /* dst = (src0 < src1) ? src0 : src1 */ +#define A0_MAX (0xf<<24) /* dst = (src0 >= src1) ? src0 : src1 */ +#define A0_FLR (0x10<<24) /* dst = floor(src0) */ +#define A0_MOD (0x11<<24) /* dst = src0 fmod 1.0 */ +#define A0_TRC (0x12<<24) /* dst = int(src0) */ +#define A0_SGE (0x13<<24) /* dst = src0 >= src1 ? 1.0 : 0.0 */ +#define A0_SLT (0x14<<24) /* dst = src0 < src1 ? 1.0 : 0.0 */ +#define A0_DEST_SATURATE (1<<22) +#define A0_DEST_TYPE_SHIFT 19 +/* Allow: R, OC, OD, U */ +#define A0_DEST_NR_SHIFT 14 +/* Allow R: 0..15, OC,OD: 0..0, U: 0..2 */ +#define A0_DEST_CHANNEL_X (1<<10) +#define A0_DEST_CHANNEL_Y (2<<10) +#define A0_DEST_CHANNEL_Z (4<<10) +#define A0_DEST_CHANNEL_W (8<<10) +#define A0_DEST_CHANNEL_ALL (0xf<<10) +#define A0_DEST_CHANNEL_SHIFT 10 +#define A0_SRC0_TYPE_SHIFT 7 +#define A0_SRC0_NR_SHIFT 2 + +#define A0_DEST_CHANNEL_XY (A0_DEST_CHANNEL_X|A0_DEST_CHANNEL_Y) +#define A0_DEST_CHANNEL_XYZ (A0_DEST_CHANNEL_XY|A0_DEST_CHANNEL_Z) + + +#define SRC_X 0 +#define SRC_Y 1 +#define SRC_Z 2 +#define SRC_W 3 +#define SRC_ZERO 4 +#define SRC_ONE 5 + +#define A1_SRC0_CHANNEL_X_NEGATE (1<<31) +#define A1_SRC0_CHANNEL_X_SHIFT 28 +#define A1_SRC0_CHANNEL_Y_NEGATE (1<<27) +#define A1_SRC0_CHANNEL_Y_SHIFT 24 +#define A1_SRC0_CHANNEL_Z_NEGATE (1<<23) +#define A1_SRC0_CHANNEL_Z_SHIFT 20 +#define A1_SRC0_CHANNEL_W_NEGATE (1<<19) +#define A1_SRC0_CHANNEL_W_SHIFT 16 +#define A1_SRC1_TYPE_SHIFT 13 +#define A1_SRC1_NR_SHIFT 8 +#define A1_SRC1_CHANNEL_X_NEGATE (1<<7) +#define A1_SRC1_CHANNEL_X_SHIFT 4 +#define A1_SRC1_CHANNEL_Y_NEGATE (1<<3) +#define A1_SRC1_CHANNEL_Y_SHIFT 0 + +#define A2_SRC1_CHANNEL_Z_NEGATE (1<<31) +#define A2_SRC1_CHANNEL_Z_SHIFT 28 +#define A2_SRC1_CHANNEL_W_NEGATE (1<<27) +#define A2_SRC1_CHANNEL_W_SHIFT 24 +#define A2_SRC2_TYPE_SHIFT 21 +#define A2_SRC2_NR_SHIFT 16 +#define A2_SRC2_CHANNEL_X_NEGATE (1<<15) +#define A2_SRC2_CHANNEL_X_SHIFT 12 +#define A2_SRC2_CHANNEL_Y_NEGATE (1<<11) +#define A2_SRC2_CHANNEL_Y_SHIFT 8 +#define A2_SRC2_CHANNEL_Z_NEGATE (1<<7) +#define A2_SRC2_CHANNEL_Z_SHIFT 4 +#define A2_SRC2_CHANNEL_W_NEGATE (1<<3) +#define A2_SRC2_CHANNEL_W_SHIFT 0 + + + +/* Texture instructions */ +#define T0_TEXLD (0x15<<24) /* Sample texture using predeclared + * sampler and address, and output + * filtered texel data to destination + * register */ +#define T0_TEXLDP (0x16<<24) /* Same as texld but performs a + * perspective divide of the texture + * coordinate .xyz values by .w before + * sampling. */ +#define T0_TEXLDB (0x17<<24) /* Same as texld but biases the + * computed LOD by w. Only S4.6 two's + * comp is used. This implies that a + * float to fixed conversion is + * done. */ +#define T0_TEXKILL (0x18<<24) /* Does not perform a sampling + * operation. Simply kills the pixel + * if any channel of the address + * register is < 0.0. */ +#define T0_DEST_TYPE_SHIFT 19 +/* Allow: R, OC, OD, U */ +/* Note: U (unpreserved) regs do not retain their values between + * phases (cannot be used for feedback) + * + * Note: oC and OD registers can only be used as the destination of a + * texture instruction once per phase (this is an implementation + * restriction). + */ +#define T0_DEST_NR_SHIFT 14 +/* Allow R: 0..15, OC,OD: 0..0, U: 0..2 */ +#define T0_SAMPLER_NR_SHIFT 0 /* This field ignored for TEXKILL */ +#define T0_SAMPLER_NR_MASK (0xf<<0) + +#define T1_ADDRESS_REG_TYPE_SHIFT 24 /* Reg to use as texture coord */ +/* Allow R, T, OC, OD -- R, OC, OD are 'dependent' reads, new program phase */ +#define T1_ADDRESS_REG_NR_SHIFT 17 +#define T2_MBZ 0 + +/* Declaration instructions */ +#define D0_DCL (0x19<<24) /* Declare a t (interpolated attrib) + * register or an s (sampler) + * register. */ +#define D0_SAMPLE_TYPE_SHIFT 22 +#define D0_SAMPLE_TYPE_2D (0x0<<22) +#define D0_SAMPLE_TYPE_CUBE (0x1<<22) +#define D0_SAMPLE_TYPE_VOLUME (0x2<<22) +#define D0_SAMPLE_TYPE_MASK (0x3<<22) + +#define D0_TYPE_SHIFT 19 +/* Allow: T, S */ +#define D0_NR_SHIFT 14 +/* Allow T: 0..10, S: 0..15 */ +#define D0_CHANNEL_X (1<<10) +#define D0_CHANNEL_Y (2<<10) +#define D0_CHANNEL_Z (4<<10) +#define D0_CHANNEL_W (8<<10) +#define D0_CHANNEL_ALL (0xf<<10) +#define D0_CHANNEL_NONE (0<<10) + +#define D0_CHANNEL_XY (D0_CHANNEL_X|D0_CHANNEL_Y) +#define D0_CHANNEL_XYZ (D0_CHANNEL_XY|D0_CHANNEL_Z) + +/* I915 Errata: Do not allow (xz), (xw), (xzw) combinations for diffuse + * or specular declarations. + * + * For T dcls, only allow: (x), (xy), (xyz), (w), (xyzw) + * + * Must be zero for S (sampler) dcls + */ +#define D1_MBZ 0 +#define D2_MBZ 0 + + + +/* p207 */ +#define _3DSTATE_MAP_STATE (CMD_3D|(0x1d<<24)|(0x0<<16)) + +#define MS1_MAPMASK_SHIFT 0 +#define MS1_MAPMASK_MASK (0x8fff<<0) + +#define MS2_UNTRUSTED_SURFACE (1<<31) +#define MS2_ADDRESS_MASK 0xfffffffc +#define MS2_VERTICAL_LINE_STRIDE (1<<1) +#define MS2_VERTICAL_OFFSET (1<<1) + +#define MS3_HEIGHT_SHIFT 21 +#define MS3_WIDTH_SHIFT 10 +#define MS3_PALETTE_SELECT (1<<9) +#define MS3_MAPSURF_FORMAT_SHIFT 7 +#define MS3_MAPSURF_FORMAT_MASK (0x7<<7) +#define MAPSURF_8BIT (1<<7) +#define MAPSURF_16BIT (2<<7) +#define MAPSURF_32BIT (3<<7) +#define MAPSURF_422 (5<<7) +#define MAPSURF_COMPRESSED (6<<7) +#define MAPSURF_4BIT_INDEXED (7<<7) +#define MS3_MT_FORMAT_MASK (0x7 << 3) +#define MS3_MT_FORMAT_SHIFT 3 +#define MT_4BIT_IDX_ARGB8888 (7<<3) /* SURFACE_4BIT_INDEXED */ +#define MT_8BIT_I8 (0<<3) /* SURFACE_8BIT */ +#define MT_8BIT_L8 (1<<3) +#define MT_8BIT_A8 (4<<3) +#define MT_8BIT_MONO8 (5<<3) +#define MT_16BIT_RGB565 (0<<3) /* SURFACE_16BIT */ +#define MT_16BIT_ARGB1555 (1<<3) +#define MT_16BIT_ARGB4444 (2<<3) +#define MT_16BIT_AY88 (3<<3) +#define MT_16BIT_88DVDU (5<<3) +#define MT_16BIT_BUMP_655LDVDU (6<<3) +#define MT_16BIT_I16 (7<<3) +#define MT_16BIT_L16 (8<<3) +#define MT_16BIT_A16 (9<<3) +#define MT_32BIT_ARGB8888 (0<<3) /* SURFACE_32BIT */ +#define MT_32BIT_ABGR8888 (1<<3) +#define MT_32BIT_XRGB8888 (2<<3) +#define MT_32BIT_XBGR8888 (3<<3) +#define MT_32BIT_QWVU8888 (4<<3) +#define MT_32BIT_AXVU8888 (5<<3) +#define MT_32BIT_LXVU8888 (6<<3) +#define MT_32BIT_XLVU8888 (7<<3) +#define MT_32BIT_ARGB2101010 (8<<3) +#define MT_32BIT_ABGR2101010 (9<<3) +#define MT_32BIT_AWVU2101010 (0xA<<3) +#define MT_32BIT_GR1616 (0xB<<3) +#define MT_32BIT_VU1616 (0xC<<3) +#define MT_32BIT_xI824 (0xD<<3) +#define MT_32BIT_xA824 (0xE<<3) +#define MT_32BIT_xL824 (0xF<<3) +#define MT_422_YCRCB_SWAPY (0<<3) /* SURFACE_422 */ +#define MT_422_YCRCB_NORMAL (1<<3) +#define MT_422_YCRCB_SWAPUV (2<<3) +#define MT_422_YCRCB_SWAPUVY (3<<3) +#define MT_COMPRESS_DXT1 (0<<3) /* SURFACE_COMPRESSED */ +#define MT_COMPRESS_DXT2_3 (1<<3) +#define MT_COMPRESS_DXT4_5 (2<<3) +#define MT_COMPRESS_FXT1 (3<<3) +#define MT_COMPRESS_DXT1_RGB (4<<3) +#define MS3_USE_FENCE_REGS (1<<2) +#define MS3_TILED_SURFACE (1<<1) +#define MS3_TILE_WALK (1<<0) + +#define MS4_PITCH_SHIFT 21 +#define MS4_CUBE_FACE_ENA_NEGX (1<<20) +#define MS4_CUBE_FACE_ENA_POSX (1<<19) +#define MS4_CUBE_FACE_ENA_NEGY (1<<18) +#define MS4_CUBE_FACE_ENA_POSY (1<<17) +#define MS4_CUBE_FACE_ENA_NEGZ (1<<16) +#define MS4_CUBE_FACE_ENA_POSZ (1<<15) +#define MS4_CUBE_FACE_ENA_MASK (0x3f<<15) +#define MS4_MAX_LOD_SHIFT 9 +#define MS4_MAX_LOD_MASK (0x3f<<9) +#define MS4_MIP_LAYOUT_LEGACY (0<<8) +#define MS4_MIP_LAYOUT_BELOW_LPT (0<<8) +#define MS4_MIP_LAYOUT_RIGHT_LPT (1<<8) +#define MS4_VOLUME_DEPTH_SHIFT 0 +#define MS4_VOLUME_DEPTH_MASK (0xff<<0) + +/* p244 */ +#define _3DSTATE_SAMPLER_STATE (CMD_3D|(0x1d<<24)|(0x1<<16)) + +#define SS1_MAPMASK_SHIFT 0 +#define SS1_MAPMASK_MASK (0x8fff<<0) + +#define SS2_REVERSE_GAMMA_ENABLE (1<<31) +#define SS2_PACKED_TO_PLANAR_ENABLE (1<<30) +#define SS2_COLORSPACE_CONVERSION (1<<29) +#define SS2_CHROMAKEY_SHIFT 27 +#define SS2_BASE_MIP_LEVEL_SHIFT 22 +#define SS2_BASE_MIP_LEVEL_MASK (0x1f<<22) +#define SS2_MIP_FILTER_SHIFT 20 +#define SS2_MIP_FILTER_MASK (0x3<<20) +#define MIPFILTER_NONE 0 +#define MIPFILTER_NEAREST 1 +#define MIPFILTER_LINEAR 3 +#define SS2_MAG_FILTER_SHIFT 17 +#define SS2_MAG_FILTER_MASK (0x7<<17) +#define FILTER_NEAREST 0 +#define FILTER_LINEAR 1 +#define FILTER_ANISOTROPIC 2 +#define FILTER_4X4_1 3 +#define FILTER_4X4_2 4 +#define FILTER_4X4_FLAT 5 +#define FILTER_6X5_MONO 6 /* XXX - check */ +#define SS2_MIN_FILTER_SHIFT 14 +#define SS2_MIN_FILTER_MASK (0x7<<14) +#define SS2_LOD_BIAS_SHIFT 5 +#define SS2_LOD_BIAS_ONE (0x10<<5) +#define SS2_LOD_BIAS_MASK (0x1ff<<5) +/* Shadow requires: + * MT_X8{I,L,A}24 or MT_{I,L,A}16 texture format + * FILTER_4X4_x MIN and MAG filters + */ +#define SS2_SHADOW_ENABLE (1<<4) +#define SS2_MAX_ANISO_MASK (1<<3) +#define SS2_MAX_ANISO_2 (0<<3) +#define SS2_MAX_ANISO_4 (1<<3) +#define SS2_SHADOW_FUNC_SHIFT 0 +#define SS2_SHADOW_FUNC_MASK (0x7<<0) +/* SS2_SHADOW_FUNC values: see COMPAREFUNC_* */ + +#define SS3_MIN_LOD_SHIFT 24 +#define SS3_MIN_LOD_ONE (0x10<<24) +#define SS3_MIN_LOD_MASK (0xff<<24) +#define SS3_KILL_PIXEL_ENABLE (1<<17) +#define SS3_TCX_ADDR_MODE_SHIFT 12 +#define SS3_TCX_ADDR_MODE_MASK (0x7<<12) +#define TEXCOORDMODE_WRAP 0 +#define TEXCOORDMODE_MIRROR 1 +#define TEXCOORDMODE_CLAMP_EDGE 2 +#define TEXCOORDMODE_CUBE 3 +#define TEXCOORDMODE_CLAMP_BORDER 4 +#define TEXCOORDMODE_MIRROR_ONCE 5 +#define SS3_TCY_ADDR_MODE_SHIFT 9 +#define SS3_TCY_ADDR_MODE_MASK (0x7<<9) +#define SS3_TCZ_ADDR_MODE_SHIFT 6 +#define SS3_TCZ_ADDR_MODE_MASK (0x7<<6) +#define SS3_NORMALIZED_COORDS (1<<5) +#define SS3_TEXTUREMAP_INDEX_SHIFT 1 +#define SS3_TEXTUREMAP_INDEX_MASK (0xf<<1) +#define SS3_DEINTERLACER_ENABLE (1<<0) + +#define SS4_BORDER_COLOR_MASK (~0) + +/* 3DSTATE_SPAN_STIPPLE, p258 + */ +#define _3DSTATE_STIPPLE ((0x3<<29)|(0x1d<<24)|(0x83<<16)) +#define ST1_ENABLE (1<<16) +#define ST1_MASK (0xffff) + +#define _3DSTATE_DEFAULT_Z ((0x3<<29)|(0x1d<<24)|(0x98<<16)) +#define _3DSTATE_DEFAULT_DIFFUSE ((0x3<<29)|(0x1d<<24)|(0x99<<16)) +#define _3DSTATE_DEFAULT_SPECULAR ((0x3<<29)|(0x1d<<24)|(0x9a<<16)) + + +#define MI_FLUSH ((0<<29)|(4<<23)) +#define FLUSH_MAP_CACHE (1<<0) +#define INHIBIT_FLUSH_RENDER_CACHE (1<<2) + + +#define CMD_3D (0x3<<29) + + +#define _3DPRIMITIVE ((0x3<<29)|(0x1f<<24)) +#define PRIM_INDIRECT (1<<23) +#define PRIM_INLINE (0<<23) +#define PRIM_INDIRECT_SEQUENTIAL (0<<17) +#define PRIM_INDIRECT_ELTS (1<<17) + +#define PRIM3D_TRILIST (0x0<<18) +#define PRIM3D_TRISTRIP (0x1<<18) +#define PRIM3D_TRISTRIP_RVRSE (0x2<<18) +#define PRIM3D_TRIFAN (0x3<<18) +#define PRIM3D_POLY (0x4<<18) +#define PRIM3D_LINELIST (0x5<<18) +#define PRIM3D_LINESTRIP (0x6<<18) +#define PRIM3D_RECTLIST (0x7<<18) +#define PRIM3D_POINTLIST (0x8<<18) +#define PRIM3D_DIB (0x9<<18) +#define PRIM3D_MASK (0x1f<<18) + +#define I915PACKCOLOR4444(r,g,b,a) \ + ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4)) + +#define I915PACKCOLOR1555(r,g,b,a) \ + ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \ + ((a) ? 0x8000 : 0)) + +#define I915PACKCOLOR565(r,g,b) \ + ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3)) + +#define I915PACKCOLOR8888(r,g,b,a) \ + ((a<<24) | (r<<16) | (g<<8) | b) + + + + +#define BR00_BITBLT_CLIENT 0x40000000 +#define BR00_OP_COLOR_BLT 0x10000000 +#define BR00_OP_SRC_COPY_BLT 0x10C00000 +#define BR13_SOLID_PATTERN 0x80000000 + +#define XY_COLOR_BLT_CMD ((2<<29)|(0x50<<22)|0x4) +#define XY_COLOR_BLT_WRITE_ALPHA (1<<21) +#define XY_COLOR_BLT_WRITE_RGB (1<<20) + +#define XY_SRC_COPY_BLT_CMD ((2<<29)|(0x53<<22)|6) +#define XY_SRC_COPY_BLT_WRITE_ALPHA (1<<21) +#define XY_SRC_COPY_BLT_WRITE_RGB (1<<20) + +#define MI_WAIT_FOR_EVENT ((0x3<<23)) +#define MI_WAIT_FOR_PLANE_B_FLIP (1<<6) +#define MI_WAIT_FOR_PLANE_A_FLIP (1<<2) + +#define MI_BATCH_BUFFER (0x30<<23) +#define MI_BATCH_BUFFER_START (0x31<<23) +#define MI_BATCH_BUFFER_END (0xa<<23) + + + +#define COMPAREFUNC_ALWAYS 0 +#define COMPAREFUNC_NEVER 0x1 +#define COMPAREFUNC_LESS 0x2 +#define COMPAREFUNC_EQUAL 0x3 +#define COMPAREFUNC_LEQUAL 0x4 +#define COMPAREFUNC_GREATER 0x5 +#define COMPAREFUNC_NOTEQUAL 0x6 +#define COMPAREFUNC_GEQUAL 0x7 + +#define STENCILOP_KEEP 0 +#define STENCILOP_ZERO 0x1 +#define STENCILOP_REPLACE 0x2 +#define STENCILOP_INCRSAT 0x3 +#define STENCILOP_DECRSAT 0x4 +#define STENCILOP_INCR 0x5 +#define STENCILOP_DECR 0x6 +#define STENCILOP_INVERT 0x7 + +#define LOGICOP_CLEAR 0 +#define LOGICOP_NOR 0x1 +#define LOGICOP_AND_INV 0x2 +#define LOGICOP_COPY_INV 0x3 +#define LOGICOP_AND_RVRSE 0x4 +#define LOGICOP_INV 0x5 +#define LOGICOP_XOR 0x6 +#define LOGICOP_NAND 0x7 +#define LOGICOP_AND 0x8 +#define LOGICOP_EQUIV 0x9 +#define LOGICOP_NOOP 0xa +#define LOGICOP_OR_INV 0xb +#define LOGICOP_COPY 0xc +#define LOGICOP_OR_RVRSE 0xd +#define LOGICOP_OR 0xe +#define LOGICOP_SET 0xf + +#define BLENDFACT_ZERO 0x01 +#define BLENDFACT_ONE 0x02 +#define BLENDFACT_SRC_COLR 0x03 +#define BLENDFACT_INV_SRC_COLR 0x04 +#define BLENDFACT_SRC_ALPHA 0x05 +#define BLENDFACT_INV_SRC_ALPHA 0x06 +#define BLENDFACT_DST_ALPHA 0x07 +#define BLENDFACT_INV_DST_ALPHA 0x08 +#define BLENDFACT_DST_COLR 0x09 +#define BLENDFACT_INV_DST_COLR 0x0a +#define BLENDFACT_SRC_ALPHA_SATURATE 0x0b +#define BLENDFACT_CONST_COLOR 0x0c +#define BLENDFACT_INV_CONST_COLOR 0x0d +#define BLENDFACT_CONST_ALPHA 0x0e +#define BLENDFACT_INV_CONST_ALPHA 0x0f +#define BLENDFACT_MASK 0x0f + +#define PCI_CHIP_I915_G 0x2582 +#define PCI_CHIP_I915_GM 0x2592 +#define PCI_CHIP_I945_G 0x2772 +#define PCI_CHIP_I945_GM 0x27A2 +#define PCI_CHIP_I945_GME 0x27AE +#define PCI_CHIP_G33_G 0x29C2 +#define PCI_CHIP_Q35_G 0x29B2 +#define PCI_CHIP_Q33_G 0x29D2 + + +#endif diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c new file mode 100644 index 0000000000..abd5571b88 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state.c @@ -0,0 +1,694 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/draw/draw_context.h" +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" + +#include "i915_context.h" +#include "i915_reg.h" +#include "i915_state.h" +#include "i915_state_inlines.h" + + +/* The i915 (and related graphics cores) do not support GL_CLAMP. The + * Intel drivers for "other operating systems" implement GL_CLAMP as + * GL_CLAMP_TO_EDGE, so the same is done here. + */ +static unsigned +translate_wrap_mode(unsigned wrap) +{ + switch (wrap) { + case PIPE_TEX_WRAP_REPEAT: + return TEXCOORDMODE_WRAP; + case PIPE_TEX_WRAP_CLAMP: + return TEXCOORDMODE_CLAMP_EDGE; /* not quite correct */ + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + return TEXCOORDMODE_CLAMP_EDGE; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return TEXCOORDMODE_CLAMP_BORDER; +// case PIPE_TEX_WRAP_MIRRORED_REPEAT: +// return TEXCOORDMODE_MIRROR; + default: + return TEXCOORDMODE_WRAP; + } +} + +static unsigned translate_img_filter( unsigned filter ) +{ + switch (filter) { + case PIPE_TEX_FILTER_NEAREST: + return FILTER_NEAREST; + case PIPE_TEX_FILTER_LINEAR: + return FILTER_LINEAR; + default: + assert(0); + return FILTER_NEAREST; + } +} + +static unsigned translate_mip_filter( unsigned filter ) +{ + switch (filter) { + case PIPE_TEX_MIPFILTER_NONE: + return MIPFILTER_NONE; + case PIPE_TEX_MIPFILTER_NEAREST: + return MIPFILTER_NEAREST; + case PIPE_TEX_FILTER_LINEAR: + return MIPFILTER_LINEAR; + default: + assert(0); + return MIPFILTER_NONE; + } +} + + +/* None of this state is actually used for anything yet. + */ +static void * +i915_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + struct i915_blend_state *cso_data = CALLOC_STRUCT( i915_blend_state ); + + { + unsigned eqRGB = blend->rgb_func; + unsigned srcRGB = blend->rgb_src_factor; + unsigned dstRGB = blend->rgb_dst_factor; + + unsigned eqA = blend->alpha_func; + unsigned srcA = blend->alpha_src_factor; + unsigned dstA = blend->alpha_dst_factor; + + /* Special handling for MIN/MAX filter modes handled at + * state_tracker level. + */ + + if (srcA != srcRGB || + dstA != dstRGB || + eqA != eqRGB) { + + cso_data->iab = (_3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD | + IAB_MODIFY_ENABLE | + IAB_ENABLE | + IAB_MODIFY_FUNC | + IAB_MODIFY_SRC_FACTOR | + IAB_MODIFY_DST_FACTOR | + SRC_ABLND_FACT(i915_translate_blend_factor(srcA)) | + DST_ABLND_FACT(i915_translate_blend_factor(dstA)) | + (i915_translate_blend_func(eqA) << IAB_FUNC_SHIFT)); + } + else { + cso_data->iab = (_3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD | + IAB_MODIFY_ENABLE | + 0); + } + } + + cso_data->modes4 |= (_3DSTATE_MODES_4_CMD | + ENABLE_LOGIC_OP_FUNC | + LOGIC_OP_FUNC(i915_translate_logic_op(blend->logicop_func))); + + if (blend->logicop_enable) + cso_data->LIS5 |= S5_LOGICOP_ENABLE; + + if (blend->dither) + cso_data->LIS5 |= S5_COLOR_DITHER_ENABLE; + + if ((blend->colormask & PIPE_MASK_R) == 0) + cso_data->LIS5 |= S5_WRITEDISABLE_RED; + + if ((blend->colormask & PIPE_MASK_G) == 0) + cso_data->LIS5 |= S5_WRITEDISABLE_GREEN; + + if ((blend->colormask & PIPE_MASK_B) == 0) + cso_data->LIS5 |= S5_WRITEDISABLE_BLUE; + + if ((blend->colormask & PIPE_MASK_A) == 0) + cso_data->LIS5 |= S5_WRITEDISABLE_ALPHA; + + if (blend->blend_enable) { + unsigned funcRGB = blend->rgb_func; + unsigned srcRGB = blend->rgb_src_factor; + unsigned dstRGB = blend->rgb_dst_factor; + + cso_data->LIS6 |= (S6_CBUF_BLEND_ENABLE | + SRC_BLND_FACT(i915_translate_blend_factor(srcRGB)) | + DST_BLND_FACT(i915_translate_blend_factor(dstRGB)) | + (i915_translate_blend_func(funcRGB) << S6_CBUF_BLEND_FUNC_SHIFT)); + } + + return cso_data; +} + +static void i915_bind_blend_state(struct pipe_context *pipe, + void *blend) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->blend = (struct i915_blend_state*)blend; + + i915->dirty |= I915_NEW_BLEND; +} + + +static void i915_delete_blend_state(struct pipe_context *pipe, void *blend) +{ + FREE(blend); +} + +static void i915_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->blend_color = *blend_color; + + i915->dirty |= I915_NEW_BLEND; +} + +static void * +i915_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + struct i915_sampler_state *cso = CALLOC_STRUCT( i915_sampler_state ); + const unsigned ws = sampler->wrap_s; + const unsigned wt = sampler->wrap_t; + const unsigned wr = sampler->wrap_r; + unsigned minFilt, magFilt; + unsigned mipFilt; + + cso->templ = sampler; + + mipFilt = translate_mip_filter(sampler->min_mip_filter); + if (sampler->max_anisotropy > 1.0) { + minFilt = FILTER_ANISOTROPIC; + magFilt = FILTER_ANISOTROPIC; + if (sampler->max_anisotropy > 2.0) { + cso->state[0] |= SS2_MAX_ANISO_4; + } + } + else { + minFilt = translate_img_filter( sampler->min_img_filter ); + magFilt = translate_img_filter( sampler->mag_img_filter ); + } + + { + int b = (int) (sampler->lod_bias * 16.0); + b = CLAMP(b, -256, 255); + cso->state[0] |= ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK); + } + + /* Shadow: + */ + if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) + { + cso->state[0] |= (SS2_SHADOW_ENABLE | + i915_translate_compare_func(sampler->compare_func)); + + minFilt = FILTER_4X4_FLAT; + magFilt = FILTER_4X4_FLAT; + } + + cso->state[0] |= ((minFilt << SS2_MIN_FILTER_SHIFT) | + (mipFilt << SS2_MIP_FILTER_SHIFT) | + (magFilt << SS2_MAG_FILTER_SHIFT)); + + cso->state[1] |= + ((translate_wrap_mode(ws) << SS3_TCX_ADDR_MODE_SHIFT) | + (translate_wrap_mode(wt) << SS3_TCY_ADDR_MODE_SHIFT) | + (translate_wrap_mode(wr) << SS3_TCZ_ADDR_MODE_SHIFT)); + + if (sampler->normalized_coords) + cso->state[1] |= SS3_NORMALIZED_COORDS; + + if (0) /* XXX not tested yet */ + { + int minlod = (int) (16.0 * sampler->min_lod); + minlod = CLAMP(minlod, 0, 16 * 11); + cso->state[1] |= (minlod << SS3_MIN_LOD_SHIFT); + } + + { + ubyte r = float_to_ubyte(sampler->border_color[0]); + ubyte g = float_to_ubyte(sampler->border_color[1]); + ubyte b = float_to_ubyte(sampler->border_color[2]); + ubyte a = float_to_ubyte(sampler->border_color[3]); + cso->state[2] = I915PACKCOLOR8888(r, g, b, a); + } + return cso; +} + +static void i915_bind_sampler_state(struct pipe_context *pipe, + unsigned unit, void *sampler) +{ + struct i915_context *i915 = i915_context(pipe); + + assert(unit < PIPE_MAX_SAMPLERS); + i915->sampler[unit] = (const struct i915_sampler_state*)sampler; + + i915->dirty |= I915_NEW_SAMPLER; +} + +static void i915_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + FREE(sampler); +} + + +/** XXX move someday? Or consolidate all these simple state setters + * into one file. + */ + +static void * +i915_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + struct i915_depth_stencil_state *cso = CALLOC_STRUCT( i915_depth_stencil_state ); + + { + int testmask = depth_stencil->stencil[0].value_mask & 0xff; + int writemask = depth_stencil->stencil[0].write_mask & 0xff; + + cso->stencil_modes4 |= (_3DSTATE_MODES_4_CMD | + ENABLE_STENCIL_TEST_MASK | + STENCIL_TEST_MASK(testmask) | + ENABLE_STENCIL_WRITE_MASK | + STENCIL_WRITE_MASK(writemask)); + } + + if (depth_stencil->stencil[0].enabled) { + int test = i915_translate_compare_func(depth_stencil->stencil[0].func); + int fop = i915_translate_stencil_op(depth_stencil->stencil[0].fail_op); + int dfop = i915_translate_stencil_op(depth_stencil->stencil[0].zfail_op); + int dpop = i915_translate_stencil_op(depth_stencil->stencil[0].zpass_op); + int ref = depth_stencil->stencil[0].ref_value & 0xff; + + cso->stencil_LIS5 |= (S5_STENCIL_TEST_ENABLE | + S5_STENCIL_WRITE_ENABLE | + (ref << S5_STENCIL_REF_SHIFT) | + (test << S5_STENCIL_TEST_FUNC_SHIFT) | + (fop << S5_STENCIL_FAIL_SHIFT) | + (dfop << S5_STENCIL_PASS_Z_FAIL_SHIFT) | + (dpop << S5_STENCIL_PASS_Z_PASS_SHIFT)); + } + + if (depth_stencil->stencil[1].enabled) { + int test = i915_translate_compare_func(depth_stencil->stencil[1].func); + int fop = i915_translate_stencil_op(depth_stencil->stencil[1].fail_op); + int dfop = i915_translate_stencil_op(depth_stencil->stencil[1].zfail_op); + int dpop = i915_translate_stencil_op(depth_stencil->stencil[1].zpass_op); + int ref = depth_stencil->stencil[1].ref_value & 0xff; + int tmask = depth_stencil->stencil[1].value_mask & 0xff; + int wmask = depth_stencil->stencil[1].write_mask & 0xff; + + cso->bfo[0] = (_3DSTATE_BACKFACE_STENCIL_OPS | + BFO_ENABLE_STENCIL_FUNCS | + BFO_ENABLE_STENCIL_TWO_SIDE | + BFO_ENABLE_STENCIL_REF | + BFO_STENCIL_TWO_SIDE | + (ref << BFO_STENCIL_REF_SHIFT) | + (test << BFO_STENCIL_TEST_SHIFT) | + (fop << BFO_STENCIL_FAIL_SHIFT) | + (dfop << BFO_STENCIL_PASS_Z_FAIL_SHIFT) | + (dpop << BFO_STENCIL_PASS_Z_PASS_SHIFT)); + + cso->bfo[1] = (_3DSTATE_BACKFACE_STENCIL_MASKS | + BFM_ENABLE_STENCIL_TEST_MASK | + BFM_ENABLE_STENCIL_WRITE_MASK | + (tmask << BFM_STENCIL_TEST_MASK_SHIFT) | + (wmask << BFM_STENCIL_WRITE_MASK_SHIFT)); + } + else { + /* This actually disables two-side stencil: The bit set is a + * modify-enable bit to indicate we are changing the two-side + * setting. Then there is a symbolic zero to show that we are + * setting the flag to zero/off. + */ + cso->bfo[0] = (_3DSTATE_BACKFACE_STENCIL_OPS | + BFO_ENABLE_STENCIL_TWO_SIDE | + 0); + cso->bfo[1] = 0; + } + + if (depth_stencil->depth.enabled) { + int func = i915_translate_compare_func(depth_stencil->depth.func); + + cso->depth_LIS6 |= (S6_DEPTH_TEST_ENABLE | + (func << S6_DEPTH_TEST_FUNC_SHIFT)); + + if (depth_stencil->depth.writemask) + cso->depth_LIS6 |= S6_DEPTH_WRITE_ENABLE; + } + + if (depth_stencil->alpha.enabled) { + int test = i915_translate_compare_func(depth_stencil->alpha.func); + ubyte refByte = float_to_ubyte(depth_stencil->alpha.ref); + + cso->depth_LIS6 |= (S6_ALPHA_TEST_ENABLE | + (test << S6_ALPHA_TEST_FUNC_SHIFT) | + (((unsigned) refByte) << S6_ALPHA_REF_SHIFT)); + } + + return cso; +} + +static void i915_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->depth_stencil = (const struct i915_depth_stencil_state *)depth_stencil; + + i915->dirty |= I915_NEW_DEPTH_STENCIL; +} + +static void i915_delete_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + FREE(depth_stencil); +} + + +static void i915_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct i915_context *i915 = i915_context(pipe); + + memcpy( &i915->scissor, scissor, sizeof(*scissor) ); + i915->dirty |= I915_NEW_SCISSOR; +} + + +static void i915_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ +} + + +static void * i915_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + return 0; +} + +static void i915_bind_fs_state(struct pipe_context *pipe, void *fs) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->fs = (struct pipe_shader_state *)fs; + + i915->dirty |= I915_NEW_FS; +} + +static void i915_delete_fs_state(struct pipe_context *pipe, void *shader) +{ + /*do nothing*/ +} + +static void * +i915_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct i915_context *i915 = i915_context(pipe); + + /* just pass-through to draw module */ + return draw_create_vertex_shader(i915->draw, templ); +} + +static void i915_bind_vs_state(struct pipe_context *pipe, void *shader) +{ + struct i915_context *i915 = i915_context(pipe); + + /* just pass-through to draw module */ + draw_bind_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader); +} + +static void i915_delete_vs_state(struct pipe_context *pipe, void *shader) +{ + struct i915_context *i915 = i915_context(pipe); + + /* just pass-through to draw module */ + draw_delete_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader); +} + +static void i915_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct i915_context *i915 = i915_context(pipe); + struct pipe_winsys *ws = pipe->winsys; + + assert(shader < PIPE_SHADER_TYPES); + assert(index == 0); + + /* Make a copy of shader constants. + * During fragment program translation we may add additional + * constants to the array. + * + * We want to consider the situation where some user constants + * (ex: a material color) may change frequently but the shader program + * stays the same. In that case we should only be updating the first + * N constants, leaving any extras from shader translation alone. + */ + { + void *mapped; + if (buf->size && + (mapped = ws->buffer_map(ws, buf->buffer, + PIPE_BUFFER_USAGE_CPU_READ))) { + memcpy(i915->current.constants[shader], mapped, buf->size); + ws->buffer_unmap(ws, buf->buffer); + i915->current.num_user_constants[shader] + = buf->size / (4 * sizeof(float)); + } + else { + i915->current.num_user_constants[shader] = 0; + } + } + + i915->dirty |= I915_NEW_CONSTANTS; +} + + +static void i915_set_sampler_texture(struct pipe_context *pipe, + unsigned sampler, + struct pipe_texture *texture) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->texture[sampler] = (struct i915_texture*)texture; /* ptr, not struct */ + + i915->dirty |= I915_NEW_TEXTURE; +} + + + +static void i915_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->framebuffer = *fb; /* struct copy */ + + i915->dirty |= I915_NEW_FRAMEBUFFER; +} + + + +static void i915_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct i915_context *i915 = i915_context(pipe); + + draw_set_clip_state(i915->draw, clip); + + i915->dirty |= I915_NEW_CLIP; +} + + + +/* Called when driver state tracker notices changes to the viewport + * matrix: + */ +static void i915_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->viewport = *viewport; /* struct copy */ + + /* pass the viewport info to the draw module */ + draw_set_viewport_state(i915->draw, &i915->viewport); + + i915->dirty |= I915_NEW_VIEWPORT; +} + + +static void * +i915_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *rasterizer) +{ + struct i915_rasterizer_state *cso = CALLOC_STRUCT( i915_rasterizer_state ); + + cso->templ = rasterizer; + cso->color_interp = rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR; + cso->light_twoside = rasterizer->light_twoside; + cso->ds[0].u = _3DSTATE_DEPTH_OFFSET_SCALE; + cso->ds[1].f = rasterizer->offset_scale; + if (rasterizer->poly_stipple_enable) { + cso->st |= ST1_ENABLE; + } + + if (rasterizer->scissor) + cso->sc[0] = _3DSTATE_SCISSOR_ENABLE_CMD | ENABLE_SCISSOR_RECT; + else + cso->sc[0] = _3DSTATE_SCISSOR_ENABLE_CMD | DISABLE_SCISSOR_RECT; + + switch (rasterizer->cull_mode) { + case PIPE_WINDING_NONE: + cso->LIS4 |= S4_CULLMODE_NONE; + break; + case PIPE_WINDING_CW: + cso->LIS4 |= S4_CULLMODE_CW; + break; + case PIPE_WINDING_CCW: + cso->LIS4 |= S4_CULLMODE_CCW; + break; + case PIPE_WINDING_BOTH: + cso->LIS4 |= S4_CULLMODE_BOTH; + break; + } + + { + int line_width = CLAMP((int)(rasterizer->line_width * 2), 1, 0xf); + + cso->LIS4 |= line_width << S4_LINE_WIDTH_SHIFT; + + if (rasterizer->line_smooth) + cso->LIS4 |= S4_LINE_ANTIALIAS_ENABLE; + } + + { + int point_size = CLAMP((int) rasterizer->point_size, 1, 0xff); + + cso->LIS4 |= point_size << S4_POINT_WIDTH_SHIFT; + } + + if (rasterizer->flatshade) { + cso->LIS4 |= (S4_FLATSHADE_ALPHA | + S4_FLATSHADE_COLOR | + S4_FLATSHADE_SPECULAR); + } + + cso->LIS7 = fui( rasterizer->offset_units ); + + + return cso; +} + +static void i915_bind_rasterizer_state( struct pipe_context *pipe, + void *setup ) +{ + struct i915_context *i915 = i915_context(pipe); + + i915->rasterizer = (struct i915_rasterizer_state *)setup; + + /* pass-through to draw module */ + draw_set_rasterizer_state(i915->draw, i915->rasterizer->templ); + + i915->dirty |= I915_NEW_RASTERIZER; +} + +static void i915_delete_rasterizer_state(struct pipe_context *pipe, + void *setup) +{ + FREE(setup); +} + +static void i915_set_vertex_buffer( struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_buffer *buffer ) +{ + struct i915_context *i915 = i915_context(pipe); + i915->vertex_buffer[index] = *buffer; + /* pass-through to draw module */ + draw_set_vertex_buffer(i915->draw, index, buffer); +} + +static void i915_set_vertex_element( struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_element *element) +{ + struct i915_context *i915 = i915_context(pipe); + /* pass-through to draw module */ + draw_set_vertex_element(i915->draw, index, element); +} + + + +void +i915_init_state_functions( struct i915_context *i915 ) +{ + i915->pipe.create_blend_state = i915_create_blend_state; + i915->pipe.bind_blend_state = i915_bind_blend_state; + i915->pipe.delete_blend_state = i915_delete_blend_state; + + i915->pipe.create_sampler_state = i915_create_sampler_state; + i915->pipe.bind_sampler_state = i915_bind_sampler_state; + i915->pipe.delete_sampler_state = i915_delete_sampler_state; + + i915->pipe.create_depth_stencil_alpha_state = i915_create_depth_stencil_state; + i915->pipe.bind_depth_stencil_alpha_state = i915_bind_depth_stencil_state; + i915->pipe.delete_depth_stencil_alpha_state = i915_delete_depth_stencil_state; + + i915->pipe.create_rasterizer_state = i915_create_rasterizer_state; + i915->pipe.bind_rasterizer_state = i915_bind_rasterizer_state; + i915->pipe.delete_rasterizer_state = i915_delete_rasterizer_state; + i915->pipe.create_fs_state = i915_create_fs_state; + i915->pipe.bind_fs_state = i915_bind_fs_state; + i915->pipe.delete_fs_state = i915_delete_fs_state; + i915->pipe.create_vs_state = i915_create_vs_state; + i915->pipe.bind_vs_state = i915_bind_vs_state; + i915->pipe.delete_vs_state = i915_delete_vs_state; + + i915->pipe.set_blend_color = i915_set_blend_color; + i915->pipe.set_clip_state = i915_set_clip_state; + i915->pipe.set_constant_buffer = i915_set_constant_buffer; + i915->pipe.set_framebuffer_state = i915_set_framebuffer_state; + + i915->pipe.set_polygon_stipple = i915_set_polygon_stipple; + i915->pipe.set_scissor_state = i915_set_scissor_state; + i915->pipe.set_sampler_texture = i915_set_sampler_texture; + i915->pipe.set_viewport_state = i915_set_viewport_state; + i915->pipe.set_vertex_buffer = i915_set_vertex_buffer; + i915->pipe.set_vertex_element = i915_set_vertex_element; +} diff --git a/src/gallium/drivers/i915simple/i915_state.h b/src/gallium/drivers/i915simple/i915_state.h new file mode 100644 index 0000000000..86c6b0027d --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state.h @@ -0,0 +1,50 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef I915_STATE_H +#define I915_STATE_H + +struct i915_context; + + +struct i915_tracked_state { + unsigned dirty; + void (*update)( struct i915_context * ); +}; + +void i915_update_immediate( struct i915_context *i915 ); +void i915_update_dynamic( struct i915_context *i915 ); +void i915_update_derived( struct i915_context *i915 ); +void i915_update_samplers( struct i915_context *i915 ); +void i915_update_textures(struct i915_context *i915); + +void i915_emit_hardware_state( struct i915_context *i915 ); + +#endif diff --git a/src/gallium/drivers/i915simple/i915_state_derived.c b/src/gallium/drivers/i915simple/i915_state_derived.c new file mode 100644 index 0000000000..653983e4a9 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_derived.c @@ -0,0 +1,177 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "pipe/p_util.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_vertex.h" +#include "i915_context.h" +#include "i915_state.h" +#include "i915_reg.h" +#include "i915_fpc.h" +#include "pipe/p_shader_tokens.h" + + +/** + * Determine which post-transform / pre-rasterization vertex attributes + * we need. + * Derived from: fs, setup states. + */ +static void calculate_vertex_layout( struct i915_context *i915 ) +{ + const struct pipe_shader_state *fs = i915->fs; + const enum interp_mode colorInterp = i915->rasterizer->color_interp; + struct vertex_info vinfo; + uint front0 = 0, back0 = 0, front1 = 0, back1 = 0; + boolean needW = 0; + uint i; + boolean texCoords[8]; + uint src = 0; + + memset(texCoords, 0, sizeof(texCoords)); + memset(&vinfo, 0, sizeof(vinfo)); + + /* pos */ + draw_emit_vertex_attr(&vinfo, EMIT_3F, INTERP_LINEAR, src++); + /* Note: we'll set the S4_VFMT_XYZ[W] bits below */ + + for (i = 0; i < fs->num_inputs; i++) { + switch (fs->input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + break; + case TGSI_SEMANTIC_COLOR: + if (fs->input_semantic_index[i] == 0) { + front0 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++); + vinfo.hwfmt[0] |= S4_VFMT_COLOR; + } + else { + assert(fs->input_semantic_index[i] == 1); + front1 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++); + vinfo.hwfmt[0] |= S4_VFMT_SPEC_FOG; + } + break; + case TGSI_SEMANTIC_GENERIC: + /* usually a texcoord */ + { + const uint unit = fs->input_semantic_index[i]; + uint hwtc; + texCoords[unit] = TRUE; + draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE, src++); + hwtc = TEXCOORDFMT_4D; + needW = TRUE; + vinfo.hwfmt[1] |= hwtc << (unit * 4); + } + break; + case TGSI_SEMANTIC_FOG: + debug_printf("i915 fogcoord not implemented yet\n"); + draw_emit_vertex_attr(&vinfo, EMIT_1F, INTERP_PERSPECTIVE, src++); + break; + default: + assert(0); + } + + } + + /* finish up texcoord fields */ + for (i = 0; i < 8; i++) { + if (!texCoords[i]) { + const uint hwtc = TEXCOORDFMT_NOT_PRESENT; + vinfo.hwfmt[1] |= hwtc << (i* 4); + } + } + + /* go back and fill in the vertex position info now that we have needW */ + if (needW) { + vinfo.hwfmt[0] |= S4_VFMT_XYZW; + vinfo.emit[0] = EMIT_4F; + } + else { + vinfo.hwfmt[0] |= S4_VFMT_XYZ; + vinfo.emit[0] = EMIT_3F; + } + + /* Additional attributes required for setup: Just twosided + * lighting. Edgeflag is dealt with specially by setting bits in + * the vertex header. + */ + if (i915->rasterizer->light_twoside) { + if (front0) { + back0 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++); + } + if (back0) { + back1 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++); + } + } + + draw_compute_vertex_size(&vinfo); + + if (memcmp(&i915->current.vertex_info, &vinfo, sizeof(vinfo))) { + /* Need to set this flag so that the LIS2/4 registers get set. + * It also means the i915_update_immediate() function must be called + * after this one, in i915_update_derived(). + */ + i915->dirty |= I915_NEW_VERTEX_FORMAT; + + memcpy(&i915->current.vertex_info, &vinfo, sizeof(vinfo)); + } +} + + + + +/* Hopefully this will remain quite simple, otherwise need to pull in + * something like the state tracker mechanism. + */ +void i915_update_derived( struct i915_context *i915 ) +{ + if (i915->dirty & (I915_NEW_RASTERIZER | I915_NEW_FS)) + calculate_vertex_layout( i915 ); + + if (i915->dirty & (I915_NEW_SAMPLER | I915_NEW_TEXTURE)) + i915_update_samplers(i915); + + if (i915->dirty & I915_NEW_TEXTURE) + i915_update_textures(i915); + + if (i915->dirty) + i915_update_immediate( i915 ); + + if (i915->dirty) + i915_update_dynamic( i915 ); + + if (i915->dirty & I915_NEW_FS) { + i915_translate_fragment_program(i915); + i915->hardware_dirty |= I915_HW_PROGRAM; /* XXX right? */ + } + + /* HW emit currently references framebuffer state directly: + */ + if (i915->dirty & I915_NEW_FRAMEBUFFER) + i915->hardware_dirty |= I915_HW_STATIC; + + i915->dirty = 0; +} diff --git a/src/gallium/drivers/i915simple/i915_state_dynamic.c b/src/gallium/drivers/i915simple/i915_state_dynamic.c new file mode 100644 index 0000000000..8cfbdddd19 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_dynamic.c @@ -0,0 +1,308 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "i915_batch.h" +#include "i915_state_inlines.h" +#include "i915_context.h" +#include "i915_reg.h" +#include "i915_state.h" +#include "pipe/p_util.h" + +#define FILE_DEBUG_FLAG DEBUG_STATE + +/* State that we have chosen to store in the DYNAMIC segment of the + * i915 indirect state mechanism. + * + * Can't cache these in the way we do the static state, as there is no + * start/size in the command packet, instead an 'end' value that gets + * incremented. + * + * Additionally, there seems to be a requirement to re-issue the full + * (active) state every time a 4kb boundary is crossed. + */ + +static INLINE void set_dynamic_indirect( struct i915_context *i915, + unsigned offset, + const unsigned *src, + unsigned dwords ) +{ + unsigned i; + + for (i = 0; i < dwords; i++) + i915->current.dynamic[offset + i] = src[i]; + + i915->hardware_dirty |= I915_HW_DYNAMIC; +} + + +/*********************************************************************** + * Modes4: stencil masks and logicop + */ +static void upload_MODES4( struct i915_context *i915 ) +{ + unsigned modes4 = 0; + + /* I915_NEW_STENCIL */ + modes4 |= i915->depth_stencil->stencil_modes4; + /* I915_NEW_BLEND */ + modes4 |= i915->blend->modes4; + + /* Always, so that we know when state is in-active: + */ + set_dynamic_indirect( i915, + I915_DYNAMIC_MODES4, + &modes4, + 1 ); +} + +const struct i915_tracked_state i915_upload_MODES4 = { + I915_NEW_BLEND | I915_NEW_DEPTH_STENCIL, + upload_MODES4 +}; + + + + +/*********************************************************************** + */ + +static void upload_BFO( struct i915_context *i915 ) +{ + set_dynamic_indirect( i915, + I915_DYNAMIC_BFO_0, + &(i915->depth_stencil->bfo[0]), + 2 ); +} + +const struct i915_tracked_state i915_upload_BFO = { + I915_NEW_DEPTH_STENCIL, + upload_BFO +}; + + +/*********************************************************************** + */ + + +static void upload_BLENDCOLOR( struct i915_context *i915 ) +{ + unsigned bc[2]; + + memset( bc, 0, sizeof(bc) ); + + /* I915_NEW_BLEND {_COLOR} + */ + { + const float *color = i915->blend_color.color; + + bc[0] = _3DSTATE_CONST_BLEND_COLOR_CMD; + bc[1] = pack_ui32_float4( color[0], + color[1], + color[2], + color[3] ); + } + + set_dynamic_indirect( i915, + I915_DYNAMIC_BC_0, + bc, + 2 ); +} + +const struct i915_tracked_state i915_upload_BLENDCOLOR = { + I915_NEW_BLEND, + upload_BLENDCOLOR +}; + +/*********************************************************************** + */ + + +static void upload_IAB( struct i915_context *i915 ) +{ + unsigned iab = i915->blend->iab; + + + set_dynamic_indirect( i915, + I915_DYNAMIC_IAB, + &iab, + 1 ); +} + +const struct i915_tracked_state i915_upload_IAB = { + I915_NEW_BLEND, + upload_IAB +}; + + +/*********************************************************************** + */ + + + +static void upload_DEPTHSCALE( struct i915_context *i915 ) +{ + set_dynamic_indirect( i915, + I915_DYNAMIC_DEPTHSCALE_0, + &(i915->rasterizer->ds[0].u), + 2 ); +} + +const struct i915_tracked_state i915_upload_DEPTHSCALE = { + I915_NEW_RASTERIZER, + upload_DEPTHSCALE +}; + + + +/*********************************************************************** + * Polygon stipple + * + * The i915 supports a 4x4 stipple natively, GL wants 32x32. + * Fortunately stipple is usually a repeating pattern. + * + * XXX: does stipple pattern need to be adjusted according to + * the window position? + * + * XXX: possibly need workaround for conform paths test. + */ + +static void upload_STIPPLE( struct i915_context *i915 ) +{ + unsigned st[2]; + + st[0] = _3DSTATE_STIPPLE; + st[1] = 0; + + /* I915_NEW_RASTERIZER + */ + st[1] |= i915->rasterizer->st; + + + /* I915_NEW_STIPPLE + */ + { + const ubyte *mask = (const ubyte *)i915->poly_stipple.stipple; + ubyte p[4]; + + p[0] = mask[12] & 0xf; + p[1] = mask[8] & 0xf; + p[2] = mask[4] & 0xf; + p[3] = mask[0] & 0xf; + + /* Not sure what to do about fallbacks, so for now just dont: + */ + st[1] |= ((p[0] << 0) | + (p[1] << 4) | + (p[2] << 8) | + (p[3] << 12)); + } + + + set_dynamic_indirect( i915, + I915_DYNAMIC_STP_0, + &st[0], + 2 ); +} + + +const struct i915_tracked_state i915_upload_STIPPLE = { + I915_NEW_RASTERIZER | I915_NEW_STIPPLE, + upload_STIPPLE +}; + + + +/*********************************************************************** + * Scissor. + */ +static void upload_SCISSOR_ENABLE( struct i915_context *i915 ) +{ + set_dynamic_indirect( i915, + I915_DYNAMIC_SC_ENA_0, + &(i915->rasterizer->sc[0]), + 1 ); +} + +const struct i915_tracked_state i915_upload_SCISSOR_ENABLE = { + I915_NEW_RASTERIZER, + upload_SCISSOR_ENABLE +}; + + + +static void upload_SCISSOR_RECT( struct i915_context *i915 ) +{ + unsigned x1 = i915->scissor.minx; + unsigned y1 = i915->scissor.miny; + unsigned x2 = i915->scissor.maxx; + unsigned y2 = i915->scissor.maxy; + unsigned sc[3]; + + sc[0] = _3DSTATE_SCISSOR_RECT_0_CMD; + sc[1] = (y1 << 16) | (x1 & 0xffff); + sc[2] = (y2 << 16) | (x2 & 0xffff); + + set_dynamic_indirect( i915, + I915_DYNAMIC_SC_RECT_0, + &sc[0], + 3 ); +} + + +const struct i915_tracked_state i915_upload_SCISSOR_RECT = { + I915_NEW_SCISSOR, + upload_SCISSOR_RECT +}; + + + + + + +static const struct i915_tracked_state *atoms[] = { + &i915_upload_MODES4, + &i915_upload_BFO, + &i915_upload_BLENDCOLOR, + &i915_upload_IAB, + &i915_upload_DEPTHSCALE, + &i915_upload_STIPPLE, + &i915_upload_SCISSOR_ENABLE, + &i915_upload_SCISSOR_RECT +}; + +/* These will be dynamic indirect state commands, but for now just end + * up on the batch buffer with everything else. + */ +void i915_update_dynamic( struct i915_context *i915 ) +{ + int i; + + for (i = 0; i < Elements(atoms); i++) + if (i915->dirty & atoms[i]->dirty) + atoms[i]->update( i915 ); +} + diff --git a/src/gallium/drivers/i915simple/i915_state_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c new file mode 100644 index 0000000000..3339287f49 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_emit.c @@ -0,0 +1,374 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "i915_reg.h" +#include "i915_context.h" +#include "i915_winsys.h" +#include "i915_batch.h" +#include "i915_reg.h" + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" + +static unsigned translate_format( enum pipe_format format ) +{ + switch (format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + return COLOR_BUF_ARGB8888; + case PIPE_FORMAT_R5G6B5_UNORM: + return COLOR_BUF_RGB565; + default: + assert(0); + return 0; + } +} + +static unsigned translate_depth_format( enum pipe_format zformat ) +{ + switch (zformat) { + case PIPE_FORMAT_S8Z24_UNORM: + return DEPTH_FRMT_24_FIXED_8_OTHER; + case PIPE_FORMAT_Z16_UNORM: + return DEPTH_FRMT_16_FIXED; + default: + assert(0); + return 0; + } +} + + +/** + * Examine framebuffer state to determine width, height. + */ +static boolean +framebuffer_size(const struct pipe_framebuffer_state *fb, + uint *width, uint *height) +{ + if (fb->cbufs[0]) { + *width = fb->cbufs[0]->width; + *height = fb->cbufs[0]->height; + return TRUE; + } + else if (fb->zsbuf) { + *width = fb->zsbuf->width; + *height = fb->zsbuf->height; + return TRUE; + } + else { + *width = *height = 0; + return FALSE; + } +} + + +/* Push the state into the sarea and/or texture memory. + */ +void +i915_emit_hardware_state(struct i915_context *i915 ) +{ + /* XXX: there must be an easier way */ + const unsigned dwords = ( 14 + + 7 + + I915_MAX_DYNAMIC + + 8 + + 2 + I915_TEX_UNITS*3 + + 2 + I915_TEX_UNITS*3 + + 2 + I915_MAX_CONSTANT*4 + + i915->current.program_len + + 6 + ) * 3/2; /* plus 50% margin */ + const unsigned relocs = ( I915_TEX_UNITS + + 3 + ) * 3/2; /* plus 50% margin */ + +#if 0 + debug_printf("i915_emit_hardware_state: %d dwords, %d relocs\n", dwords, relocs); +#endif + + if(!BEGIN_BATCH(dwords, relocs)) { + FLUSH_BATCH(); + assert(BEGIN_BATCH(dwords, relocs)); + } + + /* 14 dwords, 0 relocs */ + if (i915->hardware_dirty & I915_HW_INVARIENT) + { + OUT_BATCH(_3DSTATE_AA_CMD | + AA_LINE_ECAAR_WIDTH_ENABLE | + AA_LINE_ECAAR_WIDTH_1_0 | + AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0); + + OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD); + OUT_BATCH(0); + + OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD); + OUT_BATCH(0); + + OUT_BATCH(_3DSTATE_DFLT_Z_CMD); + OUT_BATCH(0); + + OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS | + CSB_TCB(0, 0) | + CSB_TCB(1, 1) | + CSB_TCB(2, 2) | + CSB_TCB(3, 3) | + CSB_TCB(4, 4) | + CSB_TCB(5, 5) | + CSB_TCB(6, 6) | + CSB_TCB(7, 7)); + + OUT_BATCH(_3DSTATE_RASTER_RULES_CMD | + ENABLE_POINT_RASTER_RULE | + OGL_POINT_RASTER_RULE | + ENABLE_LINE_STRIP_PROVOKE_VRTX | + ENABLE_TRI_FAN_PROVOKE_VRTX | + LINE_STRIP_PROVOKE_VRTX(1) | + TRI_FAN_PROVOKE_VRTX(2) | + ENABLE_TEXKILL_3D_4D | + TEXKILL_4D); + + /* Need to initialize this to zero. + */ + OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (0)); + OUT_BATCH(0); + + OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE); + + /* disable indirect state for now + */ + OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0); + OUT_BATCH(0); + } + + /* 7 dwords, 1 relocs */ + if (i915->hardware_dirty & I915_HW_IMMEDIATE) + { + OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | + I1_LOAD_S(0) | + I1_LOAD_S(1) | + I1_LOAD_S(2) | + I1_LOAD_S(4) | + I1_LOAD_S(5) | + I1_LOAD_S(6) | + (5)); + + if(i915->vbo) + OUT_RELOC(i915->vbo, + I915_BUFFER_ACCESS_READ, + i915->current.immediate[I915_IMMEDIATE_S0]); + else + /* FIXME: we should not do this */ + OUT_BATCH(0); + OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S1]); + OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S2]); + OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S4]); + OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S5]); + OUT_BATCH(i915->current.immediate[I915_IMMEDIATE_S6]); + } + + /* I915_MAX_DYNAMIC dwords, 0 relocs */ + if (i915->hardware_dirty & I915_HW_DYNAMIC) + { + int i; + for (i = 0; i < I915_MAX_DYNAMIC; i++) { + OUT_BATCH(i915->current.dynamic[i]); + } + } + + /* 8 dwords, 2 relocs */ + if (i915->hardware_dirty & I915_HW_STATIC) + { + struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0]; + struct pipe_surface *depth_surface = i915->framebuffer.zsbuf; + + if (cbuf_surface) { + unsigned pitch = (cbuf_surface->pitch * cbuf_surface->cpp); + + OUT_BATCH(_3DSTATE_BUF_INFO_CMD); + + OUT_BATCH(BUF_3D_ID_COLOR_BACK | + BUF_3D_PITCH(pitch) | /* pitch in bytes */ + BUF_3D_USE_FENCE); + + OUT_RELOC(cbuf_surface->buffer, + I915_BUFFER_ACCESS_WRITE, + 0); + } + + /* What happens if no zbuf?? + */ + if (depth_surface) { + unsigned zpitch = (depth_surface->pitch * depth_surface->cpp); + + OUT_BATCH(_3DSTATE_BUF_INFO_CMD); + + OUT_BATCH(BUF_3D_ID_DEPTH | + BUF_3D_PITCH(zpitch) | /* pitch in bytes */ + BUF_3D_USE_FENCE); + + OUT_RELOC(depth_surface->buffer, + I915_BUFFER_ACCESS_WRITE, + 0); + } + + { + unsigned cformat, zformat = 0; + + if (cbuf_surface) + cformat = cbuf_surface->format; + else + cformat = PIPE_FORMAT_A8R8G8B8_UNORM; /* arbitrary */ + cformat = translate_format(cformat); + + if (depth_surface) + zformat = translate_depth_format( i915->framebuffer.zsbuf->format ); + + OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD); + OUT_BATCH(DSTORG_HORT_BIAS(0x8) | /* .5 */ + DSTORG_VERT_BIAS(0x8) | /* .5 */ + LOD_PRECLAMP_OGL | + TEX_DEFAULT_COLOR_OGL | + cformat | + zformat ); + } + } + +#if 01 + /* texture images */ + /* 2 + I915_TEX_UNITS*3 dwords, I915_TEX_UNITS relocs */ + if (i915->hardware_dirty & (I915_HW_MAP | I915_HW_SAMPLER)) + { + /* XXX: we were refering to sampler state + * (current.sampler_enable_nr) below, but only checking + * I915_HW_MAP above. Should probably calculate the enabled + * flags separately - but there will be further rework of + * state so perhaps not necessary yet. + */ + const uint nr = i915->current.sampler_enable_nr; + if (nr) { + const uint enabled = i915->current.sampler_enable_flags; + uint unit; + uint count = 0; + OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr)); + OUT_BATCH(enabled); + for (unit = 0; unit < I915_TEX_UNITS; unit++) { + if (enabled & (1 << unit)) { + struct pipe_buffer *buf = + i915->texture[unit]->buffer; + uint offset = 0; + assert(buf); + + count++; + + OUT_RELOC(buf, + I915_BUFFER_ACCESS_READ, + offset); + OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */ + OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */ + } + } + assert(count == nr); + } + } +#endif + +#if 01 + /* samplers */ + /* 2 + I915_TEX_UNITS*3 dwords, 0 relocs */ + if (i915->hardware_dirty & I915_HW_SAMPLER) + { + if (i915->current.sampler_enable_nr) { + int i; + + OUT_BATCH( _3DSTATE_SAMPLER_STATE | + (3 * i915->current.sampler_enable_nr) ); + + OUT_BATCH( i915->current.sampler_enable_flags ); + + for (i = 0; i < I915_TEX_UNITS; i++) { + if (i915->current.sampler_enable_flags & (1<<i)) { + OUT_BATCH( i915->current.sampler[i][0] ); + OUT_BATCH( i915->current.sampler[i][1] ); + OUT_BATCH( i915->current.sampler[i][2] ); + } + } + } + } +#endif + + /* constants */ + /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */ + if (i915->hardware_dirty & I915_HW_PROGRAM) + { + const uint nr = i915->current.num_constants[PIPE_SHADER_FRAGMENT]; + assert(nr <= I915_MAX_CONSTANT); + if (nr > 0) { + const uint *c + = (const uint *) i915->current.constants[PIPE_SHADER_FRAGMENT]; + uint i; + OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) ); + OUT_BATCH( (1 << (nr - 1)) | ((1 << (nr - 1)) - 1) ); + for (i = 0; i < nr; i++) { + OUT_BATCH(*c++); + OUT_BATCH(*c++); + OUT_BATCH(*c++); + OUT_BATCH(*c++); + } + } + } + + /* Fragment program */ + /* i915->current.program_len dwords, 0 relocs */ + if (i915->hardware_dirty & I915_HW_PROGRAM) + { + uint i; + /* we should always have, at least, a pass-through program */ + assert(i915->current.program_len > 0); + for (i = 0; i < i915->current.program_len; i++) { + OUT_BATCH(i915->current.program[i]); + } + } + + /* drawing surface size */ + /* 6 dwords, 0 relocs */ + { + uint w, h; + boolean k = framebuffer_size(&i915->framebuffer, &w, &h); + assert(k); + + OUT_BATCH(_3DSTATE_DRAW_RECT_CMD); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(((w - 1) & 0xffff) | ((h - 1) << 16)); + OUT_BATCH(0); + OUT_BATCH(0); + } + + + i915->hardware_dirty = 0; +} diff --git a/src/gallium/drivers/i915simple/i915_state_immediate.c b/src/gallium/drivers/i915simple/i915_state_immediate.c new file mode 100644 index 0000000000..07031fc6c5 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_immediate.c @@ -0,0 +1,221 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "i915_state_inlines.h" +#include "i915_context.h" +#include "i915_state.h" +#include "i915_reg.h" +#include "p_util.h" + + +/* All state expressable with the LOAD_STATE_IMMEDIATE_1 packet. + * Would like to opportunistically recombine all these fragments into + * a single packet containing only what has changed, but for now emit + * as multiple packets. + */ + + + + +/*********************************************************************** + * S0,S1: Vertex buffer state. + */ +static void upload_S0S1(struct i915_context *i915) +{ + unsigned LIS0, LIS1; + + /* INTEL_NEW_VBO */ + /* TODO: re-use vertex buffers here? */ + LIS0 = 0; + + /* INTEL_NEW_VERTEX_SIZE -- do this where the vertex size is calculated! + */ + { + unsigned vertex_size = i915->current.vertex_info.size; + + LIS1 = ((vertex_size << 24) | + (vertex_size << 16)); + } + + /* INTEL_NEW_VBO */ + /* TODO: use a vertex generation number to track vbo changes */ + if (1 || + i915->current.immediate[I915_IMMEDIATE_S0] != LIS0 || + i915->current.immediate[I915_IMMEDIATE_S1] != LIS1) + { + i915->current.immediate[I915_IMMEDIATE_S0] = LIS0; + i915->current.immediate[I915_IMMEDIATE_S1] = LIS1; + i915->hardware_dirty |= I915_HW_IMMEDIATE; + } +} + +const struct i915_tracked_state i915_upload_S0S1 = { + I915_NEW_VBO | I915_NEW_VERTEX_FORMAT, + upload_S0S1 +}; + + + + +/*********************************************************************** + * S4: Vertex format, rasterization state + */ +static void upload_S2S4(struct i915_context *i915) +{ + unsigned LIS2, LIS4; + + /* I915_NEW_VERTEX_FORMAT */ + { + LIS2 = i915->current.vertex_info.hwfmt[1]; + LIS4 = i915->current.vertex_info.hwfmt[0]; + /* + debug_printf("LIS2: 0x%x LIS4: 0x%x\n", LIS2, LIS4); + */ + assert(LIS4); /* should never be zero? */ + } + + LIS4 |= i915->rasterizer->LIS4; + + if (LIS2 != i915->current.immediate[I915_IMMEDIATE_S2] || + LIS4 != i915->current.immediate[I915_IMMEDIATE_S4]) { + + i915->current.immediate[I915_IMMEDIATE_S2] = LIS2; + i915->current.immediate[I915_IMMEDIATE_S4] = LIS4; + i915->hardware_dirty |= I915_HW_IMMEDIATE; + } +} + + +const struct i915_tracked_state i915_upload_S2S4 = { + I915_NEW_RASTERIZER | I915_NEW_VERTEX_FORMAT, + upload_S2S4 +}; + + + +/*********************************************************************** + * + */ +static void upload_S5( struct i915_context *i915 ) +{ + unsigned LIS5 = 0; + + LIS5 |= i915->depth_stencil->stencil_LIS5; + + LIS5 |= i915->blend->LIS5; + +#if 0 + /* I915_NEW_RASTERIZER */ + if (i915->state.Polygon->OffsetFill) { + LIS5 |= S5_GLOBAL_DEPTH_OFFSET_ENABLE; + } +#endif + + + if (LIS5 != i915->current.immediate[I915_IMMEDIATE_S5]) { + i915->current.immediate[I915_IMMEDIATE_S5] = LIS5; + i915->hardware_dirty |= I915_HW_IMMEDIATE; + } +} + +const struct i915_tracked_state i915_upload_S5 = { + (I915_NEW_DEPTH_STENCIL | I915_NEW_BLEND | I915_NEW_RASTERIZER), + upload_S5 +}; + + +/*********************************************************************** + */ +static void upload_S6( struct i915_context *i915 ) +{ + unsigned LIS6 = (S6_COLOR_WRITE_ENABLE | + (2 << S6_TRISTRIP_PV_SHIFT)); + + /* I915_NEW_BLEND + */ + LIS6 |= i915->blend->LIS6; + + /* I915_NEW_DEPTH + */ + LIS6 |= i915->depth_stencil->depth_LIS6; + + if (LIS6 != i915->current.immediate[I915_IMMEDIATE_S6]) { + i915->current.immediate[I915_IMMEDIATE_S6] = LIS6; + i915->hardware_dirty |= I915_HW_IMMEDIATE; + } +} + +const struct i915_tracked_state i915_upload_S6 = { + I915_NEW_BLEND | I915_NEW_DEPTH_STENCIL, + upload_S6 +}; + + +/*********************************************************************** + */ +static void upload_S7( struct i915_context *i915 ) +{ + unsigned LIS7; + + /* I915_NEW_RASTERIZER + */ + LIS7 = i915->rasterizer->LIS7; + + if (LIS7 != i915->current.immediate[I915_IMMEDIATE_S7]) { + i915->current.immediate[I915_IMMEDIATE_S7] = LIS7; + i915->hardware_dirty |= I915_HW_IMMEDIATE; + } +} + +const struct i915_tracked_state i915_upload_S7 = { + I915_NEW_RASTERIZER, + upload_S7 +}; + + +static const struct i915_tracked_state *atoms[] = { + &i915_upload_S0S1, + &i915_upload_S2S4, + &i915_upload_S5, + &i915_upload_S6, + &i915_upload_S7 +}; + +/* + */ +void i915_update_immediate( struct i915_context *i915 ) +{ + int i; + + for (i = 0; i < Elements(atoms); i++) + if (i915->dirty & atoms[i]->dirty) + atoms[i]->update( i915 ); +} diff --git a/src/gallium/drivers/i915simple/i915_state_inlines.h b/src/gallium/drivers/i915simple/i915_state_inlines.h new file mode 100644 index 0000000000..0934ac79a4 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_inlines.h @@ -0,0 +1,230 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef I915_STATE_INLINES_H +#define I915_STATE_INLINES_H + +#include "p_compiler.h" +#include "p_defines.h" +#include "i915_reg.h" + + +static INLINE unsigned +i915_translate_compare_func(unsigned func) +{ + switch (func) { + case PIPE_FUNC_NEVER: + return COMPAREFUNC_NEVER; + case PIPE_FUNC_LESS: + return COMPAREFUNC_LESS; + case PIPE_FUNC_LEQUAL: + return COMPAREFUNC_LEQUAL; + case PIPE_FUNC_GREATER: + return COMPAREFUNC_GREATER; + case PIPE_FUNC_GEQUAL: + return COMPAREFUNC_GEQUAL; + case PIPE_FUNC_NOTEQUAL: + return COMPAREFUNC_NOTEQUAL; + case PIPE_FUNC_EQUAL: + return COMPAREFUNC_EQUAL; + case PIPE_FUNC_ALWAYS: + return COMPAREFUNC_ALWAYS; + default: + return COMPAREFUNC_ALWAYS; + } +} + +static INLINE unsigned +i915_translate_stencil_op(unsigned op) +{ + switch (op) { + case PIPE_STENCIL_OP_KEEP: + return STENCILOP_KEEP; + case PIPE_STENCIL_OP_ZERO: + return STENCILOP_ZERO; + case PIPE_STENCIL_OP_REPLACE: + return STENCILOP_REPLACE; + case PIPE_STENCIL_OP_INCR: + return STENCILOP_INCRSAT; + case PIPE_STENCIL_OP_DECR: + return STENCILOP_DECRSAT; + case PIPE_STENCIL_OP_INCR_WRAP: + return STENCILOP_INCR; + case PIPE_STENCIL_OP_DECR_WRAP: + return STENCILOP_DECR; + case PIPE_STENCIL_OP_INVERT: + return STENCILOP_INVERT; + default: + return STENCILOP_ZERO; + } +} + +static INLINE unsigned +i915_translate_blend_factor(unsigned factor) +{ + switch (factor) { + case PIPE_BLENDFACTOR_ZERO: + return BLENDFACT_ZERO; + case PIPE_BLENDFACTOR_SRC_ALPHA: + return BLENDFACT_SRC_ALPHA; + case PIPE_BLENDFACTOR_ONE: + return BLENDFACT_ONE; + case PIPE_BLENDFACTOR_SRC_COLOR: + return BLENDFACT_SRC_COLR; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + return BLENDFACT_INV_SRC_COLR; + case PIPE_BLENDFACTOR_DST_COLOR: + return BLENDFACT_DST_COLR; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + return BLENDFACT_INV_DST_COLR; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + return BLENDFACT_INV_SRC_ALPHA; + case PIPE_BLENDFACTOR_DST_ALPHA: + return BLENDFACT_DST_ALPHA; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + return BLENDFACT_INV_DST_ALPHA; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + return BLENDFACT_SRC_ALPHA_SATURATE; + case PIPE_BLENDFACTOR_CONST_COLOR: + return BLENDFACT_CONST_COLOR; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + return BLENDFACT_INV_CONST_COLOR; + case PIPE_BLENDFACTOR_CONST_ALPHA: + return BLENDFACT_CONST_ALPHA; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + return BLENDFACT_INV_CONST_ALPHA; + default: + return BLENDFACT_ZERO; + } +} + +static INLINE unsigned +i915_translate_blend_func(unsigned mode) +{ + switch (mode) { + case PIPE_BLEND_ADD: + return BLENDFUNC_ADD; + case PIPE_BLEND_MIN: + return BLENDFUNC_MIN; + case PIPE_BLEND_MAX: + return BLENDFUNC_MAX; + case PIPE_BLEND_SUBTRACT: + return BLENDFUNC_SUBTRACT; + case PIPE_BLEND_REVERSE_SUBTRACT: + return BLENDFUNC_REVERSE_SUBTRACT; + default: + return 0; + } +} + + +static INLINE unsigned +i915_translate_logic_op(unsigned opcode) +{ + switch (opcode) { + case PIPE_LOGICOP_CLEAR: + return LOGICOP_CLEAR; + case PIPE_LOGICOP_AND: + return LOGICOP_AND; + case PIPE_LOGICOP_AND_REVERSE: + return LOGICOP_AND_RVRSE; + case PIPE_LOGICOP_COPY: + return LOGICOP_COPY; + case PIPE_LOGICOP_COPY_INVERTED: + return LOGICOP_COPY_INV; + case PIPE_LOGICOP_AND_INVERTED: + return LOGICOP_AND_INV; + case PIPE_LOGICOP_NOOP: + return LOGICOP_NOOP; + case PIPE_LOGICOP_XOR: + return LOGICOP_XOR; + case PIPE_LOGICOP_OR: + return LOGICOP_OR; + case PIPE_LOGICOP_OR_INVERTED: + return LOGICOP_OR_INV; + case PIPE_LOGICOP_NOR: + return LOGICOP_NOR; + case PIPE_LOGICOP_EQUIV: + return LOGICOP_EQUIV; + case PIPE_LOGICOP_INVERT: + return LOGICOP_INV; + case PIPE_LOGICOP_OR_REVERSE: + return LOGICOP_OR_RVRSE; + case PIPE_LOGICOP_NAND: + return LOGICOP_NAND; + case PIPE_LOGICOP_SET: + return LOGICOP_SET; + default: + return LOGICOP_SET; + } +} + + + +static INLINE boolean i915_validate_vertices( unsigned hw_prim, unsigned nr ) +{ + boolean ok; + + switch (hw_prim) { + case PRIM3D_POINTLIST: + ok = (nr >= 1); + assert(ok); + break; + case PRIM3D_LINELIST: + ok = (nr >= 2) && (nr % 2) == 0; + assert(ok); + break; + case PRIM3D_LINESTRIP: + ok = (nr >= 2); + assert(ok); + break; + case PRIM3D_TRILIST: + ok = (nr >= 3) && (nr % 3) == 0; + assert(ok); + break; + case PRIM3D_TRISTRIP: + ok = (nr >= 3); + assert(ok); + break; + case PRIM3D_TRIFAN: + ok = (nr >= 3); + assert(ok); + break; + case PRIM3D_POLY: + ok = (nr >= 3); + assert(ok); + break; + default: + assert(0); + ok = 0; + break; + } + + return ok; +} + +#endif diff --git a/src/gallium/drivers/i915simple/i915_state_sampler.c b/src/gallium/drivers/i915simple/i915_state_sampler.c new file mode 100644 index 0000000000..9c1a5bbbd6 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_state_sampler.c @@ -0,0 +1,231 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_context.h" +#include "pipe/p_state.h" +#include "pipe/p_util.h" + +#include "i915_state_inlines.h" +#include "i915_context.h" +#include "i915_reg.h" +#include "i915_state.h" + + +/** + * Compute i915 texture sampling state. + * + * Recalculate all state from scratch. Perhaps not the most + * efficient, but this has gotten complex enough that we need + * something which is understandable and reliable. + * \param state returns the 3 words of compute state + */ +static void update_sampler(struct i915_context *i915, + uint unit, + const struct i915_sampler_state *sampler, + const struct i915_texture *tex, + unsigned state[3] ) +{ + const struct pipe_texture *pt = &tex->base; + + /* Need to do this after updating the maps, which call the + * intel_finalize_mipmap_tree and hence can update firstLevel: + */ + state[0] = sampler->state[0]; + state[1] = sampler->state[1]; + state[2] = sampler->state[2]; + + if (pt->format == PIPE_FORMAT_YCBCR || + pt->format == PIPE_FORMAT_YCBCR_REV) + state[0] |= SS2_COLORSPACE_CONVERSION; + + /* 3D textures don't seem to respect the border color. + * Fallback if there's ever a danger that they might refer to + * it. + * + * Effectively this means fallback on 3D clamp or + * clamp_to_border. + * + * XXX: Check if this is true on i945. + * XXX: Check if this bug got fixed in release silicon. + */ +#if 0 + { + const unsigned ws = sampler->templ->wrap_s; + const unsigned wt = sampler->templ->wrap_t; + const unsigned wr = sampler->templ->wrap_r; + if (pt->target == PIPE_TEXTURE_3D && + (sampler->templ->min_img_filter != PIPE_TEX_FILTER_NEAREST || + sampler->templ->mag_img_filter != PIPE_TEX_FILTER_NEAREST) && + (ws == PIPE_TEX_WRAP_CLAMP || + wt == PIPE_TEX_WRAP_CLAMP || + wr == PIPE_TEX_WRAP_CLAMP || + ws == PIPE_TEX_WRAP_CLAMP_TO_BORDER || + wt == PIPE_TEX_WRAP_CLAMP_TO_BORDER || + wr == PIPE_TEX_WRAP_CLAMP_TO_BORDER)) { + if (i915->strict_conformance) { + assert(0); + /* sampler->fallback = true; */ + /* TODO */ + } + } + } +#endif + + state[1] |= (unit << SS3_TEXTUREMAP_INDEX_SHIFT); +} + + +void i915_update_samplers( struct i915_context *i915 ) +{ + uint unit; + + i915->current.sampler_enable_nr = 0; + i915->current.sampler_enable_flags = 0x0; + + for (unit = 0; unit < I915_TEX_UNITS; unit++) { + /* determine unit enable/disable by looking for a bound texture */ + /* could also examine the fragment program? */ + if (i915->texture[unit]) { + update_sampler( i915, + unit, + i915->sampler[unit], /* sampler state */ + i915->texture[unit], /* texture */ + i915->current.sampler[unit] /* the result */ + ); + + i915->current.sampler_enable_nr++; + i915->current.sampler_enable_flags |= (1 << unit); + } + } + + i915->hardware_dirty |= I915_HW_SAMPLER; +} + + +static uint +translate_texture_format(enum pipe_format pipeFormat) +{ + switch (pipeFormat) { + case PIPE_FORMAT_U_L8: + return MAPSURF_8BIT | MT_8BIT_L8; + case PIPE_FORMAT_U_I8: + return MAPSURF_8BIT | MT_8BIT_I8; + case PIPE_FORMAT_U_A8: + return MAPSURF_8BIT | MT_8BIT_A8; + case PIPE_FORMAT_U_A8_L8: + return MAPSURF_16BIT | MT_16BIT_AY88; + case PIPE_FORMAT_R5G6B5_UNORM: + return MAPSURF_16BIT | MT_16BIT_RGB565; + case PIPE_FORMAT_A1R5G5B5_UNORM: + return MAPSURF_16BIT | MT_16BIT_ARGB1555; + case PIPE_FORMAT_A4R4G4B4_UNORM: + return MAPSURF_16BIT | MT_16BIT_ARGB4444; + case PIPE_FORMAT_A8R8G8B8_UNORM: + return MAPSURF_32BIT | MT_32BIT_ARGB8888; + case PIPE_FORMAT_YCBCR_REV: + return (MAPSURF_422 | MT_422_YCRCB_NORMAL); + case PIPE_FORMAT_YCBCR: + return (MAPSURF_422 | MT_422_YCRCB_SWAPY); +#if 0 + case PIPE_FORMAT_RGB_FXT1: + case PIPE_FORMAT_RGBA_FXT1: + return (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1); +#endif + case PIPE_FORMAT_Z16_UNORM: + return (MAPSURF_16BIT | MT_16BIT_L16); +#if 0 + case PIPE_FORMAT_RGBA_DXT1: + case PIPE_FORMAT_RGB_DXT1: + return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1); + case PIPE_FORMAT_RGBA_DXT3: + return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3); + case PIPE_FORMAT_RGBA_DXT5: + return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5); +#endif + case PIPE_FORMAT_S8Z24_UNORM: + return (MAPSURF_32BIT | MT_32BIT_xL824); + default: + debug_printf("i915: translate_texture_format() bad image format %x\n", + pipeFormat); + assert(0); + return 0; + } +} + + +static void +i915_update_texture(struct i915_context *i915, uint unit, + uint state[6]) +{ + const struct i915_texture *tex = i915->texture[unit]; + const struct pipe_texture *pt = &tex->base; + uint format, pitch; + const uint width = pt->width[0], height = pt->height[0], depth = pt->depth[0]; + const uint num_levels = pt->last_level; + + assert(tex); + assert(width); + assert(height); + assert(depth); + + format = translate_texture_format(pt->format); + pitch = tex->pitch * pt->cpp; + + assert(format); + assert(pitch); + + /* MS3 state */ + state[0] = + (((height - 1) << MS3_HEIGHT_SHIFT) + | ((width - 1) << MS3_WIDTH_SHIFT) + | format + | MS3_USE_FENCE_REGS); + + /* MS4 state */ + state[1] = + ((((pitch / 4) - 1) << MS4_PITCH_SHIFT) + | MS4_CUBE_FACE_ENA_MASK + | ((num_levels * 4) << MS4_MAX_LOD_SHIFT) + | ((depth - 1) << MS4_VOLUME_DEPTH_SHIFT)); +} + + +void +i915_update_textures(struct i915_context *i915) +{ + uint unit; + + for (unit = 0; unit < I915_TEX_UNITS; unit++) { + /* determine unit enable/disable by looking for a bound texture */ + /* could also examine the fragment program? */ + if (i915->texture[unit]) { + i915_update_texture(i915, unit, i915->current.texbuffer[unit]); + } + } + + i915->hardware_dirty |= I915_HW_MAP; +} diff --git a/src/gallium/drivers/i915simple/i915_strings.c b/src/gallium/drivers/i915simple/i915_strings.c new file mode 100644 index 0000000000..c713bf7208 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_strings.c @@ -0,0 +1,83 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "i915_context.h" +#include "i915_reg.h" + + +static const char *i915_get_vendor( struct pipe_context *pipe ) +{ + return "Tungsten Graphics, Inc."; +} + + +static const char *i915_get_name( struct pipe_context *pipe ) +{ + static char buffer[128]; + const char *chipset; + + switch (i915_context(pipe)->pci_id) { + case PCI_CHIP_I915_G: + chipset = "915G"; + break; + case PCI_CHIP_I915_GM: + chipset = "915GM"; + break; + case PCI_CHIP_I945_G: + chipset = "945G"; + break; + case PCI_CHIP_I945_GM: + chipset = "945GM"; + break; + case PCI_CHIP_I945_GME: + chipset = "945GME"; + break; + case PCI_CHIP_G33_G: + chipset = "G33"; + break; + case PCI_CHIP_Q35_G: + chipset = "Q35"; + break; + case PCI_CHIP_Q33_G: + chipset = "Q33"; + break; + default: + chipset = "unknown"; + break; + } + + sprintf(buffer, "pipe/i915 (chipset: %s)", chipset); + return buffer; +} + + +void +i915_init_string_functions(struct i915_context *i915) +{ + i915->pipe.get_name = i915_get_name; + i915->pipe.get_vendor = i915_get_vendor; +} diff --git a/src/gallium/drivers/i915simple/i915_surface.c b/src/gallium/drivers/i915simple/i915_surface.c new file mode 100644 index 0000000000..de0cc5fe06 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_surface.c @@ -0,0 +1,191 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "i915_context.h" +#include "i915_blit.h" +#include "i915_state.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/util/p_tile.h" + + +/* + * XXX note: same as code in sp_surface.c + */ +static struct pipe_surface * +i915_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice) +{ + struct i915_texture *tex = (struct i915_texture *)pt; + struct pipe_surface *ps; + unsigned offset; /* in bytes */ + + offset = tex->level_offset[level]; + + if (pt->target == PIPE_TEXTURE_CUBE) { + offset += tex->image_offset[level][face] * pt->cpp; + } + else if (pt->target == PIPE_TEXTURE_3D) { + offset += tex->image_offset[level][zslice] * pt->cpp; + } + else { + assert(face == 0); + assert(zslice == 0); + } + + ps = pipe->winsys->surface_alloc(pipe->winsys); + if (ps) { + assert(ps->refcount); + assert(ps->winsys); + pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer); + ps->format = pt->format; + ps->cpp = pt->cpp; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->pitch = tex->pitch; + ps->offset = offset; + } + return ps; +} + + + +/* Assumes all values are within bounds -- no checking at this level - + * do it higher up if required. + */ +static void +i915_surface_copy(struct pipe_context *pipe, + unsigned do_flip, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + struct pipe_surface *src, + unsigned srcx, unsigned srcy, unsigned width, unsigned height) +{ + assert( dst != src ); + assert( dst->cpp == src->cpp ); + + if (0) { + pipe_copy_rect(pipe_surface_map(dst), + dst->cpp, + dst->pitch, + dstx, dsty, + width, height, + pipe_surface_map(src), + do_flip ? -(int) src->pitch : src->pitch, + srcx, do_flip ? 1 - srcy - height : srcy); + + pipe_surface_unmap(src); + pipe_surface_unmap(dst); + } + else { + i915_copy_blit( i915_context(pipe), + do_flip, + dst->cpp, + (short) src->pitch, src->buffer, src->offset, + (short) dst->pitch, dst->buffer, dst->offset, + (short) srcx, (short) srcy, (short) dstx, (short) dsty, (short) width, (short) height ); + } +} + +/* Fill a rectangular sub-region. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +static void * +get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y) +{ + return (char *)dst_map + (y * dst->pitch + x) * dst->cpp; +} + + +static void +i915_surface_fill(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, unsigned value) +{ + if (0) { + unsigned i, j; + void *dst_map = pipe_surface_map(dst); + + switch (dst->cpp) { + case 1: { + ubyte *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + memset(row, value, width); + row += dst->pitch; + } + } + break; + case 2: { + ushort *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = (ushort) value; + row += dst->pitch; + } + } + break; + case 4: { + unsigned *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = value; + row += dst->pitch; + } + } + break; + default: + assert(0); + break; + } + + pipe_surface_unmap( dst ); + } + else { + i915_fill_blit( i915_context(pipe), + dst->cpp, + (short) dst->pitch, + dst->buffer, dst->offset, + (short) dstx, (short) dsty, + (short) width, (short) height, + value ); + } +} + + +void +i915_init_surface_functions(struct i915_context *i915) +{ + i915->pipe.get_tex_surface = i915_get_tex_surface; + + i915->pipe.surface_copy = i915_surface_copy; + i915->pipe.surface_fill = i915_surface_fill; +} diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c new file mode 100644 index 0000000000..6d37ae3d74 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_texture.c @@ -0,0 +1,536 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + * Michel Dänzer <michel@tungstengraphics.com> + */ + +#include "pipe/p_state.h" +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "pipe/p_winsys.h" + +#include "i915_context.h" +#include "i915_texture.h" +#include "i915_debug.h" + + +static unsigned minify( unsigned d ) +{ + return MAX2(1, d>>1); +} + + + +static void +i915_miptree_set_level_info(struct i915_texture *tex, + unsigned level, + unsigned nr_images, + unsigned x, unsigned y, unsigned w, unsigned h, unsigned d) +{ + struct pipe_texture *pt = &tex->base; + + assert(level < PIPE_MAX_TEXTURE_LEVELS); + + pt->width[level] = w; + pt->height[level] = h; + pt->depth[level] = d; + + tex->level_offset[level] = (x + y * tex->pitch) * pt->cpp; + tex->nr_images[level] = nr_images; + + /* + DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__, + level, w, h, d, x, y, tex->level_offset[level]); + */ + + /* Not sure when this would happen, but anyway: + */ + if (tex->image_offset[level]) { + FREE(tex->image_offset[level]); + tex->image_offset[level] = NULL; + } + + assert(nr_images); + assert(!tex->image_offset[level]); + + tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned)); + tex->image_offset[level][0] = 0; +} + + +static void +i915_miptree_set_image_offset(struct i915_texture *tex, + unsigned level, unsigned img, unsigned x, unsigned y) +{ + if (img == 0 && level == 0) + assert(x == 0 && y == 0); + + assert(img < tex->nr_images[level]); + + tex->image_offset[level][img] = (x + y * tex->pitch); + + /* + DBG("%s level %d img %d pos %d,%d image_offset %x\n", + __FUNCTION__, level, img, x, y, tex->image_offset[level][img]); + */ +} + + +static void +i945_miptree_layout_2d( struct i915_texture *tex ) +{ + struct pipe_texture *pt = &tex->base; + int align_h = 2, align_w = 4; + unsigned level; + unsigned x = 0; + unsigned y = 0; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + + tex->pitch = pt->width[0]; + + /* May need to adjust pitch to accomodate the placement of + * the 2nd mipmap level. This occurs when the alignment + * constraints of mipmap placement push the right edge of the + * 2nd mipmap level out past the width of its parent. + */ + if (pt->last_level > 0) { + unsigned mip1_width = align_int(minify(pt->width[0]), align_w) + + minify(minify(pt->width[0])); + + if (mip1_width > pt->width[0]) + tex->pitch = mip1_width; + } + + /* Pitch must be a whole number of dwords, even though we + * express it in texels. + */ + tex->pitch = align_int(tex->pitch * pt->cpp, 4) / pt->cpp; + tex->total_height = 0; + + for (level = 0; level <= pt->last_level; level++) { + unsigned img_height; + + i915_miptree_set_level_info(tex, level, 1, x, y, width, height, 1); + + if (pt->compressed) + img_height = MAX2(1, height/4); + else + img_height = align_int(height, align_h); + + + /* Because the images are packed better, the final offset + * might not be the maximal one: + */ + tex->total_height = MAX2(tex->total_height, y + img_height); + + /* Layout_below: step right after second mipmap level. + */ + if (level == 1) { + x += align_int(width, align_w); + } + else { + y += img_height; + } + + width = minify(width); + height = minify(height); + } +} + + +static const int initial_offsets[6][2] = { + {0, 0}, + {0, 2}, + {1, 0}, + {1, 2}, + {1, 1}, + {1, 3} +}; + +static const int step_offsets[6][2] = { + {0, 2}, + {0, 2}, + {-1, 2}, + {-1, 2}, + {-1, 1}, + {-1, 1} +}; + + +static boolean +i915_miptree_layout(struct pipe_context *pipe, struct i915_texture * tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned level; + + switch (pt->target) { + case PIPE_TEXTURE_CUBE: { + const unsigned dim = pt->width[0]; + unsigned face; + unsigned lvlWidth = pt->width[0], lvlHeight = pt->height[0]; + + assert(lvlWidth == lvlHeight); /* cubemap images are square */ + + /* double pitch for cube layouts */ + tex->pitch = ((dim * pt->cpp * 2 + 3) & ~3) / pt->cpp; + tex->total_height = dim * 4; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_level_info(tex, level, 6, + 0, 0, + /*OLD: tex->pitch, tex->total_height,*/ + lvlWidth, lvlHeight, + 1); + lvlWidth /= 2; + lvlHeight /= 2; + } + + for (face = 0; face < 6; face++) { + unsigned x = initial_offsets[face][0] * dim; + unsigned y = initial_offsets[face][1] * dim; + unsigned d = dim; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_image_offset(tex, level, face, x, y); + d >>= 1; + x += step_offsets[face][0] * d; + y += step_offsets[face][1] * d; + } + } + break; + } + case PIPE_TEXTURE_3D:{ + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + unsigned stack_height = 0; + + /* Calculate the size of a single slice. + */ + tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp; + + /* XXX: hardware expects/requires 9 levels at minimum. + */ + for (level = 0; level <= MAX2(8, pt->last_level); + level++) { + i915_miptree_set_level_info(tex, level, depth, 0, tex->total_height, + width, height, depth); + + + stack_height += MAX2(2, height); + + width = minify(width); + height = minify(height); + depth = minify(depth); + } + + /* Fixup depth image_offsets: + */ + depth = pt->depth[0]; + for (level = 0; level <= pt->last_level; level++) { + unsigned i; + for (i = 0; i < depth; i++) + i915_miptree_set_image_offset(tex, level, i, + 0, i * stack_height); + + depth = minify(depth); + } + + + /* Multiply slice size by texture depth for total size. It's + * remarkable how wasteful of memory the i915 texture layouts + * are. They are largely fixed in the i945. + */ + tex->total_height = stack_height * pt->depth[0]; + break; + } + + default:{ + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned img_height; + + tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp; + tex->total_height = 0; + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_level_info(tex, level, 1, + 0, tex->total_height, + width, height, 1); + + if (pt->compressed) + img_height = MAX2(1, height / 4); + else + img_height = (MAX2(2, height) + 1) & ~1; + + tex->total_height += img_height; + + width = minify(width); + height = minify(height); + } + break; + } + } + /* + DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, + tex->pitch, + tex->total_height, pt->cpp, tex->pitch * tex->total_height * pt->cpp); + */ + + return TRUE; +} + + +static boolean +i945_miptree_layout(struct pipe_context *pipe, struct i915_texture * tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned level; + + switch (pt->target) { + case PIPE_TEXTURE_CUBE:{ + const unsigned dim = pt->width[0]; + unsigned face; + unsigned lvlWidth = pt->width[0], lvlHeight = pt->height[0]; + + assert(lvlWidth == lvlHeight); /* cubemap images are square */ + + /* Depending on the size of the largest images, pitch can be + * determined either by the old-style packing of cubemap faces, + * or the final row of 4x4, 2x2 and 1x1 faces below this. + */ + if (dim > 32) + tex->pitch = ((dim * pt->cpp * 2 + 3) & ~3) / pt->cpp; + else + tex->pitch = 14 * 8; + + tex->total_height = dim * 4 + 4; + + /* Set all the levels to effectively occupy the whole rectangular region. + */ + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_level_info(tex, level, 6, + 0, 0, + lvlWidth, lvlHeight, 1); + lvlWidth /= 2; + lvlHeight /= 2; + } + + + for (face = 0; face < 6; face++) { + unsigned x = initial_offsets[face][0] * dim; + unsigned y = initial_offsets[face][1] * dim; + unsigned d = dim; + + if (dim == 4 && face >= 4) { + y = tex->total_height - 4; + x = (face - 4) * 8; + } + else if (dim < 4 && (face > 0)) { + y = tex->total_height - 4; + x = face * 8; + } + + for (level = 0; level <= pt->last_level; level++) { + i915_miptree_set_image_offset(tex, level, face, x, y); + + d >>= 1; + + switch (d) { + case 4: + switch (face) { + case PIPE_TEX_FACE_POS_X: + case PIPE_TEX_FACE_NEG_X: + x += step_offsets[face][0] * d; + y += step_offsets[face][1] * d; + break; + case PIPE_TEX_FACE_POS_Y: + case PIPE_TEX_FACE_NEG_Y: + y += 12; + x -= 8; + break; + case PIPE_TEX_FACE_POS_Z: + case PIPE_TEX_FACE_NEG_Z: + y = tex->total_height - 4; + x = (face - 4) * 8; + break; + } + + case 2: + y = tex->total_height - 4; + x = 16 + face * 8; + break; + + case 1: + x += 48; + break; + + default: + x += step_offsets[face][0] * d; + y += step_offsets[face][1] * d; + break; + } + } + } + break; + } + case PIPE_TEXTURE_3D:{ + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + unsigned pack_x_pitch, pack_x_nr; + unsigned pack_y_pitch; + unsigned level; + + tex->pitch = ((pt->width[0] * pt->cpp + 3) & ~3) / pt->cpp; + tex->total_height = 0; + + pack_y_pitch = MAX2(pt->height[0], 2); + pack_x_pitch = tex->pitch; + pack_x_nr = 1; + + for (level = 0; level <= pt->last_level; level++) { + unsigned nr_images = pt->target == PIPE_TEXTURE_3D ? depth : 6; + int x = 0; + int y = 0; + unsigned q, j; + + i915_miptree_set_level_info(tex, level, nr_images, + 0, tex->total_height, + width, height, depth); + + for (q = 0; q < nr_images;) { + for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) { + i915_miptree_set_image_offset(tex, level, q, x, y); + x += pack_x_pitch; + } + + x = 0; + y += pack_y_pitch; + } + + + tex->total_height += y; + + if (pack_x_pitch > 4) { + pack_x_pitch >>= 1; + pack_x_nr <<= 1; + assert(pack_x_pitch * pack_x_nr <= tex->pitch); + } + + if (pack_y_pitch > 2) { + pack_y_pitch >>= 1; + } + + width = minify(width); + height = minify(height); + depth = minify(depth); + } + break; + } + + case PIPE_TEXTURE_1D: + case PIPE_TEXTURE_2D: +// case PIPE_TEXTURE_RECTANGLE: + i945_miptree_layout_2d(tex); + break; + default: + assert(0); + return FALSE; + } + + /* + DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, + tex->pitch, + tex->total_height, pt->cpp, tex->pitch * tex->total_height * pt->cpp); + */ + + return TRUE; +} + + +struct pipe_texture * +i915_texture_create(struct pipe_context *pipe, + const struct pipe_texture *templat) +{ + struct i915_texture *tex = CALLOC_STRUCT(i915_texture); + + if (tex) { + struct i915_context *i915 = i915_context(pipe); + + tex->base = *templat; + + if (i915->flags.is_i945 ? i945_miptree_layout(pipe, tex) : + i915_miptree_layout(pipe, tex)) + tex->buffer = pipe->winsys->buffer_create(pipe->winsys, 64, + PIPE_BUFFER_USAGE_PIXEL, + tex->pitch * tex->base.cpp * + tex->total_height); + + if (!tex->buffer) { + FREE(tex); + return NULL; + } + } + + return &tex->base; +} + + +void +i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt) +{ + if (!*pt) + return; + + /* + DBG("%s %p refcount will be %d\n", + __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); + */ + if (--(*pt)->refcount <= 0) { + struct i915_texture *tex = (struct i915_texture *)*pt; + uint i; + + /* + DBG("%s deleting %p\n", __FUNCTION__, (void *) tex); + */ + + pipe_buffer_reference(pipe->winsys, &tex->buffer, NULL); + + for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++) + if (tex->image_offset[i]) + FREE(tex->image_offset[i]); + + FREE(tex); + } + *pt = NULL; +} diff --git a/src/gallium/drivers/i915simple/i915_texture.h b/src/gallium/drivers/i915simple/i915_texture.h new file mode 100644 index 0000000000..330d111dc7 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_texture.h @@ -0,0 +1,17 @@ + +#ifndef I915_TEXTURE_H +#define I915_TEXTURE_H + +struct pipe_context; +struct pipe_texture; + + +struct pipe_texture * +i915_texture_create(struct pipe_context *pipe, + const struct pipe_texture *templat); + +extern void +i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt); + + +#endif /* I915_TEXTURE_H */ diff --git a/src/gallium/drivers/i915simple/i915_winsys.h b/src/gallium/drivers/i915simple/i915_winsys.h new file mode 100644 index 0000000000..fe49710852 --- /dev/null +++ b/src/gallium/drivers/i915simple/i915_winsys.h @@ -0,0 +1,115 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \file + * This is the interface that i915simple requires any window system + * hosting it to implement. This is the only include file in i915simple + * which is public. + * + */ + +#ifndef I915_WINSYS_H +#define I915_WINSYS_H + + +#include "pipe/p_defines.h" + + +/* Pipe drivers are (meant to be!) independent of both GL and the + * window system. The window system provides a buffer manager and a + * set of additional hooks for things like command buffer submission, + * etc. + * + * There clearly has to be some agreement between the window system + * driver and the hardware driver about the format of command buffers, + * etc. + */ + +struct pipe_buffer; +struct pipe_winsys; + + +/** + * Additional winsys interface for i915simple. + * + * It is an over-simple batchbuffer mechanism. Will want to improve the + * performance of this, perhaps based on the cmdstream stuff. It + * would be pretty impossible to implement swz on top of this + * interface. + * + * Will also need additions/changes to implement static/dynamic + * indirect state. + */ +struct i915_winsys { + + /** + * Reserve space on batch buffer. + * + * Returns a null pointer if there is insufficient space in the batch buffer + * to hold the requested number of dwords and relocations. + * + * The number of dwords should also include the number of relocations. + */ + unsigned *(*batch_start)( struct i915_winsys *sws, + unsigned dwords, + unsigned relocs ); + + void (*batch_dword)( struct i915_winsys *sws, + unsigned dword ); + + /** + * Emit a relocation to a buffer. + * + * Used not only when the buffer addresses are not pinned, but also to + * ensure refered buffers will not be destroyed until the current batch + * buffer execution is finished. + * + * The access flags is a combination of I915_BUFFER_ACCESS_WRITE and + * I915_BUFFER_ACCESS_READ macros. + */ + void (*batch_reloc)( struct i915_winsys *sws, + struct pipe_buffer *buf, + unsigned access_flags, + unsigned delta ); + + void (*batch_flush)( struct i915_winsys *sws ); + void (*batch_finish)( struct i915_winsys *sws ); +}; + +#define I915_BUFFER_ACCESS_WRITE 0x1 +#define I915_BUFFER_ACCESS_READ 0x2 + +#define I915_BUFFER_USAGE_LIT_VERTEX (PIPE_BUFFER_USAGE_CUSTOM << 0) + + +struct pipe_context *i915_create( struct pipe_winsys *, + struct i915_winsys *, + unsigned pci_id ); + + +#endif diff --git a/src/gallium/drivers/i965simple/Makefile b/src/gallium/drivers/i965simple/Makefile new file mode 100644 index 0000000000..48c00ab50b --- /dev/null +++ b/src/gallium/drivers/i965simple/Makefile @@ -0,0 +1,66 @@ + +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = i965simple + +DRIVER_SOURCES = \ + brw_blit.c \ + brw_flush.c \ + brw_strings.c \ + brw_surface.c \ + brw_cc.c \ + brw_clip.c \ + brw_clip_line.c \ + brw_clip_point.c \ + brw_clip_state.c \ + brw_clip_tri.c \ + brw_clip_util.c \ + brw_context.c \ + brw_curbe.c \ + brw_draw.c \ + brw_draw_upload.c \ + brw_eu.c \ + brw_eu_debug.c \ + brw_eu_emit.c \ + brw_eu_util.c \ + brw_gs.c \ + brw_gs_emit.c \ + brw_gs_state.c \ + brw_misc_state.c \ + brw_sf.c \ + brw_sf_emit.c \ + brw_sf_state.c \ + brw_shader_info.c \ + brw_state.c \ + brw_state_batch.c \ + brw_state_cache.c \ + brw_state_pool.c \ + brw_state_upload.c \ + brw_tex_layout.c \ + brw_urb.c \ + brw_util.c \ + brw_vs.c \ + brw_vs_emit.c \ + brw_vs_state.c \ + brw_wm.c \ + brw_wm_iz.c \ + brw_wm_decl.c \ + brw_wm_glsl.c \ + brw_wm_sampler_state.c \ + brw_wm_state.c \ + brw_wm_surface_state.c + +C_SOURCES = \ + $(COMMON_SOURCES) \ + $(COMMON_BM_SOURCES) \ + $(MINIGLX_SOURCES) \ + $(DRIVER_SOURCES) + +ASM_SOURCES = + +DRIVER_DEFINES = -I. + +include ../Makefile.template + +symlinks: diff --git a/src/gallium/drivers/i965simple/SConscript b/src/gallium/drivers/i965simple/SConscript new file mode 100644 index 0000000000..74621de84c --- /dev/null +++ b/src/gallium/drivers/i965simple/SConscript @@ -0,0 +1,55 @@ +Import('*') + +env = env.Clone() + +i965simple = env.ConvenienceLibrary( + target = 'i965simple', + source = [ + 'brw_blit.c', + 'brw_cc.c', + 'brw_clip.c', + 'brw_clip_line.c', + 'brw_clip_point.c', + 'brw_clip_state.c', + 'brw_clip_tri.c', + 'brw_clip_util.c', + 'brw_context.c', + 'brw_curbe.c', + 'brw_draw.c', + 'brw_draw_upload.c', + 'brw_eu.c', + 'brw_eu_debug.c', + 'brw_eu_emit.c', + 'brw_eu_util.c', + 'brw_flush.c', + 'brw_gs.c', + 'brw_gs_emit.c', + 'brw_gs_state.c', + 'brw_misc_state.c', + 'brw_sf.c', + 'brw_sf_emit.c', + 'brw_sf_state.c', + 'brw_shader_info.c', + 'brw_state.c', + 'brw_state_batch.c', + 'brw_state_cache.c', + 'brw_state_pool.c', + 'brw_state_upload.c', + 'brw_strings.c', + 'brw_surface.c', + 'brw_tex_layout.c', + 'brw_urb.c', + 'brw_util.c', + 'brw_vs.c', + 'brw_vs_emit.c', + 'brw_vs_state.c', + 'brw_wm.c', + 'brw_wm_decl.c', + 'brw_wm_glsl.c', + 'brw_wm_iz.c', + 'brw_wm_sampler_state.c', + 'brw_wm_state.c', + 'brw_wm_surface_state.c', + ]) + +Export('i965simple') diff --git a/src/gallium/drivers/i965simple/brw_batch.h b/src/gallium/drivers/i965simple/brw_batch.h new file mode 100644 index 0000000000..5f5932a488 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_batch.h @@ -0,0 +1,59 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef BRW_BATCH_H +#define BRW_BATCH_H + +#include "brw_winsys.h" + +#define BATCH_LOCALS + +#define INTEL_BATCH_NO_CLIPRECTS 0x1 +#define INTEL_BATCH_CLIPRECTS 0x2 + +#define BEGIN_BATCH( dwords, relocs ) \ + brw->winsys->batch_start(brw->winsys, dwords, relocs) + +#define OUT_BATCH( dword ) \ + brw->winsys->batch_dword(brw->winsys, dword) + +#define OUT_RELOC( buf, flags, delta ) \ + brw->winsys->batch_reloc(brw->winsys, buf, flags, delta) + +#define ADVANCE_BATCH() \ + brw->winsys->batch_end( brw->winsys ) + +/* XXX: this is bogus - need proper handling for out-of-memory in batchbuffer. + */ +#define FLUSH_BATCH(fence) do { \ + brw->winsys->batch_flush(brw->winsys, fence); \ + brw->hardware_dirty = ~0; \ +} while (0) + +#define BRW_BATCH_STRUCT(brw, s) brw_batchbuffer_data( brw->winsys, (s), sizeof(*(s))) + +#endif diff --git a/src/gallium/drivers/i965simple/brw_blit.c b/src/gallium/drivers/i965simple/brw_blit.c new file mode 100644 index 0000000000..8494f70493 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_blit.c @@ -0,0 +1,218 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include <stdio.h> +#include <errno.h> + +#include "brw_batch.h" +#include "brw_blit.h" +#include "brw_context.h" +#include "brw_reg.h" + +#include "pipe/p_context.h" +#include "pipe/p_winsys.h" + +#define FILE_DEBUG_FLAG DEBUG_BLIT + +void brw_fill_blit(struct brw_context *brw, + unsigned cpp, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + boolean dst_tiled, + short x, short y, + short w, short h, + unsigned color) +{ + unsigned BR13, CMD; + BATCH_LOCALS; + + dst_pitch *= cpp; + + switch(cpp) { + case 1: + case 2: + case 3: + BR13 = (0xF0 << 16) | (1<<24); + CMD = XY_COLOR_BLT_CMD; + break; + case 4: + BR13 = (0xF0 << 16) | (1<<24) | (1<<25); + CMD = XY_COLOR_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + break; + default: + return; + } + + if (dst_tiled) { + CMD |= XY_DST_TILED; + dst_pitch /= 4; + } + + BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH( CMD ); + OUT_BATCH( dst_pitch | BR13 ); + OUT_BATCH( (y << 16) | x ); + OUT_BATCH( ((y+h) << 16) | (x+w) ); + OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE, dst_offset ); + OUT_BATCH( color ); + ADVANCE_BATCH(); +} + +static unsigned translate_raster_op(unsigned logicop) +{ + switch(logicop) { + case PIPE_LOGICOP_CLEAR: return 0x00; + case PIPE_LOGICOP_AND: return 0x88; + case PIPE_LOGICOP_AND_REVERSE: return 0x44; + case PIPE_LOGICOP_COPY: return 0xCC; + case PIPE_LOGICOP_AND_INVERTED: return 0x22; + case PIPE_LOGICOP_NOOP: return 0xAA; + case PIPE_LOGICOP_XOR: return 0x66; + case PIPE_LOGICOP_OR: return 0xEE; + case PIPE_LOGICOP_NOR: return 0x11; + case PIPE_LOGICOP_EQUIV: return 0x99; + case PIPE_LOGICOP_INVERT: return 0x55; + case PIPE_LOGICOP_OR_REVERSE: return 0xDD; + case PIPE_LOGICOP_COPY_INVERTED: return 0x33; + case PIPE_LOGICOP_OR_INVERTED: return 0xBB; + case PIPE_LOGICOP_NAND: return 0x77; + case PIPE_LOGICOP_SET: return 0xFF; + default: return 0; + } +} + + +/* Copy BitBlt + */ +void brw_copy_blit(struct brw_context *brw, + unsigned do_flip, + unsigned cpp, + short src_pitch, + struct pipe_buffer *src_buffer, + unsigned src_offset, + boolean src_tiled, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + boolean dst_tiled, + short src_x, short src_y, + short dst_x, short dst_y, + short w, short h, + unsigned logic_op) +{ + unsigned CMD, BR13; + int dst_y2 = dst_y + h; + int dst_x2 = dst_x + w; + BATCH_LOCALS; + + + DBG("%s src:buf(%d)/%d %d,%d dst:buf(%d)/%d %d,%d sz:%dx%d op:%d\n", + __FUNCTION__, + src_buffer, src_pitch, src_x, src_y, + dst_buffer, dst_pitch, dst_x, dst_y, + w,h,logic_op); + + assert( logic_op - PIPE_LOGICOP_CLEAR >= 0 ); + assert( logic_op - PIPE_LOGICOP_CLEAR < 0x10 ); + + src_pitch *= cpp; + dst_pitch *= cpp; + + switch(cpp) { + case 1: + case 2: + case 3: + BR13 = (translate_raster_op(logic_op) << 16) | (1<<24); + CMD = XY_SRC_COPY_BLT_CMD; + break; + case 4: + BR13 = (translate_raster_op(logic_op) << 16) | (1<<24) | + (1<<25); + CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB; + break; + default: + return; + } + + if (src_tiled) { + CMD |= XY_SRC_TILED; + src_pitch /= 4; + } + + if (dst_tiled) { + CMD |= XY_DST_TILED; + dst_pitch /= 4; + } + + if (dst_y2 < dst_y || + dst_x2 < dst_x) { + return; + } + + dst_pitch &= 0xffff; + src_pitch &= 0xffff; + + /* Initial y values don't seem to work with negative pitches. If + * we adjust the offsets manually (below), it seems to work fine. + * + * On the other hand, if we always adjust, the hardware doesn't + * know which blit directions to use, so overlapping copypixels get + * the wrong result. + */ + if (dst_pitch > 0 && src_pitch > 0) { + BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH( CMD ); + OUT_BATCH( dst_pitch | BR13 ); + OUT_BATCH( (dst_y << 16) | dst_x ); + OUT_BATCH( (dst_y2 << 16) | dst_x2 ); + OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE, + dst_offset ); + OUT_BATCH( (src_y << 16) | src_x ); + OUT_BATCH( src_pitch ); + OUT_RELOC( src_buffer, BRW_BUFFER_ACCESS_READ, + src_offset ); + ADVANCE_BATCH(); + } + else { + BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH( CMD ); + OUT_BATCH( (dst_pitch & 0xffff) | BR13 ); + OUT_BATCH( (0 << 16) | dst_x ); + OUT_BATCH( (h << 16) | dst_x2 ); + OUT_RELOC( dst_buffer, BRW_BUFFER_ACCESS_WRITE, + dst_offset + dst_y * dst_pitch ); + OUT_BATCH( (src_pitch & 0xffff) ); + OUT_RELOC( src_buffer, BRW_BUFFER_ACCESS_READ, + src_offset + src_y * src_pitch ); + ADVANCE_BATCH(); + } +} + + + diff --git a/src/gallium/drivers/i965simple/brw_blit.h b/src/gallium/drivers/i965simple/brw_blit.h new file mode 100644 index 0000000000..111c5d91d3 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_blit.h @@ -0,0 +1,33 @@ +#ifndef BRW_BLIT_H +#define BRW_BLIT_H + +#include "pipe/p_compiler.h" + +struct pipe_buffer; +struct brw_context; + +void brw_fill_blit(struct brw_context *intel, + unsigned cpp, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + boolean dst_tiled, + short x, short y, + short w, short h, + unsigned color); +void brw_copy_blit(struct brw_context *intel, + unsigned do_flip, + unsigned cpp, + short src_pitch, + struct pipe_buffer *src_buffer, + unsigned src_offset, + boolean src_tiled, + short dst_pitch, + struct pipe_buffer *dst_buffer, + unsigned dst_offset, + boolean dst_tiled, + short src_x, short src_y, + short dst_x, short dst_y, + short w, short h, + unsigned logic_op); +#endif diff --git a/src/gallium/drivers/i965simple/brw_cc.c b/src/gallium/drivers/i965simple/brw_cc.c new file mode 100644 index 0000000000..337e4f95f6 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_cc.c @@ -0,0 +1,269 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "pipe/p_util.h" + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "brw_util.h" + + +static int brw_translate_compare_func(int func) +{ + switch(func) { + case PIPE_FUNC_NEVER: + return BRW_COMPAREFUNCTION_NEVER; + case PIPE_FUNC_LESS: + return BRW_COMPAREFUNCTION_LESS; + case PIPE_FUNC_LEQUAL: + return BRW_COMPAREFUNCTION_LEQUAL; + case PIPE_FUNC_GREATER: + return BRW_COMPAREFUNCTION_GREATER; + case PIPE_FUNC_GEQUAL: + return BRW_COMPAREFUNCTION_GEQUAL; + case PIPE_FUNC_NOTEQUAL: + return BRW_COMPAREFUNCTION_NOTEQUAL; + case PIPE_FUNC_EQUAL: + return BRW_COMPAREFUNCTION_EQUAL; + case PIPE_FUNC_ALWAYS: + return BRW_COMPAREFUNCTION_ALWAYS; + } + + debug_printf("Unknown value in %s: %x\n", __FUNCTION__, func); + return BRW_COMPAREFUNCTION_ALWAYS; +} + +static int brw_translate_stencil_op(int op) +{ + switch(op) { + case PIPE_STENCIL_OP_KEEP: + return BRW_STENCILOP_KEEP; + case PIPE_STENCIL_OP_ZERO: + return BRW_STENCILOP_ZERO; + case PIPE_STENCIL_OP_REPLACE: + return BRW_STENCILOP_REPLACE; + case PIPE_STENCIL_OP_INCR: + return BRW_STENCILOP_INCRSAT; + case PIPE_STENCIL_OP_DECR: + return BRW_STENCILOP_DECRSAT; + case PIPE_STENCIL_OP_INCR_WRAP: + return BRW_STENCILOP_INCR; + case PIPE_STENCIL_OP_DECR_WRAP: + return BRW_STENCILOP_DECR; + case PIPE_STENCIL_OP_INVERT: + return BRW_STENCILOP_INVERT; + default: + return BRW_STENCILOP_ZERO; + } +} + + +static int brw_translate_logic_op(int opcode) +{ + switch(opcode) { + case PIPE_LOGICOP_CLEAR: + return BRW_LOGICOPFUNCTION_CLEAR; + case PIPE_LOGICOP_AND: + return BRW_LOGICOPFUNCTION_AND; + case PIPE_LOGICOP_AND_REVERSE: + return BRW_LOGICOPFUNCTION_AND_REVERSE; + case PIPE_LOGICOP_COPY: + return BRW_LOGICOPFUNCTION_COPY; + case PIPE_LOGICOP_COPY_INVERTED: + return BRW_LOGICOPFUNCTION_COPY_INVERTED; + case PIPE_LOGICOP_AND_INVERTED: + return BRW_LOGICOPFUNCTION_AND_INVERTED; + case PIPE_LOGICOP_NOOP: + return BRW_LOGICOPFUNCTION_NOOP; + case PIPE_LOGICOP_XOR: + return BRW_LOGICOPFUNCTION_XOR; + case PIPE_LOGICOP_OR: + return BRW_LOGICOPFUNCTION_OR; + case PIPE_LOGICOP_OR_INVERTED: + return BRW_LOGICOPFUNCTION_OR_INVERTED; + case PIPE_LOGICOP_NOR: + return BRW_LOGICOPFUNCTION_NOR; + case PIPE_LOGICOP_EQUIV: + return BRW_LOGICOPFUNCTION_EQUIV; + case PIPE_LOGICOP_INVERT: + return BRW_LOGICOPFUNCTION_INVERT; + case PIPE_LOGICOP_OR_REVERSE: + return BRW_LOGICOPFUNCTION_OR_REVERSE; + case PIPE_LOGICOP_NAND: + return BRW_LOGICOPFUNCTION_NAND; + case PIPE_LOGICOP_SET: + return BRW_LOGICOPFUNCTION_SET; + default: + return BRW_LOGICOPFUNCTION_SET; + } +} + + +static void upload_cc_vp( struct brw_context *brw ) +{ + struct brw_cc_viewport ccv; + + memset(&ccv, 0, sizeof(ccv)); + + ccv.min_depth = 0.0; + ccv.max_depth = 1.0; + + brw->cc.vp_gs_offset = brw_cache_data( &brw->cache[BRW_CC_VP], &ccv ); +} + +const struct brw_tracked_state brw_cc_vp = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_cc_vp +}; + + +static void upload_cc_unit( struct brw_context *brw ) +{ + struct brw_cc_unit_state cc; + + memset(&cc, 0, sizeof(cc)); + + /* BRW_NEW_DEPTH_STENCIL */ + if (brw->attribs.DepthStencil->stencil[0].enabled) { + cc.cc0.stencil_enable = brw->attribs.DepthStencil->stencil[0].enabled; + cc.cc0.stencil_func = brw_translate_compare_func(brw->attribs.DepthStencil->stencil[0].func); + cc.cc0.stencil_fail_op = brw_translate_stencil_op(brw->attribs.DepthStencil->stencil[0].fail_op); + cc.cc0.stencil_pass_depth_fail_op = brw_translate_stencil_op( + brw->attribs.DepthStencil->stencil[0].zfail_op); + cc.cc0.stencil_pass_depth_pass_op = brw_translate_stencil_op( + brw->attribs.DepthStencil->stencil[0].zpass_op); + cc.cc1.stencil_ref = brw->attribs.DepthStencil->stencil[0].ref_value; + cc.cc1.stencil_write_mask = brw->attribs.DepthStencil->stencil[0].write_mask; + cc.cc1.stencil_test_mask = brw->attribs.DepthStencil->stencil[0].value_mask; + + if (brw->attribs.DepthStencil->stencil[1].enabled) { + cc.cc0.bf_stencil_enable = brw->attribs.DepthStencil->stencil[1].enabled; + cc.cc0.bf_stencil_func = brw_translate_compare_func( + brw->attribs.DepthStencil->stencil[1].func); + cc.cc0.bf_stencil_fail_op = brw_translate_stencil_op( + brw->attribs.DepthStencil->stencil[1].fail_op); + cc.cc0.bf_stencil_pass_depth_fail_op = brw_translate_stencil_op( + brw->attribs.DepthStencil->stencil[1].zfail_op); + cc.cc0.bf_stencil_pass_depth_pass_op = brw_translate_stencil_op( + brw->attribs.DepthStencil->stencil[1].zpass_op); + cc.cc1.bf_stencil_ref = brw->attribs.DepthStencil->stencil[1].ref_value; + cc.cc2.bf_stencil_write_mask = brw->attribs.DepthStencil->stencil[1].write_mask; + cc.cc2.bf_stencil_test_mask = brw->attribs.DepthStencil->stencil[1].value_mask; + } + + /* Not really sure about this: + */ + if (brw->attribs.DepthStencil->stencil[0].write_mask || + brw->attribs.DepthStencil->stencil[1].write_mask) + cc.cc0.stencil_write_enable = 1; + } + + /* BRW_NEW_BLEND */ + if (brw->attribs.Blend->logicop_enable) { + cc.cc2.logicop_enable = 1; + cc.cc5.logicop_func = brw_translate_logic_op( brw->attribs.Blend->logicop_func ); + } + else if (brw->attribs.Blend->blend_enable) { + int eqRGB = brw->attribs.Blend->rgb_func; + int eqA = brw->attribs.Blend->alpha_func; + int srcRGB = brw->attribs.Blend->rgb_src_factor; + int dstRGB = brw->attribs.Blend->rgb_dst_factor; + int srcA = brw->attribs.Blend->alpha_src_factor; + int dstA = brw->attribs.Blend->alpha_dst_factor; + + if (eqRGB == PIPE_BLEND_MIN || eqRGB == PIPE_BLEND_MAX) { + srcRGB = dstRGB = PIPE_BLENDFACTOR_ONE; + } + + if (eqA == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MAX) { + srcA = dstA = PIPE_BLENDFACTOR_ONE; + } + + cc.cc6.dest_blend_factor = brw_translate_blend_factor(dstRGB); + cc.cc6.src_blend_factor = brw_translate_blend_factor(srcRGB); + cc.cc6.blend_function = brw_translate_blend_equation( eqRGB ); + + cc.cc5.ia_dest_blend_factor = brw_translate_blend_factor(dstA); + cc.cc5.ia_src_blend_factor = brw_translate_blend_factor(srcA); + cc.cc5.ia_blend_function = brw_translate_blend_equation( eqA ); + + cc.cc3.blend_enable = 1; + cc.cc3.ia_blend_enable = (srcA != srcRGB || + dstA != dstRGB || + eqA != eqRGB); + } + + /* BRW_NEW_ALPHATEST + */ + if (brw->attribs.DepthStencil->alpha.enabled) { + cc.cc3.alpha_test = 1; + cc.cc3.alpha_test_func = + brw_translate_compare_func(brw->attribs.DepthStencil->alpha.func); + + UNCLAMPED_FLOAT_TO_UBYTE(cc.cc7.alpha_ref.ub[0], + brw->attribs.DepthStencil->alpha.ref); + + cc.cc3.alpha_test_format = BRW_ALPHATEST_FORMAT_UNORM8; + } + + if (brw->attribs.Blend->dither) { + cc.cc5.dither_enable = 1; + cc.cc6.y_dither_offset = 0; + cc.cc6.x_dither_offset = 0; + } + + if (brw->attribs.DepthStencil->depth.enabled) { + cc.cc2.depth_test = brw->attribs.DepthStencil->depth.enabled; + cc.cc2.depth_test_function = brw_translate_compare_func(brw->attribs.DepthStencil->depth.func); + cc.cc2.depth_write_enable = brw->attribs.DepthStencil->depth.writemask; + } + + /* CACHE_NEW_CC_VP */ + cc.cc4.cc_viewport_state_offset = brw->cc.vp_gs_offset >> 5; + + if (BRW_DEBUG & DEBUG_STATS) + cc.cc5.statistics_enable = 1; + + brw->cc.state_gs_offset = brw_cache_data( &brw->cache[BRW_CC_UNIT], &cc ); +} + +const struct brw_tracked_state brw_cc_unit = { + .dirty = { + .brw = BRW_NEW_DEPTH_STENCIL | BRW_NEW_BLEND | BRW_NEW_ALPHA_TEST, + .cache = CACHE_NEW_CC_VP + }, + .update = upload_cc_unit +}; + diff --git a/src/gallium/drivers/i965simple/brw_clip.c b/src/gallium/drivers/i965simple/brw_clip.c new file mode 100644 index 0000000000..268124cc53 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip.c @@ -0,0 +1,206 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_state.h" +#include "brw_clip.h" + +#define FRONT_UNFILLED_BIT 0x1 +#define BACK_UNFILLED_BIT 0x2 + + +static void compile_clip_prog( struct brw_context *brw, + struct brw_clip_prog_key *key ) +{ + struct brw_clip_compile c; + const unsigned *program; + unsigned program_size; + unsigned delta; + unsigned i; + + memset(&c, 0, sizeof(c)); + + /* Begin the compilation: + */ + brw_init_compile(&c.func); + + c.func.single_program_flow = 1; + + c.key = *key; + + + /* Need to locate the two positions present in vertex + header. + * These are currently hardcoded: + */ + c.header_position_offset = ATTR_SIZE; + + for (i = 0, delta = REG_SIZE; i < PIPE_MAX_SHADER_OUTPUTS; i++) + if (c.key.attrs & (1<<i)) { + c.offset[i] = delta; + delta += ATTR_SIZE; + } + + c.nr_attrs = brw_count_bits(c.key.attrs); + c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ + c.nr_bytes = c.nr_regs * REG_SIZE; + + c.prog_data.clip_mode = c.key.clip_mode; /* XXX */ + + /* For some reason the thread is spawned with only 4 channels + * unmasked. + */ + brw_set_mask_control(&c.func, BRW_MASK_DISABLE); + + + /* Would ideally have the option of producing a program which could + * do all three: + */ + switch (key->primitive) { + case PIPE_PRIM_TRIANGLES: +#if 0 + if (key->do_unfilled) + brw_emit_unfilled_clip( &c ); + else +#endif + brw_emit_tri_clip( &c ); + break; + case PIPE_PRIM_LINES: + brw_emit_line_clip( &c ); + break; + case PIPE_PRIM_POINTS: + brw_emit_point_clip( &c ); + break; + default: + assert(0); + return; + } + + + + /* get the program + */ + program = brw_get_program(&c.func, &program_size); + + /* Upload + */ + brw->clip.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_CLIP_PROG], + &c.key, + sizeof(c.key), + program, + program_size, + &c.prog_data, + &brw->clip.prog_data ); +} + + +static boolean search_cache( struct brw_context *brw, + struct brw_clip_prog_key *key ) +{ + return brw_search_cache(&brw->cache[BRW_CLIP_PROG], + key, sizeof(*key), + &brw->clip.prog_data, + &brw->clip.prog_gs_offset); +} + + + + +/* Calculate interpolants for triangle and line rasterization. + */ +static void upload_clip_prog(struct brw_context *brw) +{ + struct brw_clip_prog_key key; + + memset(&key, 0, sizeof(key)); + + /* Populate the key: + */ + /* BRW_NEW_REDUCED_PRIMITIVE */ + key.primitive = brw->reduced_primitive; + /* CACHE_NEW_VS_PROG */ + key.attrs = brw->vs.prog_data->outputs_written; + /* BRW_NEW_RASTER */ + key.do_flat_shading = (brw->attribs.Raster->flatshade); + /* BRW_NEW_CLIP */ + key.nr_userclip = brw->attribs.Clip.nr; /* XXX */ + +#if 0 + key.clip_mode = BRW_CLIPMODE_NORMAL; + + if (key.primitive == PIPE_PRIM_TRIANGLES) { + if (brw->attribs.Raster->cull_mode == PIPE_WINDING_BOTH) + key.clip_mode = BRW_CLIPMODE_REJECT_ALL; + else { + if (brw->attribs.Raster->fill_cw != PIPE_POLYGON_MODE_FILL || + brw->attribs.Raster->fill_ccw != PIPE_POLYGON_MODE_FILL) + key.do_unfilled = 1; + + /* Most cases the fixed function units will handle. Cases where + * one or more polygon faces are unfilled will require help: + */ + if (key.do_unfilled) { + key.clip_mode = BRW_CLIPMODE_CLIP_NON_REJECTED; + + if (brw->attribs.Raster->offset_cw || + brw->attribs.Raster->offset_ccw) { + key.offset_units = brw->attribs.Raster->offset_units; + key.offset_factor = brw->attribs.Raster->offset_scale; + } + key.fill_ccw = brw->attribs.Raster->fill_ccw; + key.fill_cw = brw->attribs.Raster->fill_cw; + key.offset_ccw = brw->attribs.Raster->offset_ccw; + key.offset_cw = brw->attribs.Raster->offset_cw; + if (brw->attribs.Raster->light_twoside && + key.fill_cw != CLIP_CULL) + key.copy_bfc_cw = 1; + } + } + } +#else + key.clip_mode = BRW_CLIPMODE_ACCEPT_ALL; +#endif + + if (!search_cache(brw, &key)) + compile_clip_prog( brw, &key ); +} + +const struct brw_tracked_state brw_clip_prog = { + .dirty = { + .brw = (BRW_NEW_RASTERIZER | + BRW_NEW_CLIP | + BRW_NEW_REDUCED_PRIMITIVE), + .cache = CACHE_NEW_VS_PROG + }, + .update = upload_clip_prog +}; diff --git a/src/gallium/drivers/i965simple/brw_clip.h b/src/gallium/drivers/i965simple/brw_clip.h new file mode 100644 index 0000000000..a89d08b791 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip.h @@ -0,0 +1,170 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef BRW_CLIP_H +#define BRW_CLIP_H + + +#include "brw_context.h" +#include "brw_eu.h" + +#define MAX_VERTS (3+6+6) + +/* Note that if unfilled primitives are being emitted, we have to fix + * up polygon offset and flatshading at this point: + */ +struct brw_clip_prog_key { + unsigned attrs:32; + unsigned primitive:4; + unsigned nr_userclip:3; + unsigned do_flat_shading:1; + unsigned do_unfilled:1; + unsigned fill_cw:2; /* includes cull information */ + unsigned fill_ccw:2; /* includes cull information */ + unsigned offset_cw:1; + unsigned offset_ccw:1; + unsigned pad0:17; + + unsigned copy_bfc_cw:1; + unsigned copy_bfc_ccw:1; + unsigned clip_mode:3; + unsigned pad1:27; + + float offset_factor; + float offset_units; +}; + + +#define CLIP_LINE 0 +#define CLIP_POINT 1 +#define CLIP_FILL 2 +#define CLIP_CULL 3 + + +#define PRIM_MASK (0x1f) + +struct brw_clip_compile { + struct brw_compile func; + struct brw_clip_prog_key key; + struct brw_clip_prog_data prog_data; + + struct { + struct brw_reg R0; + struct brw_reg vertex[MAX_VERTS]; + + struct brw_reg t; + struct brw_reg t0, t1; + struct brw_reg dp0, dp1; + + struct brw_reg dpPrev; + struct brw_reg dp; + struct brw_reg loopcount; + struct brw_reg nr_verts; + struct brw_reg planemask; + + struct brw_reg inlist; + struct brw_reg outlist; + struct brw_reg freelist; + + struct brw_reg dir; + struct brw_reg tmp0, tmp1; + struct brw_reg offset; + + struct brw_reg fixed_planes; + struct brw_reg plane_equation; + } reg; + + /* 3 different ways of expressing vertex size: + */ + unsigned nr_attrs; + unsigned nr_regs; + unsigned nr_bytes; + + unsigned first_tmp; + unsigned last_tmp; + + boolean need_direction; + + unsigned last_mrf; + + unsigned header_position_offset; + unsigned offset[PIPE_ATTRIB_MAX]; +}; + +#define ATTR_SIZE (4*4) + +/* Points are only culled, so no need for a clip routine, however it + * works out easier to have a dummy one. + */ +void brw_emit_unfilled_clip( struct brw_clip_compile *c ); +void brw_emit_tri_clip( struct brw_clip_compile *c ); +void brw_emit_line_clip( struct brw_clip_compile *c ); +void brw_emit_point_clip( struct brw_clip_compile *c ); + +/* brw_clip_tri.c, for use by the unfilled clip routine: + */ +void brw_clip_tri_init_vertices( struct brw_clip_compile *c ); +void brw_clip_tri_flat_shade( struct brw_clip_compile *c ); +void brw_clip_tri( struct brw_clip_compile *c ); +void brw_clip_tri_emit_polygon( struct brw_clip_compile *c ); +void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, + unsigned nr_verts ); + + +/* Utils: + */ + +void brw_clip_interp_vertex( struct brw_clip_compile *c, + struct brw_indirect dest_ptr, + struct brw_indirect v0_ptr, /* from */ + struct brw_indirect v1_ptr, /* to */ + struct brw_reg t0, + boolean force_edgeflag ); + +void brw_clip_init_planes( struct brw_clip_compile *c ); + +void brw_clip_emit_vue(struct brw_clip_compile *c, + struct brw_indirect vert, + boolean allocate, + boolean eot, + unsigned header); + +void brw_clip_kill_thread(struct brw_clip_compile *c); + +struct brw_reg brw_clip_plane_stride( struct brw_clip_compile *c ); +struct brw_reg brw_clip_plane0_address( struct brw_clip_compile *c ); + +void brw_clip_copy_colors( struct brw_clip_compile *c, + unsigned to, unsigned from ); + +void brw_clip_init_clipmask( struct brw_clip_compile *c ); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_clip_line.c b/src/gallium/drivers/i965simple/brw_clip_line.c new file mode 100644 index 0000000000..75d9e5fcda --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_line.c @@ -0,0 +1,245 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_clip.h" + + + +static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) +{ + unsigned i = 0,j; + + /* Register usage is static, precompute here: + */ + c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; + + if (c->key.nr_userclip) { + c->reg.fixed_planes = brw_vec4_grf(i, 0); + i += (6 + c->key.nr_userclip + 1) / 2; + + c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2; + } + else + c->prog_data.curb_read_length = 0; + + + /* Payload vertices plus space for more generated vertices: + */ + for (j = 0; j < 4; j++) { + c->reg.vertex[j] = brw_vec4_grf(i, 0); + i += c->nr_regs; + } + + c->reg.t = brw_vec1_grf(i, 0); + c->reg.t0 = brw_vec1_grf(i, 1); + c->reg.t1 = brw_vec1_grf(i, 2); + c->reg.planemask = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD); + c->reg.plane_equation = brw_vec4_grf(i, 4); + i++; + + c->reg.dp0 = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */ + c->reg.dp1 = brw_vec1_grf(i, 4); + i++; + + if (!c->key.nr_userclip) { + c->reg.fixed_planes = brw_vec8_grf(i, 0); + i++; + } + + + c->first_tmp = i; + c->last_tmp = i; + + c->prog_data.urb_read_length = c->nr_regs; /* ? */ + c->prog_data.total_grf = i; +} + + + +/* Line clipping, more or less following the following algorithm: + * + * for (p=0;p<MAX_PLANES;p++) { + * if (clipmask & (1 << p)) { + * float dp0 = DOTPROD( vtx0, plane[p] ); + * float dp1 = DOTPROD( vtx1, plane[p] ); + * + * if (IS_NEGATIVE(dp1)) { + * float t = dp1 / (dp1 - dp0); + * if (t > t1) t1 = t; + * } else { + * float t = dp0 / (dp0 - dp1); + * if (t > t0) t0 = t; + * } + * + * if (t0 + t1 >= 1.0) + * return; + * } + * } + * + * interp( ctx, newvtx0, vtx0, vtx1, t0 ); + * interp( ctx, newvtx1, vtx1, vtx0, t1 ); + * + */ +static void clip_and_emit_line( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_indirect vtx0 = brw_indirect(0, 0); + struct brw_indirect vtx1 = brw_indirect(1, 0); + struct brw_indirect newvtx0 = brw_indirect(2, 0); + struct brw_indirect newvtx1 = brw_indirect(3, 0); + struct brw_indirect plane_ptr = brw_indirect(4, 0); + struct brw_instruction *plane_loop; + struct brw_instruction *plane_active; + struct brw_instruction *is_negative; + struct brw_instruction *is_neg2; + struct brw_instruction *not_culled; + struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD); + + brw_MOV(p, get_addr_reg(vtx0), brw_address(c->reg.vertex[0])); + brw_MOV(p, get_addr_reg(vtx1), brw_address(c->reg.vertex[1])); + brw_MOV(p, get_addr_reg(newvtx0), brw_address(c->reg.vertex[2])); + brw_MOV(p, get_addr_reg(newvtx1), brw_address(c->reg.vertex[3])); + brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c)); + + /* Note: init t0, t1 together: + */ + brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0)); + + brw_clip_init_planes(c); + brw_clip_init_clipmask(c); + + /* -ve rhw workaround */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), + brw_imm_ud(1<<20)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + plane_loop = brw_DO(p, BRW_EXECUTE_1); + { + /* if (planemask & 1) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1)); + + plane_active = brw_IF(p, BRW_EXECUTE_1); + { + if (c->key.nr_userclip) + brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0)); + else + brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0)); + +#if 0 + /* dp = DP4(vtx->position, plane) + */ + brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation); + + /* if (IS_NEGATIVE(dp1)) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_L); + brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation); +#else + #warning "disabled" +#endif + is_negative = brw_IF(p, BRW_EXECUTE_1); + { + brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0)); + brw_math_invert(p, c->reg.t, c->reg.t); + brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1); + + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 ); + brw_MOV(p, c->reg.t1, c->reg.t); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } + is_negative = brw_ELSE(p, is_negative); + { + /* Coming back in. We know that both cannot be negative + * because the line would have been culled in that case. + */ + + /* If both are positive, do nothing */ + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0)); + is_neg2 = brw_IF(p, BRW_EXECUTE_1); + { + brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1)); + brw_math_invert(p, c->reg.t, c->reg.t); + brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0); + + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 ); + brw_MOV(p, c->reg.t0, c->reg.t); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } + brw_ENDIF(p, is_neg2); + } + brw_ENDIF(p, is_negative); + } + brw_ENDIF(p, plane_active); + + /* plane_ptr++; + */ + brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c)); + + /* while (planemask>>=1) != 0 + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1)); + } + brw_WHILE(p, plane_loop); + + brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1); + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0)); + not_culled = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, FALSE); + brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, FALSE); + + brw_clip_emit_vue(c, newvtx0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START); + brw_clip_emit_vue(c, newvtx1, 0, 1, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END); + } + brw_ENDIF(p, not_culled); + brw_clip_kill_thread(c); +} + + + +void brw_emit_line_clip( struct brw_clip_compile *c ) +{ + brw_clip_line_alloc_regs(c); + + if (c->key.do_flat_shading) + brw_clip_copy_colors(c, 0, 1); + + clip_and_emit_line(c); +} diff --git a/src/gallium/drivers/i965simple/brw_clip_point.c b/src/gallium/drivers/i965simple/brw_clip_point.c new file mode 100644 index 0000000000..6fce7210d1 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_point.c @@ -0,0 +1,47 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_clip.h" + + +/* Point clipping, nothing to do? + */ +void brw_emit_point_clip( struct brw_clip_compile *c ) +{ + /* Send an empty message to kill the thread: + */ + brw_clip_tri_alloc_regs(c, 0); + brw_clip_kill_thread(c); +} diff --git a/src/gallium/drivers/i965simple/brw_clip_state.c b/src/gallium/drivers/i965simple/brw_clip_state.c new file mode 100644 index 0000000000..ea5c05a279 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_state.c @@ -0,0 +1,92 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "pipe/p_util.h" + + +static void upload_clip_unit( struct brw_context *brw ) +{ + struct brw_clip_unit_state clip; + + memset(&clip, 0, sizeof(clip)); + + /* CACHE_NEW_CLIP_PROG */ + clip.thread0.grf_reg_count = + align(brw->clip.prog_data->total_grf, 16) / 16 - 1; + clip.thread0.kernel_start_pointer = brw->clip.prog_gs_offset >> 6; + clip.thread3.urb_entry_read_length = brw->clip.prog_data->urb_read_length; + clip.thread3.const_urb_entry_read_length = brw->clip.prog_data->curb_read_length; + clip.clip5.clip_mode = brw->clip.prog_data->clip_mode; + + /* BRW_NEW_CURBE_OFFSETS */ + clip.thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2; + + /* BRW_NEW_URB_FENCE */ + clip.thread4.nr_urb_entries = brw->urb.nr_clip_entries; + clip.thread4.urb_entry_allocation_size = brw->urb.vsize - 1; + clip.thread4.max_threads = 1; /* 2 threads */ + + if (BRW_DEBUG & DEBUG_STATS) + clip.thread4.stats_enable = 1; + + /* CONSTANT */ + clip.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; + clip.thread1.single_program_flow = 1; + clip.thread3.dispatch_grf_start_reg = 1; + clip.thread3.urb_entry_read_offset = 0; + clip.clip5.userclip_enable_flags = 0x7f; + clip.clip5.userclip_must_clip = 1; + clip.clip5.guard_band_enable = 0; + clip.clip5.viewport_z_clip_enable = 1; + clip.clip5.viewport_xy_clip_enable = 1; + clip.clip5.vertex_position_space = BRW_CLIP_NDCSPACE; + clip.clip5.api_mode = BRW_CLIP_API_OGL; + clip.clip6.clipper_viewport_state_ptr = 0; + clip.viewport_xmin = -1; + clip.viewport_xmax = 1; + clip.viewport_ymin = -1; + clip.viewport_ymax = 1; + + brw->clip.state_gs_offset = brw_cache_data( &brw->cache[BRW_CLIP_UNIT], &clip ); +} + + +const struct brw_tracked_state brw_clip_unit = { + .dirty = { + .brw = (BRW_NEW_CURBE_OFFSETS | + BRW_NEW_URB_FENCE), + .cache = CACHE_NEW_CLIP_PROG + }, + .update = upload_clip_unit +}; diff --git a/src/gallium/drivers/i965simple/brw_clip_tri.c b/src/gallium/drivers/i965simple/brw_clip_tri.c new file mode 100644 index 0000000000..c5da7b825e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_tri.c @@ -0,0 +1,566 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_clip.h" + +static struct brw_reg get_tmp( struct brw_clip_compile *c ) +{ + struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0); + + if (++c->last_tmp > c->prog_data.total_grf) + c->prog_data.total_grf = c->last_tmp; + + return tmp; +} + +static void release_tmps( struct brw_clip_compile *c ) +{ + c->last_tmp = c->first_tmp; +} + + +void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, + unsigned nr_verts ) +{ + unsigned i = 0,j; + + /* Register usage is static, precompute here: + */ + c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; + + if (c->key.nr_userclip) { + c->reg.fixed_planes = brw_vec4_grf(i, 0); + i += (6 + c->key.nr_userclip + 1) / 2; + + c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2; + } + else + c->prog_data.curb_read_length = 0; + + + /* Payload vertices plus space for more generated vertices: + */ + for (j = 0; j < nr_verts; j++) { + c->reg.vertex[j] = brw_vec4_grf(i, 0); + i += c->nr_regs; + } + + if (c->nr_attrs & 1) { + for (j = 0; j < 3; j++) { + unsigned delta = c->nr_attrs*16 + 32; + brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0)); + } + } + + c->reg.t = brw_vec1_grf(i, 0); + c->reg.loopcount = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_UD); + c->reg.nr_verts = retype(brw_vec1_grf(i, 2), BRW_REGISTER_TYPE_UD); + c->reg.planemask = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD); + c->reg.plane_equation = brw_vec4_grf(i, 4); + i++; + + c->reg.dpPrev = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */ + c->reg.dp = brw_vec1_grf(i, 4); + i++; + + c->reg.inlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0); + i++; + + c->reg.outlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0); + i++; + + c->reg.freelist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0); + i++; + + if (!c->key.nr_userclip) { + c->reg.fixed_planes = brw_vec8_grf(i, 0); + i++; + } + + if (c->key.do_unfilled) { + c->reg.dir = brw_vec4_grf(i, 0); + c->reg.offset = brw_vec4_grf(i, 4); + i++; + c->reg.tmp0 = brw_vec4_grf(i, 0); + c->reg.tmp1 = brw_vec4_grf(i, 4); + i++; + } + + c->first_tmp = i; + c->last_tmp = i; + + c->prog_data.urb_read_length = c->nr_regs; /* ? */ + c->prog_data.total_grf = i; +} + + + +void brw_clip_tri_init_vertices( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */ + struct brw_instruction *is_rev; + + /* Initial list of indices for incoming vertexes: + */ + brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_EQ, + tmp0, + brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE)); + + /* XXX: Is there an easier way to do this? Need to reverse every + * second tristrip element: Can ignore sometimes? + */ + is_rev = brw_IF(p, BRW_EXECUTE_1); + { + brw_MOV(p, get_element(c->reg.inlist, 0), brw_address(c->reg.vertex[1]) ); + brw_MOV(p, get_element(c->reg.inlist, 1), brw_address(c->reg.vertex[0]) ); + if (c->need_direction) + brw_MOV(p, c->reg.dir, brw_imm_f(-1)); + } + is_rev = brw_ELSE(p, is_rev); + { + brw_MOV(p, get_element(c->reg.inlist, 0), brw_address(c->reg.vertex[0]) ); + brw_MOV(p, get_element(c->reg.inlist, 1), brw_address(c->reg.vertex[1]) ); + if (c->need_direction) + brw_MOV(p, c->reg.dir, brw_imm_f(1)); + } + brw_ENDIF(p, is_rev); + + brw_MOV(p, get_element(c->reg.inlist, 2), brw_address(c->reg.vertex[2]) ); + brw_MOV(p, brw_vec8_grf(c->reg.outlist.nr, 0), brw_imm_f(0)); + brw_MOV(p, c->reg.nr_verts, brw_imm_ud(3)); +} + + + +void brw_clip_tri_flat_shade( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *is_poly; + struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */ + + brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_EQ, + tmp0, + brw_imm_ud(_3DPRIM_POLYGON)); + + is_poly = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_copy_colors(c, 1, 0); + brw_clip_copy_colors(c, 2, 0); + } + is_poly = brw_ELSE(p, is_poly); + { + brw_clip_copy_colors(c, 0, 2); + brw_clip_copy_colors(c, 1, 2); + } + brw_ENDIF(p, is_poly); +} + + + +/* Use mesa's clipping algorithms, translated to GEN4 assembly. + */ +void brw_clip_tri( struct brw_clip_compile *c ) +{ +#if 0 + struct brw_compile *p = &c->func; + struct brw_indirect vtx = brw_indirect(0, 0); + struct brw_indirect vtxPrev = brw_indirect(1, 0); + struct brw_indirect vtxOut = brw_indirect(2, 0); + struct brw_indirect plane_ptr = brw_indirect(3, 0); + struct brw_indirect inlist_ptr = brw_indirect(4, 0); + struct brw_indirect outlist_ptr = brw_indirect(5, 0); + struct brw_indirect freelist_ptr = brw_indirect(6, 0); + struct brw_instruction *plane_loop; + struct brw_instruction *plane_active; + struct brw_instruction *vertex_loop; + struct brw_instruction *next_test; + struct brw_instruction *prev_test; + + brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) ); + brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c)); + brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); + brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); + + brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) ); + + plane_loop = brw_DO(p, BRW_EXECUTE_1); + { + /* if (planemask & 1) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1)); + + plane_active = brw_IF(p, BRW_EXECUTE_1); + { + /* vtxOut = freelist_ptr++ + */ + brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) ); + brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE)); + + if (c->key.nr_userclip) + brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0)); + else + brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0)); + + brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); + brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0)); + + vertex_loop = brw_DO(p, BRW_EXECUTE_1); + { + /* vtx = *input_ptr; + */ + brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0)); + + /* IS_NEGATIVE(prev) */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_L); + brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation); + prev_test = brw_IF(p, BRW_EXECUTE_1); + { + /* IS_POSITIVE(next) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_GE); + brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation); + next_test = brw_IF(p, BRW_EXECUTE_1); + { + + /* Coming back in. + */ + brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp)); + brw_math_invert(p, c->reg.t, c->reg.t); + brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev); + + /* If (vtxOut == 0) vtxOut = vtxPrev + */ + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); + brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) ); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, FALSE); + + /* *outlist_ptr++ = vtxOut; + * nr_verts++; + * vtxOut = 0; + */ + brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); + brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); + brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); + brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); + } + brw_ENDIF(p, next_test); + + } + prev_test = brw_ELSE(p, prev_test); + { + /* *outlist_ptr++ = vtxPrev; + * nr_verts++; + */ + brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev)); + brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); + brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); + + /* IS_NEGATIVE(next) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_L); + brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation); + next_test = brw_IF(p, BRW_EXECUTE_1); + { + /* Going out of bounds. Avoid division by zero as we + * know dp != dpPrev from DIFFERENT_SIGNS, above. + */ + brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev)); + brw_math_invert(p, c->reg.t, c->reg.t); + brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp); + + /* If (vtxOut == 0) vtxOut = vtx + */ + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) ); + brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) ); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, TRUE); + + /* *outlist_ptr++ = vtxOut; + * nr_verts++; + * vtxOut = 0; + */ + brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut)); + brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short))); + brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1)); + brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) ); + } + brw_ENDIF(p, next_test); + } + brw_ENDIF(p, prev_test); + + /* vtxPrev = vtx; + * inlist_ptr++; + */ + brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx)); + brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short))); + + /* while (--loopcount != 0) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); + } + brw_WHILE(p, vertex_loop); + + /* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1] + * inlist = outlist + * inlist_ptr = &inlist[0] + * outlist_ptr = &outlist[0] + */ + brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2)); + brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0)); + brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0)); + brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist)); + brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist)); + } + brw_ENDIF(p, plane_active); + + /* plane_ptr++; + */ + brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c)); + + /* nr_verts >= 3 + */ + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_GE, + c->reg.nr_verts, + brw_imm_ud(3)); + + /* && (planemask>>=1) != 0 + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1)); + } + brw_WHILE(p, plane_loop); +#else + #warning "disabled" +#endif +} + + + +void brw_clip_tri_emit_polygon(struct brw_clip_compile *c) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *loop, *if_insn; + + /* for (loopcount = nr_verts-2; loopcount > 0; loopcount--) + */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_G); + brw_ADD(p, + c->reg.loopcount, + c->reg.nr_verts, + brw_imm_d(-2)); + + if_insn = brw_IF(p, BRW_EXECUTE_1); + { + struct brw_indirect v0 = brw_indirect(0, 0); + struct brw_indirect vptr = brw_indirect(1, 0); + + brw_MOV(p, get_addr_reg(vptr), brw_address(c->reg.inlist)); + brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0)); + + brw_clip_emit_vue(c, v0, 1, 0, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_START)); + + brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2)); + brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0)); + + loop = brw_DO(p, BRW_EXECUTE_1); + { + brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_TRIFAN << 2)); + + brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2)); + brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0)); + + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); + } + brw_WHILE(p, loop); + + brw_clip_emit_vue(c, v0, 0, 1, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_END)); + } + brw_ENDIF(p, if_insn); +} + +static void do_clip_tri( struct brw_clip_compile *c ) +{ + brw_clip_init_planes(c); + + brw_clip_tri(c); +} + + +static void maybe_do_clip_tri( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *do_clip; + + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0)); + do_clip = brw_IF(p, BRW_EXECUTE_1); + { + do_clip_tri(c); + } + brw_ENDIF(p, do_clip); +} + +static void brw_clip_test( struct brw_clip_compile *c ) +{ +#if 0 + struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); + struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); + struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); + struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); + + struct brw_reg v0 = get_tmp(c); + struct brw_reg v1 = get_tmp(c); + struct brw_reg v2 = get_tmp(c); + + struct brw_indirect vt0 = brw_indirect(0, 0); + struct brw_indirect vt1 = brw_indirect(1, 0); + struct brw_indirect vt2 = brw_indirect(2, 0); + + struct brw_compile *p = &c->func; + + brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0])); + brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1])); + brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2])); + brw_MOV(p, v0, deref_4f(vt0, c->offset[VERT_RESULT_HPOS])); + brw_MOV(p, v1, deref_4f(vt1, c->offset[VERT_RESULT_HPOS])); + brw_MOV(p, v2, deref_4f(vt2, c->offset[VERT_RESULT_HPOS])); + + /* test nearz, xmin, ymin plane */ + brw_CMP(p, t1, BRW_CONDITIONAL_LE, negate(v0), get_element(v0, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, t2, BRW_CONDITIONAL_LE, negate(v1), get_element(v1, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, t3, BRW_CONDITIONAL_LE, negate(v2), get_element(v2, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_XOR(p, t, t1, t2); + brw_XOR(p, t1, t2, t3); + brw_OR(p, t, t, t1); + + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 0), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 1), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 2), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + /* test farz, xmax, ymax plane */ + brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, get_element(v0, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, get_element(v1, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, get_element(v2, 3)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + brw_XOR(p, t, t1, t2); + brw_XOR(p, t1, t2, t3); + brw_OR(p, t, t, t1); + + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 0), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 1), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, + get_element(t, 2), brw_imm_ud(0)); + brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + release_tmps(c); +#else + #warning "disabled" +#endif +} + + +void brw_emit_tri_clip( struct brw_clip_compile *c ) +{ + struct brw_instruction *neg_rhw; + struct brw_compile *p = &c->func; + brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); + brw_clip_tri_init_vertices(c); + brw_clip_init_clipmask(c); + + /* if -ve rhw workaround bit is set, + do cliptest */ + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), + brw_imm_ud(1<<20)); + neg_rhw = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_test(c); + } + brw_ENDIF(p, neg_rhw); + + /* Can't push into do_clip_tri because with polygon (or quad) + * flatshading, need to apply the flatshade here because we don't + * respect the PV when converting to trifan for emit: + */ + if (c->key.do_flat_shading) + brw_clip_tri_flat_shade(c); + + if (c->key.clip_mode == BRW_CLIPMODE_NORMAL) + do_clip_tri(c); + else + maybe_do_clip_tri(c); + + brw_clip_tri_emit_polygon(c); + + /* Send an empty message to kill the thread: + */ + brw_clip_kill_thread(c); +} diff --git a/src/gallium/drivers/i965simple/brw_clip_unfilled.c b/src/gallium/drivers/i965simple/brw_clip_unfilled.c new file mode 100644 index 0000000000..b774a76dd6 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_unfilled.c @@ -0,0 +1,477 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_clip.h" + + + +/* This is performed against the original triangles, so no indirection + * required: +BZZZT! + */ +static void compute_tri_direction( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_reg e = c->reg.tmp0; + struct brw_reg f = c->reg.tmp1; + struct brw_reg v0 = byte_offset(c->reg.vertex[0], c->offset[VERT_RESULT_HPOS]); + struct brw_reg v1 = byte_offset(c->reg.vertex[1], c->offset[VERT_RESULT_HPOS]); + struct brw_reg v2 = byte_offset(c->reg.vertex[2], c->offset[VERT_RESULT_HPOS]); + + + /* Calculate the vectors of two edges of the triangle: + */ + brw_ADD(p, e, v0, negate(v2)); + brw_ADD(p, f, v1, negate(v2)); + + /* Take their crossproduct: + */ + brw_set_access_mode(p, BRW_ALIGN_16); + brw_MUL(p, vec4(brw_null_reg()), brw_swizzle(e, 1,2,0,3), brw_swizzle(f,2,0,1,3)); + brw_MAC(p, vec4(e), negate(brw_swizzle(e, 2,0,1,3)), brw_swizzle(f,1,2,0,3)); + brw_set_access_mode(p, BRW_ALIGN_1); + + brw_MUL(p, c->reg.dir, c->reg.dir, vec4(e)); +} + + +static void cull_direction( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *ccw; + unsigned conditional; + + assert (!(c->key.fill_ccw == CLIP_CULL && + c->key.fill_cw == CLIP_CULL)); + + if (c->key.fill_ccw == CLIP_CULL) + conditional = BRW_CONDITIONAL_GE; + else + conditional = BRW_CONDITIONAL_L; + + brw_CMP(p, + vec1(brw_null_reg()), + conditional, + get_element(c->reg.dir, 2), + brw_imm_f(0)); + + ccw = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_kill_thread(c); + } + brw_ENDIF(p, ccw); +} + + + +static void copy_bfc( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *ccw; + unsigned conditional; + + /* Do we have any colors to copy? + */ + if (!(c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0]) && + !(c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1])) + return; + + /* In some wierd degnerate cases we can end up testing the + * direction twice, once for culling and once for bfc copying. Oh + * well, that's what you get for setting wierd GL state. + */ + if (c->key.copy_bfc_ccw) + conditional = BRW_CONDITIONAL_GE; + else + conditional = BRW_CONDITIONAL_L; + + brw_CMP(p, + vec1(brw_null_reg()), + conditional, + get_element(c->reg.dir, 2), + brw_imm_f(0)); + + ccw = brw_IF(p, BRW_EXECUTE_1); + { + unsigned i; + + for (i = 0; i < 3; i++) { + if (c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0]) + brw_MOV(p, + byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL0]), + byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC0])); + + if (c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1]) + brw_MOV(p, + byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL1]), + byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC1])); + } + } + brw_ENDIF(p, ccw); +} + + + + +/* + float iz = 1.0 / dir.z; + float ac = dir.x * iz; + float bc = dir.y * iz; + offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE; + offset += MAX2( abs(ac), abs(bc) ) * ctx->Polygon.OffsetFactor; + offset *= MRD; +*/ +static void compute_offset( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_reg off = c->reg.offset; + struct brw_reg dir = c->reg.dir; + + brw_math_invert(p, get_element(off, 2), get_element(dir, 2)); + brw_MUL(p, vec2(off), dir, get_element(off, 2)); + + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_GE, + brw_abs(get_element(off, 0)), + brw_abs(get_element(off, 1))); + + brw_SEL(p, vec1(off), brw_abs(get_element(off, 0)), brw_abs(get_element(off, 1))); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + brw_MUL(p, vec1(off), off, brw_imm_f(c->key.offset_factor)); + brw_ADD(p, vec1(off), off, brw_imm_f(c->key.offset_units)); +} + + +static void merge_edgeflags( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *is_poly; + struct brw_reg tmp0 = get_element_ud(c->reg.tmp0, 0); + + brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK)); + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_EQ, + tmp0, + brw_imm_ud(_3DPRIM_POLYGON)); + + /* Get away with using reg.vertex because we know that this is not + * a _3DPRIM_TRISTRIP_REVERSE: + */ + is_poly = brw_IF(p, BRW_EXECUTE_1); + { + brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ); + brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8)); + brw_MOV(p, byte_offset(c->reg.vertex[0], c->offset[VERT_RESULT_EDGE]), brw_imm_f(0)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ); + brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9)); + brw_MOV(p, byte_offset(c->reg.vertex[2], c->offset[VERT_RESULT_EDGE]), brw_imm_f(0)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } + brw_ENDIF(p, is_poly); +} + + + +static void apply_one_offset( struct brw_clip_compile *c, + struct brw_indirect vert ) +{ + struct brw_compile *p = &c->func; + struct brw_reg pos = deref_4f(vert, c->offset[VERT_RESULT_HPOS]); + struct brw_reg z = get_element(pos, 2); + + brw_ADD(p, z, z, vec1(c->reg.offset)); +} + + + +/*********************************************************************** + * Output clipped polygon as an unfilled primitive: + */ +static void emit_lines(struct brw_clip_compile *c, + boolean do_offset) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *loop; + struct brw_instruction *draw_edge; + struct brw_indirect v0 = brw_indirect(0, 0); + struct brw_indirect v1 = brw_indirect(1, 0); + struct brw_indirect v0ptr = brw_indirect(2, 0); + struct brw_indirect v1ptr = brw_indirect(3, 0); + + /* Need a seperate loop for offset: + */ + if (do_offset) { + brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); + brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist)); + + loop = brw_DO(p, BRW_EXECUTE_1); + { + brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0)); + brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2)); + + apply_one_offset(c, v0); + + brw_set_conditionalmod(p, BRW_CONDITIONAL_G); + brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); + } + brw_WHILE(p, loop); + } + + /* v1ptr = &inlist[nr_verts] + * *v1ptr = v0 + */ + brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); + brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist)); + brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW)); + brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW)); + brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0)); + + loop = brw_DO(p, BRW_EXECUTE_1); + { + brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0)); + brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2)); + brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2)); + + /* draw edge if edgeflag != 0 */ + brw_CMP(p, + vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, + deref_1f(v0, c->offset[VERT_RESULT_EDGE]), + brw_imm_f(0)); + draw_edge = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START); + brw_clip_emit_vue(c, v1, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END); + } + brw_ENDIF(p, draw_edge); + + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); + } + brw_WHILE(p, loop); +} + + + +static void emit_points(struct brw_clip_compile *c, + boolean do_offset ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *loop; + struct brw_instruction *draw_point; + + struct brw_indirect v0 = brw_indirect(0, 0); + struct brw_indirect v0ptr = brw_indirect(2, 0); + + brw_MOV(p, c->reg.loopcount, c->reg.nr_verts); + brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist)); + + loop = brw_DO(p, BRW_EXECUTE_1); + { + brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0)); + brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2)); + + /* draw if edgeflag != 0 + */ + brw_CMP(p, + vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, + deref_1f(v0, c->offset[VERT_RESULT_EDGE]), + brw_imm_f(0)); + draw_point = brw_IF(p, BRW_EXECUTE_1); + { + if (do_offset) + apply_one_offset(c, v0); + + brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END); + } + brw_ENDIF(p, draw_point); + + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1)); + } + brw_WHILE(p, loop); +} + + + + + + + +static void emit_primitives( struct brw_clip_compile *c, + unsigned mode, + boolean do_offset ) +{ + switch (mode) { + case CLIP_FILL: + brw_clip_tri_emit_polygon(c); + break; + + case CLIP_LINE: + emit_lines(c, do_offset); + break; + + case CLIP_POINT: + emit_points(c, do_offset); + break; + + case CLIP_CULL: + assert(0); + break; + } +} + + + +static void emit_unfilled_primitives( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *ccw; + + /* Direction culling has already been done. + */ + if (c->key.fill_ccw != c->key.fill_cw && + c->key.fill_ccw != CLIP_CULL && + c->key.fill_cw != CLIP_CULL) + { + brw_CMP(p, + vec1(brw_null_reg()), + BRW_CONDITIONAL_GE, + get_element(c->reg.dir, 2), + brw_imm_f(0)); + + ccw = brw_IF(p, BRW_EXECUTE_1); + { + emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw); + } + ccw = brw_ELSE(p, ccw); + { + emit_primitives(c, c->key.fill_cw, c->key.offset_cw); + } + brw_ENDIF(p, ccw); + } + else if (c->key.fill_cw != CLIP_CULL) { + emit_primitives(c, c->key.fill_cw, c->key.offset_cw); + } + else if (c->key.fill_ccw != CLIP_CULL) { + emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw); + } +} + + + + +static void check_nr_verts( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *if_insn; + + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.nr_verts, brw_imm_d(3)); + if_insn = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_kill_thread(c); + } + brw_ENDIF(p, if_insn); +} + + +void brw_emit_unfilled_clip( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *do_clip; + + + c->need_direction = ((c->key.offset_ccw || c->key.offset_cw) || + (c->key.fill_ccw != c->key.fill_cw) || + c->key.fill_ccw == CLIP_CULL || + c->key.fill_cw == CLIP_CULL || + c->key.copy_bfc_cw || + c->key.copy_bfc_ccw); + + brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); + brw_clip_tri_init_vertices(c); + + assert(c->offset[VERT_RESULT_EDGE]); + + if (c->key.fill_ccw == CLIP_CULL && + c->key.fill_cw == CLIP_CULL) { + brw_clip_kill_thread(c); + return; + } + + merge_edgeflags(c); + + /* Need to use the inlist indirection here: + */ + if (c->need_direction) + compute_tri_direction(c); + + if (c->key.fill_ccw == CLIP_CULL || + c->key.fill_cw == CLIP_CULL) + cull_direction(c); + + if (c->key.offset_ccw || + c->key.offset_cw) + compute_offset(c); + + if (c->key.copy_bfc_ccw || + c->key.copy_bfc_cw) + copy_bfc(c); + + /* Need to do this whether we clip or not: + */ + if (c->key.do_flat_shading) + brw_clip_tri_flat_shade(c); + + brw_clip_init_clipmask(c); + brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0)); + do_clip = brw_IF(p, BRW_EXECUTE_1); + { + brw_clip_init_planes(c); + brw_clip_tri(c); + check_nr_verts(c); + } + brw_ENDIF(p, do_clip); + + emit_unfilled_primitives(c); + brw_clip_kill_thread(c); +} + + + diff --git a/src/gallium/drivers/i965simple/brw_clip_util.c b/src/gallium/drivers/i965simple/brw_clip_util.c new file mode 100644 index 0000000000..6d58ceafff --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_util.c @@ -0,0 +1,351 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_clip.h" + + + + + +static struct brw_reg get_tmp( struct brw_clip_compile *c ) +{ + struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0); + + if (++c->last_tmp > c->prog_data.total_grf) + c->prog_data.total_grf = c->last_tmp; + + return tmp; +} + +static void release_tmp( struct brw_clip_compile *c, struct brw_reg tmp ) +{ + if (tmp.nr == c->last_tmp-1) + c->last_tmp--; +} + + +static struct brw_reg make_plane_ud(unsigned x, unsigned y, unsigned z, unsigned w) +{ + return brw_imm_ud((w<<24) | (z<<16) | (y<<8) | x); +} + + +void brw_clip_init_planes( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + + if (!c->key.nr_userclip) { + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 0), make_plane_ud( 0, 0, 0xff, 1)); + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 1), make_plane_ud( 0, 0, 1, 1)); + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 2), make_plane_ud( 0, 0xff, 0, 1)); + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 3), make_plane_ud( 0, 1, 0, 1)); + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 4), make_plane_ud(0xff, 0, 0, 1)); + brw_MOV(p, get_element_ud(c->reg.fixed_planes, 5), make_plane_ud( 1, 0, 0, 1)); + } +} + + + +#define W 3 + +/* Project 'pos' to screen space (or back again), overwrite with results: + */ +static void brw_clip_project_position(struct brw_clip_compile *c, struct brw_reg pos ) +{ + struct brw_compile *p = &c->func; + + /* calc rhw + */ + brw_math_invert(p, get_element(pos, W), get_element(pos, W)); + + /* value.xyz *= value.rhw + */ + brw_set_access_mode(p, BRW_ALIGN_16); + brw_MUL(p, brw_writemask(pos, TGSI_WRITEMASK_XYZ), pos, brw_swizzle1(pos, W)); + brw_set_access_mode(p, BRW_ALIGN_1); +} + + +static void brw_clip_project_vertex( struct brw_clip_compile *c, + struct brw_indirect vert_addr ) +{ +#if 0 + struct brw_compile *p = &c->func; + struct brw_reg tmp = get_tmp(c); + + /* Fixup position. Extract from the original vertex and re-project + * to screen space: + */ + brw_MOV(p, tmp, deref_4f(vert_addr, c->offset[VERT_RESULT_HPOS])); + brw_clip_project_position(c, tmp); + brw_MOV(p, deref_4f(vert_addr, c->header_position_offset), tmp); + + release_tmp(c, tmp); +#else + #warning "disabled" +#endif +} + + + + +/* Interpolate between two vertices and put the result into a0.0. + * Increment a0.0 accordingly. + */ +void brw_clip_interp_vertex( struct brw_clip_compile *c, + struct brw_indirect dest_ptr, + struct brw_indirect v0_ptr, /* from */ + struct brw_indirect v1_ptr, /* to */ + struct brw_reg t0, + boolean force_edgeflag) +{ +#if 0 + struct brw_compile *p = &c->func; + struct brw_reg tmp = get_tmp(c); + unsigned i; + + /* Just copy the vertex header: + */ + brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); + + /* Iterate over each attribute (could be done in pairs?) + */ + for (i = 0; i < c->nr_attrs; i++) { + unsigned delta = i*16 + 32; + + if (delta == c->offset[VERT_RESULT_EDGE]) { + if (force_edgeflag) + brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1)); + else + brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta)); + } + else { + /* Interpolate: + * + * New = attr0 + t*attr1 - t*attr0 + */ + brw_MUL(p, + vec4(brw_null_reg()), + deref_4f(v1_ptr, delta), + t0); + + brw_MAC(p, + tmp, + negate(deref_4f(v0_ptr, delta)), + t0); + + brw_ADD(p, + deref_4f(dest_ptr, delta), + deref_4f(v0_ptr, delta), + tmp); + } + } + + if (i & 1) { + unsigned delta = i*16 + 32; + brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); + } + + release_tmp(c, tmp); + + /* Recreate the projected (NDC) coordinate in the new vertex + * header: + */ + brw_clip_project_vertex(c, dest_ptr ); +#else + #warning "disabled" +#endif +} + + + + +#define MAX_MRF 16 + +void brw_clip_emit_vue(struct brw_clip_compile *c, + struct brw_indirect vert, + boolean allocate, + boolean eot, + unsigned header) +{ + struct brw_compile *p = &c->func; + unsigned start = c->last_mrf; + + assert(!(allocate && eot)); + + /* Cycle through mrf regs - probably futile as we have to wait for + * the allocation response anyway. Also, the order this function + * is invoked doesn't correspond to the order the instructions will + * be executed, so it won't have any effect in many cases. + */ +#if 0 + if (start + c->nr_regs + 1 >= MAX_MRF) + start = 0; + + c->last_mrf = start + c->nr_regs + 1; +#endif + + /* Copy the vertex from vertn into m1..mN+1: + */ + brw_copy_from_indirect(p, brw_message_reg(start+1), vert, c->nr_regs); + + /* Overwrite PrimType and PrimStart in the message header, for + * each vertex in turn: + */ + brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header)); + + + /* Send each vertex as a seperate write to the urb. This + * is different to the concept in brw_sf_emit.c, where + * subsequent writes are used to build up a single urb + * entry. Each of these writes instantiates a seperate + * urb entry - (I think... what about 'allocate'?) + */ + brw_urb_WRITE(p, + allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD), + start, + c->reg.R0, + allocate, + 1, /* used */ + c->nr_regs + 1, /* msg length */ + allocate ? 1 : 0, /* response_length */ + eot, /* eot */ + 1, /* writes_complete */ + 0, /* urb offset */ + BRW_URB_SWIZZLE_NONE); +} + + + +void brw_clip_kill_thread(struct brw_clip_compile *c) +{ + struct brw_compile *p = &c->func; + + /* Send an empty message to kill the thread and release any + * allocated urb entry: + */ + brw_urb_WRITE(p, + retype(brw_null_reg(), BRW_REGISTER_TYPE_UD), + 0, + c->reg.R0, + 0, /* allocate */ + 0, /* used */ + 0, /* msg len */ + 0, /* response len */ + 1, /* eot */ + 1, /* writes complete */ + 0, + BRW_URB_SWIZZLE_NONE); +} + + + + +struct brw_reg brw_clip_plane0_address( struct brw_clip_compile *c ) +{ + return brw_address(c->reg.fixed_planes); +} + + +struct brw_reg brw_clip_plane_stride( struct brw_clip_compile *c ) +{ + if (c->key.nr_userclip) { + return brw_imm_uw(16); + } + else { + return brw_imm_uw(4); + } +} + + +/* If flatshading, distribute color from provoking vertex prior to + * clipping. + */ +void brw_clip_copy_colors( struct brw_clip_compile *c, + unsigned to, unsigned from ) +{ +#if 0 + struct brw_compile *p = &c->func; + + if (c->offset[VERT_RESULT_COL0]) + brw_MOV(p, + byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL0]), + byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL0])); + + if (c->offset[VERT_RESULT_COL1]) + brw_MOV(p, + byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_COL1]), + byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_COL1])); + + if (c->offset[VERT_RESULT_BFC0]) + brw_MOV(p, + byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC0]), + byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC0])); + + if (c->offset[VERT_RESULT_BFC1]) + brw_MOV(p, + byte_offset(c->reg.vertex[to], c->offset[VERT_RESULT_BFC1]), + byte_offset(c->reg.vertex[from], c->offset[VERT_RESULT_BFC1])); +#else + #warning "disabled" +#endif +} + + + +void brw_clip_init_clipmask( struct brw_clip_compile *c ) +{ + struct brw_compile *p = &c->func; + struct brw_reg incoming = get_element_ud(c->reg.R0, 2); + + /* Shift so that lowest outcode bit is rightmost: + */ + brw_SHR(p, c->reg.planemask, incoming, brw_imm_ud(26)); + + if (c->key.nr_userclip) { + struct brw_reg tmp = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UD); + + /* Rearrange userclip outcodes so that they come directly after + * the fixed plane bits. + */ + brw_AND(p, tmp, incoming, brw_imm_ud(0x3f<<14)); + brw_SHR(p, tmp, tmp, brw_imm_ud(8)); + brw_OR(p, c->reg.planemask, c->reg.planemask, tmp); + + release_tmp(c, tmp); + } +} + diff --git a/src/gallium/drivers/i965simple/brw_context.c b/src/gallium/drivers/i965simple/brw_context.c new file mode 100644 index 0000000000..5e58701e91 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_context.c @@ -0,0 +1,245 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_draw.h" +#include "brw_vs.h" +#include "brw_tex_layout.h" +#include "brw_winsys.h" + +#include "pipe/p_winsys.h" +#include "pipe/p_context.h" +#include "pipe/p_util.h" + +/*************************************** + * Mesa's Driver Functions + ***************************************/ + +#ifndef BRW_DEBUG +int BRW_DEBUG = (0); +#endif + +static void brw_destroy(struct pipe_context *pipe) +{ + struct brw_context *brw = brw_context(pipe); + + FREE(brw); +} + +static void brw_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue) +{ + int x, y, w, h; + /* FIXME: corny... */ + + x = 0; + y = 0; + w = ps->width; + h = ps->height; + + pipe->surface_fill(pipe, ps, x, y, w, h, clearValue); +} + + +static int +brw_get_param(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return 8; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 0; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 0; + case PIPE_CAP_POINT_SPRITE: + return 0; + case PIPE_CAP_MAX_RENDER_TARGETS: + return 1; + case PIPE_CAP_OCCLUSION_QUERY: + return 0; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 11; /* max 1024x1024 */ + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 8; /* max 128x128x128 */ + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 11; /* max 1024x1024 */ + default: + return 0; + } +} + + +static float +brw_get_paramf(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 7.5; + + case PIPE_CAP_MAX_POINT_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 255.0; + + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 4.0; + + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; + + default: + return 0; + } +} + +static boolean +brw_is_format_supported( struct pipe_context *pipe, + enum pipe_format format, uint type ) +{ +#if 0 + /* XXX: This is broken -- rewrite if still needed. */ + static const unsigned tex_supported[] = { + PIPE_FORMAT_U_R8_G8_B8_A8, + PIPE_FORMAT_U_A8_R8_G8_B8, + PIPE_FORMAT_U_R5_G6_B5, + PIPE_FORMAT_U_L8, + PIPE_FORMAT_U_A8, + PIPE_FORMAT_U_I8, + PIPE_FORMAT_U_L8_A8, + PIPE_FORMAT_YCBCR, + PIPE_FORMAT_YCBCR_REV, + PIPE_FORMAT_S8_Z24, + }; + + + /* Actually a lot more than this - add later: + */ + static const unsigned render_supported[] = { + PIPE_FORMAT_U_A8_R8_G8_B8, + PIPE_FORMAT_U_R5_G6_B5, + }; + + /* + */ + static const unsigned z_stencil_supported[] = { + PIPE_FORMAT_U_Z16, + PIPE_FORMAT_U_Z32, + PIPE_FORMAT_S8_Z24, + }; + + switch (type) { + case PIPE_RENDER_FORMAT: + *numFormats = Elements(render_supported); + return render_supported; + + case PIPE_TEX_FORMAT: + *numFormats = Elements(tex_supported); + return render_supported; + + case PIPE_Z_STENCIL_FORMAT: + *numFormats = Elements(render_supported); + return render_supported; + + default: + *numFormats = 0; + return NULL; + } +#else + switch (format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_R5G6B5_UNORM: + case PIPE_FORMAT_S8Z24_UNORM: + return TRUE; + default: + return FALSE; + }; + return FALSE; +#endif +} + + + + +struct pipe_context *brw_create(struct pipe_winsys *pipe_winsys, + struct brw_winsys *brw_winsys, + unsigned pci_id) +{ + struct brw_context *brw; + + pipe_winsys->printf(pipe_winsys, + "%s: creating brw_context with pci id 0x%x\n", + __FUNCTION__, pci_id); + + brw = CALLOC_STRUCT(brw_context); + if (brw == NULL) + return NULL; + + brw->winsys = brw_winsys; + brw->pipe.winsys = pipe_winsys; + + brw->pipe.destroy = brw_destroy; + brw->pipe.is_format_supported = brw_is_format_supported; + brw->pipe.get_param = brw_get_param; + brw->pipe.get_paramf = brw_get_paramf; + brw->pipe.clear = brw_clear; + brw->pipe.texture_create = brw_texture_create; + brw->pipe.texture_release = brw_texture_release; + + brw_init_surface_functions(brw); + brw_init_state_functions(brw); + brw_init_flush_functions(brw); + brw_init_string_functions(brw); + brw_init_draw_functions( brw ); + + + brw_init_state( brw ); + + brw->pci_id = pci_id; + brw->dirty = ~0; + brw->hardware_dirty = ~0; + + memset(&brw->wm.bind, ~0, sizeof(brw->wm.bind)); + + return &brw->pipe; +} + diff --git a/src/gallium/drivers/i965simple/brw_context.h b/src/gallium/drivers/i965simple/brw_context.h new file mode 100644 index 0000000000..65664d853d --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_context.h @@ -0,0 +1,690 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRWCONTEXT_INC +#define BRWCONTEXT_INC + + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_state.h" + +#include "brw_structs.h" +#include "brw_winsys.h" + + +/* Glossary: + * + * URB - uniform resource buffer. A mid-sized buffer which is + * partitioned between the fixed function units and used for passing + * values (vertices, primitives, constants) between them. + * + * CURBE - constant URB entry. An urb region (entry) used to hold + * constant values which the fixed function units can be instructed to + * preload into the GRF when spawining a thread. + * + * VUE - vertex URB entry. An urb entry holding a vertex and usually + * a vertex header. The header contains control information and + * things like primitive type, Begin/end flags and clip codes. + * + * PUE - primitive URB entry. An urb entry produced by the setup (SF) + * unit holding rasterization and interpolation parameters. + * + * GRF - general register file. One of several register files + * addressable by programmed threads. The inputs (r0, payload, curbe, + * urb) of the thread are preloaded to this area before the thread is + * spawned. The registers are individually 8 dwords wide and suitable + * for general usage. Registers holding thread input values are not + * special and may be overwritten. + * + * MRF - message register file. Threads communicate (and terminate) + * by sending messages. Message parameters are placed in contigous + * MRF registers. All program output is via these messages. URB + * entries are populated by sending a message to the shared URB + * function containing the new data, together with a control word, + * often an unmodified copy of R0. + * + * R0 - GRF register 0. Typically holds control information used when + * sending messages to other threads. + * + * EU or GEN4 EU: The name of the programmable subsystem of the + * i965 hardware. Threads are executed by the EU, the registers + * described above are part of the EU architecture. + * + * Fixed function units: + * + * CS - Command streamer. Notional first unit, little software + * interaction. Holds the URB entries used for constant data, ie the + * CURBEs. + * + * VF/VS - Vertex Fetch / Vertex Shader. The fixed function part of + * this unit is responsible for pulling vertices out of vertex buffers + * in vram and injecting them into the processing pipe as VUEs. If + * enabled, it first passes them to a VS thread which is a good place + * for the driver to implement any active vertex shader. + * + * GS - Geometry Shader. This corresponds to a new DX10 concept. If + * enabled, incoming strips etc are passed to GS threads in individual + * line/triangle/point units. The GS thread may perform arbitary + * computation and emit whatever primtives with whatever vertices it + * chooses. This makes GS an excellent place to implement GL's + * unfilled polygon modes, though of course it is capable of much + * more. Additionally, GS is used to translate away primitives not + * handled by latter units, including Quads and Lineloops. + * + * CS - Clipper. Mesa's clipping algorithms are imported to run on + * this unit. The fixed function part performs cliptesting against + * the 6 fixed clipplanes and makes descisions on whether or not the + * incoming primitive needs to be passed to a thread for clipping. + * User clip planes are handled via cooperation with the VS thread. + * + * SF - Strips Fans or Setup: Triangles are prepared for + * rasterization. Interpolation coefficients are calculated. + * Flatshading and two-side lighting usually performed here. + * + * WM - Windower. Interpolation of vertex attributes performed here. + * Fragment shader implemented here. SIMD aspects of EU taken full + * advantage of, as pixels are processed in blocks of 16. + * + * CC - Color Calculator. No EU threads associated with this unit. + * Handles blending and (presumably) depth and stencil testing. + */ + +#define BRW_MAX_CURBE (32*16) + +struct brw_context; +struct brw_winsys; + + +/* Raised when we receive new state across the pipe interface: + */ +#define BRW_NEW_VIEWPORT 0x1 +#define BRW_NEW_RASTERIZER 0x2 +#define BRW_NEW_FS 0x4 +#define BRW_NEW_BLEND 0x8 +#define BRW_NEW_CLIP 0x10 +#define BRW_NEW_SCISSOR 0x20 +#define BRW_NEW_STIPPLE 0x40 +#define BRW_NEW_FRAMEBUFFER 0x80 +#define BRW_NEW_ALPHA_TEST 0x100 +#define BRW_NEW_DEPTH_STENCIL 0x200 +#define BRW_NEW_SAMPLER 0x400 +#define BRW_NEW_TEXTURE 0x800 +#define BRW_NEW_CONSTANTS 0x1000 +#define BRW_NEW_VBO 0x2000 +#define BRW_NEW_VS 0x4000 + +/* Raised for other internal events: + */ +#define BRW_NEW_URB_FENCE 0x10000 +#define BRW_NEW_PSP 0x20000 +#define BRW_NEW_CURBE_OFFSETS 0x40000 +#define BRW_NEW_REDUCED_PRIMITIVE 0x80000 +#define BRW_NEW_PRIMITIVE 0x100000 +#define BRW_NEW_SCENE 0x200000 +#define BRW_NEW_SF_LINKAGE 0x400000 + +extern int BRW_DEBUG; + +#define DEBUG_TEXTURE 0x1 +#define DEBUG_STATE 0x2 +#define DEBUG_IOCTL 0x4 +#define DEBUG_PRIMS 0x8 +#define DEBUG_VERTS 0x10 +#define DEBUG_FALLBACKS 0x20 +#define DEBUG_VERBOSE 0x40 +#define DEBUG_DRI 0x80 +#define DEBUG_DMA 0x100 +#define DEBUG_SANITY 0x200 +#define DEBUG_SYNC 0x400 +#define DEBUG_SLEEP 0x800 +#define DEBUG_PIXEL 0x1000 +#define DEBUG_STATS 0x2000 +#define DEBUG_TILE 0x4000 +#define DEBUG_SINGLE_THREAD 0x8000 +#define DEBUG_WM 0x10000 +#define DEBUG_URB 0x20000 +#define DEBUG_VS 0x40000 +#define DEBUG_BATCH 0x80000 +#define DEBUG_BUFMGR 0x100000 +#define DEBUG_BLIT 0x200000 +#define DEBUG_REGION 0x400000 +#define DEBUG_MIPTREE 0x800000 + +#define DBG(...) do { \ + if (BRW_DEBUG & FILE_DEBUG_FLAG) \ + brw->pipe.winsys->printf(brw->pipe.winsys, __VA_ARGS__); \ +} while(0) + +#define PRINT(...) do { \ + brw->pipe.winsys->printf(brw->pipe.winsys, __VA_ARGS__); \ +} while(0) + +struct brw_state_flags { + unsigned cache; + unsigned brw; +}; + + +struct brw_shader_info { + int nr_regs[8]; /* TGSI_FILE_* */ +}; + + + +struct brw_vertex_program { + struct pipe_shader_state program; + struct brw_shader_info info; + int id; +}; + + + +struct brw_fragment_program { + struct pipe_shader_state program; + struct brw_shader_info info; + + boolean UsesDepth; + boolean UsesKill; + boolean ComputesDepth; + int id; +}; + + + + +struct pipe_setup_linkage { + struct { + unsigned vp_output:5; + unsigned interp_mode:4; + unsigned bf_vp_output:5; + } fp_input[PIPE_MAX_SHADER_INPUTS]; + + unsigned fp_input_count:5; + unsigned max_vp_output:5; +}; + + + +struct brw_texture { + struct pipe_texture base; + + /* Derived from the above: + */ + unsigned pitch; + unsigned depth_pitch; /* per-image on i945? */ + unsigned total_height; + + unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS]; + + /* Explicitly store the offset of each image for each cube face or + * depth value. Pretty much have to accept that hardware formats + * are going to be so diverse that there is no unified way to + * compute the offsets of depth/cube images within a mipmap level, + * so have to store them as a lookup table: + */ + unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS]; /**< array [depth] of offsets */ + + /* Includes image offset tables: + */ + unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; +}; + +/* Data about a particular attempt to compile a program. Note that + * there can be many of these, each in a different GL state + * corresponding to a different brw_wm_prog_key struct, with different + * compiled programs: + */ +/* Data about a particular attempt to compile a program. Note that + * there can be many of these, each in a different GL state + * corresponding to a different brw_wm_prog_key struct, with different + * compiled programs: + */ + +struct brw_wm_prog_data { + unsigned curb_read_length; + unsigned urb_read_length; + + unsigned first_curbe_grf; + unsigned total_grf; + unsigned total_scratch; + + /* Internally generated constants for the CURBE. These are loaded + * ahead of the data from the constant buffer. + */ + const float internal_const[8]; + unsigned nr_internal_consts; + unsigned max_const; + + boolean error; +}; + +struct brw_sf_prog_data { + unsigned urb_read_length; + unsigned total_grf; + + /* Each vertex may have upto 12 attributes, 4 components each, + * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11 + * rows. + * + * Actually we use 4 for each, so call it 12 rows. + */ + unsigned urb_entry_size; +}; + +struct brw_clip_prog_data { + unsigned curb_read_length; /* user planes? */ + unsigned clip_mode; + unsigned urb_read_length; + unsigned total_grf; +}; + +struct brw_gs_prog_data { + unsigned urb_read_length; + unsigned total_grf; +}; + +struct brw_vs_prog_data { + unsigned curb_read_length; + unsigned urb_read_length; + unsigned total_grf; + unsigned outputs_written; + + unsigned inputs_read; + + unsigned max_const; + + float imm_buf[PIPE_MAX_CONSTANT][4]; + unsigned num_imm; + unsigned num_consts; + + /* Used for calculating urb partitions: + */ + unsigned urb_entry_size; +}; + + +#define BRW_MAX_TEX_UNIT 8 +#define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1 + +/* Create a fixed sized struct for caching binding tables: + */ +struct brw_surface_binding_table { + unsigned surf_ss_offset[BRW_WM_MAX_SURF]; +}; + + +struct brw_cache; + +struct brw_mem_pool { + struct pipe_buffer *buffer; + + unsigned size; + unsigned offset; /* offset of first free byte */ + + struct brw_context *brw; +}; + +struct brw_cache_item { + unsigned hash; + unsigned key_size; /* for variable-sized keys */ + const void *key; + + unsigned offset; /* offset within pool's buffer */ + unsigned data_size; + + struct brw_cache_item *next; +}; + + + +struct brw_cache { + unsigned id; + + const char *name; + + struct brw_context *brw; + struct brw_mem_pool *pool; + + struct brw_cache_item **items; + unsigned size, n_items; + + unsigned key_size; /* for fixed-size keys */ + unsigned aux_size; + + unsigned last_addr; /* offset of active item */ +}; + + + + +/* Considered adding a member to this struct to document which flags + * an update might raise so that ordering of the state atoms can be + * checked or derived at runtime. Dropped the idea in favor of having + * a debug mode where the state is monitored for flags which are + * raised that have already been tested against. + */ +struct brw_tracked_state { + struct brw_state_flags dirty; + void (*update)( struct brw_context *brw ); +}; + + +/* Flags for brw->state.cache. + */ +#define CACHE_NEW_CC_VP (1<<BRW_CC_VP) +#define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT) +#define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG) +#define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR) +#define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER) +#define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT) +#define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG) +#define CACHE_NEW_SF_VP (1<<BRW_SF_VP) +#define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT) +#define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT) +#define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG) +#define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT) +#define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG) +#define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP) +#define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT) +#define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG) +#define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE) +#define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND) + + + + +enum brw_mempool_id { + BRW_GS_POOL, + BRW_SS_POOL, + BRW_MAX_POOL +}; + + +struct brw_cached_batch_item { + struct header *header; + unsigned sz; + struct brw_cached_batch_item *next; +}; + + + +/* Protect against a future where PIPE_ATTRIB_MAX > 32. Wouldn't life + * be easier if C allowed arrays of packed elements? + */ +#define ATTRIB_BIT_DWORDS ((PIPE_ATTRIB_MAX+31)/32) + + + + +struct brw_vertex_info { + unsigned varying; /* varying:1[PIPE_ATTRIB_MAX] */ + unsigned sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[PIPE_ATTRIB_MAX] */ +}; + + + + + +struct brw_context +{ + struct pipe_context pipe; + struct brw_winsys *winsys; + + unsigned primitive; + unsigned reduced_primitive; + + boolean emit_state_always; + + struct { + struct brw_state_flags dirty; + } state; + + + struct { + const struct pipe_blend_state *Blend; + const struct pipe_depth_stencil_alpha_state *DepthStencil; + const struct pipe_poly_stipple *PolygonStipple; + const struct pipe_rasterizer_state *Raster; + const struct pipe_sampler_state *Samplers[PIPE_MAX_SAMPLERS]; + const struct brw_vertex_program *VertexProgram; + const struct brw_fragment_program *FragmentProgram; + + struct pipe_clip_state Clip; + struct pipe_blend_color BlendColor; + struct pipe_scissor_state Scissor; + struct pipe_viewport_state Viewport; + struct pipe_framebuffer_state FrameBuffer; + + const struct pipe_constant_buffer *Constants[2]; + const struct brw_texture *Texture[PIPE_MAX_SAMPLERS]; + } attribs; + + struct brw_mem_pool pool[BRW_MAX_POOL]; + struct brw_cache cache[BRW_MAX_CACHE]; + struct brw_cached_batch_item *cached_batch_items; + + struct { + + /* Arrays with buffer objects to copy non-bufferobj arrays into + * for upload: + */ + struct pipe_vertex_buffer *vbo_array[PIPE_ATTRIB_MAX]; + + struct brw_vertex_element_state inputs[PIPE_ATTRIB_MAX]; + +#define BRW_NR_UPLOAD_BUFS 17 +#define BRW_UPLOAD_INIT_SIZE (128*1024) + + /* Summary of size and varying of active arrays, so we can check + * for changes to this state: + */ + struct brw_vertex_info info; + } vb; + + + unsigned hardware_dirty; + unsigned dirty; + unsigned pci_id; + /* BRW_NEW_URB_ALLOCATIONS: + */ + struct { + unsigned vsize; /* vertex size plus header in urb registers */ + unsigned csize; /* constant buffer size in urb registers */ + unsigned sfsize; /* setup data size in urb registers */ + + boolean constrained; + + unsigned nr_vs_entries; + unsigned nr_gs_entries; + unsigned nr_clip_entries; + unsigned nr_sf_entries; + unsigned nr_cs_entries; + +/* unsigned vs_size; */ +/* unsigned gs_size; */ +/* unsigned clip_size; */ +/* unsigned sf_size; */ +/* unsigned cs_size; */ + + unsigned vs_start; + unsigned gs_start; + unsigned clip_start; + unsigned sf_start; + unsigned cs_start; + } urb; + + + /* BRW_NEW_CURBE_OFFSETS: + */ + struct { + unsigned wm_start; + unsigned wm_size; + unsigned clip_start; + unsigned clip_size; + unsigned vs_start; + unsigned vs_size; + unsigned total_size; + + unsigned gs_offset; + + float *last_buf; + unsigned last_bufsz; + } curbe; + + struct { + struct brw_vs_prog_data *prog_data; + + unsigned prog_gs_offset; + unsigned state_gs_offset; + } vs; + + struct { + struct brw_gs_prog_data *prog_data; + + boolean prog_active; + unsigned prog_gs_offset; + unsigned state_gs_offset; + } gs; + + struct { + struct brw_clip_prog_data *prog_data; + + unsigned prog_gs_offset; + unsigned vp_gs_offset; + unsigned state_gs_offset; + } clip; + + + struct { + struct brw_sf_prog_data *prog_data; + + struct pipe_setup_linkage linkage; + + unsigned prog_gs_offset; + unsigned vp_gs_offset; + unsigned state_gs_offset; + } sf; + + struct { + struct brw_wm_prog_data *prog_data; + +// struct brw_wm_compiler *compile_data; + + + /** + * Array of sampler state uploaded at sampler_gs_offset of BRW_SAMPLER + * cache + */ + struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT]; + + unsigned render_surf; + unsigned nr_surfaces; + + unsigned max_threads; + struct pipe_buffer *scratch_buffer; + unsigned scratch_buffer_size; + + unsigned sampler_count; + unsigned sampler_gs_offset; + + struct brw_surface_binding_table bind; + unsigned bind_ss_offset; + + unsigned prog_gs_offset; + unsigned state_gs_offset; + } wm; + + + struct { + unsigned vp_gs_offset; + unsigned state_gs_offset; + } cc; + + + /* Used to give every program string a unique id + */ + unsigned program_id; +}; + + +#define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a) + + +/*====================================================================== + * brw_vtbl.c + */ +void brw_do_flush( struct brw_context *brw, + unsigned flags ); + + +/*====================================================================== + * brw_state.c + */ +void brw_validate_state(struct brw_context *brw); +void brw_init_state(struct brw_context *brw); +void brw_destroy_state(struct brw_context *brw); + + +/*====================================================================== + * brw_tex.c + */ +void brwUpdateTextureState( struct brw_context *brw ); + + +/* brw_urb.c + */ +void brw_upload_urb_fence(struct brw_context *brw); + +void brw_upload_constant_buffer_state(struct brw_context *brw); + +void brw_init_surface_functions(struct brw_context *brw); +void brw_init_state_functions(struct brw_context *brw); +void brw_init_flush_functions(struct brw_context *brw); +void brw_init_string_functions(struct brw_context *brw); + +/*====================================================================== + * Inline conversion functions. These are better-typed than the + * macros used previously: + */ +static inline struct brw_context * +brw_context( struct pipe_context *ctx ) +{ + return (struct brw_context *)ctx; +} + +#endif + diff --git a/src/gallium/drivers/i965simple/brw_curbe.c b/src/gallium/drivers/i965simple/brw_curbe.c new file mode 100644 index 0000000000..52bbd525c1 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_curbe.c @@ -0,0 +1,368 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_state.h" +#include "brw_batch.h" +#include "brw_util.h" +#include "brw_wm.h" +#include "pipe/p_state.h" +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" + +#define FILE_DEBUG_FLAG DEBUG_FALLBACKS + +/* Partition the CURBE between the various users of constant values: + */ +static void calculate_curbe_offsets( struct brw_context *brw ) +{ + /* CACHE_NEW_WM_PROG */ + unsigned nr_fp_regs = align(brw->wm.prog_data->max_const, 16); + + /* BRW_NEW_VERTEX_PROGRAM */ + unsigned nr_vp_regs = align(brw->vs.prog_data->max_const, 16); + unsigned nr_clip_regs = 0; + unsigned total_regs; + +#if 0 + /* BRW_NEW_CLIP ? */ + if (brw->attribs.Transform->ClipPlanesEnabled) { + unsigned nr_planes = 6 + brw_count_bits(brw->attribs.Transform->ClipPlanesEnabled); + nr_clip_regs = align(nr_planes * 4, 16); + } +#endif + + + total_regs = nr_fp_regs + nr_vp_regs + nr_clip_regs; + + /* This can happen - what to do? Probably rather than falling + * back, the best thing to do is emit programs which code the + * constants as immediate values. Could do this either as a static + * cap on WM and VS, or adaptively. + * + * Unfortunately, this is currently dependent on the results of the + * program generation process (in the case of wm), so this would + * introduce the need to re-generate programs in the event of a + * curbe allocation failure. + */ + /* Max size is 32 - just large enough to + * hold the 128 parameters allowed by + * the fragment and vertex program + * api's. It's not clear what happens + * when both VP and FP want to use 128 + * parameters, though. + */ + assert(total_regs <= 32); + + /* Lazy resize: + */ + if (nr_fp_regs > brw->curbe.wm_size || + nr_vp_regs > brw->curbe.vs_size || + nr_clip_regs != brw->curbe.clip_size || + (total_regs < brw->curbe.total_size / 4 && + brw->curbe.total_size > 16)) { + + unsigned reg = 0; + + /* Calculate a new layout: + */ + reg = 0; + brw->curbe.wm_start = reg; + brw->curbe.wm_size = nr_fp_regs; reg += nr_fp_regs; + brw->curbe.clip_start = reg; + brw->curbe.clip_size = nr_clip_regs; reg += nr_clip_regs; + brw->curbe.vs_start = reg; + brw->curbe.vs_size = nr_vp_regs; reg += nr_vp_regs; + brw->curbe.total_size = reg; + +#if 0 + if (0) + DBG("curbe wm %d+%d clip %d+%d vs %d+%d\n", + brw->curbe.wm_start, + brw->curbe.wm_size, + brw->curbe.clip_start, + brw->curbe.clip_size, + brw->curbe.vs_start, + brw->curbe.vs_size ); +#endif + + brw->state.dirty.brw |= BRW_NEW_CURBE_OFFSETS; + } +} + + +const struct brw_tracked_state brw_curbe_offsets = { + .dirty = { + .brw = (BRW_NEW_CLIP | + BRW_NEW_VS), + .cache = CACHE_NEW_WM_PROG + }, + .update = calculate_curbe_offsets +}; + + + +/* Define the number of curbes within CS's urb allocation. Multiple + * urb entries -> multiple curbes. These will be used by + * fixed-function hardware in a double-buffering scheme to avoid a + * pipeline stall each time the contents of the curbe is changed. + */ +void brw_upload_constant_buffer_state(struct brw_context *brw) +{ + struct brw_constant_buffer_state cbs; + memset(&cbs, 0, sizeof(cbs)); + + /* It appears that this is the state packet for the CS unit, ie. the + * urb entries detailed here are housed in the CS range from the + * URB_FENCE command. + */ + cbs.header.opcode = CMD_CONST_BUFFER_STATE; + cbs.header.length = sizeof(cbs)/4 - 2; + + /* BRW_NEW_URB_FENCE */ + cbs.bits0.nr_urb_entries = brw->urb.nr_cs_entries; + cbs.bits0.urb_entry_size = brw->urb.csize - 1; + + assert(brw->urb.nr_cs_entries); + BRW_CACHED_BATCH_STRUCT(brw, &cbs); +} + + +static float fixed_plane[6][4] = { + { 0, 0, -1, 1 }, + { 0, 0, 1, 1 }, + { 0, -1, 0, 1 }, + { 0, 1, 0, 1 }, + {-1, 0, 0, 1 }, + { 1, 0, 0, 1 } +}; + +/* Upload a new set of constants. Too much variability to go into the + * cache mechanism, but maybe would benefit from a comparison against + * the current uploaded set of constants. + */ +static void upload_constant_buffer(struct brw_context *brw) +{ + struct brw_mem_pool *pool = &brw->pool[BRW_GS_POOL]; + unsigned sz = brw->curbe.total_size; + unsigned bufsz = sz * sizeof(float); + float *buf; + unsigned i; + + + if (sz == 0) { + struct brw_constant_buffer cb; + cb.header.opcode = CMD_CONST_BUFFER; + cb.header.length = sizeof(cb)/4 - 2; + cb.header.valid = 0; + cb.bits0.buffer_length = 0; + cb.bits0.buffer_address = 0; + BRW_BATCH_STRUCT(brw, &cb); + + if (brw->curbe.last_buf) { + free(brw->curbe.last_buf); + brw->curbe.last_buf = NULL; + brw->curbe.last_bufsz = 0; + } + + return; + } + + buf = (float *)malloc(bufsz); + + memset(buf, 0, bufsz); + + if (brw->curbe.wm_size) { + unsigned offset = brw->curbe.wm_start * 16; + + /* First the constant buffer constants: + */ + + /* Then any internally generated constants: + */ + for (i = 0; i < brw->wm.prog_data->nr_internal_consts; i++) + buf[offset + i] = brw->wm.prog_data->internal_const[i]; + + assert(brw->wm.prog_data->max_const == + brw->wm.prog_data->nr_internal_consts); + } + + + /* The clipplanes are actually delivered to both CLIP and VS units. + * VS uses them to calculate the outcode bitmasks. + */ + if (brw->curbe.clip_size) { + unsigned offset = brw->curbe.clip_start * 16; + unsigned j; + + /* If any planes are going this way, send them all this way: + */ + for (i = 0; i < 6; i++) { + buf[offset + i * 4 + 0] = fixed_plane[i][0]; + buf[offset + i * 4 + 1] = fixed_plane[i][1]; + buf[offset + i * 4 + 2] = fixed_plane[i][2]; + buf[offset + i * 4 + 3] = fixed_plane[i][3]; + } + + /* Clip planes: BRW_NEW_CLIP: + */ + for (j = 0; j < brw->attribs.Clip.nr; j++) { + buf[offset + i * 4 + 0] = brw->attribs.Clip.ucp[j][0]; + buf[offset + i * 4 + 1] = brw->attribs.Clip.ucp[j][1]; + buf[offset + i * 4 + 2] = brw->attribs.Clip.ucp[j][2]; + buf[offset + i * 4 + 3] = brw->attribs.Clip.ucp[j][3]; + i++; + } + } + + + if (brw->curbe.vs_size) { + unsigned offset = brw->curbe.vs_start * 16; + /*unsigned nr = vp->max_const;*/ + const struct pipe_constant_buffer *cbuffer = brw->attribs.Constants[0]; + struct pipe_winsys *ws = brw->pipe.winsys; + /* FIXME: buffer size is num_consts + num_immediates */ + if (brw->vs.prog_data->num_consts) { + /* map the vertex constant buffer and copy to curbe: */ + void *data = ws->buffer_map(ws, cbuffer->buffer, 0); + /* FIXME: this is wrong. the cbuffer->size currently + * represents size of consts + immediates. so if we'll + * have both we'll copy over the end of the buffer + * with the subsequent memcpy */ + memcpy(&buf[offset], data, cbuffer->size); + ws->buffer_unmap(ws, cbuffer->buffer); + offset += cbuffer->size; + } + /*immediates*/ + if (brw->vs.prog_data->num_imm) { + memcpy(&buf[offset], brw->vs.prog_data->imm_buf, + brw->vs.prog_data->num_imm * 4 * sizeof(float)); + } + } + + if (1) { + for (i = 0; i < sz; i+=4) + debug_printf("curbe %d.%d: %f %f %f %f\n", i/8, i&4, + buf[i+0], buf[i+1], buf[i+2], buf[i+3]); + + debug_printf("last_buf %p buf %p sz %d/%d cmp %d\n", + brw->curbe.last_buf, buf, + bufsz, brw->curbe.last_bufsz, + brw->curbe.last_buf ? memcmp(buf, brw->curbe.last_buf, bufsz) : -1); + } + + if (brw->curbe.last_buf && + bufsz == brw->curbe.last_bufsz && + memcmp(buf, brw->curbe.last_buf, bufsz) == 0) { + free(buf); +/* return; */ + } + else { + if (brw->curbe.last_buf) + free(brw->curbe.last_buf); + brw->curbe.last_buf = buf; + brw->curbe.last_bufsz = bufsz; + + + if (!brw_pool_alloc(pool, + bufsz, + 1 << 6, + &brw->curbe.gs_offset)) { + debug_printf("out of GS memory for curbe\n"); + assert(0); + return; + } + + + /* Copy data to the buffer: + */ + brw->winsys->buffer_subdata_typed(brw->winsys, + pool->buffer, + brw->curbe.gs_offset, + bufsz, + buf, + BRW_CONSTANT_BUFFER ); + } + + /* TODO: only emit the constant_buffer packet when necessary, ie: + - contents have changed + - offset has changed + - hw requirements due to other packets emitted. + */ + { + struct brw_constant_buffer cb; + + memset(&cb, 0, sizeof(cb)); + + cb.header.opcode = CMD_CONST_BUFFER; + cb.header.length = sizeof(cb)/4 - 2; + cb.header.valid = 1; + cb.bits0.buffer_length = sz - 1; + cb.bits0.buffer_address = brw->curbe.gs_offset >> 6; + + /* Because this provokes an action (ie copy the constants into the + * URB), it shouldn't be shortcircuited if identical to the + * previous time - because eg. the urb destination may have + * changed, or the urb contents different to last time. + * + * Note that the data referred to is actually copied internally, + * not just used in place according to passed pointer. + * + * It appears that the CS unit takes care of using each available + * URB entry (Const URB Entry == CURBE) in turn, and issuing + * flushes as necessary when doublebuffering of CURBEs isn't + * possible. + */ + BRW_BATCH_STRUCT(brw, &cb); + } +} + +/* This tracked state is unique in that the state it monitors varies + * dynamically depending on the parameters tracked by the fragment and + * vertex programs. This is the template used as a starting point, + * each context will maintain a copy of this internally and update as + * required. + */ +const struct brw_tracked_state brw_constant_buffer = { + .dirty = { + .brw = (BRW_NEW_CLIP | + BRW_NEW_CONSTANTS | + BRW_NEW_URB_FENCE | /* Implicit - hardware requires this, not used above */ + BRW_NEW_PSP | /* Implicit - hardware requires this, not used above */ + BRW_NEW_CURBE_OFFSETS), + .cache = (CACHE_NEW_WM_PROG) + }, + .update = upload_constant_buffer +}; + diff --git a/src/gallium/drivers/i965simple/brw_defines.h b/src/gallium/drivers/i965simple/brw_defines.h new file mode 100644 index 0000000000..9379a397f6 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_defines.h @@ -0,0 +1,852 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_DEFINES_H +#define BRW_DEFINES_H + +/* + */ +#define MI_NOOP 0x00 +#define MI_USER_INTERRUPT 0x02 +#define MI_WAIT_FOR_EVENT 0x03 +#define MI_FLUSH 0x04 +#define MI_REPORT_HEAD 0x07 +#define MI_ARB_ON_OFF 0x08 +#define MI_BATCH_BUFFER_END 0x0A +#define MI_OVERLAY_FLIP 0x11 +#define MI_LOAD_SCAN_LINES_INCL 0x12 +#define MI_LOAD_SCAN_LINES_EXCL 0x13 +#define MI_DISPLAY_BUFFER_INFO 0x14 +#define MI_SET_CONTEXT 0x18 +#define MI_STORE_DATA_IMM 0x20 +#define MI_STORE_DATA_INDEX 0x21 +#define MI_LOAD_REGISTER_IMM 0x22 +#define MI_STORE_REGISTER_MEM 0x24 +#define MI_BATCH_BUFFER_START 0x31 + +#define MI_SYNCHRONOUS_FLIP 0x0 +#define MI_ASYNCHRONOUS_FLIP 0x1 + +#define MI_BUFFER_SECURE 0x0 +#define MI_BUFFER_NONSECURE 0x1 + +#define MI_ARBITRATE_AT_CHAIN_POINTS 0x0 +#define MI_ARBITRATE_BETWEEN_INSTS 0x1 +#define MI_NO_ARBITRATION 0x3 + +#define MI_CONDITION_CODE_WAIT_DISABLED 0x0 +#define MI_CONDITION_CODE_WAIT_0 0x1 +#define MI_CONDITION_CODE_WAIT_1 0x2 +#define MI_CONDITION_CODE_WAIT_2 0x3 +#define MI_CONDITION_CODE_WAIT_3 0x4 +#define MI_CONDITION_CODE_WAIT_4 0x5 + +#define MI_DISPLAY_PIPE_A 0x0 +#define MI_DISPLAY_PIPE_B 0x1 + +#define MI_DISPLAY_PLANE_A 0x0 +#define MI_DISPLAY_PLANE_B 0x1 +#define MI_DISPLAY_PLANE_C 0x2 + +#define MI_STANDARD_FLIP 0x0 +#define MI_ENQUEUE_FLIP_PERFORM_BASE_FRAME_NUMBER_LOAD 0x1 +#define MI_ENQUEUE_FLIP_TARGET_FRAME_NUMBER_RELATIVE 0x2 +#define MI_ENQUEUE_FLIP_ABSOLUTE_TARGET_FRAME_NUMBER 0x3 + +#define MI_PHYSICAL_ADDRESS 0x0 +#define MI_VIRTUAL_ADDRESS 0x1 + +#define MI_BUFFER_MEMORY_MAIN 0x0 +#define MI_BUFFER_MEMORY_GTT 0x2 +#define MI_BUFFER_MEMORY_PER_PROCESS_GTT 0x3 + +#define MI_FLIP_CONTINUE 0x0 +#define MI_FLIP_ON 0x1 +#define MI_FLIP_OFF 0x2 + +#define MI_UNTRUSTED_REGISTER_SPACE 0x0 +#define MI_TRUSTED_REGISTER_SPACE 0x1 + +/* 3D state: + */ +#define _3DOP_3DSTATE_PIPELINED 0x0 +#define _3DOP_3DSTATE_NONPIPELINED 0x1 +#define _3DOP_3DCONTROL 0x2 +#define _3DOP_3DPRIMITIVE 0x3 + +#define _3DSTATE_PIPELINED_POINTERS 0x00 +#define _3DSTATE_BINDING_TABLE_POINTERS 0x01 +#define _3DSTATE_VERTEX_BUFFERS 0x08 +#define _3DSTATE_VERTEX_ELEMENTS 0x09 +#define _3DSTATE_INDEX_BUFFER 0x0A +#define _3DSTATE_VF_STATISTICS 0x0B +#define _3DSTATE_DRAWING_RECTANGLE 0x00 +#define _3DSTATE_CONSTANT_COLOR 0x01 +#define _3DSTATE_SAMPLER_PALETTE_LOAD 0x02 +#define _3DSTATE_CHROMA_KEY 0x04 +#define _3DSTATE_DEPTH_BUFFER 0x05 +#define _3DSTATE_POLY_STIPPLE_OFFSET 0x06 +#define _3DSTATE_POLY_STIPPLE_PATTERN 0x07 +#define _3DSTATE_LINE_STIPPLE 0x08 +#define _3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP 0x09 +#define _3DCONTROL 0x00 +#define _3DPRIMITIVE 0x00 + +#define PIPE_CONTROL_NOWRITE 0x00 +#define PIPE_CONTROL_WRITEIMMEDIATE 0x01 +#define PIPE_CONTROL_WRITEDEPTH 0x02 +#define PIPE_CONTROL_WRITETIMESTAMP 0x03 + +#define PIPE_CONTROL_GTTWRITE_PROCESS_LOCAL 0x00 +#define PIPE_CONTROL_GTTWRITE_GLOBAL 0x01 + +#define _3DPRIM_POINTLIST 0x01 +#define _3DPRIM_LINELIST 0x02 +#define _3DPRIM_LINESTRIP 0x03 +#define _3DPRIM_TRILIST 0x04 +#define _3DPRIM_TRISTRIP 0x05 +#define _3DPRIM_TRIFAN 0x06 +#define _3DPRIM_QUADLIST 0x07 +#define _3DPRIM_QUADSTRIP 0x08 +#define _3DPRIM_LINELIST_ADJ 0x09 +#define _3DPRIM_LINESTRIP_ADJ 0x0A +#define _3DPRIM_TRILIST_ADJ 0x0B +#define _3DPRIM_TRISTRIP_ADJ 0x0C +#define _3DPRIM_TRISTRIP_REVERSE 0x0D +#define _3DPRIM_POLYGON 0x0E +#define _3DPRIM_RECTLIST 0x0F +#define _3DPRIM_LINELOOP 0x10 +#define _3DPRIM_POINTLIST_BF 0x11 +#define _3DPRIM_LINESTRIP_CONT 0x12 +#define _3DPRIM_LINESTRIP_BF 0x13 +#define _3DPRIM_LINESTRIP_CONT_BF 0x14 +#define _3DPRIM_TRIFAN_NOSTIPPLE 0x15 + +#define _3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL 0 +#define _3DPRIM_VERTEXBUFFER_ACCESS_RANDOM 1 + +#define BRW_ANISORATIO_2 0 +#define BRW_ANISORATIO_4 1 +#define BRW_ANISORATIO_6 2 +#define BRW_ANISORATIO_8 3 +#define BRW_ANISORATIO_10 4 +#define BRW_ANISORATIO_12 5 +#define BRW_ANISORATIO_14 6 +#define BRW_ANISORATIO_16 7 + +#define BRW_BLENDFACTOR_ONE 0x1 +#define BRW_BLENDFACTOR_SRC_COLOR 0x2 +#define BRW_BLENDFACTOR_SRC_ALPHA 0x3 +#define BRW_BLENDFACTOR_DST_ALPHA 0x4 +#define BRW_BLENDFACTOR_DST_COLOR 0x5 +#define BRW_BLENDFACTOR_SRC_ALPHA_SATURATE 0x6 +#define BRW_BLENDFACTOR_CONST_COLOR 0x7 +#define BRW_BLENDFACTOR_CONST_ALPHA 0x8 +#define BRW_BLENDFACTOR_SRC1_COLOR 0x9 +#define BRW_BLENDFACTOR_SRC1_ALPHA 0x0A +#define BRW_BLENDFACTOR_ZERO 0x11 +#define BRW_BLENDFACTOR_INV_SRC_COLOR 0x12 +#define BRW_BLENDFACTOR_INV_SRC_ALPHA 0x13 +#define BRW_BLENDFACTOR_INV_DST_ALPHA 0x14 +#define BRW_BLENDFACTOR_INV_DST_COLOR 0x15 +#define BRW_BLENDFACTOR_INV_CONST_COLOR 0x17 +#define BRW_BLENDFACTOR_INV_CONST_ALPHA 0x18 +#define BRW_BLENDFACTOR_INV_SRC1_COLOR 0x19 +#define BRW_BLENDFACTOR_INV_SRC1_ALPHA 0x1A + +#define BRW_BLENDFUNCTION_ADD 0 +#define BRW_BLENDFUNCTION_SUBTRACT 1 +#define BRW_BLENDFUNCTION_REVERSE_SUBTRACT 2 +#define BRW_BLENDFUNCTION_MIN 3 +#define BRW_BLENDFUNCTION_MAX 4 + +#define BRW_ALPHATEST_FORMAT_UNORM8 0 +#define BRW_ALPHATEST_FORMAT_FLOAT32 1 + +#define BRW_CHROMAKEY_KILL_ON_ANY_MATCH 0 +#define BRW_CHROMAKEY_REPLACE_BLACK 1 + +#define BRW_CLIP_API_OGL 0 +#define BRW_CLIP_API_DX 1 + +#define BRW_CLIPMODE_NORMAL 0 +#define BRW_CLIPMODE_CLIP_ALL 1 +#define BRW_CLIPMODE_CLIP_NON_REJECTED 2 +#define BRW_CLIPMODE_REJECT_ALL 3 +#define BRW_CLIPMODE_ACCEPT_ALL 4 + +#define BRW_CLIP_NDCSPACE 0 +#define BRW_CLIP_SCREENSPACE 1 + +#define BRW_COMPAREFUNCTION_ALWAYS 0 +#define BRW_COMPAREFUNCTION_NEVER 1 +#define BRW_COMPAREFUNCTION_LESS 2 +#define BRW_COMPAREFUNCTION_EQUAL 3 +#define BRW_COMPAREFUNCTION_LEQUAL 4 +#define BRW_COMPAREFUNCTION_GREATER 5 +#define BRW_COMPAREFUNCTION_NOTEQUAL 6 +#define BRW_COMPAREFUNCTION_GEQUAL 7 + +#define BRW_COVERAGE_PIXELS_HALF 0 +#define BRW_COVERAGE_PIXELS_1 1 +#define BRW_COVERAGE_PIXELS_2 2 +#define BRW_COVERAGE_PIXELS_4 3 + +#define BRW_CULLMODE_BOTH 0 +#define BRW_CULLMODE_NONE 1 +#define BRW_CULLMODE_FRONT 2 +#define BRW_CULLMODE_BACK 3 + +#define BRW_DEFAULTCOLOR_R8G8B8A8_UNORM 0 +#define BRW_DEFAULTCOLOR_R32G32B32A32_FLOAT 1 + +#define BRW_DEPTHFORMAT_D32_FLOAT_S8X24_UINT 0 +#define BRW_DEPTHFORMAT_D32_FLOAT 1 +#define BRW_DEPTHFORMAT_D24_UNORM_S8_UINT 2 +#define BRW_DEPTHFORMAT_D16_UNORM 5 + +#define BRW_FLOATING_POINT_IEEE_754 0 +#define BRW_FLOATING_POINT_NON_IEEE_754 1 + +#define BRW_FRONTWINDING_CW 0 +#define BRW_FRONTWINDING_CCW 1 + +#define BRW_SPRITE_POINT_ENABLE 16 + +#define BRW_INDEX_BYTE 0 +#define BRW_INDEX_WORD 1 +#define BRW_INDEX_DWORD 2 + +#define BRW_LOGICOPFUNCTION_CLEAR 0 +#define BRW_LOGICOPFUNCTION_NOR 1 +#define BRW_LOGICOPFUNCTION_AND_INVERTED 2 +#define BRW_LOGICOPFUNCTION_COPY_INVERTED 3 +#define BRW_LOGICOPFUNCTION_AND_REVERSE 4 +#define BRW_LOGICOPFUNCTION_INVERT 5 +#define BRW_LOGICOPFUNCTION_XOR 6 +#define BRW_LOGICOPFUNCTION_NAND 7 +#define BRW_LOGICOPFUNCTION_AND 8 +#define BRW_LOGICOPFUNCTION_EQUIV 9 +#define BRW_LOGICOPFUNCTION_NOOP 10 +#define BRW_LOGICOPFUNCTION_OR_INVERTED 11 +#define BRW_LOGICOPFUNCTION_COPY 12 +#define BRW_LOGICOPFUNCTION_OR_REVERSE 13 +#define BRW_LOGICOPFUNCTION_OR 14 +#define BRW_LOGICOPFUNCTION_SET 15 + +#define BRW_MAPFILTER_NEAREST 0x0 +#define BRW_MAPFILTER_LINEAR 0x1 +#define BRW_MAPFILTER_ANISOTROPIC 0x2 + +#define BRW_MIPFILTER_NONE 0 +#define BRW_MIPFILTER_NEAREST 1 +#define BRW_MIPFILTER_LINEAR 3 + +#define BRW_POLYGON_FRONT_FACING 0 +#define BRW_POLYGON_BACK_FACING 1 + +#define BRW_PREFILTER_ALWAYS 0x0 +#define BRW_PREFILTER_NEVER 0x1 +#define BRW_PREFILTER_LESS 0x2 +#define BRW_PREFILTER_EQUAL 0x3 +#define BRW_PREFILTER_LEQUAL 0x4 +#define BRW_PREFILTER_GREATER 0x5 +#define BRW_PREFILTER_NOTEQUAL 0x6 +#define BRW_PREFILTER_GEQUAL 0x7 + +#define BRW_PROVOKING_VERTEX_0 0 +#define BRW_PROVOKING_VERTEX_1 1 +#define BRW_PROVOKING_VERTEX_2 2 + +#define BRW_RASTRULE_UPPER_LEFT 0 +#define BRW_RASTRULE_UPPER_RIGHT 1 + +#define BRW_RENDERTARGET_CLAMPRANGE_UNORM 0 +#define BRW_RENDERTARGET_CLAMPRANGE_SNORM 1 +#define BRW_RENDERTARGET_CLAMPRANGE_FORMAT 2 + +#define BRW_STENCILOP_KEEP 0 +#define BRW_STENCILOP_ZERO 1 +#define BRW_STENCILOP_REPLACE 2 +#define BRW_STENCILOP_INCRSAT 3 +#define BRW_STENCILOP_DECRSAT 4 +#define BRW_STENCILOP_INCR 5 +#define BRW_STENCILOP_DECR 6 +#define BRW_STENCILOP_INVERT 7 + +#define BRW_SURFACE_MIPMAPLAYOUT_BELOW 0 +#define BRW_SURFACE_MIPMAPLAYOUT_RIGHT 1 + +#define BRW_SURFACEFORMAT_R32G32B32A32_FLOAT 0x000 +#define BRW_SURFACEFORMAT_R32G32B32A32_SINT 0x001 +#define BRW_SURFACEFORMAT_R32G32B32A32_UINT 0x002 +#define BRW_SURFACEFORMAT_R32G32B32A32_UNORM 0x003 +#define BRW_SURFACEFORMAT_R32G32B32A32_SNORM 0x004 +#define BRW_SURFACEFORMAT_R64G64_FLOAT 0x005 +#define BRW_SURFACEFORMAT_R32G32B32X32_FLOAT 0x006 +#define BRW_SURFACEFORMAT_R32G32B32A32_SSCALED 0x007 +#define BRW_SURFACEFORMAT_R32G32B32A32_USCALED 0x008 +#define BRW_SURFACEFORMAT_R32G32B32_FLOAT 0x040 +#define BRW_SURFACEFORMAT_R32G32B32_SINT 0x041 +#define BRW_SURFACEFORMAT_R32G32B32_UINT 0x042 +#define BRW_SURFACEFORMAT_R32G32B32_UNORM 0x043 +#define BRW_SURFACEFORMAT_R32G32B32_SNORM 0x044 +#define BRW_SURFACEFORMAT_R32G32B32_SSCALED 0x045 +#define BRW_SURFACEFORMAT_R32G32B32_USCALED 0x046 +#define BRW_SURFACEFORMAT_R16G16B16A16_UNORM 0x080 +#define BRW_SURFACEFORMAT_R16G16B16A16_SNORM 0x081 +#define BRW_SURFACEFORMAT_R16G16B16A16_SINT 0x082 +#define BRW_SURFACEFORMAT_R16G16B16A16_UINT 0x083 +#define BRW_SURFACEFORMAT_R16G16B16A16_FLOAT 0x084 +#define BRW_SURFACEFORMAT_R32G32_FLOAT 0x085 +#define BRW_SURFACEFORMAT_R32G32_SINT 0x086 +#define BRW_SURFACEFORMAT_R32G32_UINT 0x087 +#define BRW_SURFACEFORMAT_R32_FLOAT_X8X24_TYPELESS 0x088 +#define BRW_SURFACEFORMAT_X32_TYPELESS_G8X24_UINT 0x089 +#define BRW_SURFACEFORMAT_L32A32_FLOAT 0x08A +#define BRW_SURFACEFORMAT_R32G32_UNORM 0x08B +#define BRW_SURFACEFORMAT_R32G32_SNORM 0x08C +#define BRW_SURFACEFORMAT_R64_FLOAT 0x08D +#define BRW_SURFACEFORMAT_R16G16B16X16_UNORM 0x08E +#define BRW_SURFACEFORMAT_R16G16B16X16_FLOAT 0x08F +#define BRW_SURFACEFORMAT_A32X32_FLOAT 0x090 +#define BRW_SURFACEFORMAT_L32X32_FLOAT 0x091 +#define BRW_SURFACEFORMAT_I32X32_FLOAT 0x092 +#define BRW_SURFACEFORMAT_R16G16B16A16_SSCALED 0x093 +#define BRW_SURFACEFORMAT_R16G16B16A16_USCALED 0x094 +#define BRW_SURFACEFORMAT_R32G32_SSCALED 0x095 +#define BRW_SURFACEFORMAT_R32G32_USCALED 0x096 +#define BRW_SURFACEFORMAT_B8G8R8A8_UNORM 0x0C0 +#define BRW_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB 0x0C1 +#define BRW_SURFACEFORMAT_R10G10B10A2_UNORM 0x0C2 +#define BRW_SURFACEFORMAT_R10G10B10A2_UNORM_SRGB 0x0C3 +#define BRW_SURFACEFORMAT_R10G10B10A2_UINT 0x0C4 +#define BRW_SURFACEFORMAT_R10G10B10_SNORM_A2_UNORM 0x0C5 +#define BRW_SURFACEFORMAT_R8G8B8A8_UNORM 0x0C7 +#define BRW_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB 0x0C8 +#define BRW_SURFACEFORMAT_R8G8B8A8_SNORM 0x0C9 +#define BRW_SURFACEFORMAT_R8G8B8A8_SINT 0x0CA +#define BRW_SURFACEFORMAT_R8G8B8A8_UINT 0x0CB +#define BRW_SURFACEFORMAT_R16G16_UNORM 0x0CC +#define BRW_SURFACEFORMAT_R16G16_SNORM 0x0CD +#define BRW_SURFACEFORMAT_R16G16_SINT 0x0CE +#define BRW_SURFACEFORMAT_R16G16_UINT 0x0CF +#define BRW_SURFACEFORMAT_R16G16_FLOAT 0x0D0 +#define BRW_SURFACEFORMAT_B10G10R10A2_UNORM 0x0D1 +#define BRW_SURFACEFORMAT_B10G10R10A2_UNORM_SRGB 0x0D2 +#define BRW_SURFACEFORMAT_R11G11B10_FLOAT 0x0D3 +#define BRW_SURFACEFORMAT_R32_SINT 0x0D6 +#define BRW_SURFACEFORMAT_R32_UINT 0x0D7 +#define BRW_SURFACEFORMAT_R32_FLOAT 0x0D8 +#define BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS 0x0D9 +#define BRW_SURFACEFORMAT_X24_TYPELESS_G8_UINT 0x0DA +#define BRW_SURFACEFORMAT_L16A16_UNORM 0x0DF +#define BRW_SURFACEFORMAT_I24X8_UNORM 0x0E0 +#define BRW_SURFACEFORMAT_L24X8_UNORM 0x0E1 +#define BRW_SURFACEFORMAT_A24X8_UNORM 0x0E2 +#define BRW_SURFACEFORMAT_I32_FLOAT 0x0E3 +#define BRW_SURFACEFORMAT_L32_FLOAT 0x0E4 +#define BRW_SURFACEFORMAT_A32_FLOAT 0x0E5 +#define BRW_SURFACEFORMAT_B8G8R8X8_UNORM 0x0E9 +#define BRW_SURFACEFORMAT_B8G8R8X8_UNORM_SRGB 0x0EA +#define BRW_SURFACEFORMAT_R8G8B8X8_UNORM 0x0EB +#define BRW_SURFACEFORMAT_R8G8B8X8_UNORM_SRGB 0x0EC +#define BRW_SURFACEFORMAT_R9G9B9E5_SHAREDEXP 0x0ED +#define BRW_SURFACEFORMAT_B10G10R10X2_UNORM 0x0EE +#define BRW_SURFACEFORMAT_L16A16_FLOAT 0x0F0 +#define BRW_SURFACEFORMAT_R32_UNORM 0x0F1 +#define BRW_SURFACEFORMAT_R32_SNORM 0x0F2 +#define BRW_SURFACEFORMAT_R10G10B10X2_USCALED 0x0F3 +#define BRW_SURFACEFORMAT_R8G8B8A8_SSCALED 0x0F4 +#define BRW_SURFACEFORMAT_R8G8B8A8_USCALED 0x0F5 +#define BRW_SURFACEFORMAT_R16G16_SSCALED 0x0F6 +#define BRW_SURFACEFORMAT_R16G16_USCALED 0x0F7 +#define BRW_SURFACEFORMAT_R32_SSCALED 0x0F8 +#define BRW_SURFACEFORMAT_R32_USCALED 0x0F9 +#define BRW_SURFACEFORMAT_B5G6R5_UNORM 0x100 +#define BRW_SURFACEFORMAT_B5G6R5_UNORM_SRGB 0x101 +#define BRW_SURFACEFORMAT_B5G5R5A1_UNORM 0x102 +#define BRW_SURFACEFORMAT_B5G5R5A1_UNORM_SRGB 0x103 +#define BRW_SURFACEFORMAT_B4G4R4A4_UNORM 0x104 +#define BRW_SURFACEFORMAT_B4G4R4A4_UNORM_SRGB 0x105 +#define BRW_SURFACEFORMAT_R8G8_UNORM 0x106 +#define BRW_SURFACEFORMAT_R8G8_SNORM 0x107 +#define BRW_SURFACEFORMAT_R8G8_SINT 0x108 +#define BRW_SURFACEFORMAT_R8G8_UINT 0x109 +#define BRW_SURFACEFORMAT_R16_UNORM 0x10A +#define BRW_SURFACEFORMAT_R16_SNORM 0x10B +#define BRW_SURFACEFORMAT_R16_SINT 0x10C +#define BRW_SURFACEFORMAT_R16_UINT 0x10D +#define BRW_SURFACEFORMAT_R16_FLOAT 0x10E +#define BRW_SURFACEFORMAT_I16_UNORM 0x111 +#define BRW_SURFACEFORMAT_L16_UNORM 0x112 +#define BRW_SURFACEFORMAT_A16_UNORM 0x113 +#define BRW_SURFACEFORMAT_L8A8_UNORM 0x114 +#define BRW_SURFACEFORMAT_I16_FLOAT 0x115 +#define BRW_SURFACEFORMAT_L16_FLOAT 0x116 +#define BRW_SURFACEFORMAT_A16_FLOAT 0x117 +#define BRW_SURFACEFORMAT_R5G5_SNORM_B6_UNORM 0x119 +#define BRW_SURFACEFORMAT_B5G5R5X1_UNORM 0x11A +#define BRW_SURFACEFORMAT_B5G5R5X1_UNORM_SRGB 0x11B +#define BRW_SURFACEFORMAT_R8G8_SSCALED 0x11C +#define BRW_SURFACEFORMAT_R8G8_USCALED 0x11D +#define BRW_SURFACEFORMAT_R16_SSCALED 0x11E +#define BRW_SURFACEFORMAT_R16_USCALED 0x11F +#define BRW_SURFACEFORMAT_R8_UNORM 0x140 +#define BRW_SURFACEFORMAT_R8_SNORM 0x141 +#define BRW_SURFACEFORMAT_R8_SINT 0x142 +#define BRW_SURFACEFORMAT_R8_UINT 0x143 +#define BRW_SURFACEFORMAT_A8_UNORM 0x144 +#define BRW_SURFACEFORMAT_I8_UNORM 0x145 +#define BRW_SURFACEFORMAT_L8_UNORM 0x146 +#define BRW_SURFACEFORMAT_P4A4_UNORM 0x147 +#define BRW_SURFACEFORMAT_A4P4_UNORM 0x148 +#define BRW_SURFACEFORMAT_R8_SSCALED 0x149 +#define BRW_SURFACEFORMAT_R8_USCALED 0x14A +#define BRW_SURFACEFORMAT_R1_UINT 0x181 +#define BRW_SURFACEFORMAT_YCRCB_NORMAL 0x182 +#define BRW_SURFACEFORMAT_YCRCB_SWAPUVY 0x183 +#define BRW_SURFACEFORMAT_BC1_UNORM 0x186 +#define BRW_SURFACEFORMAT_BC2_UNORM 0x187 +#define BRW_SURFACEFORMAT_BC3_UNORM 0x188 +#define BRW_SURFACEFORMAT_BC4_UNORM 0x189 +#define BRW_SURFACEFORMAT_BC5_UNORM 0x18A +#define BRW_SURFACEFORMAT_BC1_UNORM_SRGB 0x18B +#define BRW_SURFACEFORMAT_BC2_UNORM_SRGB 0x18C +#define BRW_SURFACEFORMAT_BC3_UNORM_SRGB 0x18D +#define BRW_SURFACEFORMAT_MONO8 0x18E +#define BRW_SURFACEFORMAT_YCRCB_SWAPUV 0x18F +#define BRW_SURFACEFORMAT_YCRCB_SWAPY 0x190 +#define BRW_SURFACEFORMAT_DXT1_RGB 0x191 +#define BRW_SURFACEFORMAT_FXT1 0x192 +#define BRW_SURFACEFORMAT_R8G8B8_UNORM 0x193 +#define BRW_SURFACEFORMAT_R8G8B8_SNORM 0x194 +#define BRW_SURFACEFORMAT_R8G8B8_SSCALED 0x195 +#define BRW_SURFACEFORMAT_R8G8B8_USCALED 0x196 +#define BRW_SURFACEFORMAT_R64G64B64A64_FLOAT 0x197 +#define BRW_SURFACEFORMAT_R64G64B64_FLOAT 0x198 +#define BRW_SURFACEFORMAT_BC4_SNORM 0x199 +#define BRW_SURFACEFORMAT_BC5_SNORM 0x19A +#define BRW_SURFACEFORMAT_R16G16B16_UNORM 0x19C +#define BRW_SURFACEFORMAT_R16G16B16_SNORM 0x19D +#define BRW_SURFACEFORMAT_R16G16B16_SSCALED 0x19E +#define BRW_SURFACEFORMAT_R16G16B16_USCALED 0x19F + +#define BRW_SURFACERETURNFORMAT_FLOAT32 0 +#define BRW_SURFACERETURNFORMAT_S1 1 + +#define BRW_SURFACE_1D 0 +#define BRW_SURFACE_2D 1 +#define BRW_SURFACE_3D 2 +#define BRW_SURFACE_CUBE 3 +#define BRW_SURFACE_BUFFER 4 +#define BRW_SURFACE_NULL 7 + +#define BRW_TEXCOORDMODE_WRAP 0 +#define BRW_TEXCOORDMODE_MIRROR 1 +#define BRW_TEXCOORDMODE_CLAMP 2 +#define BRW_TEXCOORDMODE_CUBE 3 +#define BRW_TEXCOORDMODE_CLAMP_BORDER 4 +#define BRW_TEXCOORDMODE_MIRROR_ONCE 5 + +#define BRW_THREAD_PRIORITY_NORMAL 0 +#define BRW_THREAD_PRIORITY_HIGH 1 + +#define BRW_TILEWALK_XMAJOR 0 +#define BRW_TILEWALK_YMAJOR 1 + +#define BRW_VERTEX_SUBPIXEL_PRECISION_8BITS 0 +#define BRW_VERTEX_SUBPIXEL_PRECISION_4BITS 1 + +#define BRW_VERTEXBUFFER_ACCESS_VERTEXDATA 0 +#define BRW_VERTEXBUFFER_ACCESS_INSTANCEDATA 1 + +#define BRW_VFCOMPONENT_NOSTORE 0 +#define BRW_VFCOMPONENT_STORE_SRC 1 +#define BRW_VFCOMPONENT_STORE_0 2 +#define BRW_VFCOMPONENT_STORE_1_FLT 3 +#define BRW_VFCOMPONENT_STORE_1_INT 4 +#define BRW_VFCOMPONENT_STORE_VID 5 +#define BRW_VFCOMPONENT_STORE_IID 6 +#define BRW_VFCOMPONENT_STORE_PID 7 + + + +/* Execution Unit (EU) defines + */ + +#define BRW_ALIGN_1 0 +#define BRW_ALIGN_16 1 + +#define BRW_ADDRESS_DIRECT 0 +#define BRW_ADDRESS_REGISTER_INDIRECT_REGISTER 1 + +#define BRW_CHANNEL_X 0 +#define BRW_CHANNEL_Y 1 +#define BRW_CHANNEL_Z 2 +#define BRW_CHANNEL_W 3 + +#define BRW_COMPRESSION_NONE 0 +#define BRW_COMPRESSION_2NDHALF 1 +#define BRW_COMPRESSION_COMPRESSED 2 + +#define BRW_CONDITIONAL_NONE 0 +#define BRW_CONDITIONAL_Z 1 +#define BRW_CONDITIONAL_NZ 2 +#define BRW_CONDITIONAL_EQ 1 /* Z */ +#define BRW_CONDITIONAL_NEQ 2 /* NZ */ +#define BRW_CONDITIONAL_G 3 +#define BRW_CONDITIONAL_GE 4 +#define BRW_CONDITIONAL_L 5 +#define BRW_CONDITIONAL_LE 6 +#define BRW_CONDITIONAL_C 7 +#define BRW_CONDITIONAL_O 8 + +#define BRW_DEBUG_NONE 0 +#define BRW_DEBUG_BREAKPOINT 1 + +#define BRW_DEPENDENCY_NORMAL 0 +#define BRW_DEPENDENCY_NOTCLEARED 1 +#define BRW_DEPENDENCY_NOTCHECKED 2 +#define BRW_DEPENDENCY_DISABLE 3 + +#define BRW_EXECUTE_1 0 +#define BRW_EXECUTE_2 1 +#define BRW_EXECUTE_4 2 +#define BRW_EXECUTE_8 3 +#define BRW_EXECUTE_16 4 +#define BRW_EXECUTE_32 5 + +#define BRW_HORIZONTAL_STRIDE_0 0 +#define BRW_HORIZONTAL_STRIDE_1 1 +#define BRW_HORIZONTAL_STRIDE_2 2 +#define BRW_HORIZONTAL_STRIDE_4 3 + +#define BRW_INSTRUCTION_NORMAL 0 +#define BRW_INSTRUCTION_SATURATE 1 + +#define BRW_MASK_ENABLE 0 +#define BRW_MASK_DISABLE 1 + +#define BRW_OPCODE_MOV 1 +#define BRW_OPCODE_SEL 2 +#define BRW_OPCODE_NOT 4 +#define BRW_OPCODE_AND 5 +#define BRW_OPCODE_OR 6 +#define BRW_OPCODE_XOR 7 +#define BRW_OPCODE_SHR 8 +#define BRW_OPCODE_SHL 9 +#define BRW_OPCODE_RSR 10 +#define BRW_OPCODE_RSL 11 +#define BRW_OPCODE_ASR 12 +#define BRW_OPCODE_CMP 16 +#define BRW_OPCODE_JMPI 32 +#define BRW_OPCODE_IF 34 +#define BRW_OPCODE_IFF 35 +#define BRW_OPCODE_ELSE 36 +#define BRW_OPCODE_ENDIF 37 +#define BRW_OPCODE_DO 38 +#define BRW_OPCODE_WHILE 39 +#define BRW_OPCODE_BREAK 40 +#define BRW_OPCODE_CONTINUE 41 +#define BRW_OPCODE_HALT 42 +#define BRW_OPCODE_MSAVE 44 +#define BRW_OPCODE_MRESTORE 45 +#define BRW_OPCODE_PUSH 46 +#define BRW_OPCODE_POP 47 +#define BRW_OPCODE_WAIT 48 +#define BRW_OPCODE_SEND 49 +#define BRW_OPCODE_ADD 64 +#define BRW_OPCODE_MUL 65 +#define BRW_OPCODE_AVG 66 +#define BRW_OPCODE_FRC 67 +#define BRW_OPCODE_RNDU 68 +#define BRW_OPCODE_RNDD 69 +#define BRW_OPCODE_RNDE 70 +#define BRW_OPCODE_RNDZ 71 +#define BRW_OPCODE_MAC 72 +#define BRW_OPCODE_MACH 73 +#define BRW_OPCODE_LZD 74 +#define BRW_OPCODE_SAD2 80 +#define BRW_OPCODE_SADA2 81 +#define BRW_OPCODE_DP4 84 +#define BRW_OPCODE_DPH 85 +#define BRW_OPCODE_DP3 86 +#define BRW_OPCODE_DP2 87 +#define BRW_OPCODE_DPA2 88 +#define BRW_OPCODE_LINE 89 +#define BRW_OPCODE_NOP 126 + +#define BRW_PREDICATE_NONE 0 +#define BRW_PREDICATE_NORMAL 1 +#define BRW_PREDICATE_ALIGN1_ANYV 2 +#define BRW_PREDICATE_ALIGN1_ALLV 3 +#define BRW_PREDICATE_ALIGN1_ANY2H 4 +#define BRW_PREDICATE_ALIGN1_ALL2H 5 +#define BRW_PREDICATE_ALIGN1_ANY4H 6 +#define BRW_PREDICATE_ALIGN1_ALL4H 7 +#define BRW_PREDICATE_ALIGN1_ANY8H 8 +#define BRW_PREDICATE_ALIGN1_ALL8H 9 +#define BRW_PREDICATE_ALIGN1_ANY16H 10 +#define BRW_PREDICATE_ALIGN1_ALL16H 11 +#define BRW_PREDICATE_ALIGN16_REPLICATE_X 2 +#define BRW_PREDICATE_ALIGN16_REPLICATE_Y 3 +#define BRW_PREDICATE_ALIGN16_REPLICATE_Z 4 +#define BRW_PREDICATE_ALIGN16_REPLICATE_W 5 +#define BRW_PREDICATE_ALIGN16_ANY4H 6 +#define BRW_PREDICATE_ALIGN16_ALL4H 7 + +#define BRW_ARCHITECTURE_REGISTER_FILE 0 +#define BRW_GENERAL_REGISTER_FILE 1 +#define BRW_MESSAGE_REGISTER_FILE 2 +#define BRW_IMMEDIATE_VALUE 3 + +#define BRW_REGISTER_TYPE_UD 0 +#define BRW_REGISTER_TYPE_D 1 +#define BRW_REGISTER_TYPE_UW 2 +#define BRW_REGISTER_TYPE_W 3 +#define BRW_REGISTER_TYPE_UB 4 +#define BRW_REGISTER_TYPE_B 5 +#define BRW_REGISTER_TYPE_VF 5 /* packed float vector, immediates only? */ +#define BRW_REGISTER_TYPE_HF 6 +#define BRW_REGISTER_TYPE_V 6 /* packed int vector, immediates only, uword dest only */ +#define BRW_REGISTER_TYPE_F 7 + +#define BRW_ARF_NULL 0x00 +#define BRW_ARF_ADDRESS 0x10 +#define BRW_ARF_ACCUMULATOR 0x20 +#define BRW_ARF_FLAG 0x30 +#define BRW_ARF_MASK 0x40 +#define BRW_ARF_MASK_STACK 0x50 +#define BRW_ARF_MASK_STACK_DEPTH 0x60 +#define BRW_ARF_STATE 0x70 +#define BRW_ARF_CONTROL 0x80 +#define BRW_ARF_NOTIFICATION_COUNT 0x90 +#define BRW_ARF_IP 0xA0 + +#define BRW_AMASK 0 +#define BRW_IMASK 1 +#define BRW_LMASK 2 +#define BRW_CMASK 3 + + + +#define BRW_THREAD_NORMAL 0 +#define BRW_THREAD_ATOMIC 1 +#define BRW_THREAD_SWITCH 2 + +#define BRW_VERTICAL_STRIDE_0 0 +#define BRW_VERTICAL_STRIDE_1 1 +#define BRW_VERTICAL_STRIDE_2 2 +#define BRW_VERTICAL_STRIDE_4 3 +#define BRW_VERTICAL_STRIDE_8 4 +#define BRW_VERTICAL_STRIDE_16 5 +#define BRW_VERTICAL_STRIDE_32 6 +#define BRW_VERTICAL_STRIDE_64 7 +#define BRW_VERTICAL_STRIDE_128 8 +#define BRW_VERTICAL_STRIDE_256 9 +#define BRW_VERTICAL_STRIDE_ONE_DIMENSIONAL 0xF + +#define BRW_WIDTH_1 0 +#define BRW_WIDTH_2 1 +#define BRW_WIDTH_4 2 +#define BRW_WIDTH_8 3 +#define BRW_WIDTH_16 4 + +#define BRW_STATELESS_BUFFER_BOUNDARY_1K 0 +#define BRW_STATELESS_BUFFER_BOUNDARY_2K 1 +#define BRW_STATELESS_BUFFER_BOUNDARY_4K 2 +#define BRW_STATELESS_BUFFER_BOUNDARY_8K 3 +#define BRW_STATELESS_BUFFER_BOUNDARY_16K 4 +#define BRW_STATELESS_BUFFER_BOUNDARY_32K 5 +#define BRW_STATELESS_BUFFER_BOUNDARY_64K 6 +#define BRW_STATELESS_BUFFER_BOUNDARY_128K 7 +#define BRW_STATELESS_BUFFER_BOUNDARY_256K 8 +#define BRW_STATELESS_BUFFER_BOUNDARY_512K 9 +#define BRW_STATELESS_BUFFER_BOUNDARY_1M 10 +#define BRW_STATELESS_BUFFER_BOUNDARY_2M 11 + +#define BRW_POLYGON_FACING_FRONT 0 +#define BRW_POLYGON_FACING_BACK 1 + +#define BRW_MESSAGE_TARGET_NULL 0 +#define BRW_MESSAGE_TARGET_MATH 1 +#define BRW_MESSAGE_TARGET_SAMPLER 2 +#define BRW_MESSAGE_TARGET_GATEWAY 3 +#define BRW_MESSAGE_TARGET_DATAPORT_READ 4 +#define BRW_MESSAGE_TARGET_DATAPORT_WRITE 5 +#define BRW_MESSAGE_TARGET_URB 6 +#define BRW_MESSAGE_TARGET_THREAD_SPAWNER 7 + +#define BRW_SAMPLER_RETURN_FORMAT_FLOAT32 0 +#define BRW_SAMPLER_RETURN_FORMAT_UINT32 2 +#define BRW_SAMPLER_RETURN_FORMAT_SINT32 3 + +#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE 0 +#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE 0 +#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS 0 +#define BRW_SAMPLER_MESSAGE_SIMD8_KILLPIX 1 +#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD 1 +#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD 1 +#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS 2 +#define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS 2 +#define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE 0 +#define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE 2 +#define BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO 2 +#define BRW_SAMPLER_MESSAGE_SIMD8_RESINFO 2 +#define BRW_SAMPLER_MESSAGE_SIMD16_RESINFO 2 +#define BRW_SAMPLER_MESSAGE_SIMD4X2_LD 3 +#define BRW_SAMPLER_MESSAGE_SIMD8_LD 3 +#define BRW_SAMPLER_MESSAGE_SIMD16_LD 3 + +#define BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW 0 +#define BRW_DATAPORT_OWORD_BLOCK_1_OWORDHIGH 1 +#define BRW_DATAPORT_OWORD_BLOCK_2_OWORDS 2 +#define BRW_DATAPORT_OWORD_BLOCK_4_OWORDS 3 +#define BRW_DATAPORT_OWORD_BLOCK_8_OWORDS 4 + +#define BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD 0 +#define BRW_DATAPORT_OWORD_DUAL_BLOCK_4OWORDS 2 + +#define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_8DWORDS 2 +#define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_16DWORDS 3 + +#define BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ 0 +#define BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 1 +#define BRW_DATAPORT_READ_MESSAGE_DWORD_BLOCK_READ 2 +#define BRW_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 3 + +#define BRW_DATAPORT_READ_TARGET_DATA_CACHE 0 +#define BRW_DATAPORT_READ_TARGET_RENDER_CACHE 1 +#define BRW_DATAPORT_READ_TARGET_SAMPLER_CACHE 2 + +#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE 0 +#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED 1 +#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01 2 +#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN23 3 +#define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01 4 + +#define BRW_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 0 +#define BRW_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE 1 +#define BRW_DATAPORT_WRITE_MESSAGE_DWORD_BLOCK_WRITE 2 +#define BRW_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE 3 +#define BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE 4 +#define BRW_DATAPORT_WRITE_MESSAGE_STREAMED_VERTEX_BUFFER_WRITE 5 +#define BRW_DATAPORT_WRITE_MESSAGE_FLUSH_RENDER_CACHE 7 + +#define BRW_MATH_FUNCTION_INV 1 +#define BRW_MATH_FUNCTION_LOG 2 +#define BRW_MATH_FUNCTION_EXP 3 +#define BRW_MATH_FUNCTION_SQRT 4 +#define BRW_MATH_FUNCTION_RSQ 5 +#define BRW_MATH_FUNCTION_SIN 6 /* was 7 */ +#define BRW_MATH_FUNCTION_COS 7 /* was 8 */ +#define BRW_MATH_FUNCTION_SINCOS 8 /* was 6 */ +#define BRW_MATH_FUNCTION_TAN 9 +#define BRW_MATH_FUNCTION_POW 10 +#define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER 11 +#define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT 12 +#define BRW_MATH_FUNCTION_INT_DIV_REMAINDER 13 + +#define BRW_MATH_INTEGER_UNSIGNED 0 +#define BRW_MATH_INTEGER_SIGNED 1 + +#define BRW_MATH_PRECISION_FULL 0 +#define BRW_MATH_PRECISION_PARTIAL 1 + +#define BRW_MATH_SATURATE_NONE 0 +#define BRW_MATH_SATURATE_SATURATE 1 + +#define BRW_MATH_DATA_VECTOR 0 +#define BRW_MATH_DATA_SCALAR 1 + +#define BRW_URB_OPCODE_WRITE 0 + +#define BRW_URB_SWIZZLE_NONE 0 +#define BRW_URB_SWIZZLE_INTERLEAVE 1 +#define BRW_URB_SWIZZLE_TRANSPOSE 2 + +#define BRW_SCRATCH_SPACE_SIZE_1K 0 +#define BRW_SCRATCH_SPACE_SIZE_2K 1 +#define BRW_SCRATCH_SPACE_SIZE_4K 2 +#define BRW_SCRATCH_SPACE_SIZE_8K 3 +#define BRW_SCRATCH_SPACE_SIZE_16K 4 +#define BRW_SCRATCH_SPACE_SIZE_32K 5 +#define BRW_SCRATCH_SPACE_SIZE_64K 6 +#define BRW_SCRATCH_SPACE_SIZE_128K 7 +#define BRW_SCRATCH_SPACE_SIZE_256K 8 +#define BRW_SCRATCH_SPACE_SIZE_512K 9 +#define BRW_SCRATCH_SPACE_SIZE_1M 10 +#define BRW_SCRATCH_SPACE_SIZE_2M 11 + + + + +#define CMD_URB_FENCE 0x6000 +#define CMD_CONST_BUFFER_STATE 0x6001 +#define CMD_CONST_BUFFER 0x6002 + +#define CMD_STATE_BASE_ADDRESS 0x6101 +#define CMD_STATE_INSN_POINTER 0x6102 +#define CMD_PIPELINE_SELECT 0x6104 + +#define CMD_PIPELINED_STATE_POINTERS 0x7800 +#define CMD_BINDING_TABLE_PTRS 0x7801 +#define CMD_VERTEX_BUFFER 0x7808 +#define CMD_VERTEX_ELEMENT 0x7809 +#define CMD_INDEX_BUFFER 0x780a +#define CMD_VF_STATISTICS 0x780b + +#define CMD_DRAW_RECT 0x7900 +#define CMD_BLEND_CONSTANT_COLOR 0x7901 +#define CMD_CHROMA_KEY 0x7904 +#define CMD_DEPTH_BUFFER 0x7905 +#define CMD_POLY_STIPPLE_OFFSET 0x7906 +#define CMD_POLY_STIPPLE_PATTERN 0x7907 +#define CMD_LINE_STIPPLE_PATTERN 0x7908 +#define CMD_GLOBAL_DEPTH_OFFSET_CLAMP 0x7909 + +#define CMD_PIPE_CONTROL 0x7a00 + +#define CMD_3D_PRIM 0x7b00 + +#define CMD_MI_FLUSH 0x0200 + + +/* Various values from the R0 vertex header: + */ +#define R02_PRIM_END 0x1 +#define R02_PRIM_START 0x2 + + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_draw.c b/src/gallium/drivers/i965simple/brw_draw.c new file mode 100644 index 0000000000..7598e3dc8a --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_draw.c @@ -0,0 +1,239 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include <stdlib.h> + +#include "brw_batch.h" +#include "brw_draw.h" +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_state.h" + +#include "pipe/p_context.h" +#include "pipe/p_winsys.h" + +static unsigned hw_prim[PIPE_PRIM_POLYGON+1] = { + _3DPRIM_POINTLIST, + _3DPRIM_LINELIST, + _3DPRIM_LINELOOP, + _3DPRIM_LINESTRIP, + _3DPRIM_TRILIST, + _3DPRIM_TRISTRIP, + _3DPRIM_TRIFAN, + _3DPRIM_QUADLIST, + _3DPRIM_QUADSTRIP, + _3DPRIM_POLYGON +}; + + +static const int reduced_prim[PIPE_PRIM_POLYGON+1] = { + PIPE_PRIM_POINTS, + PIPE_PRIM_LINES, + PIPE_PRIM_LINES, + PIPE_PRIM_LINES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES +}; + + +/* When the primitive changes, set a state bit and re-validate. Not + * the nicest and would rather deal with this by having all the + * programs be immune to the active primitive (ie. cope with all + * possibilities). That may not be realistic however. + */ +static void brw_set_prim(struct brw_context *brw, int prim) +{ + PRINT("PRIM: %d\n", prim); + + /* Slight optimization to avoid the GS program when not needed: + */ + if (prim == PIPE_PRIM_QUAD_STRIP && + brw->attribs.Raster->flatshade && + brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_FILL && + brw->attribs.Raster->fill_ccw == PIPE_POLYGON_MODE_FILL) + prim = PIPE_PRIM_TRIANGLE_STRIP; + + if (prim != brw->primitive) { + brw->primitive = prim; + brw->state.dirty.brw |= BRW_NEW_PRIMITIVE; + + if (reduced_prim[prim] != brw->reduced_primitive) { + brw->reduced_primitive = reduced_prim[prim]; + brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE; + } + + brw_validate_state(brw); + } + +} + + +static unsigned trim(int prim, unsigned length) +{ + if (prim == PIPE_PRIM_QUAD_STRIP) + return length > 3 ? (length - length % 2) : 0; + else if (prim == PIPE_PRIM_QUADS) + return length - length % 4; + else + return length; +} + + + +static boolean brw_emit_prim( struct brw_context *brw, + boolean indexed, + unsigned start, + unsigned count ) + +{ + struct brw_3d_primitive prim_packet; + + if (BRW_DEBUG & DEBUG_PRIMS) + PRINT("PRIM: %d %d %d\n", brw->primitive, start, count); + + prim_packet.header.opcode = CMD_3D_PRIM; + prim_packet.header.length = sizeof(prim_packet)/4 - 2; + prim_packet.header.pad = 0; + prim_packet.header.topology = hw_prim[brw->primitive]; + prim_packet.header.indexed = indexed; + + prim_packet.verts_per_instance = trim(brw->primitive, count); + prim_packet.start_vert_location = start; + prim_packet.instance_count = 1; + prim_packet.start_instance_location = 0; + prim_packet.base_vert_location = 0; + + if (prim_packet.verts_per_instance == 0) + return TRUE; + + return brw_batchbuffer_data( brw->winsys, + &prim_packet, + sizeof(prim_packet) ); +} + + +/* May fail if out of video memory for texture or vbo upload, or on + * fallback conditions. + */ +static boolean brw_try_draw_elements( struct pipe_context *pipe, + struct pipe_buffer *index_buffer, + unsigned index_size, + unsigned mode, + unsigned start, + unsigned count ) +{ + struct brw_context *brw = brw_context(pipe); + + /* Set the first primitive ahead of validate_state: + */ + brw_set_prim(brw, mode); + + /* Upload index, vertex data: + */ + if (index_buffer && + !brw_upload_indices( brw, index_buffer, index_size, start, count )) + return FALSE; + + if (!brw_upload_vertex_buffers(brw)) + return FALSE; + + if (!brw_upload_vertex_elements( brw )) + return FALSE; + + /* XXX: Need to separate validate and upload of state. + */ + if (brw->state.dirty.brw) + brw_validate_state( brw ); + + if (!brw_emit_prim(brw, index_buffer != NULL, + start, count)) + return FALSE; + + return TRUE; +} + + + +static boolean brw_draw_elements( struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, + unsigned start, + unsigned count ) +{ + if (!brw_try_draw_elements( pipe, + indexBuffer, + indexSize, + mode, start, count )) + { + /* flush ? */ + + if (!brw_try_draw_elements( pipe, + indexBuffer, + indexSize, + mode, start, + count )) { + assert(0); + return FALSE; + } + } + + return TRUE; +} + + + +static boolean brw_draw_arrays( struct pipe_context *pipe, + unsigned mode, + unsigned start, + unsigned count ) +{ + if (!brw_try_draw_elements( pipe, NULL, 0, mode, start, count )) { + /* flush ? */ + + if (!brw_try_draw_elements( pipe, NULL, 0, mode, start, count )) { + assert(0); + return FALSE; + } + } + + return TRUE; +} + + + +void brw_init_draw_functions( struct brw_context *brw ) +{ + brw->pipe.draw_arrays = brw_draw_arrays; + brw->pipe.draw_elements = brw_draw_elements; +} + + diff --git a/src/gallium/drivers/i965simple/brw_draw.h b/src/gallium/drivers/i965simple/brw_draw.h new file mode 100644 index 0000000000..62fe0d5d0e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_draw.h @@ -0,0 +1,55 @@ + /************************************************************************** + * + * Copyright 2005 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef BRW_DRAW_H +#define BRW_DRAW_H + +#include "pipe/p_context.h" + +struct brw_context; + + + +void brw_init_draw_functions( struct brw_context *brw ); + + +boolean brw_upload_vertices( struct brw_context *brw, + unsigned min_index, + unsigned max_index ); + +boolean brw_upload_indices(struct brw_context *brw, + const struct pipe_buffer *index_buffer, + int ib_size, int start, int count); + +boolean brw_upload_vertex_buffers( struct brw_context *brw ); +boolean brw_upload_vertex_elements( struct brw_context *brw ); + +unsigned brw_translate_surface_format( unsigned id ); + + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_draw_upload.c b/src/gallium/drivers/i965simple/brw_draw_upload.c new file mode 100644 index 0000000000..aa85d93866 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_draw_upload.c @@ -0,0 +1,299 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include <stdlib.h> + +#include "brw_batch.h" +#include "brw_draw.h" +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_state.h" + +struct brw_array_state { + union header_union header; + + struct { + union { + struct { + unsigned pitch:11; + unsigned pad:15; + unsigned access_type:1; + unsigned vb_index:5; + } bits; + unsigned dword; + } vb0; + + struct pipe_buffer *buffer; + unsigned offset; + + unsigned max_index; + unsigned instance_data_step_rate; + + } vb[BRW_VBP_MAX]; +}; + + + +unsigned brw_translate_surface_format( unsigned id ) +{ + switch (id) { + case PIPE_FORMAT_R64_FLOAT: + return BRW_SURFACEFORMAT_R64_FLOAT; + case PIPE_FORMAT_R64G64_FLOAT: + return BRW_SURFACEFORMAT_R64G64_FLOAT; + case PIPE_FORMAT_R64G64B64_FLOAT: + return BRW_SURFACEFORMAT_R64G64B64_FLOAT; + case PIPE_FORMAT_R64G64B64A64_FLOAT: + return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT; + + case PIPE_FORMAT_R32_FLOAT: + return BRW_SURFACEFORMAT_R32_FLOAT; + case PIPE_FORMAT_R32G32_FLOAT: + return BRW_SURFACEFORMAT_R32G32_FLOAT; + case PIPE_FORMAT_R32G32B32_FLOAT: + return BRW_SURFACEFORMAT_R32G32B32_FLOAT; + case PIPE_FORMAT_R32G32B32A32_FLOAT: + return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT; + + case PIPE_FORMAT_R32_UNORM: + return BRW_SURFACEFORMAT_R32_UNORM; + case PIPE_FORMAT_R32G32_UNORM: + return BRW_SURFACEFORMAT_R32G32_UNORM; + case PIPE_FORMAT_R32G32B32_UNORM: + return BRW_SURFACEFORMAT_R32G32B32_UNORM; + case PIPE_FORMAT_R32G32B32A32_UNORM: + return BRW_SURFACEFORMAT_R32G32B32A32_UNORM; + + case PIPE_FORMAT_R32_USCALED: + return BRW_SURFACEFORMAT_R32_USCALED; + case PIPE_FORMAT_R32G32_USCALED: + return BRW_SURFACEFORMAT_R32G32_USCALED; + case PIPE_FORMAT_R32G32B32_USCALED: + return BRW_SURFACEFORMAT_R32G32B32_USCALED; + case PIPE_FORMAT_R32G32B32A32_USCALED: + return BRW_SURFACEFORMAT_R32G32B32A32_USCALED; + + case PIPE_FORMAT_R32_SNORM: + return BRW_SURFACEFORMAT_R32_SNORM; + case PIPE_FORMAT_R32G32_SNORM: + return BRW_SURFACEFORMAT_R32G32_SNORM; + case PIPE_FORMAT_R32G32B32_SNORM: + return BRW_SURFACEFORMAT_R32G32B32_SNORM; + case PIPE_FORMAT_R32G32B32A32_SNORM: + return BRW_SURFACEFORMAT_R32G32B32A32_SNORM; + + case PIPE_FORMAT_R32_SSCALED: + return BRW_SURFACEFORMAT_R32_SSCALED; + case PIPE_FORMAT_R32G32_SSCALED: + return BRW_SURFACEFORMAT_R32G32_SSCALED; + case PIPE_FORMAT_R32G32B32_SSCALED: + return BRW_SURFACEFORMAT_R32G32B32_SSCALED; + case PIPE_FORMAT_R32G32B32A32_SSCALED: + return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED; + + case PIPE_FORMAT_R16_UNORM: + return BRW_SURFACEFORMAT_R16_UNORM; + case PIPE_FORMAT_R16G16_UNORM: + return BRW_SURFACEFORMAT_R16G16_UNORM; + case PIPE_FORMAT_R16G16B16_UNORM: + return BRW_SURFACEFORMAT_R16G16B16_UNORM; + case PIPE_FORMAT_R16G16B16A16_UNORM: + return BRW_SURFACEFORMAT_R16G16B16A16_UNORM; + + case PIPE_FORMAT_R16_USCALED: + return BRW_SURFACEFORMAT_R16_USCALED; + case PIPE_FORMAT_R16G16_USCALED: + return BRW_SURFACEFORMAT_R16G16_USCALED; + case PIPE_FORMAT_R16G16B16_USCALED: + return BRW_SURFACEFORMAT_R16G16B16_USCALED; + case PIPE_FORMAT_R16G16B16A16_USCALED: + return BRW_SURFACEFORMAT_R16G16B16A16_USCALED; + + case PIPE_FORMAT_R16_SNORM: + return BRW_SURFACEFORMAT_R16_SNORM; + case PIPE_FORMAT_R16G16_SNORM: + return BRW_SURFACEFORMAT_R16G16_SNORM; + case PIPE_FORMAT_R16G16B16_SNORM: + return BRW_SURFACEFORMAT_R16G16B16_SNORM; + case PIPE_FORMAT_R16G16B16A16_SNORM: + return BRW_SURFACEFORMAT_R16G16B16A16_SNORM; + + case PIPE_FORMAT_R16_SSCALED: + return BRW_SURFACEFORMAT_R16_SSCALED; + case PIPE_FORMAT_R16G16_SSCALED: + return BRW_SURFACEFORMAT_R16G16_SSCALED; + case PIPE_FORMAT_R16G16B16_SSCALED: + return BRW_SURFACEFORMAT_R16G16B16_SSCALED; + case PIPE_FORMAT_R16G16B16A16_SSCALED: + return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED; + + case PIPE_FORMAT_R8_UNORM: + return BRW_SURFACEFORMAT_R8_UNORM; + case PIPE_FORMAT_R8G8_UNORM: + return BRW_SURFACEFORMAT_R8G8_UNORM; + case PIPE_FORMAT_R8G8B8_UNORM: + return BRW_SURFACEFORMAT_R8G8B8_UNORM; + case PIPE_FORMAT_R8G8B8A8_UNORM: + return BRW_SURFACEFORMAT_R8G8B8A8_UNORM; + + case PIPE_FORMAT_R8_USCALED: + return BRW_SURFACEFORMAT_R8_USCALED; + case PIPE_FORMAT_R8G8_USCALED: + return BRW_SURFACEFORMAT_R8G8_USCALED; + case PIPE_FORMAT_R8G8B8_USCALED: + return BRW_SURFACEFORMAT_R8G8B8_USCALED; + case PIPE_FORMAT_R8G8B8A8_USCALED: + return BRW_SURFACEFORMAT_R8G8B8A8_USCALED; + + case PIPE_FORMAT_R8_SNORM: + return BRW_SURFACEFORMAT_R8_SNORM; + case PIPE_FORMAT_R8G8_SNORM: + return BRW_SURFACEFORMAT_R8G8_SNORM; + case PIPE_FORMAT_R8G8B8_SNORM: + return BRW_SURFACEFORMAT_R8G8B8_SNORM; + case PIPE_FORMAT_R8G8B8A8_SNORM: + return BRW_SURFACEFORMAT_R8G8B8A8_SNORM; + + case PIPE_FORMAT_R8_SSCALED: + return BRW_SURFACEFORMAT_R8_SSCALED; + case PIPE_FORMAT_R8G8_SSCALED: + return BRW_SURFACEFORMAT_R8G8_SSCALED; + case PIPE_FORMAT_R8G8B8_SSCALED: + return BRW_SURFACEFORMAT_R8G8B8_SSCALED; + case PIPE_FORMAT_R8G8B8A8_SSCALED: + return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED; + + default: + assert(0); + return 0; + } +} + +static unsigned get_index_type(int type) +{ + switch (type) { + case 1: return BRW_INDEX_BYTE; + case 2: return BRW_INDEX_WORD; + case 4: return BRW_INDEX_DWORD; + default: assert(0); return 0; + } +} + + +boolean brw_upload_vertex_buffers( struct brw_context *brw ) +{ + struct brw_array_state vbp; + unsigned nr_enabled = 0; + unsigned i; + + memset(&vbp, 0, sizeof(vbp)); + + /* This is a hardware limit: + */ + + for (i = 0; i < BRW_VEP_MAX; i++) + { + if (brw->vb.vbo_array[i] == NULL) { + nr_enabled = i; + break; + } + + vbp.vb[i].vb0.bits.pitch = brw->vb.vbo_array[i]->pitch; + vbp.vb[i].vb0.bits.pad = 0; + vbp.vb[i].vb0.bits.access_type = BRW_VERTEXBUFFER_ACCESS_VERTEXDATA; + vbp.vb[i].vb0.bits.vb_index = i; + vbp.vb[i].offset = brw->vb.vbo_array[i]->buffer_offset; + vbp.vb[i].buffer = brw->vb.vbo_array[i]->buffer; + vbp.vb[i].max_index = brw->vb.vbo_array[i]->max_index; + } + + + vbp.header.bits.length = (1 + nr_enabled * 4) - 2; + vbp.header.bits.opcode = CMD_VERTEX_BUFFER; + + BEGIN_BATCH(vbp.header.bits.length+2, 0); + OUT_BATCH( vbp.header.dword ); + + for (i = 0; i < nr_enabled; i++) { + OUT_BATCH( vbp.vb[i].vb0.dword ); + OUT_RELOC( vbp.vb[i].buffer, PIPE_BUFFER_USAGE_GPU_READ, + vbp.vb[i].offset); + OUT_BATCH( vbp.vb[i].max_index ); + OUT_BATCH( vbp.vb[i].instance_data_step_rate ); + } + ADVANCE_BATCH(); + return TRUE; +} + + + +boolean brw_upload_vertex_elements( struct brw_context *brw ) +{ + struct brw_vertex_element_packet vep; + + unsigned i; + unsigned nr_enabled = brw->attribs.VertexProgram->program.num_inputs; + + memset(&vep, 0, sizeof(vep)); + + for (i = 0; i < nr_enabled; i++) + vep.ve[i] = brw->vb.inputs[i]; + + + vep.header.length = (1 + nr_enabled * sizeof(vep.ve[0])/4) - 2; + vep.header.opcode = CMD_VERTEX_ELEMENT; + brw_cached_batch_struct(brw, &vep, 4 + nr_enabled * sizeof(vep.ve[0])); + + return TRUE; +} + +boolean brw_upload_indices( struct brw_context *brw, + const struct pipe_buffer *index_buffer, + int ib_size, int start, int count) +{ + /* Emit the indexbuffer packet: + */ + { + struct brw_indexbuffer ib; + + memset(&ib, 0, sizeof(ib)); + + ib.header.bits.opcode = CMD_INDEX_BUFFER; + ib.header.bits.length = sizeof(ib)/4 - 2; + ib.header.bits.index_format = get_index_type(ib_size); + ib.header.bits.cut_index_enable = 0; + + + BEGIN_BATCH(4, 0); + OUT_BATCH( ib.header.dword ); + OUT_RELOC( index_buffer, PIPE_BUFFER_USAGE_GPU_READ, start); + OUT_RELOC( index_buffer, PIPE_BUFFER_USAGE_GPU_READ, start + count); + OUT_BATCH( 0 ); + ADVANCE_BATCH(); + } + return TRUE; +} diff --git a/src/gallium/drivers/i965simple/brw_eu.c b/src/gallium/drivers/i965simple/brw_eu.c new file mode 100644 index 0000000000..e2002d1821 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_eu.c @@ -0,0 +1,130 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_eu.h" + + + +/* How does predicate control work when execution_size != 8? Do I + * need to test/set for 0xffff when execution_size is 16? + */ +void brw_set_predicate_control_flag_value( struct brw_compile *p, unsigned value ) +{ + p->current->header.predicate_control = BRW_PREDICATE_NONE; + + if (value != 0xff) { + if (value != p->flag_value) { + brw_push_insn_state(p); + brw_MOV(p, brw_flag_reg(), brw_imm_uw(value)); + p->flag_value = value; + brw_pop_insn_state(p); + } + + p->current->header.predicate_control = BRW_PREDICATE_NORMAL; + } +} + +void brw_set_predicate_control( struct brw_compile *p, unsigned pc ) +{ + p->current->header.predicate_control = pc; +} + +void brw_set_conditionalmod( struct brw_compile *p, unsigned conditional ) +{ + p->current->header.destreg__conditonalmod = conditional; +} + +void brw_set_access_mode( struct brw_compile *p, unsigned access_mode ) +{ + p->current->header.access_mode = access_mode; +} + +void brw_set_compression_control( struct brw_compile *p, boolean compression_control ) +{ + p->current->header.compression_control = compression_control; +} + +void brw_set_mask_control( struct brw_compile *p, unsigned value ) +{ + p->current->header.mask_control = value; +} + +void brw_set_saturate( struct brw_compile *p, unsigned value ) +{ + p->current->header.saturate = value; +} + +void brw_push_insn_state( struct brw_compile *p ) +{ + assert(p->current != &p->stack[BRW_EU_MAX_INSN_STACK-1]); + memcpy(p->current+1, p->current, sizeof(struct brw_instruction)); + p->current++; +} + +void brw_pop_insn_state( struct brw_compile *p ) +{ + assert(p->current != p->stack); + p->current--; +} + + +/*********************************************************************** + */ +void brw_init_compile( struct brw_compile *p ) +{ + p->nr_insn = 0; + p->current = p->stack; + memset(p->current, 0, sizeof(p->current[0])); + + /* Some defaults? + */ + brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */ + brw_set_saturate(p, 0); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_set_predicate_control_flag_value(p, 0xff); +} + + +const unsigned *brw_get_program( struct brw_compile *p, + unsigned *sz ) +{ + unsigned i; + + for (i = 0; i < 8; i++) + brw_NOP(p); + + *sz = p->nr_insn * sizeof(struct brw_instruction); + return (const unsigned *)p->store; +} + diff --git a/src/gallium/drivers/i965simple/brw_eu.h b/src/gallium/drivers/i965simple/brw_eu.h new file mode 100644 index 0000000000..23151ae9ed --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_eu.h @@ -0,0 +1,888 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_EU_H +#define BRW_EU_H + +#include "brw_structs.h" +#include "brw_defines.h" + +#include "pipe/p_compiler.h" +#include "pipe/p_shader_tokens.h" + +#define BRW_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<2) | ((c)<<4) | ((d)<<6)) +#define BRW_GET_SWZ(swz, idx) (((swz) >> ((idx)*2)) & 0x3) + +#define BRW_SWIZZLE_NOOP BRW_SWIZZLE4(0,1,2,3) +#define BRW_SWIZZLE_XYZW BRW_SWIZZLE4(0,1,2,3) +#define BRW_SWIZZLE_XXXX BRW_SWIZZLE4(0,0,0,0) +#define BRW_SWIZZLE_XYXY BRW_SWIZZLE4(0,1,0,1) + + +#define REG_SIZE (8*4) + + +/* These aren't hardware structs, just something useful for us to pass around: + * + * Align1 operation has a lot of control over input ranges. Used in + * WM programs to implement shaders decomposed into "channel serial" + * or "structure of array" form: + */ +struct brw_reg +{ + unsigned type:4; + unsigned file:2; + unsigned nr:8; + unsigned subnr:5; /* :1 in align16 */ + unsigned negate:1; /* source only */ + unsigned abs:1; /* source only */ + unsigned vstride:4; /* source only */ + unsigned width:3; /* src only, align1 only */ + unsigned hstride:2; /* src only, align1 only */ + unsigned address_mode:1; /* relative addressing, hopefully! */ + unsigned pad0:1; + + union { + struct { + unsigned swizzle:8; /* src only, align16 only */ + unsigned writemask:4; /* dest only, align16 only */ + int indirect_offset:10; /* relative addressing offset */ + unsigned pad1:10; /* two dwords total */ + } bits; + + float f; + int d; + unsigned ud; + } dw1; +}; + + +struct brw_indirect { + unsigned addr_subnr:4; + int addr_offset:10; + unsigned pad:18; +}; + + +#define BRW_EU_MAX_INSN_STACK 5 +#define BRW_EU_MAX_INSN 1200 + +struct brw_compile { + struct brw_instruction store[BRW_EU_MAX_INSN]; + unsigned nr_insn; + + /* Allow clients to push/pop instruction state: + */ + struct brw_instruction stack[BRW_EU_MAX_INSN_STACK]; + struct brw_instruction *current; + + unsigned flag_value; + boolean single_program_flow; +}; + + + +static __inline int type_sz( unsigned type ) +{ + switch( type ) { + case BRW_REGISTER_TYPE_UD: + case BRW_REGISTER_TYPE_D: + case BRW_REGISTER_TYPE_F: + return 4; + case BRW_REGISTER_TYPE_HF: + case BRW_REGISTER_TYPE_UW: + case BRW_REGISTER_TYPE_W: + return 2; + case BRW_REGISTER_TYPE_UB: + case BRW_REGISTER_TYPE_B: + return 1; + default: + return 0; + } +} + +static __inline struct brw_reg brw_reg( unsigned file, + unsigned nr, + unsigned subnr, + unsigned type, + unsigned vstride, + unsigned width, + unsigned hstride, + unsigned swizzle, + unsigned writemask) +{ + + struct brw_reg reg; + reg.type = type; + reg.file = file; + reg.nr = nr; + reg.subnr = subnr * type_sz(type); + reg.negate = 0; + reg.abs = 0; + reg.vstride = vstride; + reg.width = width; + reg.hstride = hstride; + reg.address_mode = BRW_ADDRESS_DIRECT; + reg.pad0 = 0; + + /* Could do better: If the reg is r5.3<0;1,0>, we probably want to + * set swizzle and writemask to W, as the lower bits of subnr will + * be lost when converted to align16. This is probably too much to + * keep track of as you'd want it adjusted by suboffset(), etc. + * Perhaps fix up when converting to align16? + */ + reg.dw1.bits.swizzle = swizzle; + reg.dw1.bits.writemask = writemask; + reg.dw1.bits.indirect_offset = 0; + reg.dw1.bits.pad1 = 0; + return reg; +} + +static __inline struct brw_reg brw_vec16_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return brw_reg(file, + nr, + subnr, + BRW_REGISTER_TYPE_F, + BRW_VERTICAL_STRIDE_16, + BRW_WIDTH_16, + BRW_HORIZONTAL_STRIDE_1, + BRW_SWIZZLE_XYZW, + TGSI_WRITEMASK_XYZW); +} + +static __inline struct brw_reg brw_vec8_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return brw_reg(file, + nr, + subnr, + BRW_REGISTER_TYPE_F, + BRW_VERTICAL_STRIDE_8, + BRW_WIDTH_8, + BRW_HORIZONTAL_STRIDE_1, + BRW_SWIZZLE_XYZW, + TGSI_WRITEMASK_XYZW); +} + + +static __inline struct brw_reg brw_vec4_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return brw_reg(file, + nr, + subnr, + BRW_REGISTER_TYPE_F, + BRW_VERTICAL_STRIDE_4, + BRW_WIDTH_4, + BRW_HORIZONTAL_STRIDE_1, + BRW_SWIZZLE_XYZW, + TGSI_WRITEMASK_XYZW); +} + + +static __inline struct brw_reg brw_vec2_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return brw_reg(file, + nr, + subnr, + BRW_REGISTER_TYPE_F, + BRW_VERTICAL_STRIDE_2, + BRW_WIDTH_2, + BRW_HORIZONTAL_STRIDE_1, + BRW_SWIZZLE_XYXY, + TGSI_WRITEMASK_XY); +} + +static __inline struct brw_reg brw_vec1_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return brw_reg(file, + nr, + subnr, + BRW_REGISTER_TYPE_F, + BRW_VERTICAL_STRIDE_0, + BRW_WIDTH_1, + BRW_HORIZONTAL_STRIDE_0, + BRW_SWIZZLE_XXXX, + TGSI_WRITEMASK_X); +} + + +static __inline struct brw_reg retype( struct brw_reg reg, + unsigned type ) +{ + reg.type = type; + return reg; +} + +static __inline struct brw_reg suboffset( struct brw_reg reg, + unsigned delta ) +{ + reg.subnr += delta * type_sz(reg.type); + return reg; +} + + +static __inline struct brw_reg offset( struct brw_reg reg, + unsigned delta ) +{ + reg.nr += delta; + return reg; +} + + +static __inline struct brw_reg byte_offset( struct brw_reg reg, + unsigned bytes ) +{ + unsigned newoffset = reg.nr * REG_SIZE + reg.subnr + bytes; + reg.nr = newoffset / REG_SIZE; + reg.subnr = newoffset % REG_SIZE; + return reg; +} + + +static __inline struct brw_reg brw_uw16_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return suboffset(retype(brw_vec16_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); +} + +static __inline struct brw_reg brw_uw8_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return suboffset(retype(brw_vec8_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); +} + +static __inline struct brw_reg brw_uw1_reg( unsigned file, + unsigned nr, + unsigned subnr ) +{ + return suboffset(retype(brw_vec1_reg(file, nr, 0), BRW_REGISTER_TYPE_UW), subnr); +} + +static __inline struct brw_reg brw_imm_reg( unsigned type ) +{ + return brw_reg( BRW_IMMEDIATE_VALUE, + 0, + 0, + type, + BRW_VERTICAL_STRIDE_0, + BRW_WIDTH_1, + BRW_HORIZONTAL_STRIDE_0, + 0, + 0); +} + +static __inline struct brw_reg brw_imm_f( float f ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_F); + imm.dw1.f = f; + return imm; +} + +static __inline struct brw_reg brw_imm_d( int d ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_D); + imm.dw1.d = d; + return imm; +} + +static __inline struct brw_reg brw_imm_ud( unsigned ud ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UD); + imm.dw1.ud = ud; + return imm; +} + +static __inline struct brw_reg brw_imm_uw( ushort uw ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_UW); + imm.dw1.ud = uw; + return imm; +} + +static __inline struct brw_reg brw_imm_w( short w ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_W); + imm.dw1.d = w; + return imm; +} + +/* brw_imm_b and brw_imm_ub aren't supported by hardware - the type + * numbers alias with _V and _VF below: + */ + +/* Vector of eight signed half-byte values: + */ +static __inline struct brw_reg brw_imm_v( unsigned v ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_V); + imm.vstride = BRW_VERTICAL_STRIDE_0; + imm.width = BRW_WIDTH_8; + imm.hstride = BRW_HORIZONTAL_STRIDE_1; + imm.dw1.ud = v; + return imm; +} + +/* Vector of four 8-bit float values: + */ +static __inline struct brw_reg brw_imm_vf( unsigned v ) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF); + imm.vstride = BRW_VERTICAL_STRIDE_0; + imm.width = BRW_WIDTH_4; + imm.hstride = BRW_HORIZONTAL_STRIDE_1; + imm.dw1.ud = v; + return imm; +} + +#define VF_ZERO 0x0 +#define VF_ONE 0x30 +#define VF_NEG (1<<7) + +static __inline struct brw_reg brw_imm_vf4( unsigned v0, + unsigned v1, + unsigned v2, + unsigned v3) +{ + struct brw_reg imm = brw_imm_reg(BRW_REGISTER_TYPE_VF); + imm.vstride = BRW_VERTICAL_STRIDE_0; + imm.width = BRW_WIDTH_4; + imm.hstride = BRW_HORIZONTAL_STRIDE_1; + imm.dw1.ud = ((v0 << 0) | + (v1 << 8) | + (v2 << 16) | + (v3 << 24)); + return imm; +} + + +static __inline struct brw_reg brw_address( struct brw_reg reg ) +{ + return brw_imm_uw(reg.nr * REG_SIZE + reg.subnr); +} + + +static __inline struct brw_reg brw_vec1_grf( unsigned nr, + unsigned subnr ) +{ + return brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); +} + +static __inline struct brw_reg brw_vec8_grf( unsigned nr, + unsigned subnr ) +{ + return brw_vec8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); +} + +static __inline struct brw_reg brw_vec4_grf( unsigned nr, + unsigned subnr ) +{ + return brw_vec4_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); +} + + +static __inline struct brw_reg brw_vec2_grf( unsigned nr, + unsigned subnr ) +{ + return brw_vec2_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); +} + +static __inline struct brw_reg brw_uw8_grf( unsigned nr, + unsigned subnr ) +{ + return brw_uw8_reg(BRW_GENERAL_REGISTER_FILE, nr, subnr); +} + +static __inline struct brw_reg brw_null_reg( void ) +{ + return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_NULL, + 0); +} + +static __inline struct brw_reg brw_address_reg( unsigned subnr ) +{ + return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_ADDRESS, + subnr); +} + +/* If/else instructions break in align16 mode if writemask & swizzle + * aren't xyzw. This goes against the convention for other scalar + * regs: + */ +static __inline struct brw_reg brw_ip_reg( void ) +{ + return brw_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_IP, + 0, + BRW_REGISTER_TYPE_UD, + BRW_VERTICAL_STRIDE_4, /* ? */ + BRW_WIDTH_1, + BRW_HORIZONTAL_STRIDE_0, + BRW_SWIZZLE_XYZW, /* NOTE! */ + TGSI_WRITEMASK_XYZW); /* NOTE! */ +} + +static __inline struct brw_reg brw_acc_reg( void ) +{ + return brw_vec8_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_ACCUMULATOR, + 0); +} + + +static __inline struct brw_reg brw_flag_reg( void ) +{ + return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_FLAG, + 0); +} + + +static __inline struct brw_reg brw_mask_reg( unsigned subnr ) +{ + return brw_uw1_reg(BRW_ARCHITECTURE_REGISTER_FILE, + BRW_ARF_MASK, + subnr); +} + +static __inline struct brw_reg brw_message_reg( unsigned nr ) +{ + return brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, + nr, + 0); +} + + + + +/* This is almost always called with a numeric constant argument, so + * make things easy to evaluate at compile time: + */ +static __inline unsigned cvt( unsigned val ) +{ + switch (val) { + case 0: return 0; + case 1: return 1; + case 2: return 2; + case 4: return 3; + case 8: return 4; + case 16: return 5; + case 32: return 6; + } + return 0; +} + +static __inline struct brw_reg stride( struct brw_reg reg, + unsigned vstride, + unsigned width, + unsigned hstride ) +{ + + reg.vstride = cvt(vstride); + reg.width = cvt(width) - 1; + reg.hstride = cvt(hstride); + return reg; +} + +static __inline struct brw_reg vec16( struct brw_reg reg ) +{ + return stride(reg, 16,16,1); +} + +static __inline struct brw_reg vec8( struct brw_reg reg ) +{ + return stride(reg, 8,8,1); +} + +static __inline struct brw_reg vec4( struct brw_reg reg ) +{ + return stride(reg, 4,4,1); +} + +static __inline struct brw_reg vec2( struct brw_reg reg ) +{ + return stride(reg, 2,2,1); +} + +static __inline struct brw_reg vec1( struct brw_reg reg ) +{ + return stride(reg, 0,1,0); +} + +static __inline struct brw_reg get_element( struct brw_reg reg, unsigned elt ) +{ + return vec1(suboffset(reg, elt)); +} + +static __inline struct brw_reg get_element_ud( struct brw_reg reg, unsigned elt ) +{ + return vec1(suboffset(retype(reg, BRW_REGISTER_TYPE_UD), elt)); +} + + +static __inline struct brw_reg brw_swizzle( struct brw_reg reg, + unsigned x, + unsigned y, + unsigned z, + unsigned w) +{ + reg.dw1.bits.swizzle = BRW_SWIZZLE4(BRW_GET_SWZ(reg.dw1.bits.swizzle, x), + BRW_GET_SWZ(reg.dw1.bits.swizzle, y), + BRW_GET_SWZ(reg.dw1.bits.swizzle, z), + BRW_GET_SWZ(reg.dw1.bits.swizzle, w)); + return reg; +} + + +static __inline struct brw_reg brw_swizzle1( struct brw_reg reg, + unsigned x ) +{ + return brw_swizzle(reg, x, x, x, x); +} + +static __inline struct brw_reg brw_writemask( struct brw_reg reg, + unsigned mask ) +{ + reg.dw1.bits.writemask &= mask; + return reg; +} + +static __inline struct brw_reg brw_set_writemask( struct brw_reg reg, + unsigned mask ) +{ + reg.dw1.bits.writemask = mask; + return reg; +} + +static __inline struct brw_reg negate( struct brw_reg reg ) +{ + reg.negate ^= 1; + return reg; +} + +static __inline struct brw_reg brw_abs( struct brw_reg reg ) +{ + reg.abs = 1; + return reg; +} + +/*********************************************************************** + */ +static __inline struct brw_reg brw_vec4_indirect( unsigned subnr, + int offset ) +{ + struct brw_reg reg = brw_vec4_grf(0, 0); + reg.subnr = subnr; + reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER; + reg.dw1.bits.indirect_offset = offset; + return reg; +} + +static __inline struct brw_reg brw_vec1_indirect( unsigned subnr, + int offset ) +{ + struct brw_reg reg = brw_vec1_grf(0, 0); + reg.subnr = subnr; + reg.address_mode = BRW_ADDRESS_REGISTER_INDIRECT_REGISTER; + reg.dw1.bits.indirect_offset = offset; + return reg; +} + +static __inline struct brw_reg deref_4f(struct brw_indirect ptr, int offset) +{ + return brw_vec4_indirect(ptr.addr_subnr, ptr.addr_offset + offset); +} + +static __inline struct brw_reg deref_1f(struct brw_indirect ptr, int offset) +{ + return brw_vec1_indirect(ptr.addr_subnr, ptr.addr_offset + offset); +} + +static __inline struct brw_reg deref_4b(struct brw_indirect ptr, int offset) +{ + return retype(deref_4f(ptr, offset), BRW_REGISTER_TYPE_B); +} + +static __inline struct brw_reg deref_1uw(struct brw_indirect ptr, int offset) +{ + return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UW); +} + +static __inline struct brw_reg deref_1ud(struct brw_indirect ptr, int offset) +{ + return retype(deref_1f(ptr, offset), BRW_REGISTER_TYPE_UD); +} + +static __inline struct brw_reg get_addr_reg(struct brw_indirect ptr) +{ + return brw_address_reg(ptr.addr_subnr); +} + +static __inline struct brw_indirect brw_indirect_offset( struct brw_indirect ptr, int offset ) +{ + ptr.addr_offset += offset; + return ptr; +} + +static __inline struct brw_indirect brw_indirect( unsigned addr_subnr, int offset ) +{ + struct brw_indirect ptr; + ptr.addr_subnr = addr_subnr; + ptr.addr_offset = offset; + ptr.pad = 0; + return ptr; +} + +static __inline struct brw_instruction *current_insn( struct brw_compile *p) +{ + return &p->store[p->nr_insn]; +} + +void brw_pop_insn_state( struct brw_compile *p ); +void brw_push_insn_state( struct brw_compile *p ); +void brw_set_mask_control( struct brw_compile *p, unsigned value ); +void brw_set_saturate( struct brw_compile *p, unsigned value ); +void brw_set_access_mode( struct brw_compile *p, unsigned access_mode ); +void brw_set_compression_control( struct brw_compile *p, boolean control ); +void brw_set_predicate_control_flag_value( struct brw_compile *p, unsigned value ); +void brw_set_predicate_control( struct brw_compile *p, unsigned pc ); +void brw_set_conditionalmod( struct brw_compile *p, unsigned conditional ); + +void brw_init_compile( struct brw_compile *p ); +const unsigned *brw_get_program( struct brw_compile *p, unsigned *sz ); + + +struct brw_instruction *brw_alu1( struct brw_compile *p, + unsigned opcode, + struct brw_reg dest, + struct brw_reg src ); + +struct brw_instruction *brw_alu2(struct brw_compile *p, + unsigned opcode, + struct brw_reg dest, + struct brw_reg src0, + struct brw_reg src1 ); + +/* Helpers for regular instructions: + */ +#define ALU1(OP) \ +struct brw_instruction *brw_##OP(struct brw_compile *p, \ + struct brw_reg dest, \ + struct brw_reg src0); + +#define ALU2(OP) \ +struct brw_instruction *brw_##OP(struct brw_compile *p, \ + struct brw_reg dest, \ + struct brw_reg src0, \ + struct brw_reg src1); + +ALU1(MOV) +ALU2(SEL) +ALU1(NOT) +ALU2(AND) +ALU2(OR) +ALU2(XOR) +ALU2(SHR) +ALU2(SHL) +ALU2(RSR) +ALU2(RSL) +ALU2(ASR) +ALU2(JMPI) +ALU2(ADD) +ALU2(MUL) +ALU1(FRC) +ALU1(RNDD) +ALU2(MAC) +ALU2(MACH) +ALU1(LZD) +ALU2(DP4) +ALU2(DPH) +ALU2(DP3) +ALU2(DP2) +ALU2(LINE) + +#undef ALU1 +#undef ALU2 + + + +/* Helpers for SEND instruction: + */ +void brw_urb_WRITE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + boolean allocate, + boolean used, + unsigned msg_length, + unsigned response_length, + boolean eot, + boolean writes_complete, + unsigned offset, + unsigned swizzle); + +void brw_fb_WRITE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + unsigned binding_table_index, + unsigned msg_length, + unsigned response_length, + boolean eot); + +void brw_SAMPLE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + unsigned binding_table_index, + unsigned sampler, + unsigned writemask, + unsigned msg_type, + unsigned response_length, + unsigned msg_length, + boolean eot); + +void brw_math_16( struct brw_compile *p, + struct brw_reg dest, + unsigned function, + unsigned saturate, + unsigned msg_reg_nr, + struct brw_reg src, + unsigned precision ); + +void brw_math( struct brw_compile *p, + struct brw_reg dest, + unsigned function, + unsigned saturate, + unsigned msg_reg_nr, + struct brw_reg src, + unsigned data_type, + unsigned precision ); + +void brw_dp_READ_16( struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + unsigned scratch_offset ); + +void brw_dp_WRITE_16( struct brw_compile *p, + struct brw_reg src, + unsigned msg_reg_nr, + unsigned scratch_offset ); + +/* If/else/endif. Works by manipulating the execution flags on each + * channel. + */ +struct brw_instruction *brw_IF(struct brw_compile *p, + unsigned execute_size); + +struct brw_instruction *brw_ELSE(struct brw_compile *p, + struct brw_instruction *if_insn); + +void brw_ENDIF(struct brw_compile *p, + struct brw_instruction *if_or_else_insn); + + +/* DO/WHILE loops: + */ +struct brw_instruction *brw_DO(struct brw_compile *p, + unsigned execute_size); + +struct brw_instruction *brw_WHILE(struct brw_compile *p, + struct brw_instruction *patch_insn); + +struct brw_instruction *brw_BREAK(struct brw_compile *p); +struct brw_instruction *brw_CONT(struct brw_compile *p); +/* Forward jumps: + */ +void brw_land_fwd_jump(struct brw_compile *p, + struct brw_instruction *jmp_insn); + + + +void brw_NOP(struct brw_compile *p); + +/* Special case: there is never a destination, execution size will be + * taken from src0: + */ +void brw_CMP(struct brw_compile *p, + struct brw_reg dest, + unsigned conditional, + struct brw_reg src0, + struct brw_reg src1); + +void brw_print_reg( struct brw_reg reg ); + + +/*********************************************************************** + * brw_eu_util.c: + */ + +void brw_copy_indirect_to_indirect(struct brw_compile *p, + struct brw_indirect dst_ptr, + struct brw_indirect src_ptr, + unsigned count); + +void brw_copy_from_indirect(struct brw_compile *p, + struct brw_reg dst, + struct brw_indirect ptr, + unsigned count); + +void brw_copy4(struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src, + unsigned count); + +void brw_copy8(struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src, + unsigned count); + +void brw_math_invert( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src); + +void brw_set_src1( struct brw_instruction *insn, + struct brw_reg reg ); +#endif diff --git a/src/gallium/drivers/i965simple/brw_eu_debug.c b/src/gallium/drivers/i965simple/brw_eu_debug.c new file mode 100644 index 0000000000..4a94ddefa6 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_eu_debug.c @@ -0,0 +1,90 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_debug.h" + +#include "brw_eu.h" + +void brw_print_reg( struct brw_reg hwreg ) +{ + static const char *file[] = { + "arf", + "grf", + "msg", + "imm" + }; + + static const char *type[] = { + "ud", + "d", + "uw", + "w", + "ub", + "vf", + "hf", + "f" + }; + + debug_printf("%s%s", + hwreg.abs ? "abs/" : "", + hwreg.negate ? "-" : ""); + + if (hwreg.file == BRW_GENERAL_REGISTER_FILE && + hwreg.nr % 2 == 0 && + hwreg.subnr == 0 && + hwreg.vstride == BRW_VERTICAL_STRIDE_8 && + hwreg.width == BRW_WIDTH_8 && + hwreg.hstride == BRW_HORIZONTAL_STRIDE_1 && + hwreg.type == BRW_REGISTER_TYPE_F) { + debug_printf("vec%d", hwreg.nr); + } + else if (hwreg.file == BRW_GENERAL_REGISTER_FILE && + hwreg.vstride == BRW_VERTICAL_STRIDE_0 && + hwreg.width == BRW_WIDTH_1 && + hwreg.hstride == BRW_HORIZONTAL_STRIDE_0 && + hwreg.type == BRW_REGISTER_TYPE_F) { + debug_printf("scl%d.%d", hwreg.nr, hwreg.subnr / 4); + } + else { + debug_printf("%s%d.%d<%d;%d,%d>:%s", + file[hwreg.file], + hwreg.nr, + hwreg.subnr / type_sz(hwreg.type), + hwreg.vstride ? (1<<(hwreg.vstride-1)) : 0, + 1<<hwreg.width, + hwreg.hstride ? (1<<(hwreg.hstride-1)) : 0, + type[hwreg.type]); + } +} + + + diff --git a/src/gallium/drivers/i965simple/brw_eu_emit.c b/src/gallium/drivers/i965simple/brw_eu_emit.c new file mode 100644 index 0000000000..400a80b6fb --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_eu_emit.c @@ -0,0 +1,1080 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_eu.h" + + + + +/*********************************************************************** + * Internal helper for constructing instructions + */ + +static void guess_execution_size( struct brw_instruction *insn, + struct brw_reg reg ) +{ + if (reg.width == BRW_WIDTH_8 && + insn->header.compression_control == BRW_COMPRESSION_COMPRESSED) + insn->header.execution_size = BRW_EXECUTE_16; + else + insn->header.execution_size = reg.width; /* note - definitions are compatible */ +} + + +static void brw_set_dest( struct brw_instruction *insn, + struct brw_reg dest ) +{ + insn->bits1.da1.dest_reg_file = dest.file; + insn->bits1.da1.dest_reg_type = dest.type; + insn->bits1.da1.dest_address_mode = dest.address_mode; + + if (dest.address_mode == BRW_ADDRESS_DIRECT) { + insn->bits1.da1.dest_reg_nr = dest.nr; + + if (insn->header.access_mode == BRW_ALIGN_1) { + insn->bits1.da1.dest_subreg_nr = dest.subnr; + insn->bits1.da1.dest_horiz_stride = BRW_HORIZONTAL_STRIDE_1; + } + else { + insn->bits1.da16.dest_subreg_nr = dest.subnr / 16; + insn->bits1.da16.dest_writemask = dest.dw1.bits.writemask; + } + } + else { + insn->bits1.ia1.dest_subreg_nr = dest.subnr; + + /* These are different sizes in align1 vs align16: + */ + if (insn->header.access_mode == BRW_ALIGN_1) { + insn->bits1.ia1.dest_indirect_offset = dest.dw1.bits.indirect_offset; + insn->bits1.ia1.dest_horiz_stride = BRW_HORIZONTAL_STRIDE_1; + } + else { + insn->bits1.ia16.dest_indirect_offset = dest.dw1.bits.indirect_offset; + } + } + + /* NEW: Set the execution size based on dest.width and + * insn->compression_control: + */ + guess_execution_size(insn, dest); +} + +static void brw_set_src0( struct brw_instruction *insn, + struct brw_reg reg ) +{ + assert(reg.file != BRW_MESSAGE_REGISTER_FILE); + + insn->bits1.da1.src0_reg_file = reg.file; + insn->bits1.da1.src0_reg_type = reg.type; + insn->bits2.da1.src0_abs = reg.abs; + insn->bits2.da1.src0_negate = reg.negate; + insn->bits2.da1.src0_address_mode = reg.address_mode; + + if (reg.file == BRW_IMMEDIATE_VALUE) { + insn->bits3.ud = reg.dw1.ud; + + /* Required to set some fields in src1 as well: + */ + insn->bits1.da1.src1_reg_file = 0; /* arf */ + insn->bits1.da1.src1_reg_type = reg.type; + } + else + { + if (reg.address_mode == BRW_ADDRESS_DIRECT) { + if (insn->header.access_mode == BRW_ALIGN_1) { + insn->bits2.da1.src0_subreg_nr = reg.subnr; + insn->bits2.da1.src0_reg_nr = reg.nr; + } + else { + insn->bits2.da16.src0_subreg_nr = reg.subnr / 16; + insn->bits2.da16.src0_reg_nr = reg.nr; + } + } + else { + insn->bits2.ia1.src0_subreg_nr = reg.subnr; + + if (insn->header.access_mode == BRW_ALIGN_1) { + insn->bits2.ia1.src0_indirect_offset = reg.dw1.bits.indirect_offset; + } + else { + insn->bits2.ia16.src0_subreg_nr = reg.dw1.bits.indirect_offset; + } + } + + if (insn->header.access_mode == BRW_ALIGN_1) { + if (reg.width == BRW_WIDTH_1 && + insn->header.execution_size == BRW_EXECUTE_1) { + insn->bits2.da1.src0_horiz_stride = BRW_HORIZONTAL_STRIDE_0; + insn->bits2.da1.src0_width = BRW_WIDTH_1; + insn->bits2.da1.src0_vert_stride = BRW_VERTICAL_STRIDE_0; + } + else { + insn->bits2.da1.src0_horiz_stride = reg.hstride; + insn->bits2.da1.src0_width = reg.width; + insn->bits2.da1.src0_vert_stride = reg.vstride; + } + } + else { + insn->bits2.da16.src0_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X); + insn->bits2.da16.src0_swz_y = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Y); + insn->bits2.da16.src0_swz_z = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Z); + insn->bits2.da16.src0_swz_w = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_W); + + /* This is an oddity of the fact we're using the same + * descriptions for registers in align_16 as align_1: + */ + if (reg.vstride == BRW_VERTICAL_STRIDE_8) + insn->bits2.da16.src0_vert_stride = BRW_VERTICAL_STRIDE_4; + else + insn->bits2.da16.src0_vert_stride = reg.vstride; + } + } +} + + +void brw_set_src1( struct brw_instruction *insn, + struct brw_reg reg ) +{ + assert(reg.file != BRW_MESSAGE_REGISTER_FILE); + + insn->bits1.da1.src1_reg_file = reg.file; + insn->bits1.da1.src1_reg_type = reg.type; + insn->bits3.da1.src1_abs = reg.abs; + insn->bits3.da1.src1_negate = reg.negate; + + /* Only src1 can be immediate in two-argument instructions. + */ + assert(insn->bits1.da1.src0_reg_file != BRW_IMMEDIATE_VALUE); + + if (reg.file == BRW_IMMEDIATE_VALUE) { + insn->bits3.ud = reg.dw1.ud; + } + else { + /* This is a hardware restriction, which may or may not be lifted + * in the future: + */ + assert (reg.address_mode == BRW_ADDRESS_DIRECT); + //assert (reg.file == BRW_GENERAL_REGISTER_FILE); + + if (insn->header.access_mode == BRW_ALIGN_1) { + insn->bits3.da1.src1_subreg_nr = reg.subnr; + insn->bits3.da1.src1_reg_nr = reg.nr; + } + else { + insn->bits3.da16.src1_subreg_nr = reg.subnr / 16; + insn->bits3.da16.src1_reg_nr = reg.nr; + } + + if (insn->header.access_mode == BRW_ALIGN_1) { + if (reg.width == BRW_WIDTH_1 && + insn->header.execution_size == BRW_EXECUTE_1) { + insn->bits3.da1.src1_horiz_stride = BRW_HORIZONTAL_STRIDE_0; + insn->bits3.da1.src1_width = BRW_WIDTH_1; + insn->bits3.da1.src1_vert_stride = BRW_VERTICAL_STRIDE_0; + } + else { + insn->bits3.da1.src1_horiz_stride = reg.hstride; + insn->bits3.da1.src1_width = reg.width; + insn->bits3.da1.src1_vert_stride = reg.vstride; + } + } + else { + insn->bits3.da16.src1_swz_x = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_X); + insn->bits3.da16.src1_swz_y = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Y); + insn->bits3.da16.src1_swz_z = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_Z); + insn->bits3.da16.src1_swz_w = BRW_GET_SWZ(reg.dw1.bits.swizzle, BRW_CHANNEL_W); + + /* This is an oddity of the fact we're using the same + * descriptions for registers in align_16 as align_1: + */ + if (reg.vstride == BRW_VERTICAL_STRIDE_8) + insn->bits3.da16.src1_vert_stride = BRW_VERTICAL_STRIDE_4; + else + insn->bits3.da16.src1_vert_stride = reg.vstride; + } + } +} + + + +static void brw_set_math_message( struct brw_instruction *insn, + unsigned msg_length, + unsigned response_length, + unsigned function, + unsigned integer_type, + boolean low_precision, + boolean saturate, + unsigned dataType ) +{ + brw_set_src1(insn, brw_imm_d(0)); + + insn->bits3.math.function = function; + insn->bits3.math.int_type = integer_type; + insn->bits3.math.precision = low_precision; + insn->bits3.math.saturate = saturate; + insn->bits3.math.data_type = dataType; + insn->bits3.math.response_length = response_length; + insn->bits3.math.msg_length = msg_length; + insn->bits3.math.msg_target = BRW_MESSAGE_TARGET_MATH; + insn->bits3.math.end_of_thread = 0; +} + +static void brw_set_urb_message( struct brw_instruction *insn, + boolean allocate, + boolean used, + unsigned msg_length, + unsigned response_length, + boolean end_of_thread, + boolean complete, + unsigned offset, + unsigned swizzle_control ) +{ + brw_set_src1(insn, brw_imm_d(0)); + + insn->bits3.urb.opcode = 0; /* ? */ + insn->bits3.urb.offset = offset; + insn->bits3.urb.swizzle_control = swizzle_control; + insn->bits3.urb.allocate = allocate; + insn->bits3.urb.used = used; /* ? */ + insn->bits3.urb.complete = complete; + insn->bits3.urb.response_length = response_length; + insn->bits3.urb.msg_length = msg_length; + insn->bits3.urb.msg_target = BRW_MESSAGE_TARGET_URB; + insn->bits3.urb.end_of_thread = end_of_thread; +} + +static void brw_set_dp_write_message( struct brw_instruction *insn, + unsigned binding_table_index, + unsigned msg_control, + unsigned msg_type, + unsigned msg_length, + unsigned pixel_scoreboard_clear, + unsigned response_length, + unsigned end_of_thread ) +{ + brw_set_src1(insn, brw_imm_d(0)); + + insn->bits3.dp_write.binding_table_index = binding_table_index; + insn->bits3.dp_write.msg_control = msg_control; + insn->bits3.dp_write.pixel_scoreboard_clear = pixel_scoreboard_clear; + insn->bits3.dp_write.msg_type = msg_type; + insn->bits3.dp_write.send_commit_msg = 0; + insn->bits3.dp_write.response_length = response_length; + insn->bits3.dp_write.msg_length = msg_length; + insn->bits3.dp_write.msg_target = BRW_MESSAGE_TARGET_DATAPORT_WRITE; + insn->bits3.urb.end_of_thread = end_of_thread; +} + +static void brw_set_dp_read_message( struct brw_instruction *insn, + unsigned binding_table_index, + unsigned msg_control, + unsigned msg_type, + unsigned target_cache, + unsigned msg_length, + unsigned response_length, + unsigned end_of_thread ) +{ + brw_set_src1(insn, brw_imm_d(0)); + + insn->bits3.dp_read.binding_table_index = binding_table_index; + insn->bits3.dp_read.msg_control = msg_control; + insn->bits3.dp_read.msg_type = msg_type; + insn->bits3.dp_read.target_cache = target_cache; + insn->bits3.dp_read.response_length = response_length; + insn->bits3.dp_read.msg_length = msg_length; + insn->bits3.dp_read.msg_target = BRW_MESSAGE_TARGET_DATAPORT_READ; + insn->bits3.dp_read.end_of_thread = end_of_thread; +} + +static void brw_set_sampler_message( struct brw_instruction *insn, + unsigned binding_table_index, + unsigned sampler, + unsigned msg_type, + unsigned response_length, + unsigned msg_length, + boolean eot) +{ + brw_set_src1(insn, brw_imm_d(0)); + + insn->bits3.sampler.binding_table_index = binding_table_index; + insn->bits3.sampler.sampler = sampler; + insn->bits3.sampler.msg_type = msg_type; + insn->bits3.sampler.return_format = BRW_SAMPLER_RETURN_FORMAT_FLOAT32; + insn->bits3.sampler.response_length = response_length; + insn->bits3.sampler.msg_length = msg_length; + insn->bits3.sampler.end_of_thread = eot; + insn->bits3.sampler.msg_target = BRW_MESSAGE_TARGET_SAMPLER; +} + + + +static struct brw_instruction *next_insn( struct brw_compile *p, + unsigned opcode ) +{ + struct brw_instruction *insn; + + assert(p->nr_insn + 1 < BRW_EU_MAX_INSN); + + insn = &p->store[p->nr_insn++]; + memcpy(insn, p->current, sizeof(*insn)); + + /* Reset this one-shot flag: + */ + + if (p->current->header.destreg__conditonalmod) { + p->current->header.destreg__conditonalmod = 0; + p->current->header.predicate_control = BRW_PREDICATE_NORMAL; + } + + insn->header.opcode = opcode; + return insn; +} + + +struct brw_instruction *brw_alu1( struct brw_compile *p, + unsigned opcode, + struct brw_reg dest, + struct brw_reg src ) +{ + struct brw_instruction *insn = next_insn(p, opcode); + brw_set_dest(insn, dest); + brw_set_src0(insn, src); + return insn; +} + +struct brw_instruction *brw_alu2(struct brw_compile *p, + unsigned opcode, + struct brw_reg dest, + struct brw_reg src0, + struct brw_reg src1 ) +{ + struct brw_instruction *insn = next_insn(p, opcode); + brw_set_dest(insn, dest); + brw_set_src0(insn, src0); + brw_set_src1(insn, src1); + return insn; +} + + +/*********************************************************************** + * Convenience routines. + */ +#define ALU1(OP) \ +struct brw_instruction *brw_##OP(struct brw_compile *p, \ + struct brw_reg dest, \ + struct brw_reg src0) \ +{ \ + return brw_alu1(p, BRW_OPCODE_##OP, dest, src0); \ +} + +#define ALU2(OP) \ +struct brw_instruction *brw_##OP(struct brw_compile *p, \ + struct brw_reg dest, \ + struct brw_reg src0, \ + struct brw_reg src1) \ +{ \ + return brw_alu2(p, BRW_OPCODE_##OP, dest, src0, src1); \ +} + + +ALU1(MOV) +ALU2(SEL) +ALU1(NOT) +ALU2(AND) +ALU2(OR) +ALU2(XOR) +ALU2(SHR) +ALU2(SHL) +ALU2(RSR) +ALU2(RSL) +ALU2(ASR) +ALU2(ADD) +ALU2(MUL) +ALU1(FRC) +ALU1(RNDD) +ALU2(MAC) +ALU2(MACH) +ALU1(LZD) +ALU2(DP4) +ALU2(DPH) +ALU2(DP3) +ALU2(DP2) +ALU2(LINE) + + + + +void brw_NOP(struct brw_compile *p) +{ + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_NOP); + brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src1(insn, brw_imm_ud(0x0)); +} + + + + + +/*********************************************************************** + * Comparisons, if/else/endif + */ + +struct brw_instruction *brw_JMPI(struct brw_compile *p, + struct brw_reg dest, + struct brw_reg src0, + struct brw_reg src1) +{ + struct brw_instruction *insn = brw_alu2(p, BRW_OPCODE_JMPI, dest, src0, src1); + + p->current->header.predicate_control = BRW_PREDICATE_NONE; + + return insn; +} + +/* EU takes the value from the flag register and pushes it onto some + * sort of a stack (presumably merging with any flag value already on + * the stack). Within an if block, the flags at the top of the stack + * control execution on each channel of the unit, eg. on each of the + * 16 pixel values in our wm programs. + * + * When the matching 'else' instruction is reached (presumably by + * countdown of the instruction count patched in by our ELSE/ENDIF + * functions), the relevent flags are inverted. + * + * When the matching 'endif' instruction is reached, the flags are + * popped off. If the stack is now empty, normal execution resumes. + * + * No attempt is made to deal with stack overflow (14 elements?). + */ +struct brw_instruction *brw_IF(struct brw_compile *p, unsigned execute_size) +{ + struct brw_instruction *insn; + + if (p->single_program_flow) { + assert(execute_size == BRW_EXECUTE_1); + + insn = next_insn(p, BRW_OPCODE_ADD); + insn->header.predicate_inverse = 1; + } else { + insn = next_insn(p, BRW_OPCODE_IF); + } + + /* Override the defaults for this instruction: + */ + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + + insn->header.execution_size = execute_size; + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.predicate_control = BRW_PREDICATE_NORMAL; + insn->header.mask_control = BRW_MASK_ENABLE; + + p->current->header.predicate_control = BRW_PREDICATE_NONE; + + return insn; +} + + +struct brw_instruction *brw_ELSE(struct brw_compile *p, + struct brw_instruction *if_insn) +{ + struct brw_instruction *insn; + + if (p->single_program_flow) { + insn = next_insn(p, BRW_OPCODE_ADD); + } else { + insn = next_insn(p, BRW_OPCODE_ELSE); + } + + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.execution_size = if_insn->header.execution_size; + insn->header.mask_control = BRW_MASK_ENABLE; + + /* Patch the if instruction to point at this instruction. + */ + if (p->single_program_flow) { + assert(if_insn->header.opcode == BRW_OPCODE_ADD); + + if_insn->bits3.ud = (insn - if_insn + 1) * 16; + } else { + assert(if_insn->header.opcode == BRW_OPCODE_IF); + + if_insn->bits3.if_else.jump_count = insn - if_insn; + if_insn->bits3.if_else.pop_count = 1; + if_insn->bits3.if_else.pad0 = 0; + } + + return insn; +} + +void brw_ENDIF(struct brw_compile *p, + struct brw_instruction *patch_insn) +{ + if (p->single_program_flow) { + /* In single program flow mode, there's no need to execute an ENDIF, + * since we don't need to do any stack operations, and if we're executing + * currently, we want to just continue executing. + */ + struct brw_instruction *next = &p->store[p->nr_insn]; + + assert(patch_insn->header.opcode == BRW_OPCODE_ADD); + + patch_insn->bits3.ud = (next - patch_insn) * 16; + } else { + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_ENDIF); + + brw_set_dest(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src0(insn, retype(brw_vec4_grf(0,0), BRW_REGISTER_TYPE_UD)); + brw_set_src1(insn, brw_imm_d(0x0)); + + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.execution_size = patch_insn->header.execution_size; + insn->header.mask_control = BRW_MASK_ENABLE; + + assert(patch_insn->bits3.if_else.jump_count == 0); + + /* Patch the if or else instructions to point at this or the next + * instruction respectively. + */ + if (patch_insn->header.opcode == BRW_OPCODE_IF) { + /* Automagically turn it into an IFF: + */ + patch_insn->header.opcode = BRW_OPCODE_IFF; + patch_insn->bits3.if_else.jump_count = insn - patch_insn + 1; + patch_insn->bits3.if_else.pop_count = 0; + patch_insn->bits3.if_else.pad0 = 0; + } else if (patch_insn->header.opcode == BRW_OPCODE_ELSE) { + patch_insn->bits3.if_else.jump_count = insn - patch_insn + 1; + patch_insn->bits3.if_else.pop_count = 1; + patch_insn->bits3.if_else.pad0 = 0; + } else { + assert(0); + } + + /* Also pop item off the stack in the endif instruction: + */ + insn->bits3.if_else.jump_count = 0; + insn->bits3.if_else.pop_count = 1; + insn->bits3.if_else.pad0 = 0; + } +} + +struct brw_instruction *brw_BREAK(struct brw_compile *p) +{ + struct brw_instruction *insn; + insn = next_insn(p, BRW_OPCODE_BREAK); + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.execution_size = BRW_EXECUTE_8; + insn->header.mask_control = BRW_MASK_DISABLE; + insn->bits3.if_else.pad0 = 0; + return insn; +} + +struct brw_instruction *brw_CONT(struct brw_compile *p) +{ + struct brw_instruction *insn; + insn = next_insn(p, BRW_OPCODE_CONTINUE); + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.execution_size = BRW_EXECUTE_8; + insn->header.mask_control = BRW_MASK_DISABLE; + insn->bits3.if_else.pad0 = 0; + return insn; +} + +/* DO/WHILE loop: + */ +struct brw_instruction *brw_DO(struct brw_compile *p, unsigned execute_size) +{ + if (p->single_program_flow) { + return &p->store[p->nr_insn]; + } else { + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_DO); + + /* Override the defaults for this instruction: + */ + brw_set_dest(insn, brw_null_reg()); + brw_set_src0(insn, brw_null_reg()); + brw_set_src1(insn, brw_null_reg()); + + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.execution_size = execute_size; + insn->header.predicate_control = BRW_PREDICATE_NONE; + /* insn->header.mask_control = BRW_MASK_ENABLE; */ + insn->header.mask_control = BRW_MASK_DISABLE; + + return insn; + } +} + + + +struct brw_instruction *brw_WHILE(struct brw_compile *p, + struct brw_instruction *do_insn) +{ + struct brw_instruction *insn; + + if (p->single_program_flow) + insn = next_insn(p, BRW_OPCODE_ADD); + else + insn = next_insn(p, BRW_OPCODE_WHILE); + + brw_set_dest(insn, brw_ip_reg()); + brw_set_src0(insn, brw_ip_reg()); + brw_set_src1(insn, brw_imm_d(0x0)); + + insn->header.compression_control = BRW_COMPRESSION_NONE; + + if (p->single_program_flow) { + insn->header.execution_size = BRW_EXECUTE_1; + + insn->bits3.d = (do_insn - insn) * 16; + } else { + insn->header.execution_size = do_insn->header.execution_size; + + assert(do_insn->header.opcode == BRW_OPCODE_DO); + insn->bits3.if_else.jump_count = do_insn - insn; + insn->bits3.if_else.pop_count = 0; + insn->bits3.if_else.pad0 = 0; + } + +/* insn->header.mask_control = BRW_MASK_ENABLE; */ + + insn->header.mask_control = BRW_MASK_DISABLE; + p->current->header.predicate_control = BRW_PREDICATE_NONE; + return insn; +} + + +/* FORWARD JUMPS: + */ +void brw_land_fwd_jump(struct brw_compile *p, + struct brw_instruction *jmp_insn) +{ + struct brw_instruction *landing = &p->store[p->nr_insn]; + + assert(jmp_insn->header.opcode == BRW_OPCODE_JMPI); + assert(jmp_insn->bits1.da1.src1_reg_file = BRW_IMMEDIATE_VALUE); + + jmp_insn->bits3.ud = (landing - jmp_insn) - 1; +} + + + +/* To integrate with the above, it makes sense that the comparison + * instruction should populate the flag register. It might be simpler + * just to use the flag reg for most WM tasks? + */ +void brw_CMP(struct brw_compile *p, + struct brw_reg dest, + unsigned conditional, + struct brw_reg src0, + struct brw_reg src1) +{ + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_CMP); + + insn->header.destreg__conditonalmod = conditional; + brw_set_dest(insn, dest); + brw_set_src0(insn, src0); + brw_set_src1(insn, src1); + +/* guess_execution_size(insn, src0); */ + + + /* Make it so that future instructions will use the computed flag + * value until brw_set_predicate_control_flag_value() is called + * again. + */ + if (dest.file == BRW_ARCHITECTURE_REGISTER_FILE && + dest.nr == 0) { + p->current->header.predicate_control = BRW_PREDICATE_NORMAL; + p->flag_value = 0xff; + } +} + + + +/*********************************************************************** + * Helpers for the various SEND message types: + */ + +/* Invert 8 values + */ +void brw_math( struct brw_compile *p, + struct brw_reg dest, + unsigned function, + unsigned saturate, + unsigned msg_reg_nr, + struct brw_reg src, + unsigned data_type, + unsigned precision ) +{ + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + unsigned msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1; + unsigned response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1; + + /* Example code doesn't set predicate_control for send + * instructions. + */ + insn->header.predicate_control = 0; + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); + brw_set_src0(insn, src); + brw_set_math_message(insn, + msg_length, response_length, + function, + BRW_MATH_INTEGER_UNSIGNED, + precision, + saturate, + data_type); +} + +/* Use 2 send instructions to invert 16 elements + */ +void brw_math_16( struct brw_compile *p, + struct brw_reg dest, + unsigned function, + unsigned saturate, + unsigned msg_reg_nr, + struct brw_reg src, + unsigned precision ) +{ + struct brw_instruction *insn; + unsigned msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1; + unsigned response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1; + + /* First instruction: + */ + brw_push_insn_state(p); + brw_set_predicate_control_flag_value(p, 0xff); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + + insn = next_insn(p, BRW_OPCODE_SEND); + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); + brw_set_src0(insn, src); + brw_set_math_message(insn, + msg_length, response_length, + function, + BRW_MATH_INTEGER_UNSIGNED, + precision, + saturate, + BRW_MATH_DATA_VECTOR); + + /* Second instruction: + */ + insn = next_insn(p, BRW_OPCODE_SEND); + insn->header.compression_control = BRW_COMPRESSION_2NDHALF; + insn->header.destreg__conditonalmod = msg_reg_nr+1; + + brw_set_dest(insn, offset(dest,1)); + brw_set_src0(insn, src); + brw_set_math_message(insn, + msg_length, response_length, + function, + BRW_MATH_INTEGER_UNSIGNED, + precision, + saturate, + BRW_MATH_DATA_VECTOR); + + brw_pop_insn_state(p); +} + + + + +void brw_dp_WRITE_16( struct brw_compile *p, + struct brw_reg src, + unsigned msg_reg_nr, + unsigned scratch_offset ) +{ + { + brw_push_insn_state(p); + brw_set_mask_control(p, BRW_MASK_DISABLE); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + + brw_MOV(p, + retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D), + brw_imm_d(scratch_offset)); + + brw_pop_insn_state(p); + } + + { + unsigned msg_length = 3; + struct brw_reg dest = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW); + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + insn->header.predicate_control = 0; /* XXX */ + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); + brw_set_src0(insn, src); + + brw_set_dp_write_message(insn, + 255, /* bti */ + BRW_DATAPORT_OWORD_BLOCK_4_OWORDS, /* msg_control */ + BRW_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE, /* msg_type */ + msg_length, + 0, /* pixel scoreboard */ + 0, /* response_length */ + 0); /* eot */ + } + +} + + +void brw_dp_READ_16( struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + unsigned scratch_offset ) +{ + { + brw_push_insn_state(p); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_set_mask_control(p, BRW_MASK_DISABLE); + + brw_MOV(p, + retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D), + brw_imm_d(scratch_offset)); + + brw_pop_insn_state(p); + } + + { + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + insn->header.predicate_control = 0; /* XXX */ + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); /* UW? */ + brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW)); + + brw_set_dp_read_message(insn, + 255, /* bti */ + 3, /* msg_control */ + BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */ + 1, /* target cache */ + 1, /* msg_length */ + 2, /* response_length */ + 0); /* eot */ + } +} + + +void brw_fb_WRITE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + unsigned binding_table_index, + unsigned msg_length, + unsigned response_length, + boolean eot) +{ + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + insn->header.predicate_control = 0; /* XXX */ + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); + brw_set_src0(insn, src0); + brw_set_dp_write_message(insn, + binding_table_index, + BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE, /* msg_control */ + BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE, /* msg_type */ + msg_length, + 1, /* pixel scoreboard */ + response_length, + eot); +} + + + +void brw_SAMPLE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + unsigned binding_table_index, + unsigned sampler, + unsigned writemask, + unsigned msg_type, + unsigned response_length, + unsigned msg_length, + boolean eot) +{ + boolean need_stall = 0; + + if(writemask == 0) { +/* debug_printf("%s: zero writemask??\n", __FUNCTION__); */ + return; + } + + /* Hardware doesn't do destination dependency checking on send + * instructions properly. Add a workaround which generates the + * dependency by other means. In practice it seems like this bug + * only crops up for texture samples, and only where registers are + * written by the send and then written again later without being + * read in between. Luckily for us, we already track that + * information and use it to modify the writemask for the + * instruction, so that is a guide for whether a workaround is + * needed. + */ + if (writemask != TGSI_WRITEMASK_XYZW) { + unsigned dst_offset = 0; + unsigned i, newmask = 0, len = 0; + + for (i = 0; i < 4; i++) { + if (writemask & (1<<i)) + break; + dst_offset += 2; + } + for (; i < 4; i++) { + if (!(writemask & (1<<i))) + break; + newmask |= 1<<i; + len++; + } + + if (newmask != writemask) { + need_stall = 1; +/* debug_printf("need stall %x %x\n", newmask , writemask); */ + } + else { + struct brw_reg m1 = brw_message_reg(msg_reg_nr); + + newmask = ~newmask & TGSI_WRITEMASK_XYZW; + + brw_push_insn_state(p); + + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_set_mask_control(p, BRW_MASK_DISABLE); + + brw_MOV(p, m1, brw_vec8_grf(0,0)); + brw_MOV(p, get_element_ud(m1, 2), brw_imm_ud(newmask << 12)); + + brw_pop_insn_state(p); + + src0 = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW); + dest = offset(dest, dst_offset); + response_length = len * 2; + } + } + + { + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + insn->header.predicate_control = 0; /* XXX */ + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_dest(insn, dest); + brw_set_src0(insn, src0); + brw_set_sampler_message(insn, + binding_table_index, + sampler, + msg_type, + response_length, + msg_length, + eot); + } + + if (need_stall) + { + struct brw_reg reg = vec8(offset(dest, response_length-1)); + + /* mov (8) r9.0<1>:f r9.0<8;8,1>:f { Align1 } + */ + brw_push_insn_state(p); + brw_set_compression_control(p, FALSE); + brw_MOV(p, reg, reg); + brw_pop_insn_state(p); + } + +} + +/* All these variables are pretty confusing - we might be better off + * using bitmasks and macros for this, in the old style. Or perhaps + * just having the caller instantiate the fields in dword3 itself. + */ +void brw_urb_WRITE(struct brw_compile *p, + struct brw_reg dest, + unsigned msg_reg_nr, + struct brw_reg src0, + boolean allocate, + boolean used, + unsigned msg_length, + unsigned response_length, + boolean eot, + boolean writes_complete, + unsigned offset, + unsigned swizzle) +{ + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + assert(msg_length < 16); + + brw_set_dest(insn, dest); + brw_set_src0(insn, src0); + brw_set_src1(insn, brw_imm_d(0)); + + insn->header.destreg__conditonalmod = msg_reg_nr; + + brw_set_urb_message(insn, + allocate, + used, + msg_length, + response_length, + eot, + writes_complete, + offset, + swizzle); +} + diff --git a/src/gallium/drivers/i965simple/brw_eu_util.c b/src/gallium/drivers/i965simple/brw_eu_util.c new file mode 100644 index 0000000000..3a65b141f0 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_eu_util.c @@ -0,0 +1,126 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_eu.h" + + +void brw_math_invert( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src) +{ + brw_math( p, + dst, + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 0, + src, + BRW_MATH_PRECISION_FULL, + BRW_MATH_DATA_VECTOR ); +} + + + +void brw_copy4(struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src, + unsigned count) +{ + unsigned i; + + dst = vec4(dst); + src = vec4(src); + + for (i = 0; i < count; i++) + { + unsigned delta = i*32; + brw_MOV(p, byte_offset(dst, delta), byte_offset(src, delta)); + brw_MOV(p, byte_offset(dst, delta+16), byte_offset(src, delta+16)); + } +} + + +void brw_copy8(struct brw_compile *p, + struct brw_reg dst, + struct brw_reg src, + unsigned count) +{ + unsigned i; + + dst = vec8(dst); + src = vec8(src); + + for (i = 0; i < count; i++) + { + unsigned delta = i*32; + brw_MOV(p, byte_offset(dst, delta), byte_offset(src, delta)); + } +} + + +void brw_copy_indirect_to_indirect(struct brw_compile *p, + struct brw_indirect dst_ptr, + struct brw_indirect src_ptr, + unsigned count) +{ + unsigned i; + + for (i = 0; i < count; i++) + { + unsigned delta = i*32; + brw_MOV(p, deref_4f(dst_ptr, delta), deref_4f(src_ptr, delta)); + brw_MOV(p, deref_4f(dst_ptr, delta+16), deref_4f(src_ptr, delta+16)); + } +} + + +void brw_copy_from_indirect(struct brw_compile *p, + struct brw_reg dst, + struct brw_indirect ptr, + unsigned count) +{ + unsigned i; + + dst = vec4(dst); + + for (i = 0; i < count; i++) + { + unsigned delta = i*32; + brw_MOV(p, byte_offset(dst, delta), deref_4f(ptr, delta)); + brw_MOV(p, byte_offset(dst, delta+16), deref_4f(ptr, delta+16)); + } +} + + + + diff --git a/src/gallium/drivers/i965simple/brw_flush.c b/src/gallium/drivers/i965simple/brw_flush.c new file mode 100644 index 0000000000..5216c680cf --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_flush.c @@ -0,0 +1,80 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_defines.h" +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_batch.h" + + +/** + * In future we may want a fence-like interface instead of finish. + */ +static void brw_flush( struct pipe_context *pipe, + unsigned flags ) +{ + struct brw_context *brw = brw_context(pipe); + struct pipe_fence_handle *fence; + + /* Do we need to emit an MI_FLUSH command to flush the hardware + * caches? + */ + if (flags & (PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE)) { + struct brw_mi_flush flush; + + memset(&flush, 0, sizeof(flush)); + flush.opcode = CMD_MI_FLUSH; + + if (!(flags & PIPE_FLUSH_RENDER_CACHE)) + flush.flags |= BRW_INHIBIT_FLUSH_RENDER_CACHE; + + if (flags & PIPE_FLUSH_TEXTURE_CACHE) + flush.flags |= BRW_FLUSH_READ_CACHE; + + BRW_BATCH_STRUCT(brw, &flush); + } + + /* If there are no flags, just flush pending commands to hardware: + */ + FLUSH_BATCH( &fence ); + + if (flags & PIPE_FLUSH_WAIT) { +// brw->winsys->wait_fence(brw->winsys, fence); + } +} + + + +void brw_init_flush_functions( struct brw_context *brw ) +{ + brw->pipe.flush = brw_flush; +} diff --git a/src/gallium/drivers/i965simple/brw_gs.c b/src/gallium/drivers/i965simple/brw_gs.c new file mode 100644 index 0000000000..de60868ccc --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_gs.c @@ -0,0 +1,196 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_state.h" +#include "brw_gs.h" + + + +static void compile_gs_prog( struct brw_context *brw, + struct brw_gs_prog_key *key ) +{ + struct brw_gs_compile c; + const unsigned *program; + unsigned program_size; + + memset(&c, 0, sizeof(c)); + + c.key = *key; + + /* Need to locate the two positions present in vertex + header. + * These are currently hardcoded: + */ + c.nr_attrs = brw_count_bits(c.key.attrs); + c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ + c.nr_bytes = c.nr_regs * REG_SIZE; + + + /* Begin the compilation: + */ + brw_init_compile(&c.func); + + c.func.single_program_flow = 1; + + /* For some reason the thread is spawned with only 4 channels + * unmasked. + */ + brw_set_mask_control(&c.func, BRW_MASK_DISABLE); + + + /* Note that primitives which don't require a GS program have + * already been weeded out by this stage: + */ + switch (key->primitive) { + case PIPE_PRIM_QUADS: + brw_gs_quads( &c ); + break; + case PIPE_PRIM_QUAD_STRIP: + brw_gs_quad_strip( &c ); + break; + case PIPE_PRIM_LINE_LOOP: + brw_gs_lines( &c ); + break; + case PIPE_PRIM_LINES: + if (key->hint_gs_always) + brw_gs_lines( &c ); + else { + return; + } + break; + case PIPE_PRIM_TRIANGLES: + if (key->hint_gs_always) + brw_gs_tris( &c ); + else { + return; + } + break; + case PIPE_PRIM_POINTS: + if (key->hint_gs_always) + brw_gs_points( &c ); + else { + return; + } + break; + default: + return; + } + + /* get the program + */ + program = brw_get_program(&c.func, &program_size); + + /* Upload + */ + brw->gs.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_GS_PROG], + &c.key, + sizeof(c.key), + program, + program_size, + &c.prog_data, + &brw->gs.prog_data ); +} + + +static boolean search_cache( struct brw_context *brw, + struct brw_gs_prog_key *key ) +{ + return brw_search_cache(&brw->cache[BRW_GS_PROG], + key, sizeof(*key), + &brw->gs.prog_data, + &brw->gs.prog_gs_offset); +} + + +static const int gs_prim[PIPE_PRIM_POLYGON+1] = { + PIPE_PRIM_POINTS, + PIPE_PRIM_LINES, + PIPE_PRIM_LINE_LOOP, + PIPE_PRIM_LINES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_TRIANGLES, + PIPE_PRIM_QUADS, + PIPE_PRIM_QUAD_STRIP, + PIPE_PRIM_TRIANGLES +}; + +static void populate_key( struct brw_context *brw, + struct brw_gs_prog_key *key ) +{ + memset(key, 0, sizeof(*key)); + + /* CACHE_NEW_VS_PROG */ + key->attrs = brw->vs.prog_data->outputs_written; + + /* BRW_NEW_PRIMITIVE */ + key->primitive = gs_prim[brw->primitive]; + + key->hint_gs_always = 0; /* debug code? */ + + key->need_gs_prog = (key->hint_gs_always || + brw->primitive == PIPE_PRIM_QUADS || + brw->primitive == PIPE_PRIM_QUAD_STRIP || + brw->primitive == PIPE_PRIM_LINE_LOOP); +} + +/* Calculate interpolants for triangle and line rasterization. + */ +static void upload_gs_prog( struct brw_context *brw ) +{ + struct brw_gs_prog_key key; + + /* Populate the key: + */ + populate_key(brw, &key); + + if (brw->gs.prog_active != key.need_gs_prog) { + brw->state.dirty.cache |= CACHE_NEW_GS_PROG; + brw->gs.prog_active = key.need_gs_prog; + } + + if (brw->gs.prog_active) { + if (!search_cache(brw, &key)) + compile_gs_prog( brw, &key ); + } +} + + +const struct brw_tracked_state brw_gs_prog = { + .dirty = { + .brw = BRW_NEW_PRIMITIVE, + .cache = CACHE_NEW_VS_PROG + }, + .update = upload_gs_prog +}; diff --git a/src/gallium/drivers/i965simple/brw_gs.h b/src/gallium/drivers/i965simple/brw_gs.h new file mode 100644 index 0000000000..f09141c6aa --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_gs.h @@ -0,0 +1,75 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_GS_H +#define BRW_GS_H + + +#include "brw_context.h" +#include "brw_eu.h" + +#define MAX_GS_VERTS (4) + +struct brw_gs_prog_key { + unsigned attrs:32; + unsigned primitive:4; + unsigned hint_gs_always:1; + unsigned need_gs_prog:1; + unsigned pad:26; +}; + +struct brw_gs_compile { + struct brw_compile func; + struct brw_gs_prog_key key; + struct brw_gs_prog_data prog_data; + + struct { + struct brw_reg R0; + struct brw_reg vertex[MAX_GS_VERTS]; + } reg; + + /* 3 different ways of expressing vertex size: + */ + unsigned nr_attrs; + unsigned nr_regs; + unsigned nr_bytes; +}; + +#define ATTR_SIZE (4*4) + +void brw_gs_quads( struct brw_gs_compile *c ); +void brw_gs_quad_strip( struct brw_gs_compile *c ); +void brw_gs_tris( struct brw_gs_compile *c ); +void brw_gs_lines( struct brw_gs_compile *c ); +void brw_gs_points( struct brw_gs_compile *c ); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_gs_emit.c b/src/gallium/drivers/i965simple/brw_gs_emit.c new file mode 100644 index 0000000000..c3cc90b10f --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_gs_emit.c @@ -0,0 +1,148 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_gs.h" + +static void brw_gs_alloc_regs( struct brw_gs_compile *c, + unsigned nr_verts ) +{ + unsigned i = 0,j; + + /* Register usage is static, precompute here: + */ + c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; + + /* Payload vertices plus space for more generated vertices: + */ + for (j = 0; j < nr_verts; j++) { + c->reg.vertex[j] = brw_vec4_grf(i, 0); + i += c->nr_regs; + } + + c->prog_data.urb_read_length = c->nr_regs; + c->prog_data.total_grf = i; +} + + +static void brw_gs_emit_vue(struct brw_gs_compile *c, + struct brw_reg vert, + boolean last, + unsigned header) +{ + struct brw_compile *p = &c->func; + boolean allocate = !last; + + /* Overwrite PrimType and PrimStart in the message header, for + * each vertex in turn: + */ + brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header)); + + /* Copy the vertex from vertn into m1..mN+1: + */ + brw_copy8(p, brw_message_reg(1), vert, c->nr_regs); + + /* Send each vertex as a seperate write to the urb. This is + * different to the concept in brw_sf_emit.c, where subsequent + * writes are used to build up a single urb entry. Each of these + * writes instantiates a seperate urb entry, and a new one must be + * allocated each time. + */ + brw_urb_WRITE(p, + allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD), + 0, + c->reg.R0, + allocate, + 1, /* used */ + c->nr_regs + 1, /* msg length */ + allocate ? 1 : 0, /* response length */ + allocate ? 0 : 1, /* eot */ + 1, /* writes_complete */ + 0, /* urb offset */ + BRW_URB_SWIZZLE_NONE); +} + + + +void brw_gs_quads( struct brw_gs_compile *c ) +{ + brw_gs_alloc_regs(c, 4); + + /* Use polygons for correct edgeflag behaviour. Note that vertex 3 + * is the PV for quads, but vertex 0 for polygons: + */ + brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); +} + +void brw_gs_quad_strip( struct brw_gs_compile *c ) +{ + brw_gs_alloc_regs(c, 4); + + brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); + brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END)); +} + +void brw_gs_tris( struct brw_gs_compile *c ) +{ + brw_gs_alloc_regs(c, 3); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_TRILIST << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_TRILIST << 2)); + brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_TRILIST << 2) | R02_PRIM_END)); +} + +void brw_gs_lines( struct brw_gs_compile *c ) +{ + brw_gs_alloc_regs(c, 2); + brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_START)); + brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_END)); +} + +void brw_gs_points( struct brw_gs_compile *c ) +{ + brw_gs_alloc_regs(c, 1); + brw_gs_emit_vue(c, c->reg.vertex[0], 1, ((_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END)); +} + + + + + + + + diff --git a/src/gallium/drivers/i965simple/brw_gs_state.c b/src/gallium/drivers/i965simple/brw_gs_state.c new file mode 100644 index 0000000000..3932e9e939 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_gs_state.c @@ -0,0 +1,89 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "pipe/p_util.h" + + + +static void upload_gs_unit( struct brw_context *brw ) +{ + struct brw_gs_unit_state gs; + + memset(&gs, 0, sizeof(gs)); + + /* CACHE_NEW_GS_PROG */ + if (brw->gs.prog_active) { + gs.thread0.grf_reg_count = + align(brw->gs.prog_data->total_grf, 16) / 16 - 1; + gs.thread0.kernel_start_pointer = brw->gs.prog_gs_offset >> 6; + gs.thread3.urb_entry_read_length = brw->gs.prog_data->urb_read_length; + } + else { + gs.thread0.grf_reg_count = 0; + gs.thread0.kernel_start_pointer = 0; + gs.thread3.urb_entry_read_length = 1; + } + + /* BRW_NEW_URB_FENCE */ + gs.thread4.nr_urb_entries = brw->urb.nr_gs_entries; + gs.thread4.urb_entry_allocation_size = brw->urb.vsize - 1; + + gs.thread4.max_threads = 0; /* Hardware requirement */ + + if (BRW_DEBUG & DEBUG_STATS) + gs.thread4.stats_enable = 1; + + /* CONSTANT */ + gs.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; + gs.thread1.single_program_flow = 1; + gs.thread3.dispatch_grf_start_reg = 1; + gs.thread3.const_urb_entry_read_offset = 0; + gs.thread3.const_urb_entry_read_length = 0; + gs.thread3.urb_entry_read_offset = 0; + + + brw->gs.state_gs_offset = brw_cache_data( &brw->cache[BRW_GS_UNIT], &gs ); +} + + +const struct brw_tracked_state brw_gs_unit = { + .dirty = { + .brw = (BRW_NEW_CURBE_OFFSETS | + BRW_NEW_URB_FENCE), + .cache = CACHE_NEW_GS_PROG + }, + .update = upload_gs_unit +}; diff --git a/src/gallium/drivers/i965simple/brw_misc_state.c b/src/gallium/drivers/i965simple/brw_misc_state.c new file mode 100644 index 0000000000..925049ecc1 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_misc_state.c @@ -0,0 +1,486 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_batch.h" +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + + + + + +/*********************************************************************** + * Blend color + */ + +static void upload_blend_constant_color(struct brw_context *brw) +{ + struct brw_blend_constant_color bcc; + + memset(&bcc, 0, sizeof(bcc)); + bcc.header.opcode = CMD_BLEND_CONSTANT_COLOR; + bcc.header.length = sizeof(bcc)/4-2; + bcc.blend_constant_color[0] = brw->attribs.BlendColor.color[0]; + bcc.blend_constant_color[1] = brw->attribs.BlendColor.color[1]; + bcc.blend_constant_color[2] = brw->attribs.BlendColor.color[2]; + bcc.blend_constant_color[3] = brw->attribs.BlendColor.color[3]; + + BRW_CACHED_BATCH_STRUCT(brw, &bcc); +} + + +const struct brw_tracked_state brw_blend_constant_color = { + .dirty = { + .brw = BRW_NEW_BLEND, + .cache = 0 + }, + .update = upload_blend_constant_color +}; + + +/*********************************************************************** + * Drawing rectangle + */ +static void upload_drawing_rect(struct brw_context *brw) +{ + struct brw_drawrect bdr; + + memset(&bdr, 0, sizeof(bdr)); + bdr.header.opcode = CMD_DRAW_RECT; + bdr.header.length = sizeof(bdr)/4 - 2; + bdr.xmin = 0; + bdr.ymin = 0; + bdr.xmax = brw->attribs.FrameBuffer.cbufs[0]->width; + bdr.ymax = brw->attribs.FrameBuffer.cbufs[0]->height; + bdr.xorg = 0; + bdr.yorg = 0; + + /* Can't use BRW_CACHED_BATCH_STRUCT because this is also emitted + * uncached in brw_draw.c: + */ + BRW_BATCH_STRUCT(brw, &bdr); +} + +const struct brw_tracked_state brw_drawing_rect = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_drawing_rect +}; + +/** + * Upload the binding table pointers, which point each stage's array of surface + * state pointers. + * + * The binding table pointers are relative to the surface state base address, + * which is the BRW_SS_POOL cache buffer. + */ +static void upload_binding_table_pointers(struct brw_context *brw) +{ + struct brw_binding_table_pointers btp; + memset(&btp, 0, sizeof(btp)); + + btp.header.opcode = CMD_BINDING_TABLE_PTRS; + btp.header.length = sizeof(btp)/4 - 2; + btp.vs = 0; + btp.gs = 0; + btp.clp = 0; + btp.sf = 0; + btp.wm = brw->wm.bind_ss_offset; + + BRW_CACHED_BATCH_STRUCT(brw, &btp); +} + +const struct brw_tracked_state brw_binding_table_pointers = { + .dirty = { + .brw = 0, + .cache = CACHE_NEW_SURF_BIND + }, + .update = upload_binding_table_pointers, +}; + + +/** + * Upload pointers to the per-stage state. + * + * The state pointers in this packet are all relative to the general state + * base address set by CMD_STATE_BASE_ADDRESS, which is the BRW_GS_POOL buffer. + */ +static void upload_pipelined_state_pointers(struct brw_context *brw ) +{ + struct brw_pipelined_state_pointers psp; + memset(&psp, 0, sizeof(psp)); + + psp.header.opcode = CMD_PIPELINED_STATE_POINTERS; + psp.header.length = sizeof(psp)/4 - 2; + + psp.vs.offset = brw->vs.state_gs_offset >> 5; + psp.sf.offset = brw->sf.state_gs_offset >> 5; + psp.wm.offset = brw->wm.state_gs_offset >> 5; + psp.cc.offset = brw->cc.state_gs_offset >> 5; + + /* GS gets turned on and off regularly. Need to re-emit URB fence + * after this occurs. + */ + if (brw->gs.prog_active) { + psp.gs.offset = brw->gs.state_gs_offset >> 5; + psp.gs.enable = 1; + } + + if (0) { + psp.clp.offset = brw->clip.state_gs_offset >> 5; + psp.clp.enable = 1; + } + + + if (BRW_CACHED_BATCH_STRUCT(brw, &psp)) + brw->state.dirty.brw |= BRW_NEW_PSP; +} + +const struct brw_tracked_state brw_pipelined_state_pointers = { + .dirty = { + .brw = 0, + .cache = (CACHE_NEW_VS_UNIT | + CACHE_NEW_GS_UNIT | + CACHE_NEW_GS_PROG | + CACHE_NEW_CLIP_UNIT | + CACHE_NEW_SF_UNIT | + CACHE_NEW_WM_UNIT | + CACHE_NEW_CC_UNIT) + }, + .update = upload_pipelined_state_pointers +}; + +static void upload_psp_urb_cbs(struct brw_context *brw ) +{ + upload_pipelined_state_pointers(brw); + brw_upload_urb_fence(brw); + brw_upload_constant_buffer_state(brw); +} + + +const struct brw_tracked_state brw_psp_urb_cbs = { + .dirty = { + .brw = BRW_NEW_URB_FENCE, + .cache = (CACHE_NEW_VS_UNIT | + CACHE_NEW_GS_UNIT | + CACHE_NEW_GS_PROG | + CACHE_NEW_CLIP_UNIT | + CACHE_NEW_SF_UNIT | + CACHE_NEW_WM_UNIT | + CACHE_NEW_CC_UNIT) + }, + .update = upload_psp_urb_cbs +}; + +/** + * Upload the depthbuffer offset and format. + * + * We have to do this per state validation as we need to emit the relocation + * in the batch buffer. + */ +static void upload_depthbuffer(struct brw_context *brw) +{ + struct pipe_surface *depth_surface = brw->attribs.FrameBuffer.zsbuf; + + BEGIN_BATCH(5, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (5 - 2)); + if (depth_surface == NULL) { + OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) | + (BRW_SURFACE_NULL << 29)); + OUT_BATCH(0); + OUT_BATCH(0); + OUT_BATCH(0); + } else { + unsigned int format; + + switch (depth_surface->cpp) { + case 2: + format = BRW_DEPTHFORMAT_D16_UNORM; + break; + case 4: + if (depth_surface->format == PIPE_FORMAT_Z32_FLOAT) + format = BRW_DEPTHFORMAT_D32_FLOAT; + else + format = BRW_DEPTHFORMAT_D24_UNORM_S8_UINT; + break; + default: + assert(0); + return; + } + + OUT_BATCH(((depth_surface->pitch * depth_surface->cpp) - 1) | + (format << 18) | + (BRW_TILEWALK_YMAJOR << 26) | +// (depth_surface->region->tiled << 27) | + (BRW_SURFACE_2D << 29)); + OUT_RELOC(depth_surface->buffer, + PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE, 0); + OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) | + ((depth_surface->pitch - 1) << 6) | + ((depth_surface->height - 1) << 19)); + OUT_BATCH(0); + } + ADVANCE_BATCH(); +} + +const struct brw_tracked_state brw_depthbuffer = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_depthbuffer, +}; + + + + +/*********************************************************************** + * Polygon stipple packet + */ + +static void upload_polygon_stipple(struct brw_context *brw) +{ + struct brw_polygon_stipple bps; + unsigned i; + + memset(&bps, 0, sizeof(bps)); + bps.header.opcode = CMD_POLY_STIPPLE_PATTERN; + bps.header.length = sizeof(bps)/4-2; + + /* XXX: state tracker should send *all* state down initially! + */ + if (brw->attribs.PolygonStipple) + for (i = 0; i < 32; i++) + bps.stipple[i] = brw->attribs.PolygonStipple->stipple[31 - i]; /* invert */ + + BRW_CACHED_BATCH_STRUCT(brw, &bps); +} + +const struct brw_tracked_state brw_polygon_stipple = { + .dirty = { + .brw = BRW_NEW_STIPPLE, + .cache = 0 + }, + .update = upload_polygon_stipple +}; + + +/*********************************************************************** + * Line stipple packet + */ + +static void upload_line_stipple(struct brw_context *brw) +{ + struct brw_line_stipple bls; + float tmp; + int tmpi; + + memset(&bls, 0, sizeof(bls)); + bls.header.opcode = CMD_LINE_STIPPLE_PATTERN; + bls.header.length = sizeof(bls)/4 - 2; + + bls.bits0.pattern = brw->attribs.Raster->line_stipple_pattern; + bls.bits1.repeat_count = brw->attribs.Raster->line_stipple_factor; + + tmp = 1.0 / (float) brw->attribs.Raster->line_stipple_factor; + tmpi = tmp * (1<<13); + + + bls.bits1.inverse_repeat_count = tmpi; + + BRW_CACHED_BATCH_STRUCT(brw, &bls); +} + +const struct brw_tracked_state brw_line_stipple = { + .dirty = { + .brw = BRW_NEW_STIPPLE, + .cache = 0 + }, + .update = upload_line_stipple +}; + + +/*********************************************************************** + * Misc constant state packets + */ + +static void upload_pipe_control(struct brw_context *brw) +{ + struct brw_pipe_control pc; + + return; + + memset(&pc, 0, sizeof(pc)); + + pc.header.opcode = CMD_PIPE_CONTROL; + pc.header.length = sizeof(pc)/4 - 2; + pc.header.post_sync_operation = PIPE_CONTROL_NOWRITE; + + pc.header.instruction_state_cache_flush_enable = 1; + + pc.bits1.dest_addr_type = PIPE_CONTROL_GTTWRITE_GLOBAL; + + BRW_BATCH_STRUCT(brw, &pc); +} + +const struct brw_tracked_state brw_pipe_control = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_pipe_control +}; + + +/*********************************************************************** + * Misc invarient state packets + */ + +static void upload_invarient_state( struct brw_context *brw ) +{ + { + struct brw_mi_flush flush; + + memset(&flush, 0, sizeof(flush)); + flush.opcode = CMD_MI_FLUSH; + flush.flags = BRW_FLUSH_STATE_CACHE | BRW_FLUSH_READ_CACHE; + BRW_BATCH_STRUCT(brw, &flush); + } + + { + /* 0x61040000 Pipeline Select */ + /* PipelineSelect : 0 */ + struct brw_pipeline_select ps; + + memset(&ps, 0, sizeof(ps)); + ps.header.opcode = CMD_PIPELINE_SELECT; + ps.header.pipeline_select = 0; + BRW_BATCH_STRUCT(brw, &ps); + } + + { + struct brw_global_depth_offset_clamp gdo; + memset(&gdo, 0, sizeof(gdo)); + + /* Disable depth offset clamping. + */ + gdo.header.opcode = CMD_GLOBAL_DEPTH_OFFSET_CLAMP; + gdo.header.length = sizeof(gdo)/4 - 2; + gdo.depth_offset_clamp = 0.0; + + BRW_BATCH_STRUCT(brw, &gdo); + } + + + /* 0x61020000 State Instruction Pointer */ + { + struct brw_system_instruction_pointer sip; + memset(&sip, 0, sizeof(sip)); + + sip.header.opcode = CMD_STATE_INSN_POINTER; + sip.header.length = 0; + sip.bits0.pad = 0; + sip.bits0.system_instruction_pointer = 0; + BRW_BATCH_STRUCT(brw, &sip); + } + + + { + struct brw_vf_statistics vfs; + memset(&vfs, 0, sizeof(vfs)); + + vfs.opcode = CMD_VF_STATISTICS; + if (BRW_DEBUG & DEBUG_STATS) + vfs.statistics_enable = 1; + + BRW_BATCH_STRUCT(brw, &vfs); + } + + + { + struct brw_polygon_stipple_offset bpso; + + memset(&bpso, 0, sizeof(bpso)); + bpso.header.opcode = CMD_POLY_STIPPLE_OFFSET; + bpso.header.length = sizeof(bpso)/4-2; + bpso.bits0.x_offset = 0; + bpso.bits0.y_offset = 0; + + BRW_BATCH_STRUCT(brw, &bpso); + } +} + +const struct brw_tracked_state brw_invarient_state = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_invarient_state +}; + +/** + * Define the base addresses which some state is referenced from. + * + * This allows us to avoid having to emit relocations in many places for + * cached state, and instead emit pointers inside of large, mostly-static + * state pools. This comes at the expense of memory, and more expensive cache + * misses. + */ +static void upload_state_base_address( struct brw_context *brw ) +{ + /* Output the structure (brw_state_base_address) directly to the + * batchbuffer, so we can emit relocations inline. + */ + BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS); + OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2)); + OUT_RELOC(brw->pool[BRW_GS_POOL].buffer, + PIPE_BUFFER_USAGE_GPU_READ, + 1); /* General state base address */ + OUT_RELOC(brw->pool[BRW_SS_POOL].buffer, + PIPE_BUFFER_USAGE_GPU_READ, + 1); /* Surface state base address */ + OUT_BATCH(1); /* Indirect object base address */ + OUT_BATCH(1); /* General state upper bound */ + OUT_BATCH(1); /* Indirect object upper bound */ + ADVANCE_BATCH(); +} + + +const struct brw_tracked_state brw_state_base_address = { + .dirty = { + .brw = BRW_NEW_SCENE, + .cache = 0 + }, + .update = upload_state_base_address +}; diff --git a/src/gallium/drivers/i965simple/brw_reg.h b/src/gallium/drivers/i965simple/brw_reg.h new file mode 100644 index 0000000000..9e885c3b3b --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_reg.h @@ -0,0 +1,76 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#define CMD_MI (0x0 << 29) +#define CMD_2D (0x2 << 29) +#define CMD_3D (0x3 << 29) + +#define MI_BATCH_BUFFER_END (CMD_MI | 0xA << 23) + +/* Stalls command execution waiting for the given events to have occurred. */ +#define MI_WAIT_FOR_EVENT (CMD_MI | (0x3 << 23)) +#define MI_WAIT_FOR_PLANE_B_FLIP (1<<6) +#define MI_WAIT_FOR_PLANE_A_FLIP (1<<2) + +/* Primitive dispatch on 830-945 */ +#define _3DPRIMITIVE (CMD_3D | (0x1f << 24)) +#define PRIM_INDIRECT (1<<23) +#define PRIM_INLINE (0<<23) +#define PRIM_INDIRECT_SEQUENTIAL (0<<17) +#define PRIM_INDIRECT_ELTS (1<<17) + +#define PRIM3D_TRILIST (0x0<<18) +#define PRIM3D_TRISTRIP (0x1<<18) +#define PRIM3D_TRISTRIP_RVRSE (0x2<<18) +#define PRIM3D_TRIFAN (0x3<<18) +#define PRIM3D_POLY (0x4<<18) +#define PRIM3D_LINELIST (0x5<<18) +#define PRIM3D_LINESTRIP (0x6<<18) +#define PRIM3D_RECTLIST (0x7<<18) +#define PRIM3D_POINTLIST (0x8<<18) +#define PRIM3D_DIB (0x9<<18) +#define PRIM3D_MASK (0x1f<<18) + +#define XY_SETUP_BLT_CMD (CMD_2D | (0x01 << 22) | 6) + +#define XY_COLOR_BLT_CMD (CMD_2D | (0x50 << 22) | 4) + +#define XY_SRC_COPY_BLT_CMD (CMD_2D | (0x53 << 22) | 6) + +/* BR00 */ +#define XY_BLT_WRITE_ALPHA (1 << 21) +#define XY_BLT_WRITE_RGB (1 << 20) +#define XY_SRC_TILED (1 << 15) +#define XY_DST_TILED (1 << 11) + +/* BR13 */ +#define BR13_565 (0x1 << 24) +#define BR13_8888 (0x3 << 24) + +#define FENCE_LINEAR 0 +#define FENCE_XMAJOR 1 +#define FENCE_YMAJOR 2 diff --git a/src/gallium/drivers/i965simple/brw_sf.c b/src/gallium/drivers/i965simple/brw_sf.c new file mode 100644 index 0000000000..7c83b81c85 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_sf.c @@ -0,0 +1,351 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_sf.h" +#include "brw_state.h" +#include "tgsi/util/tgsi_parse.h" + + +static void compile_sf_prog( struct brw_context *brw, + struct brw_sf_prog_key *key ) +{ + struct brw_sf_compile c; + const unsigned *program; + unsigned program_size; + + memset(&c, 0, sizeof(c)); + + /* Begin the compilation: + */ + brw_init_compile(&c.func); + + c.key = *key; + + + c.nr_attrs = c.key.vp_output_count; + c.nr_attr_regs = (c.nr_attrs+1)/2; + + c.nr_setup_attrs = c.key.fp_input_count + 1; /* +1 for position */ + c.nr_setup_regs = (c.nr_setup_attrs+1)/2; + + c.prog_data.urb_read_length = c.nr_attr_regs; + c.prog_data.urb_entry_size = c.nr_setup_regs * 2; + + + /* Which primitive? Or all three? + */ + switch (key->primitive) { + case SF_TRIANGLES: + c.nr_verts = 3; + brw_emit_tri_setup( &c ); + break; + case SF_LINES: + c.nr_verts = 2; + brw_emit_line_setup( &c ); + break; + case SF_POINTS: + c.nr_verts = 1; + brw_emit_point_setup( &c ); + break; + + case SF_UNFILLED_TRIS: + default: + assert(0); + return; + } + + + + /* get the program + */ + program = brw_get_program(&c.func, &program_size); + + /* Upload + */ + brw->sf.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_SF_PROG], + &c.key, + sizeof(c.key), + program, + program_size, + &c.prog_data, + &brw->sf.prog_data ); +} + + +static boolean search_cache( struct brw_context *brw, + struct brw_sf_prog_key *key ) +{ + return brw_search_cache(&brw->cache[BRW_SF_PROG], + key, sizeof(*key), + &brw->sf.prog_data, + &brw->sf.prog_gs_offset); +} + + +/* Calculate interpolants for triangle and line rasterization. + */ +static void upload_sf_prog( struct brw_context *brw ) +{ + const struct brw_fragment_program *fs = brw->attribs.FragmentProgram; + struct brw_sf_prog_key key; + struct tgsi_parse_context parse; + int i, done = 0; + + + memset(&key, 0, sizeof(key)); + + /* Populate the key, noting state dependencies: + */ + /* CACHE_NEW_VS_PROG */ + key.vp_output_count = brw->vs.prog_data->outputs_written; + + /* BRW_NEW_FS */ + key.fp_input_count = brw->attribs.FragmentProgram->info.nr_regs[TGSI_FILE_INPUT]; + + + /* BRW_NEW_REDUCED_PRIMITIVE */ + switch (brw->reduced_primitive) { + case PIPE_PRIM_TRIANGLES: +// if (key.attrs & (1<<VERT_RESULT_EDGE)) +// key.primitive = SF_UNFILLED_TRIS; +// else + key.primitive = SF_TRIANGLES; + break; + case PIPE_PRIM_LINES: + key.primitive = SF_LINES; + break; + case PIPE_PRIM_POINTS: + key.primitive = SF_POINTS; + break; + } + + + + /* Scan fp inputs to figure out what interpolation modes are + * required for each incoming vp output. There is an assumption + * that the state tracker makes sure there is a 1:1 linkage between + * these sets of attributes (XXX: position??) + */ + tgsi_parse_init( &parse, fs->program.tokens ); + while( !done && + !tgsi_parse_end_of_tokens( &parse ) ) + { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_INPUT) + { + int first = parse.FullToken.FullDeclaration.u.DeclarationRange.First; + int last = parse.FullToken.FullDeclaration.u.DeclarationRange.Last; + int interp_mode = parse.FullToken.FullDeclaration.Interpolation.Interpolate; + //int semantic = parse.FullToken.FullDeclaration.Semantic.SemanticName; + //int semantic_index = parse.FullToken.FullDeclaration.Semantic.SemanticIndex; + + debug_printf("fs input %d..%d interp mode %d\n", first, last, interp_mode); + + switch (interp_mode) { + case TGSI_INTERPOLATE_CONSTANT: + for (i = first; i <= last; i++) + key.const_mask |= (1 << i); + break; + case TGSI_INTERPOLATE_LINEAR: + for (i = first; i <= last; i++) + key.linear_mask |= (1 << i); + break; + case TGSI_INTERPOLATE_PERSPECTIVE: + for (i = first; i <= last; i++) + key.persp_mask |= (1 << i); + break; + default: + break; + } + + /* Also need stuff for flat shading, twosided color. + */ + + } + break; + default: + done = 1; + break; + } + } + + /* Hack: Adjust for position. Optimize away when not required (ie + * for perspective interpolation). + */ + key.persp_mask <<= 1; + key.linear_mask <<= 1; + key.linear_mask |= 1; + key.const_mask <<= 1; + + debug_printf("key.persp_mask: %x\n", key.persp_mask); + debug_printf("key.linear_mask: %x\n", key.linear_mask); + debug_printf("key.const_mask: %x\n", key.const_mask); + + +// key.do_point_sprite = brw->attribs.Point->PointSprite; +// key.SpriteOrigin = brw->attribs.Point->SpriteOrigin; + +// key.do_flat_shading = (brw->attribs.Raster->flatshade); +// key.do_twoside_color = (brw->attribs.Light->Enabled && brw->attribs.Light->Model.TwoSide); + +// if (key.do_twoside_color) +// key.frontface_ccw = (brw->attribs.Polygon->FrontFace == GL_CCW); + + + if (!search_cache(brw, &key)) + compile_sf_prog( brw, &key ); +} + + +const struct brw_tracked_state brw_sf_prog = { + .dirty = { + .brw = (BRW_NEW_RASTERIZER | + BRW_NEW_REDUCED_PRIMITIVE | + BRW_NEW_VS | + BRW_NEW_FS), + .cache = 0, + }, + .update = upload_sf_prog +}; + + + +#if 0 +/* Build a struct like the one we'd like the state tracker to pass to + * us. + */ +static void update_sf_linkage( struct brw_context *brw ) +{ + const struct brw_vertex_program *vs = brw->attribs.VertexProgram; + const struct brw_fragment_program *fs = brw->attribs.FragmentProgram; + struct pipe_setup_linkage state; + struct tgsi_parse_context parse; + + int i, j; + int nr_vp_outputs = 0; + int done = 0; + + struct { + unsigned semantic:8; + unsigned semantic_index:16; + } fp_semantic[32], vp_semantic[32]; + + memset(&state, 0, sizeof(state)); + + state.fp_input_count = 0; + + + + + + + assert(state.fp_input_count == fs->program.num_inputs); + + + /* Then scan vp outputs + */ + done = 0; + tgsi_parse_init( &parse, vs->program.tokens ); + while( !done && + !tgsi_parse_end_of_tokens( &parse ) ) + { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_INPUT) + { + int first = parse.FullToken.FullDeclaration.u.DeclarationRange.First; + int last = parse.FullToken.FullDeclaration.u.DeclarationRange.Last; + + for (i = first; i < last; i++) { + vp_semantic[i].semantic = + parse.FullToken.FullDeclaration.Semantic.SemanticName; + vp_semantic[i].semantic_index = + parse.FullToken.FullDeclaration.Semantic.SemanticIndex; + } + + assert(last > nr_vp_outputs); + nr_vp_outputs = last; + } + break; + default: + done = 1; + break; + } + } + + + /* Now match based on semantic information. + */ + for (i = 0; i< state.fp_input_count; i++) { + for (j = 0; j < nr_vp_outputs; j++) { + if (fp_semantic[i].semantic == vp_semantic[j].semantic && + fp_semantic[i].semantic_index == vp_semantic[j].semantic_index) { + state.fp_input[i].vp_output = j; + } + } + if (fp_semantic[i].semantic == TGSI_SEMANTIC_COLOR) { + for (j = 0; j < nr_vp_outputs; j++) { + if (TGSI_SEMANTIC_BCOLOR == vp_semantic[j].semantic && + fp_semantic[i].semantic_index == vp_semantic[j].semantic_index) { + state.fp_input[i].bf_vp_output = j; + } + } + } + } + + if (memcmp(&brw->sf.linkage, &state, sizeof(state)) != 0) { + brw->sf.linkage = state; + brw->state.dirty.brw |= BRW_NEW_SF_LINKAGE; + } +} + + +const struct brw_tracked_state brw_sf_linkage = { + .dirty = { + .brw = (BRW_NEW_VS | + BRW_NEW_FS), + .cache = 0, + }, + .update = update_sf_linkage +}; + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_sf.h b/src/gallium/drivers/i965simple/brw_sf.h new file mode 100644 index 0000000000..b7ada47560 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_sf.h @@ -0,0 +1,122 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_SF_H +#define BRW_SF_H + +#include "brw_context.h" +#include "brw_eu.h" + + +#define SF_POINTS 0 +#define SF_LINES 1 +#define SF_TRIANGLES 2 +#define SF_UNFILLED_TRIS 3 + + + +struct brw_sf_prog_key { + unsigned vp_output_count:5; + unsigned fp_input_count:5; + + unsigned primitive:2; + unsigned do_twoside_color:1; + unsigned do_flat_shading:1; + unsigned frontface_ccw:1; + unsigned do_point_sprite:1; + + /* Interpolation masks; + */ + unsigned linear_mask; + unsigned persp_mask; + unsigned const_mask; + + +// int SpriteOrigin; +}; + +struct brw_sf_point_tex { + boolean CoordReplace; +}; + +struct brw_sf_compile { + struct brw_compile func; + struct brw_sf_prog_key key; + struct brw_sf_prog_data prog_data; + + struct brw_reg pv; + struct brw_reg det; + struct brw_reg dx0; + struct brw_reg dx2; + struct brw_reg dy0; + struct brw_reg dy2; + + /* z and 1/w passed in seperately: + */ + struct brw_reg z[3]; + struct brw_reg inv_w[3]; + + /* The vertices: + */ + struct brw_reg vert[3]; + + /* Temporaries, allocated after last vertex reg. + */ + struct brw_reg inv_det; + struct brw_reg a1_sub_a0; + struct brw_reg a2_sub_a0; + struct brw_reg tmp; + + struct brw_reg m1Cx; + struct brw_reg m2Cy; + struct brw_reg m3C0; + + unsigned nr_verts; + unsigned nr_attrs; + unsigned nr_attr_regs; + unsigned nr_setup_attrs; + unsigned nr_setup_regs; +#if 0 + ubyte attr_to_idx[VERT_RESULT_MAX]; + ubyte idx_to_attr[VERT_RESULT_MAX]; + struct brw_sf_point_tex point_attrs[VERT_RESULT_MAX]; +#endif +}; + + +void brw_emit_tri_setup( struct brw_sf_compile *c ); +void brw_emit_line_setup( struct brw_sf_compile *c ); +void brw_emit_point_setup( struct brw_sf_compile *c ); +void brw_emit_point_sprite_setup( struct brw_sf_compile *c ); +void brw_emit_anyprim_setup( struct brw_sf_compile *c ); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_sf_emit.c b/src/gallium/drivers/i965simple/brw_sf_emit.c new file mode 100644 index 0000000000..78d6fa5e9e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_sf_emit.c @@ -0,0 +1,382 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_defines.h" +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_util.h" +#include "brw_sf.h" + + + +/*********************************************************************** + * Triangle setup. + */ + + +static void alloc_regs( struct brw_sf_compile *c ) +{ + unsigned reg, i; + + /* Values computed by fixed function unit: + */ + c->pv = retype(brw_vec1_grf(1, 1), BRW_REGISTER_TYPE_UD); + c->det = brw_vec1_grf(1, 2); + c->dx0 = brw_vec1_grf(1, 3); + c->dx2 = brw_vec1_grf(1, 4); + c->dy0 = brw_vec1_grf(1, 5); + c->dy2 = brw_vec1_grf(1, 6); + + /* z and 1/w passed in seperately: + */ + c->z[0] = brw_vec1_grf(2, 0); + c->inv_w[0] = brw_vec1_grf(2, 1); + c->z[1] = brw_vec1_grf(2, 2); + c->inv_w[1] = brw_vec1_grf(2, 3); + c->z[2] = brw_vec1_grf(2, 4); + c->inv_w[2] = brw_vec1_grf(2, 5); + + /* The vertices: + */ + reg = 3; + for (i = 0; i < c->nr_verts; i++) { + c->vert[i] = brw_vec8_grf(reg, 0); + reg += c->nr_attr_regs; + } + + /* Temporaries, allocated after last vertex reg. + */ + c->inv_det = brw_vec1_grf(reg, 0); reg++; + c->a1_sub_a0 = brw_vec8_grf(reg, 0); reg++; + c->a2_sub_a0 = brw_vec8_grf(reg, 0); reg++; + c->tmp = brw_vec8_grf(reg, 0); reg++; + + /* Note grf allocation: + */ + c->prog_data.total_grf = reg; + + + /* Outputs of this program - interpolation coefficients for + * rasterization: + */ + c->m1Cx = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 1, 0); + c->m2Cy = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 2, 0); + c->m3C0 = brw_vec8_reg(BRW_MESSAGE_REGISTER_FILE, 3, 0); +} + + +static void copy_z_inv_w( struct brw_sf_compile *c ) +{ + struct brw_compile *p = &c->func; + unsigned i; + + brw_push_insn_state(p); + + /* Copy both scalars with a single MOV: + */ + for (i = 0; i < c->nr_verts; i++) + brw_MOV(p, vec2(suboffset(c->vert[i], 2)), vec2(c->z[i])); + + brw_pop_insn_state(p); +} + + +static void invert_det( struct brw_sf_compile *c) +{ + brw_math(&c->func, + c->inv_det, + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 0, + c->det, + BRW_MATH_DATA_SCALAR, + BRW_MATH_PRECISION_FULL); + +} + +#define NON_PERPECTIVE_ATTRS (FRAG_BIT_WPOS | \ + FRAG_BIT_COL0 | \ + FRAG_BIT_COL1) + +static boolean calculate_masks( struct brw_sf_compile *c, + unsigned reg, + ushort *pc, + ushort *pc_persp, + ushort *pc_linear) +{ + boolean is_last_attr = (reg == c->nr_setup_regs - 1); + unsigned persp_mask = c->key.persp_mask; + unsigned linear_mask = c->key.linear_mask; + + debug_printf("persp_mask: %x\n", persp_mask); + debug_printf("linear_mask: %x\n", linear_mask); + + *pc_persp = 0; + *pc_linear = 0; + *pc = 0xf; + + if (persp_mask & (1 << (reg*2))) + *pc_persp = 0xf; + + if (linear_mask & (1 << (reg*2))) + *pc_linear = 0xf; + + /* Maybe only processs one attribute on the final round: + */ + if (reg*2+1 < c->nr_setup_attrs) { + *pc |= 0xf0; + + if (persp_mask & (1 << (reg*2+1))) + *pc_persp |= 0xf0; + + if (linear_mask & (1 << (reg*2+1))) + *pc_linear |= 0xf0; + } + + debug_printf("pc: %x\n", *pc); + debug_printf("pc_persp: %x\n", *pc_persp); + debug_printf("pc_linear: %x\n", *pc_linear); + + + return is_last_attr; +} + + + +void brw_emit_tri_setup( struct brw_sf_compile *c ) +{ + struct brw_compile *p = &c->func; + unsigned i; + + debug_printf("%s START ==============\n", __FUNCTION__); + + c->nr_verts = 3; + alloc_regs(c); + invert_det(c); + copy_z_inv_w(c); + + + for (i = 0; i < c->nr_setup_regs; i++) + { + /* Pair of incoming attributes: + */ + struct brw_reg a0 = offset(c->vert[0], i); + struct brw_reg a1 = offset(c->vert[1], i); + struct brw_reg a2 = offset(c->vert[2], i); + ushort pc = 0, pc_persp = 0, pc_linear = 0; + boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear); + + if (pc_persp) + { + brw_set_predicate_control_flag_value(p, pc_persp); + brw_MUL(p, a0, a0, c->inv_w[0]); + brw_MUL(p, a1, a1, c->inv_w[1]); + brw_MUL(p, a2, a2, c->inv_w[2]); + } + + + /* Calculate coefficients for interpolated values: + */ + if (pc_linear) + { + brw_set_predicate_control_flag_value(p, pc_linear); + + brw_ADD(p, c->a1_sub_a0, a1, negate(a0)); + brw_ADD(p, c->a2_sub_a0, a2, negate(a0)); + + /* calculate dA/dx + */ + brw_MUL(p, brw_null_reg(), c->a1_sub_a0, c->dy2); + brw_MAC(p, c->tmp, c->a2_sub_a0, negate(c->dy0)); + brw_MUL(p, c->m1Cx, c->tmp, c->inv_det); + + /* calculate dA/dy + */ + brw_MUL(p, brw_null_reg(), c->a2_sub_a0, c->dx0); + brw_MAC(p, c->tmp, c->a1_sub_a0, negate(c->dx2)); + brw_MUL(p, c->m2Cy, c->tmp, c->inv_det); + } + + { + brw_set_predicate_control_flag_value(p, pc); + /* start point for interpolation + */ + brw_MOV(p, c->m3C0, a0); + + /* Copy m0..m3 to URB. m0 is implicitly copied from r0 in + * the send instruction: + */ + brw_urb_WRITE(p, + brw_null_reg(), + 0, + brw_vec8_grf(0, 0), /* r0, will be copied to m0 */ + 0, /* allocate */ + 1, /* used */ + 4, /* msg len */ + 0, /* response len */ + last, /* eot */ + last, /* writes complete */ + i*4, /* offset */ + BRW_URB_SWIZZLE_TRANSPOSE); /* XXX: Swizzle control "SF to windower" */ + } + } + + debug_printf("%s DONE ==============\n", __FUNCTION__); + +} + + + +void brw_emit_line_setup( struct brw_sf_compile *c ) +{ + struct brw_compile *p = &c->func; + unsigned i; + + + c->nr_verts = 2; + alloc_regs(c); + invert_det(c); + copy_z_inv_w(c); + + for (i = 0; i < c->nr_setup_regs; i++) + { + /* Pair of incoming attributes: + */ + struct brw_reg a0 = offset(c->vert[0], i); + struct brw_reg a1 = offset(c->vert[1], i); + ushort pc, pc_persp, pc_linear; + boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear); + + if (pc_persp) + { + brw_set_predicate_control_flag_value(p, pc_persp); + brw_MUL(p, a0, a0, c->inv_w[0]); + brw_MUL(p, a1, a1, c->inv_w[1]); + } + + /* Calculate coefficients for position, color: + */ + if (pc_linear) { + brw_set_predicate_control_flag_value(p, pc_linear); + + brw_ADD(p, c->a1_sub_a0, a1, negate(a0)); + + brw_MUL(p, c->tmp, c->a1_sub_a0, c->dx0); + brw_MUL(p, c->m1Cx, c->tmp, c->inv_det); + + brw_MUL(p, c->tmp, c->a1_sub_a0, c->dy0); + brw_MUL(p, c->m2Cy, c->tmp, c->inv_det); + } + + { + brw_set_predicate_control_flag_value(p, pc); + + /* start point for interpolation + */ + brw_MOV(p, c->m3C0, a0); + + /* Copy m0..m3 to URB. + */ + brw_urb_WRITE(p, + brw_null_reg(), + 0, + brw_vec8_grf(0, 0), + 0, /* allocate */ + 1, /* used */ + 4, /* msg len */ + 0, /* response len */ + last, /* eot */ + last, /* writes complete */ + i*4, /* urb destination offset */ + BRW_URB_SWIZZLE_TRANSPOSE); + } + } +} + + +/* Points setup - several simplifications as all attributes are + * constant across the face of the point (point sprites excluded!) + */ +void brw_emit_point_setup( struct brw_sf_compile *c ) +{ + struct brw_compile *p = &c->func; + unsigned i; + + c->nr_verts = 1; + alloc_regs(c); + copy_z_inv_w(c); + + brw_MOV(p, c->m1Cx, brw_imm_ud(0)); /* zero - move out of loop */ + brw_MOV(p, c->m2Cy, brw_imm_ud(0)); /* zero - move out of loop */ + + for (i = 0; i < c->nr_setup_regs; i++) + { + struct brw_reg a0 = offset(c->vert[0], i); + ushort pc, pc_persp, pc_linear; + boolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear); + + if (pc_persp) + { + /* This seems odd as the values are all constant, but the + * fragment shader will be expecting it: + */ + brw_set_predicate_control_flag_value(p, pc_persp); + brw_MUL(p, a0, a0, c->inv_w[0]); + } + + + /* The delta values are always zero, just send the starting + * coordinate. Again, this is to fit in with the interpolation + * code in the fragment shader. + */ + { + brw_set_predicate_control_flag_value(p, pc); + + brw_MOV(p, c->m3C0, a0); /* constant value */ + + /* Copy m0..m3 to URB. + */ + brw_urb_WRITE(p, + brw_null_reg(), + 0, + brw_vec8_grf(0, 0), + 0, /* allocate */ + 1, /* used */ + 4, /* msg len */ + 0, /* response len */ + last, /* eot */ + last, /* writes complete */ + i*4, /* urb destination offset */ + BRW_URB_SWIZZLE_TRANSPOSE); + } + } +} diff --git a/src/gallium/drivers/i965simple/brw_sf_state.c b/src/gallium/drivers/i965simple/brw_sf_state.c new file mode 100644 index 0000000000..9acd3ea61b --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_sf_state.c @@ -0,0 +1,180 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "pipe/p_util.h" + +static void upload_sf_vp(struct brw_context *brw) +{ + struct brw_sf_viewport sfv; + + memset(&sfv, 0, sizeof(sfv)); + + + /* BRW_NEW_VIEWPORT */ + { + const float *scale = brw->attribs.Viewport.scale; + const float *trans = brw->attribs.Viewport.translate; + + sfv.viewport.m00 = scale[0]; + sfv.viewport.m11 = scale[1]; + sfv.viewport.m22 = scale[2]; + sfv.viewport.m30 = trans[0]; + sfv.viewport.m31 = trans[1]; + sfv.viewport.m32 = trans[2]; + } + + /* _NEW_SCISSOR */ + sfv.scissor.xmin = brw->attribs.Scissor.minx; + sfv.scissor.xmax = brw->attribs.Scissor.maxx - 1; + sfv.scissor.ymin = brw->attribs.Scissor.miny; + sfv.scissor.ymax = brw->attribs.Scissor.maxy - 1; + + brw->sf.vp_gs_offset = brw_cache_data( &brw->cache[BRW_SF_VP], &sfv ); +} + +const struct brw_tracked_state brw_sf_vp = { + .dirty = { + .brw = (BRW_NEW_SCISSOR | + BRW_NEW_VIEWPORT), + .cache = 0 + }, + .update = upload_sf_vp +}; + +static void upload_sf_unit( struct brw_context *brw ) +{ + struct brw_sf_unit_state sf; + memset(&sf, 0, sizeof(sf)); + + /* CACHE_NEW_SF_PROG */ + sf.thread0.grf_reg_count = align(brw->sf.prog_data->total_grf, 16) / 16 - 1; + sf.thread0.kernel_start_pointer = brw->sf.prog_gs_offset >> 6; + sf.thread3.urb_entry_read_length = brw->sf.prog_data->urb_read_length; + + sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; + sf.thread3.dispatch_grf_start_reg = 3; + sf.thread3.urb_entry_read_offset = 1; + + /* BRW_NEW_URB_FENCE */ + sf.thread4.nr_urb_entries = brw->urb.nr_sf_entries; + sf.thread4.urb_entry_allocation_size = brw->urb.sfsize - 1; + sf.thread4.max_threads = MIN2(12, brw->urb.nr_sf_entries / 2) - 1; + + if (BRW_DEBUG & DEBUG_SINGLE_THREAD) + sf.thread4.max_threads = 0; + + if (BRW_DEBUG & DEBUG_STATS) + sf.thread4.stats_enable = 1; + + /* CACHE_NEW_SF_VP */ + sf.sf5.sf_viewport_state_offset = brw->sf.vp_gs_offset >> 5; + sf.sf5.viewport_transform = 1; + + /* BRW_NEW_RASTER */ + if (brw->attribs.Raster->scissor) + sf.sf6.scissor = 1; + +#if 0 + if (brw->attribs.Polygon->FrontFace == GL_CCW) + sf.sf5.front_winding = BRW_FRONTWINDING_CCW; + else + sf.sf5.front_winding = BRW_FRONTWINDING_CW; + + + if (brw->attribs.Polygon->CullFlag) { + switch (brw->attribs.Polygon->CullFaceMode) { + case GL_FRONT: + sf.sf6.cull_mode = BRW_CULLMODE_FRONT; + break; + case GL_BACK: + sf.sf6.cull_mode = BRW_CULLMODE_BACK; + break; + case GL_FRONT_AND_BACK: + sf.sf6.cull_mode = BRW_CULLMODE_BOTH; + break; + default: + assert(0); + break; + } + } + else + sf.sf6.cull_mode = BRW_CULLMODE_NONE; +#else + sf.sf5.front_winding = BRW_FRONTWINDING_CCW; + sf.sf6.cull_mode = BRW_CULLMODE_NONE; +#endif + + sf.sf6.line_width = CLAMP(brw->attribs.Raster->line_width, 1.0, 5.0) * (1<<1); + + sf.sf6.line_endcap_aa_region_width = 1; + if (brw->attribs.Raster->line_smooth) + sf.sf6.aa_enable = 1; + else if (sf.sf6.line_width <= 0x2) + sf.sf6.line_width = 0; + + sf.sf6.point_rast_rule = 1; /* opengl conventions */ + + sf.sf7.sprite_point = brw->attribs.Raster->point_sprite; + sf.sf7.point_size = CLAMP(brw->attribs.Raster->line_width, 1.0, 255.0) * (1<<3); + sf.sf7.use_point_size_state = !brw->attribs.Raster->point_size_per_vertex; + + /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons: + */ + sf.sf7.trifan_pv = 2; + sf.sf7.linestrip_pv = 1; + sf.sf7.tristrip_pv = 2; + sf.sf7.line_last_pixel_enable = 0; + + /* Set bias for OpenGL rasterization rules: + */ + sf.sf6.dest_org_vbias = 0x8; + sf.sf6.dest_org_hbias = 0x8; + + brw->sf.state_gs_offset = brw_cache_data( &brw->cache[BRW_SF_UNIT], &sf ); +} + + +const struct brw_tracked_state brw_sf_unit = { + .dirty = { + .brw = (BRW_NEW_RASTERIZER | + BRW_NEW_URB_FENCE), + .cache = (CACHE_NEW_SF_VP | + CACHE_NEW_SF_PROG) + }, + .update = upload_sf_unit +}; + + diff --git a/src/gallium/drivers/i965simple/brw_shader_info.c b/src/gallium/drivers/i965simple/brw_shader_info.c new file mode 100644 index 0000000000..431b45466a --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_shader_info.c @@ -0,0 +1,49 @@ + +#include "brw_context.h" +#include "brw_state.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" + + + + +void brw_shader_info(const struct tgsi_token *tokens, + struct brw_shader_info *info ) +{ + struct tgsi_parse_context parse; + int done = 0; + + tgsi_parse_init( &parse, tokens ); + + while( !done && + !tgsi_parse_end_of_tokens( &parse ) ) + { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + { + const struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration; + unsigned last = decl->u.DeclarationRange.Last; + + assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE ); + + // Broken by crazy wpos init: + //assert( info->nr_regs[decl->Declaration.File] <= last); + + info->nr_regs[decl->Declaration.File] = MAX2(info->nr_regs[decl->Declaration.File], + last+1); + break; + } + case TGSI_TOKEN_TYPE_IMMEDIATE: + case TGSI_TOKEN_TYPE_INSTRUCTION: + default: + done = 1; + break; + } + } + + tgsi_parse_free (&parse); + +} diff --git a/src/gallium/drivers/i965simple/brw_state.c b/src/gallium/drivers/i965simple/brw_state.c new file mode 100644 index 0000000000..95dfce88e4 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state.c @@ -0,0 +1,424 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Zack Rusin <zack@tungstengraphics.com> + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_dump.h" + +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_state.h" +#include "brw_draw.h" + + +#define DUP( TYPE, VAL ) \ +do { \ + struct TYPE *x = malloc(sizeof(*x)); \ + memcpy(x, VAL, sizeof(*x) ); \ + return x; \ +} while (0) + +/************************************************************************ + * Blend + */ +static void * +brw_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + DUP( pipe_blend_state, blend ); +} + +static void brw_bind_blend_state(struct pipe_context *pipe, + void *blend) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Blend = (struct pipe_blend_state*)blend; + brw->state.dirty.brw |= BRW_NEW_BLEND; +} + + +static void brw_delete_blend_state(struct pipe_context *pipe, void *blend) +{ + free(blend); +} + +static void brw_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.BlendColor = *blend_color; + + brw->state.dirty.brw |= BRW_NEW_BLEND; +} + +/************************************************************************ + * Sampler + */ + +static void * +brw_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + DUP( pipe_sampler_state, sampler ); +} + +static void brw_bind_sampler_state(struct pipe_context *pipe, + unsigned unit, void *sampler) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Samplers[unit] = sampler; + brw->state.dirty.brw |= BRW_NEW_SAMPLER; +} + +static void brw_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + free(sampler); +} + + +/************************************************************************ + * Depth stencil + */ + +static void * +brw_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + DUP( pipe_depth_stencil_alpha_state, depth_stencil ); +} + +static void brw_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.DepthStencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil; + + brw->state.dirty.brw |= BRW_NEW_DEPTH_STENCIL; +} + +static void brw_delete_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + free(depth_stencil); +} + +/************************************************************************ + * Scissor + */ +static void brw_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct brw_context *brw = brw_context(pipe); + + memcpy( &brw->attribs.Scissor, scissor, sizeof(*scissor) ); + brw->state.dirty.brw |= BRW_NEW_SCISSOR; +} + + +/************************************************************************ + * Stipple + */ + +static void brw_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ +} + + +/************************************************************************ + * Fragment shader + */ + +static void * brw_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *shader) +{ + struct brw_fragment_program *brw_fp = CALLOC_STRUCT(brw_fragment_program); + + /* XXX: Do I have to duplicate the tokens as well?? + */ + brw_fp->program = *shader; + brw_fp->id = brw_context(pipe)->program_id++; + + brw_shader_info(shader->tokens, + &brw_fp->info); + + tgsi_dump(shader->tokens, 0); + + + return (void *)brw_fp; +} + +static void brw_bind_fs_state(struct pipe_context *pipe, void *shader) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.FragmentProgram = (struct brw_fragment_program *)shader; + brw->state.dirty.brw |= BRW_NEW_FS; +} + +static void brw_delete_fs_state(struct pipe_context *pipe, void *shader) +{ + FREE(shader); +} + + +/************************************************************************ + * Vertex shader and other TNL state + */ + +static void *brw_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *shader) +{ + struct brw_vertex_program *brw_vp = CALLOC_STRUCT(brw_vertex_program); + + /* XXX: Do I have to duplicate the tokens as well?? + */ + brw_vp->program = *shader; + brw_vp->id = brw_context(pipe)->program_id++; + brw_shader_info(shader->tokens, + &brw_vp->info); + + tgsi_dump(shader->tokens, 0); + + return (void *)brw_vp; +} + +static void brw_bind_vs_state(struct pipe_context *pipe, void *vs) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.VertexProgram = (struct brw_vertex_program *)vs; + brw->state.dirty.brw |= BRW_NEW_VS; + + debug_printf("YYYYYYYYYYYYY BINDING VERTEX SHADER\n"); +} + +static void brw_delete_vs_state(struct pipe_context *pipe, void *shader) +{ + FREE(shader); +} + + +static void brw_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Clip = *clip; +} + + +static void brw_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Viewport = *viewport; /* struct copy */ + brw->state.dirty.brw |= BRW_NEW_VIEWPORT; + + /* pass the viewport info to the draw module */ + //draw_set_viewport_state(brw->draw, viewport); +} + + +static void brw_set_vertex_buffer( struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_buffer *buffer ) +{ + struct brw_context *brw = brw_context(pipe); + brw->vb.vbo_array[index] = buffer; +} + +static void brw_set_vertex_element(struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_element *element) +{ + /* flush ? */ + struct brw_context *brw = brw_context(pipe); + + assert(index < PIPE_ATTRIB_MAX); + struct brw_vertex_element_state el; + memset(&el, 0, sizeof(el)); + + el.ve0.src_offset = element->src_offset; + el.ve0.src_format = brw_translate_surface_format(element->src_format); + el.ve0.valid = 1; + el.ve0.vertex_buffer_index = element->vertex_buffer_index; + + el.ve1.dst_offset = index * 4; + + el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_SRC; + el.ve1.vfcomponent0 = BRW_VFCOMPONENT_STORE_SRC; + + switch (element->nr_components) { + case 1: el.ve1.vfcomponent1 = BRW_VFCOMPONENT_STORE_0; + case 2: el.ve1.vfcomponent2 = BRW_VFCOMPONENT_STORE_0; + case 3: el.ve1.vfcomponent3 = BRW_VFCOMPONENT_STORE_1_FLT; + break; + } + + brw->vb.inputs[index] = el; +} + + + +/************************************************************************ + * Constant buffers + */ + +static void brw_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct brw_context *brw = brw_context(pipe); + + assert(buf == 0 || index == 0); + + brw->attribs.Constants[shader] = buf; + brw->state.dirty.brw |= BRW_NEW_CONSTANTS; +} + + +/************************************************************************ + * Texture surfaces + */ + + +static void brw_set_sampler_texture(struct pipe_context *pipe, + unsigned unit, + struct pipe_texture *texture) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Texture[unit] = (struct brw_texture*)texture; /* ptr, not struct */ + + brw->state.dirty.brw |= BRW_NEW_TEXTURE; +} + + +/************************************************************************ + * Render targets, etc + */ + +static void brw_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.FrameBuffer = *fb; /* struct copy */ + + brw->state.dirty.brw |= BRW_NEW_FRAMEBUFFER; +} + + + +/************************************************************************ + * Rasterizer state + */ + +static void * +brw_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *rasterizer) +{ + DUP(pipe_rasterizer_state, rasterizer); +} + +static void brw_bind_rasterizer_state( struct pipe_context *pipe, + void *setup ) +{ + struct brw_context *brw = brw_context(pipe); + + brw->attribs.Raster = (struct pipe_rasterizer_state *)setup; + + /* Also pass-through to draw module: + */ + //draw_set_rasterizer_state(brw->draw, setup); + + brw->state.dirty.brw |= BRW_NEW_RASTERIZER; +} + +static void brw_delete_rasterizer_state(struct pipe_context *pipe, + void *setup) +{ + free(setup); +} + + + +void +brw_init_state_functions( struct brw_context *brw ) +{ + brw->pipe.create_blend_state = brw_create_blend_state; + brw->pipe.bind_blend_state = brw_bind_blend_state; + brw->pipe.delete_blend_state = brw_delete_blend_state; + + brw->pipe.create_sampler_state = brw_create_sampler_state; + brw->pipe.bind_sampler_state = brw_bind_sampler_state; + brw->pipe.delete_sampler_state = brw_delete_sampler_state; + + brw->pipe.create_depth_stencil_alpha_state = brw_create_depth_stencil_state; + brw->pipe.bind_depth_stencil_alpha_state = brw_bind_depth_stencil_state; + brw->pipe.delete_depth_stencil_alpha_state = brw_delete_depth_stencil_state; + + brw->pipe.create_rasterizer_state = brw_create_rasterizer_state; + brw->pipe.bind_rasterizer_state = brw_bind_rasterizer_state; + brw->pipe.delete_rasterizer_state = brw_delete_rasterizer_state; + brw->pipe.create_fs_state = brw_create_fs_state; + brw->pipe.bind_fs_state = brw_bind_fs_state; + brw->pipe.delete_fs_state = brw_delete_fs_state; + brw->pipe.create_vs_state = brw_create_vs_state; + brw->pipe.bind_vs_state = brw_bind_vs_state; + brw->pipe.delete_vs_state = brw_delete_vs_state; + + brw->pipe.set_blend_color = brw_set_blend_color; + brw->pipe.set_clip_state = brw_set_clip_state; + brw->pipe.set_constant_buffer = brw_set_constant_buffer; + brw->pipe.set_framebuffer_state = brw_set_framebuffer_state; + +// brw->pipe.set_feedback_state = brw_set_feedback_state; +// brw->pipe.set_feedback_buffer = brw_set_feedback_buffer; + + brw->pipe.set_polygon_stipple = brw_set_polygon_stipple; + brw->pipe.set_scissor_state = brw_set_scissor_state; + brw->pipe.set_sampler_texture = brw_set_sampler_texture; + brw->pipe.set_viewport_state = brw_set_viewport_state; + brw->pipe.set_vertex_buffer = brw_set_vertex_buffer; + brw->pipe.set_vertex_element = brw_set_vertex_element; +} diff --git a/src/gallium/drivers/i965simple/brw_state.h b/src/gallium/drivers/i965simple/brw_state.h new file mode 100644 index 0000000000..258e9a556e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state.h @@ -0,0 +1,158 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_STATE_H +#define BRW_STATE_H + +#include "brw_context.h" +#include "brw_winsys.h" + + +const struct brw_tracked_state brw_blend_constant_color; +const struct brw_tracked_state brw_cc_unit; +const struct brw_tracked_state brw_cc_vp; +const struct brw_tracked_state brw_clip_prog; +const struct brw_tracked_state brw_clip_unit; +const struct brw_tracked_state brw_constant_buffer_state; +const struct brw_tracked_state brw_constant_buffer; +const struct brw_tracked_state brw_curbe_offsets; +const struct brw_tracked_state brw_invarient_state; +const struct brw_tracked_state brw_gs_prog; +const struct brw_tracked_state brw_gs_unit; +const struct brw_tracked_state brw_drawing_rect; +const struct brw_tracked_state brw_line_stipple; +const struct brw_tracked_state brw_pipelined_state_pointers; +const struct brw_tracked_state brw_binding_table_pointers; +const struct brw_tracked_state brw_depthbuffer; +const struct brw_tracked_state brw_polygon_stipple_offset; +const struct brw_tracked_state brw_polygon_stipple; +const struct brw_tracked_state brw_program_parameters; +const struct brw_tracked_state brw_recalculate_urb_fence; +const struct brw_tracked_state brw_sf_prog; +const struct brw_tracked_state brw_sf_unit; +const struct brw_tracked_state brw_sf_vp; +const struct brw_tracked_state brw_state_base_address; +const struct brw_tracked_state brw_urb_fence; +const struct brw_tracked_state brw_vertex_state; +const struct brw_tracked_state brw_vs_prog; +const struct brw_tracked_state brw_vs_unit; +const struct brw_tracked_state brw_wm_prog; +const struct brw_tracked_state brw_wm_samplers; +const struct brw_tracked_state brw_wm_surfaces; +const struct brw_tracked_state brw_wm_unit; + +const struct brw_tracked_state brw_psp_urb_cbs; + +const struct brw_tracked_state brw_active_vertprog; +const struct brw_tracked_state brw_tnl_vertprog; +const struct brw_tracked_state brw_pipe_control; + +const struct brw_tracked_state brw_clear_surface_cache; +const struct brw_tracked_state brw_clear_batch_cache; + +/*********************************************************************** + * brw_state_cache.c + */ +unsigned brw_cache_data(struct brw_cache *cache, + const void *data ); + +unsigned brw_cache_data_sz(struct brw_cache *cache, + const void *data, + unsigned data_sz); + +unsigned brw_upload_cache( struct brw_cache *cache, + const void *key, + unsigned key_sz, + const void *data, + unsigned data_sz, + const void *aux, + void *aux_return ); + +boolean brw_search_cache( struct brw_cache *cache, + const void *key, + unsigned key_size, + void *aux_return, + unsigned *offset_return); + +void brw_init_caches( struct brw_context *brw ); +void brw_destroy_caches( struct brw_context *brw ); + +static inline struct pipe_buffer *brw_cache_buffer(struct brw_context *brw, + enum brw_cache_id id) +{ + return brw->cache[id].pool->buffer; +} + +/*********************************************************************** + * brw_state_batch.c + */ +#define BRW_CACHED_BATCH_STRUCT(brw, s) brw_cached_batch_struct( brw, (s), sizeof(*(s)) ) + +boolean brw_cached_batch_struct( struct brw_context *brw, + const void *data, + unsigned sz ); + +void brw_destroy_batch_cache( struct brw_context *brw ); + + +/*********************************************************************** + * brw_state_pool.c + */ +void brw_init_pools( struct brw_context *brw ); +void brw_destroy_pools( struct brw_context *brw ); + +boolean brw_pool_alloc( struct brw_mem_pool *pool, + unsigned size, + unsigned alignment, + unsigned *offset_return); + +void brw_pool_fence( struct brw_context *brw, + struct brw_mem_pool *pool, + unsigned fence ); + + +void brw_pool_check_wrap( struct brw_context *brw, + struct brw_mem_pool *pool ); + +void brw_clear_all_caches( struct brw_context *brw ); +void brw_invalidate_pools( struct brw_context *brw ); +void brw_clear_batch_cache_flush( struct brw_context *brw ); + + +/* brw_shader_info.c + */ + +void brw_shader_info(const struct tgsi_token *tokens, + struct brw_shader_info *info ); + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_state_batch.c b/src/gallium/drivers/i965simple/brw_state_batch.c new file mode 100644 index 0000000000..35db76b594 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state_batch.c @@ -0,0 +1,113 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_state.h" +#include "brw_winsys.h" + +#include "pipe/p_util.h" + +/* A facility similar to the data caching code above, which aims to + * prevent identical commands being issued repeatedly. + */ +boolean brw_cached_batch_struct( struct brw_context *brw, + const void *data, + unsigned sz ) +{ + struct brw_cached_batch_item *item = brw->cached_batch_items; + struct header *newheader = (struct header *)data; + + if (brw->emit_state_always) { + brw_batchbuffer_data(brw->winsys, data, sz); + return TRUE; + } + + while (item) { + if (item->header->opcode == newheader->opcode) { + if (item->sz == sz && memcmp(item->header, newheader, sz) == 0) + return FALSE; + if (item->sz != sz) { + FREE(item->header); + item->header = MALLOC(sz); + item->sz = sz; + } + goto emit; + } + item = item->next; + } + + assert(!item); + item = CALLOC_STRUCT(brw_cached_batch_item); + item->header = MALLOC(sz); + item->sz = sz; + item->next = brw->cached_batch_items; + brw->cached_batch_items = item; + +emit: + memcpy(item->header, newheader, sz); + brw_batchbuffer_data(brw->winsys, data, sz); + return TRUE; +} + +static void clear_batch_cache( struct brw_context *brw ) +{ + struct brw_cached_batch_item *item = brw->cached_batch_items; + + while (item) { + struct brw_cached_batch_item *next = item->next; + free((void *)item->header); + free(item); + item = next; + } + + brw->cached_batch_items = NULL; + + + brw_clear_all_caches(brw); + + brw_invalidate_pools(brw); +} + +void brw_clear_batch_cache_flush( struct brw_context *brw ) +{ + clear_batch_cache(brw); + +/* brw_do_flush(brw, BRW_FLUSH_STATE_CACHE|BRW_FLUSH_READ_CACHE); */ + + brw->state.dirty.brw |= ~0; + brw->state.dirty.cache |= ~0; +} + + + +void brw_destroy_batch_cache( struct brw_context *brw ) +{ + clear_batch_cache(brw); +} diff --git a/src/gallium/drivers/i965simple/brw_state_cache.c b/src/gallium/drivers/i965simple/brw_state_cache.c new file mode 100644 index 0000000000..b3a5124461 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state_cache.c @@ -0,0 +1,443 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_state.h" + +#include "brw_wm.h" +#include "brw_vs.h" +#include "brw_clip.h" +#include "brw_sf.h" +#include "brw_gs.h" + +#include "pipe/p_util.h" + + + +/*********************************************************************** + * Check cache for uploaded version of struct, else upload new one. + * Fail when memory is exhausted. + * + * XXX: FIXME: Currently search is so slow it would be quicker to + * regenerate the data every time... + */ + +static unsigned hash_key( const void *key, unsigned key_size ) +{ + unsigned *ikey = (unsigned *)key; + unsigned hash = 0, i; + + assert(key_size % 4 == 0); + + /* I'm sure this can be improved on: + */ + for (i = 0; i < key_size/4; i++) + hash ^= ikey[i]; + + return hash; +} + +static struct brw_cache_item *search_cache( struct brw_cache *cache, + unsigned hash, + const void *key, + unsigned key_size) +{ + struct brw_cache_item *c; + + for (c = cache->items[hash % cache->size]; c; c = c->next) { + if (c->hash == hash && + c->key_size == key_size && + memcmp(c->key, key, key_size) == 0) + return c; + } + + return NULL; +} + + +static void rehash( struct brw_cache *cache ) +{ + struct brw_cache_item **items; + struct brw_cache_item *c, *next; + unsigned size, i; + + size = cache->size * 3; + items = (struct brw_cache_item**) MALLOC(size * sizeof(*items)); + memset(items, 0, size * sizeof(*items)); + + for (i = 0; i < cache->size; i++) + for (c = cache->items[i]; c; c = next) { + next = c->next; + c->next = items[c->hash % size]; + items[c->hash % size] = c; + } + + FREE(cache->items); + cache->items = items; + cache->size = size; +} + + +boolean brw_search_cache( struct brw_cache *cache, + const void *key, + unsigned key_size, + void *aux_return, + unsigned *offset_return) +{ + struct brw_cache_item *item; + unsigned addr = 0; + unsigned hash = hash_key(key, key_size); + + item = search_cache(cache, hash, key, key_size); + + if (item) { + if (aux_return) + *(void **)aux_return = (void *)((char *)item->key + item->key_size); + + *offset_return = addr = item->offset; + } + + if (item == NULL || addr != cache->last_addr) { + cache->brw->state.dirty.cache |= 1<<cache->id; + cache->last_addr = addr; + } + + return item != NULL; +} + +unsigned brw_upload_cache( struct brw_cache *cache, + const void *key, + unsigned key_size, + const void *data, + unsigned data_size, + const void *aux, + void *aux_return ) +{ + unsigned offset; + struct brw_cache_item *item = CALLOC_STRUCT(brw_cache_item); + unsigned hash = hash_key(key, key_size); + void *tmp = MALLOC(key_size + cache->aux_size); + + if (!brw_pool_alloc(cache->pool, data_size, 1 << 6, &offset)) { + /* Should not be possible: + */ + debug_printf("brw_pool_alloc failed\n"); + exit(1); + } + + memcpy(tmp, key, key_size); + + if (cache->aux_size) + memcpy(tmp+key_size, aux, cache->aux_size); + + item->key = tmp; + item->hash = hash; + item->key_size = key_size; + item->offset = offset; + item->data_size = data_size; + + if (++cache->n_items > cache->size * 1.5) + rehash(cache); + + hash %= cache->size; + item->next = cache->items[hash]; + cache->items[hash] = item; + + if (aux_return) { + assert(cache->aux_size); + *(void **)aux_return = (void *)((char *)item->key + item->key_size); + } + + if (BRW_DEBUG & DEBUG_STATE) + debug_printf("upload %s: %d bytes to pool buffer %p offset %x\n", + cache->name, + data_size, + (void*)cache->pool->buffer, + offset); + + /* Copy data to the buffer: + */ + cache->brw->winsys->buffer_subdata_typed(cache->brw->winsys, + cache->pool->buffer, + offset, + data_size, + data, + cache->id); + + cache->brw->state.dirty.cache |= 1<<cache->id; + cache->last_addr = offset; + + return offset; +} + +/* This doesn't really work with aux data. Use search/upload instead + */ +unsigned brw_cache_data_sz(struct brw_cache *cache, + const void *data, + unsigned data_size) +{ + unsigned addr; + + if (!brw_search_cache(cache, data, data_size, NULL, &addr)) { + addr = brw_upload_cache(cache, + data, data_size, + data, data_size, + NULL, NULL); + } + + return addr; +} + +unsigned brw_cache_data(struct brw_cache *cache, + const void *data) +{ + return brw_cache_data_sz(cache, data, cache->key_size); +} + +enum pool_type { + DW_SURFACE_STATE, + DW_GENERAL_STATE +}; + +static void brw_init_cache( struct brw_context *brw, + const char *name, + unsigned id, + unsigned key_size, + unsigned aux_size, + enum pool_type pool_type) +{ + struct brw_cache *cache = &brw->cache[id]; + cache->brw = brw; + cache->id = id; + cache->name = name; + cache->items = NULL; + + cache->size = 7; + cache->n_items = 0; + cache->items = (struct brw_cache_item **) + CALLOC(cache->size, sizeof(struct brw_cache_item)); + + + cache->key_size = key_size; + cache->aux_size = aux_size; + switch (pool_type) { + case DW_GENERAL_STATE: cache->pool = &brw->pool[BRW_GS_POOL]; break; + case DW_SURFACE_STATE: cache->pool = &brw->pool[BRW_SS_POOL]; break; + default: assert(0); break; + } +} + +void brw_init_caches( struct brw_context *brw ) +{ + + brw_init_cache(brw, + "CC_VP", + BRW_CC_VP, + sizeof(struct brw_cc_viewport), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "CC_UNIT", + BRW_CC_UNIT, + sizeof(struct brw_cc_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "WM_PROG", + BRW_WM_PROG, + sizeof(struct brw_wm_prog_key), + sizeof(struct brw_wm_prog_data), + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SAMPLER_DEFAULT_COLOR", + BRW_SAMPLER_DEFAULT_COLOR, + sizeof(struct brw_sampler_default_color), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SAMPLER", + BRW_SAMPLER, + 0, /* variable key/data size */ + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "WM_UNIT", + BRW_WM_UNIT, + sizeof(struct brw_wm_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SF_PROG", + BRW_SF_PROG, + sizeof(struct brw_sf_prog_key), + sizeof(struct brw_sf_prog_data), + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SF_VP", + BRW_SF_VP, + sizeof(struct brw_sf_viewport), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SF_UNIT", + BRW_SF_UNIT, + sizeof(struct brw_sf_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "VS_UNIT", + BRW_VS_UNIT, + sizeof(struct brw_vs_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "VS_PROG", + BRW_VS_PROG, + sizeof(struct brw_vs_prog_key), + sizeof(struct brw_vs_prog_data), + DW_GENERAL_STATE); + + brw_init_cache(brw, + "CLIP_UNIT", + BRW_CLIP_UNIT, + sizeof(struct brw_clip_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "CLIP_PROG", + BRW_CLIP_PROG, + sizeof(struct brw_clip_prog_key), + sizeof(struct brw_clip_prog_data), + DW_GENERAL_STATE); + + brw_init_cache(brw, + "GS_UNIT", + BRW_GS_UNIT, + sizeof(struct brw_gs_unit_state), + 0, + DW_GENERAL_STATE); + + brw_init_cache(brw, + "GS_PROG", + BRW_GS_PROG, + sizeof(struct brw_gs_prog_key), + sizeof(struct brw_gs_prog_data), + DW_GENERAL_STATE); + + brw_init_cache(brw, + "SS_SURFACE", + BRW_SS_SURFACE, + sizeof(struct brw_surface_state), + 0, + DW_SURFACE_STATE); + + brw_init_cache(brw, + "SS_SURF_BIND", + BRW_SS_SURF_BIND, + sizeof(struct brw_surface_binding_table), + 0, + DW_SURFACE_STATE); +} + + +/* When we lose hardware context, need to invalidate the surface cache + * as these structs must be explicitly re-uploaded. They are subject + * to fixup by the memory manager as they contain absolute agp + * offsets, so we need to ensure there is a fresh version of the + * struct available to receive the fixup. + * + * XXX: Need to ensure that there aren't two versions of a surface or + * bufferobj with different backing data active in the same buffer at + * once? Otherwise the cache could confuse them. Maybe better not to + * cache at all? + * + * --> Isn't this the same as saying need to ensure batch is flushed + * before new data is uploaded to an existing buffer? We + * already try to make sure of that. + */ +static void clear_cache( struct brw_cache *cache ) +{ + struct brw_cache_item *c, *next; + unsigned i; + + for (i = 0; i < cache->size; i++) { + for (c = cache->items[i]; c; c = next) { + next = c->next; + free((void *)c->key); + free(c); + } + cache->items[i] = NULL; + } + + cache->n_items = 0; +} + +void brw_clear_all_caches( struct brw_context *brw ) +{ + int i; + + if (BRW_DEBUG & DEBUG_STATE) + debug_printf("%s\n", __FUNCTION__); + + for (i = 0; i < BRW_MAX_CACHE; i++) + clear_cache(&brw->cache[i]); + + if (brw->curbe.last_buf) { + FREE(brw->curbe.last_buf); + brw->curbe.last_buf = NULL; + } + + brw->state.dirty.brw |= ~0; + brw->state.dirty.cache |= ~0; +} + + + + + +void brw_destroy_caches( struct brw_context *brw ) +{ + unsigned i; + + for (i = 0; i < BRW_MAX_CACHE; i++) + clear_cache(&brw->cache[i]); +} diff --git a/src/gallium/drivers/i965simple/brw_state_pool.c b/src/gallium/drivers/i965simple/brw_state_pool.c new file mode 100644 index 0000000000..f3174bfe0a --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state_pool.c @@ -0,0 +1,137 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +/** @file brw_state_pool.c + * Implements the state pool allocator. + * + * For the 965, we create two state pools for state cache entries. Objects + * will be allocated into the pools depending on which state base address + * their pointer is relative to in other 965 state. + * + * The state pools are relatively simple: As objects are allocated, increment + * the offset to allocate space. When the pool is "full" (rather, close to + * full), we reset the pool and reset the state cache entries that point into + * the pool. + */ + +#include "pipe/p_winsys.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "brw_context.h" +#include "brw_state.h" + +boolean brw_pool_alloc( struct brw_mem_pool *pool, + unsigned size, + unsigned alignment, + unsigned *offset_return) +{ + unsigned fixup = align(pool->offset, alignment) - pool->offset; + + size = align(size, 4); + + if (pool->offset + fixup + size >= pool->size) { + debug_printf("%s failed\n", __FUNCTION__); + assert(0); + exit(0); + } + + pool->offset += fixup; + *offset_return = pool->offset; + pool->offset += size; + + return TRUE; +} + +static +void brw_invalidate_pool( struct brw_mem_pool *pool ) +{ + if (BRW_DEBUG & DEBUG_STATE) + debug_printf("\n\n\n %s \n\n\n", __FUNCTION__); + + pool->offset = 0; + + brw_clear_all_caches(pool->brw); +} + + +static void brw_init_pool( struct brw_context *brw, + unsigned pool_id, + unsigned size ) +{ + struct brw_mem_pool *pool = &brw->pool[pool_id]; + + pool->size = size; + pool->brw = brw; + + pool->buffer = brw->pipe.winsys->buffer_create(brw->pipe.winsys, + 4096, + 0 /* DRM_BO_FLAG_MEM_TT */, + size); +} + +static void brw_destroy_pool( struct brw_context *brw, + unsigned pool_id ) +{ + struct brw_mem_pool *pool = &brw->pool[pool_id]; + + pipe_buffer_reference( pool->brw->pipe.winsys, + &pool->buffer, + NULL ); +} + + +void brw_pool_check_wrap( struct brw_context *brw, + struct brw_mem_pool *pool ) +{ + if (pool->offset > (pool->size * 3) / 4) { + brw->state.dirty.brw |= BRW_NEW_SCENE; + } + +} + +void brw_init_pools( struct brw_context *brw ) +{ + brw_init_pool(brw, BRW_GS_POOL, 0x80000); + brw_init_pool(brw, BRW_SS_POOL, 0x80000); +} + +void brw_destroy_pools( struct brw_context *brw ) +{ + brw_destroy_pool(brw, BRW_GS_POOL); + brw_destroy_pool(brw, BRW_SS_POOL); +} + + +void brw_invalidate_pools( struct brw_context *brw ) +{ + brw_invalidate_pool(&brw->pool[BRW_GS_POOL]); + brw_invalidate_pool(&brw->pool[BRW_SS_POOL]); +} diff --git a/src/gallium/drivers/i965simple/brw_state_upload.c b/src/gallium/drivers/i965simple/brw_state_upload.c new file mode 100644 index 0000000000..e727601e1e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state_upload.c @@ -0,0 +1,202 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_state.h" + +#include "pipe/p_util.h" + +/* This is used to initialize brw->state.atoms[]. We could use this + * list directly except for a single atom, brw_constant_buffer, which + * has a .dirty value which changes according to the parameters of the + * current fragment and vertex programs, and so cannot be a static + * value. + */ +const struct brw_tracked_state *atoms[] = +{ + &brw_vs_prog, + &brw_gs_prog, + &brw_clip_prog, + &brw_sf_prog, + &brw_wm_prog, + + /* Once all the programs are done, we know how large urb entry + * sizes need to be and can decide if we need to change the urb + * layout. + */ + &brw_curbe_offsets, + &brw_recalculate_urb_fence, + + + &brw_cc_vp, + &brw_cc_unit, + + &brw_wm_surfaces, /* must do before samplers */ + &brw_wm_samplers, + + &brw_wm_unit, + &brw_sf_vp, + &brw_sf_unit, + &brw_vs_unit, /* always required, enabled or not */ + &brw_clip_unit, + &brw_gs_unit, + + /* Command packets: + */ + &brw_invarient_state, + &brw_state_base_address, + &brw_pipe_control, + + &brw_binding_table_pointers, + &brw_blend_constant_color, + + &brw_drawing_rect, + &brw_depthbuffer, + + &brw_polygon_stipple, + &brw_line_stipple, + + &brw_psp_urb_cbs, + + &brw_constant_buffer +}; + + +void brw_init_state( struct brw_context *brw ) +{ + brw_init_pools(brw); + brw_init_caches(brw); + + brw->state.dirty.brw = ~0; + brw->emit_state_always = 0; +} + + +void brw_destroy_state( struct brw_context *brw ) +{ + brw_destroy_caches(brw); + brw_destroy_batch_cache(brw); + brw_destroy_pools(brw); +} + +/*********************************************************************** + */ + +static boolean check_state( const struct brw_state_flags *a, + const struct brw_state_flags *b ) +{ + return ((a->brw & b->brw) || + (a->cache & b->cache)); +} + +static void accumulate_state( struct brw_state_flags *a, + const struct brw_state_flags *b ) +{ + a->brw |= b->brw; + a->cache |= b->cache; +} + + +static void xor_states( struct brw_state_flags *result, + const struct brw_state_flags *a, + const struct brw_state_flags *b ) +{ + result->brw = a->brw ^ b->brw; + result->cache = a->cache ^ b->cache; +} + + +/*********************************************************************** + * Emit all state: + */ +void brw_validate_state( struct brw_context *brw ) +{ + struct brw_state_flags *state = &brw->state.dirty; + unsigned i; + + if (brw->emit_state_always) + state->brw |= ~0; + + if (state->cache == 0 && + state->brw == 0) + return; + + if (brw->state.dirty.brw & BRW_NEW_SCENE) + brw_clear_batch_cache_flush(brw); + + if (BRW_DEBUG) { + /* Debug version which enforces various sanity checks on the + * state flags which are generated and checked to help ensure + * state atoms are ordered correctly in the list. + */ + struct brw_state_flags examined, prev; + memset(&examined, 0, sizeof(examined)); + prev = *state; + + for (i = 0; i < Elements(atoms); i++) { + const struct brw_tracked_state *atom = atoms[i]; + struct brw_state_flags generated; + + assert(atom->dirty.brw || + atom->dirty.cache); + assert(atom->update); + + if (check_state(state, &atom->dirty)) { + atom->update( brw ); + } + + accumulate_state(&examined, &atom->dirty); + + /* generated = (prev ^ state) + * if (examined & generated) + * fail; + */ + xor_states(&generated, &prev, state); + assert(!check_state(&examined, &generated)); + prev = *state; + } + } + else { + for (i = 0; i < Elements(atoms); i++) { + const struct brw_tracked_state *atom = atoms[i]; + + assert(atom->dirty.brw || + atom->dirty.cache); + assert(atom->update); + + if (check_state(state, &atom->dirty)) + atom->update( brw ); + } + } + + memset(state, 0, sizeof(*state)); +} diff --git a/src/gallium/drivers/i965simple/brw_strings.c b/src/gallium/drivers/i965simple/brw_strings.c new file mode 100644 index 0000000000..29a41ed1e9 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_strings.c @@ -0,0 +1,72 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "brw_context.h" +#include "brw_reg.h" + + +static const char *brw_get_vendor( struct pipe_context *pipe ) +{ + return "Tungsten Graphics, Inc."; +} + + +static const char *brw_get_name( struct pipe_context *pipe ) +{ + static char buffer[128]; + const char *chipset; + + switch (brw_context(pipe)->pci_id) { + case PCI_CHIP_I965_Q: + chipset = "Intel(R) 965Q"; + break; + case PCI_CHIP_I965_G: + case PCI_CHIP_I965_G_1: + chipset = "Intel(R) 965G"; + break; + case PCI_CHIP_I965_GM: + chipset = "Intel(R) 965GM"; + break; + case PCI_CHIP_I965_GME: + chipset = "Intel(R) 965GME/GLE"; + break; + default: + chipset = "unknown"; + break; + } + + sprintf(buffer, "pipe/i965 (chipset: %s)", chipset); + return buffer; +} + + +void +brw_init_string_functions(struct brw_context *brw) +{ + brw->pipe.get_name = brw_get_name; + brw->pipe.get_vendor = brw_get_vendor; +} diff --git a/src/gallium/drivers/i965simple/brw_structs.h b/src/gallium/drivers/i965simple/brw_structs.h new file mode 100644 index 0000000000..bbb087e95d --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_structs.h @@ -0,0 +1,1348 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_STRUCTS_H +#define BRW_STRUCTS_H + +#include "pipe/p_compiler.h" + +/* Command packets: + */ +struct header +{ + unsigned length:16; + unsigned opcode:16; +}; + + +union header_union +{ + struct header bits; + unsigned dword; +}; + +struct brw_3d_control +{ + struct + { + unsigned length:8; + unsigned notify_enable:1; + unsigned pad:3; + unsigned wc_flush_enable:1; + unsigned depth_stall_enable:1; + unsigned operation:2; + unsigned opcode:16; + } header; + + struct + { + unsigned pad:2; + unsigned dest_addr_type:1; + unsigned dest_addr:29; + } dest; + + unsigned dword2; + unsigned dword3; +}; + + +struct brw_3d_primitive +{ + struct + { + unsigned length:8; + unsigned pad:2; + unsigned topology:5; + unsigned indexed:1; + unsigned opcode:16; + } header; + + unsigned verts_per_instance; + unsigned start_vert_location; + unsigned instance_count; + unsigned start_instance_location; + unsigned base_vert_location; +}; + +/* These seem to be passed around as function args, so it works out + * better to keep them as #defines: + */ +#define BRW_FLUSH_READ_CACHE 0x1 +#define BRW_FLUSH_STATE_CACHE 0x2 +#define BRW_INHIBIT_FLUSH_RENDER_CACHE 0x4 +#define BRW_FLUSH_SNAPSHOT_COUNTERS 0x8 + +struct brw_mi_flush +{ + unsigned flags:4; + unsigned pad:12; + unsigned opcode:16; +}; + +struct brw_vf_statistics +{ + unsigned statistics_enable:1; + unsigned pad:15; + unsigned opcode:16; +}; + + + +struct brw_binding_table_pointers +{ + struct header header; + unsigned vs; + unsigned gs; + unsigned clp; + unsigned sf; + unsigned wm; +}; + + +struct brw_blend_constant_color +{ + struct header header; + float blend_constant_color[4]; +}; + + +struct brw_depthbuffer +{ + union header_union header; + + union { + struct { + unsigned pitch:18; + unsigned format:3; + unsigned pad:4; + unsigned depth_offset_disable:1; + unsigned tile_walk:1; + unsigned tiled_surface:1; + unsigned pad2:1; + unsigned surface_type:3; + } bits; + unsigned dword; + } dword1; + + unsigned dword2_base_addr; + + union { + struct { + unsigned pad:1; + unsigned mipmap_layout:1; + unsigned lod:4; + unsigned width:13; + unsigned height:13; + } bits; + unsigned dword; + } dword3; + + union { + struct { + unsigned pad:12; + unsigned min_array_element:9; + unsigned depth:11; + } bits; + unsigned dword; + } dword4; +}; + +struct brw_drawrect +{ + struct header header; + unsigned xmin:16; + unsigned ymin:16; + unsigned xmax:16; + unsigned ymax:16; + unsigned xorg:16; + unsigned yorg:16; +}; + + + + +struct brw_global_depth_offset_clamp +{ + struct header header; + float depth_offset_clamp; +}; + +struct brw_indexbuffer +{ + union { + struct + { + unsigned length:8; + unsigned index_format:2; + unsigned cut_index_enable:1; + unsigned pad:5; + unsigned opcode:16; + } bits; + unsigned dword; + + } header; + + unsigned buffer_start; + unsigned buffer_end; +}; + + +struct brw_line_stipple +{ + struct header header; + + struct + { + unsigned pattern:16; + unsigned pad:16; + } bits0; + + struct + { + unsigned repeat_count:9; + unsigned pad:7; + unsigned inverse_repeat_count:16; + } bits1; +}; + + +struct brw_pipelined_state_pointers +{ + struct header header; + + struct { + unsigned pad:5; + unsigned offset:27; + } vs; + + struct + { + unsigned enable:1; + unsigned pad:4; + unsigned offset:27; + } gs; + + struct + { + unsigned enable:1; + unsigned pad:4; + unsigned offset:27; + } clp; + + struct + { + unsigned pad:5; + unsigned offset:27; + } sf; + + struct + { + unsigned pad:5; + unsigned offset:27; + } wm; + + struct + { + unsigned pad:5; + unsigned offset:27; /* KW: check me! */ + } cc; +}; + + +struct brw_polygon_stipple_offset +{ + struct header header; + + struct { + unsigned y_offset:5; + unsigned pad:3; + unsigned x_offset:5; + unsigned pad0:19; + } bits0; +}; + + + +struct brw_polygon_stipple +{ + struct header header; + unsigned stipple[32]; +}; + + + +struct brw_pipeline_select +{ + struct + { + unsigned pipeline_select:1; + unsigned pad:15; + unsigned opcode:16; + } header; +}; + + +struct brw_pipe_control +{ + struct + { + unsigned length:8; + unsigned notify_enable:1; + unsigned pad:2; + unsigned instruction_state_cache_flush_enable:1; + unsigned write_cache_flush_enable:1; + unsigned depth_stall_enable:1; + unsigned post_sync_operation:2; + + unsigned opcode:16; + } header; + + struct + { + unsigned pad:2; + unsigned dest_addr_type:1; + unsigned dest_addr:29; + } bits1; + + unsigned data0; + unsigned data1; +}; + + +struct brw_urb_fence +{ + struct + { + unsigned length:8; + unsigned vs_realloc:1; + unsigned gs_realloc:1; + unsigned clp_realloc:1; + unsigned sf_realloc:1; + unsigned vfe_realloc:1; + unsigned cs_realloc:1; + unsigned pad:2; + unsigned opcode:16; + } header; + + struct + { + unsigned vs_fence:10; + unsigned gs_fence:10; + unsigned clp_fence:10; + unsigned pad:2; + } bits0; + + struct + { + unsigned sf_fence:10; + unsigned vf_fence:10; + unsigned cs_fence:10; + unsigned pad:2; + } bits1; +}; + +struct brw_constant_buffer_state /* previously brw_command_streamer */ +{ + struct header header; + + struct + { + unsigned nr_urb_entries:3; + unsigned pad:1; + unsigned urb_entry_size:5; + unsigned pad0:23; + } bits0; +}; + +struct brw_constant_buffer +{ + struct + { + unsigned length:8; + unsigned valid:1; + unsigned pad:7; + unsigned opcode:16; + } header; + + struct + { + unsigned buffer_length:6; + unsigned buffer_address:26; + } bits0; +}; + +struct brw_state_base_address +{ + struct header header; + + struct + { + unsigned modify_enable:1; + unsigned pad:4; + unsigned general_state_address:27; + } bits0; + + struct + { + unsigned modify_enable:1; + unsigned pad:4; + unsigned surface_state_address:27; + } bits1; + + struct + { + unsigned modify_enable:1; + unsigned pad:4; + unsigned indirect_object_state_address:27; + } bits2; + + struct + { + unsigned modify_enable:1; + unsigned pad:11; + unsigned general_state_upper_bound:20; + } bits3; + + struct + { + unsigned modify_enable:1; + unsigned pad:11; + unsigned indirect_object_state_upper_bound:20; + } bits4; +}; + +struct brw_state_prefetch +{ + struct header header; + + struct + { + unsigned prefetch_count:3; + unsigned pad:3; + unsigned prefetch_pointer:26; + } bits0; +}; + +struct brw_system_instruction_pointer +{ + struct header header; + + struct + { + unsigned pad:4; + unsigned system_instruction_pointer:28; + } bits0; +}; + + + + +/* State structs for the various fixed function units: + */ + + +struct thread0 +{ + unsigned pad0:1; + unsigned grf_reg_count:3; + unsigned pad1:2; + unsigned kernel_start_pointer:26; +}; + +struct thread1 +{ + unsigned ext_halt_exception_enable:1; + unsigned sw_exception_enable:1; + unsigned mask_stack_exception_enable:1; + unsigned timeout_exception_enable:1; + unsigned illegal_op_exception_enable:1; + unsigned pad0:3; + unsigned depth_coef_urb_read_offset:6; /* WM only */ + unsigned pad1:2; + unsigned floating_point_mode:1; + unsigned thread_priority:1; + unsigned binding_table_entry_count:8; + unsigned pad3:5; + unsigned single_program_flow:1; +}; + +struct thread2 +{ + unsigned per_thread_scratch_space:4; + unsigned pad0:6; + unsigned scratch_space_base_pointer:22; +}; + + +struct thread3 +{ + unsigned dispatch_grf_start_reg:4; + unsigned urb_entry_read_offset:6; + unsigned pad0:1; + unsigned urb_entry_read_length:6; + unsigned pad1:1; + unsigned const_urb_entry_read_offset:6; + unsigned pad2:1; + unsigned const_urb_entry_read_length:6; + unsigned pad3:1; +}; + + + +struct brw_clip_unit_state +{ + struct thread0 thread0; + struct + { + unsigned pad0:7; + unsigned sw_exception_enable:1; + unsigned pad1:3; + unsigned mask_stack_exception_enable:1; + unsigned pad2:1; + unsigned illegal_op_exception_enable:1; + unsigned pad3:2; + unsigned floating_point_mode:1; + unsigned thread_priority:1; + unsigned binding_table_entry_count:8; + unsigned pad4:5; + unsigned single_program_flow:1; + } thread1; + + struct thread2 thread2; + struct thread3 thread3; + + struct + { + unsigned pad0:9; + unsigned gs_output_stats:1; /* not always */ + unsigned stats_enable:1; + unsigned nr_urb_entries:7; + unsigned pad1:1; + unsigned urb_entry_allocation_size:5; + unsigned pad2:1; + unsigned max_threads:1; /* may be less */ + unsigned pad3:6; + } thread4; + + struct + { + unsigned pad0:13; + unsigned clip_mode:3; + unsigned userclip_enable_flags:8; + unsigned userclip_must_clip:1; + unsigned pad1:1; + unsigned guard_band_enable:1; + unsigned viewport_z_clip_enable:1; + unsigned viewport_xy_clip_enable:1; + unsigned vertex_position_space:1; + unsigned api_mode:1; + unsigned pad2:1; + } clip5; + + struct + { + unsigned pad0:5; + unsigned clipper_viewport_state_ptr:27; + } clip6; + + + float viewport_xmin; + float viewport_xmax; + float viewport_ymin; + float viewport_ymax; +}; + + + +struct brw_cc_unit_state +{ + struct + { + unsigned pad0:3; + unsigned bf_stencil_pass_depth_pass_op:3; + unsigned bf_stencil_pass_depth_fail_op:3; + unsigned bf_stencil_fail_op:3; + unsigned bf_stencil_func:3; + unsigned bf_stencil_enable:1; + unsigned pad1:2; + unsigned stencil_write_enable:1; + unsigned stencil_pass_depth_pass_op:3; + unsigned stencil_pass_depth_fail_op:3; + unsigned stencil_fail_op:3; + unsigned stencil_func:3; + unsigned stencil_enable:1; + } cc0; + + + struct + { + unsigned bf_stencil_ref:8; + unsigned stencil_write_mask:8; + unsigned stencil_test_mask:8; + unsigned stencil_ref:8; + } cc1; + + + struct + { + unsigned logicop_enable:1; + unsigned pad0:10; + unsigned depth_write_enable:1; + unsigned depth_test_function:3; + unsigned depth_test:1; + unsigned bf_stencil_write_mask:8; + unsigned bf_stencil_test_mask:8; + } cc2; + + + struct + { + unsigned pad0:8; + unsigned alpha_test_func:3; + unsigned alpha_test:1; + unsigned blend_enable:1; + unsigned ia_blend_enable:1; + unsigned pad1:1; + unsigned alpha_test_format:1; + unsigned pad2:16; + } cc3; + + struct + { + unsigned pad0:5; + unsigned cc_viewport_state_offset:27; + } cc4; + + struct + { + unsigned pad0:2; + unsigned ia_dest_blend_factor:5; + unsigned ia_src_blend_factor:5; + unsigned ia_blend_function:3; + unsigned statistics_enable:1; + unsigned logicop_func:4; + unsigned pad1:11; + unsigned dither_enable:1; + } cc5; + + struct + { + unsigned clamp_post_alpha_blend:1; + unsigned clamp_pre_alpha_blend:1; + unsigned clamp_range:2; + unsigned pad0:11; + unsigned y_dither_offset:2; + unsigned x_dither_offset:2; + unsigned dest_blend_factor:5; + unsigned src_blend_factor:5; + unsigned blend_function:3; + } cc6; + + struct { + union { + float f; + ubyte ub[4]; + } alpha_ref; + } cc7; +}; + + + +struct brw_sf_unit_state +{ + struct thread0 thread0; + struct thread1 thread1; + struct thread2 thread2; + struct thread3 thread3; + + struct + { + unsigned pad0:10; + unsigned stats_enable:1; + unsigned nr_urb_entries:7; + unsigned pad1:1; + unsigned urb_entry_allocation_size:5; + unsigned pad2:1; + unsigned max_threads:6; + unsigned pad3:1; + } thread4; + + struct + { + unsigned front_winding:1; + unsigned viewport_transform:1; + unsigned pad0:3; + unsigned sf_viewport_state_offset:27; + } sf5; + + struct + { + unsigned pad0:9; + unsigned dest_org_vbias:4; + unsigned dest_org_hbias:4; + unsigned scissor:1; + unsigned disable_2x2_trifilter:1; + unsigned disable_zero_pix_trifilter:1; + unsigned point_rast_rule:2; + unsigned line_endcap_aa_region_width:2; + unsigned line_width:4; + unsigned fast_scissor_disable:1; + unsigned cull_mode:2; + unsigned aa_enable:1; + } sf6; + + struct + { + unsigned point_size:11; + unsigned use_point_size_state:1; + unsigned subpixel_precision:1; + unsigned sprite_point:1; + unsigned pad0:11; + unsigned trifan_pv:2; + unsigned linestrip_pv:2; + unsigned tristrip_pv:2; + unsigned line_last_pixel_enable:1; + } sf7; + +}; + + +struct brw_gs_unit_state +{ + struct thread0 thread0; + struct thread1 thread1; + struct thread2 thread2; + struct thread3 thread3; + + struct + { + unsigned pad0:10; + unsigned stats_enable:1; + unsigned nr_urb_entries:7; + unsigned pad1:1; + unsigned urb_entry_allocation_size:5; + unsigned pad2:1; + unsigned max_threads:1; + unsigned pad3:6; + } thread4; + + struct + { + unsigned sampler_count:3; + unsigned pad0:2; + unsigned sampler_state_pointer:27; + } gs5; + + + struct + { + unsigned max_vp_index:4; + unsigned pad0:26; + unsigned reorder_enable:1; + unsigned pad1:1; + } gs6; +}; + + +struct brw_vs_unit_state +{ + struct thread0 thread0; + struct thread1 thread1; + struct thread2 thread2; + struct thread3 thread3; + + struct + { + unsigned pad0:10; + unsigned stats_enable:1; + unsigned nr_urb_entries:7; + unsigned pad1:1; + unsigned urb_entry_allocation_size:5; + unsigned pad2:1; + unsigned max_threads:4; + unsigned pad3:3; + } thread4; + + struct + { + unsigned sampler_count:3; + unsigned pad0:2; + unsigned sampler_state_pointer:27; + } vs5; + + struct + { + unsigned vs_enable:1; + unsigned vert_cache_disable:1; + unsigned pad0:30; + } vs6; +}; + + +struct brw_wm_unit_state +{ + struct thread0 thread0; + struct thread1 thread1; + struct thread2 thread2; + struct thread3 thread3; + + struct { + unsigned stats_enable:1; + unsigned pad0:1; + unsigned sampler_count:3; + unsigned sampler_state_pointer:27; + } wm4; + + struct + { + unsigned enable_8_pix:1; + unsigned enable_16_pix:1; + unsigned enable_32_pix:1; + unsigned pad0:7; + unsigned legacy_global_depth_bias:1; + unsigned line_stipple:1; + unsigned depth_offset:1; + unsigned polygon_stipple:1; + unsigned line_aa_region_width:2; + unsigned line_endcap_aa_region_width:2; + unsigned early_depth_test:1; + unsigned thread_dispatch_enable:1; + unsigned program_uses_depth:1; + unsigned program_computes_depth:1; + unsigned program_uses_killpixel:1; + unsigned legacy_line_rast: 1; + unsigned pad1:1; + unsigned max_threads:6; + unsigned pad2:1; + } wm5; + + float global_depth_offset_constant; + float global_depth_offset_scale; +}; + +struct brw_sampler_default_color { + float color[4]; +}; + +struct brw_sampler_state +{ + + struct + { + unsigned shadow_function:3; + unsigned lod_bias:11; + unsigned min_filter:3; + unsigned mag_filter:3; + unsigned mip_filter:2; + unsigned base_level:5; + unsigned pad:1; + unsigned lod_preclamp:1; + unsigned default_color_mode:1; + unsigned pad0:1; + unsigned disable:1; + } ss0; + + struct + { + unsigned r_wrap_mode:3; + unsigned t_wrap_mode:3; + unsigned s_wrap_mode:3; + unsigned pad:3; + unsigned max_lod:10; + unsigned min_lod:10; + } ss1; + + + struct + { + unsigned pad:5; + unsigned default_color_pointer:27; + } ss2; + + struct + { + unsigned pad:19; + unsigned max_aniso:3; + unsigned chroma_key_mode:1; + unsigned chroma_key_index:2; + unsigned chroma_key_enable:1; + unsigned monochrome_filter_width:3; + unsigned monochrome_filter_height:3; + } ss3; +}; + + +struct brw_clipper_viewport +{ + float xmin; + float xmax; + float ymin; + float ymax; +}; + +struct brw_cc_viewport +{ + float min_depth; + float max_depth; +}; + +struct brw_sf_viewport +{ + struct { + float m00; + float m11; + float m22; + float m30; + float m31; + float m32; + } viewport; + + struct { + short xmin; + short ymin; + short xmax; + short ymax; + } scissor; +}; + +/* Documented in the subsystem/shared-functions/sampler chapter... + */ +struct brw_surface_state +{ + struct { + unsigned cube_pos_z:1; + unsigned cube_neg_z:1; + unsigned cube_pos_y:1; + unsigned cube_neg_y:1; + unsigned cube_pos_x:1; + unsigned cube_neg_x:1; + unsigned pad:4; + unsigned mipmap_layout_mode:1; + unsigned vert_line_stride_ofs:1; + unsigned vert_line_stride:1; + unsigned color_blend:1; + unsigned writedisable_blue:1; + unsigned writedisable_green:1; + unsigned writedisable_red:1; + unsigned writedisable_alpha:1; + unsigned surface_format:9; + unsigned data_return_format:1; + unsigned pad0:1; + unsigned surface_type:3; + } ss0; + + struct { + unsigned base_addr; + } ss1; + + struct { + unsigned pad:2; + unsigned mip_count:4; + unsigned width:13; + unsigned height:13; + } ss2; + + struct { + unsigned tile_walk:1; + unsigned tiled_surface:1; + unsigned pad:1; + unsigned pitch:18; + unsigned depth:11; + } ss3; + + struct { + unsigned pad:19; + unsigned min_array_elt:9; + unsigned min_lod:4; + } ss4; +}; + + + +struct brw_vertex_buffer_state +{ + struct { + unsigned pitch:11; + unsigned pad:15; + unsigned access_type:1; + unsigned vb_index:5; + } vb0; + + unsigned start_addr; + unsigned max_index; +#if 1 + unsigned instance_data_step_rate; /* not included for sequential/random vertices? */ +#endif +}; + +#define BRW_VBP_MAX 17 + +struct brw_vb_array_state { + struct header header; + struct brw_vertex_buffer_state vb[BRW_VBP_MAX]; +}; + + +struct brw_vertex_element_state +{ + struct + { + unsigned src_offset:11; + unsigned pad:5; + unsigned src_format:9; + unsigned pad0:1; + unsigned valid:1; + unsigned vertex_buffer_index:5; + } ve0; + + struct + { + unsigned dst_offset:8; + unsigned pad:8; + unsigned vfcomponent3:4; + unsigned vfcomponent2:4; + unsigned vfcomponent1:4; + unsigned vfcomponent0:4; + } ve1; +}; + +#define BRW_VEP_MAX 18 + +struct brw_vertex_element_packet { + struct header header; + struct brw_vertex_element_state ve[BRW_VEP_MAX]; /* note: less than _TNL_ATTRIB_MAX */ +}; + + +struct brw_urb_immediate { + unsigned opcode:4; + unsigned offset:6; + unsigned swizzle_control:2; + unsigned pad:1; + unsigned allocate:1; + unsigned used:1; + unsigned complete:1; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; +}; + +/* Instruction format for the execution units: + */ + +struct brw_instruction +{ + struct + { + unsigned opcode:7; + unsigned pad:1; + unsigned access_mode:1; + unsigned mask_control:1; + unsigned dependency_control:2; + unsigned compression_control:2; + unsigned thread_control:2; + unsigned predicate_control:4; + unsigned predicate_inverse:1; + unsigned execution_size:3; + unsigned destreg__conditonalmod:4; /* destreg - send, conditionalmod - others */ + unsigned pad0:2; + unsigned debug_control:1; + unsigned saturate:1; + } header; + + union { + struct + { + unsigned dest_reg_file:2; + unsigned dest_reg_type:3; + unsigned src0_reg_file:2; + unsigned src0_reg_type:3; + unsigned src1_reg_file:2; + unsigned src1_reg_type:3; + unsigned pad:1; + unsigned dest_subreg_nr:5; + unsigned dest_reg_nr:8; + unsigned dest_horiz_stride:2; + unsigned dest_address_mode:1; + } da1; + + struct + { + unsigned dest_reg_file:2; + unsigned dest_reg_type:3; + unsigned src0_reg_file:2; + unsigned src0_reg_type:3; + unsigned pad:6; + int dest_indirect_offset:10; /* offset against the deref'd address reg */ + unsigned dest_subreg_nr:3; /* subnr for the address reg a0.x */ + unsigned dest_horiz_stride:2; + unsigned dest_address_mode:1; + } ia1; + + struct + { + unsigned dest_reg_file:2; + unsigned dest_reg_type:3; + unsigned src0_reg_file:2; + unsigned src0_reg_type:3; + unsigned src1_reg_file:2; + unsigned src1_reg_type:3; + unsigned pad0:1; + unsigned dest_writemask:4; + unsigned dest_subreg_nr:1; + unsigned dest_reg_nr:8; + unsigned pad1:2; + unsigned dest_address_mode:1; + } da16; + + struct + { + unsigned dest_reg_file:2; + unsigned dest_reg_type:3; + unsigned src0_reg_file:2; + unsigned src0_reg_type:3; + unsigned pad0:6; + unsigned dest_writemask:4; + int dest_indirect_offset:6; + unsigned dest_subreg_nr:3; + unsigned pad1:2; + unsigned dest_address_mode:1; + } ia16; + } bits1; + + + union { + struct + { + unsigned src0_subreg_nr:5; + unsigned src0_reg_nr:8; + unsigned src0_abs:1; + unsigned src0_negate:1; + unsigned src0_address_mode:1; + unsigned src0_horiz_stride:2; + unsigned src0_width:3; + unsigned src0_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad:6; + } da1; + + struct + { + int src0_indirect_offset:10; + unsigned src0_subreg_nr:3; + unsigned src0_abs:1; + unsigned src0_negate:1; + unsigned src0_address_mode:1; + unsigned src0_horiz_stride:2; + unsigned src0_width:3; + unsigned src0_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad:6; + } ia1; + + struct + { + unsigned src0_swz_x:2; + unsigned src0_swz_y:2; + unsigned src0_subreg_nr:1; + unsigned src0_reg_nr:8; + unsigned src0_abs:1; + unsigned src0_negate:1; + unsigned src0_address_mode:1; + unsigned src0_swz_z:2; + unsigned src0_swz_w:2; + unsigned pad0:1; + unsigned src0_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad1:6; + } da16; + + struct + { + unsigned src0_swz_x:2; + unsigned src0_swz_y:2; + int src0_indirect_offset:6; + unsigned src0_subreg_nr:3; + unsigned src0_abs:1; + unsigned src0_negate:1; + unsigned src0_address_mode:1; + unsigned src0_swz_z:2; + unsigned src0_swz_w:2; + unsigned pad0:1; + unsigned src0_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad1:6; + } ia16; + + } bits2; + + union + { + struct + { + unsigned src1_subreg_nr:5; + unsigned src1_reg_nr:8; + unsigned src1_abs:1; + unsigned src1_negate:1; + unsigned pad:1; + unsigned src1_horiz_stride:2; + unsigned src1_width:3; + unsigned src1_vert_stride:4; + unsigned pad0:7; + } da1; + + struct + { + unsigned src1_swz_x:2; + unsigned src1_swz_y:2; + unsigned src1_subreg_nr:1; + unsigned src1_reg_nr:8; + unsigned src1_abs:1; + unsigned src1_negate:1; + unsigned pad0:1; + unsigned src1_swz_z:2; + unsigned src1_swz_w:2; + unsigned pad1:1; + unsigned src1_vert_stride:4; + unsigned pad2:7; + } da16; + + struct + { + int src1_indirect_offset:10; + unsigned src1_subreg_nr:3; + unsigned src1_abs:1; + unsigned src1_negate:1; + unsigned pad0:1; + unsigned src1_horiz_stride:2; + unsigned src1_width:3; + unsigned src1_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad1:6; + } ia1; + + struct + { + unsigned src1_swz_x:2; + unsigned src1_swz_y:2; + int src1_indirect_offset:6; + unsigned src1_subreg_nr:3; + unsigned src1_abs:1; + unsigned src1_negate:1; + unsigned pad0:1; + unsigned src1_swz_z:2; + unsigned src1_swz_w:2; + unsigned pad1:1; + unsigned src1_vert_stride:4; + unsigned flag_reg_nr:1; + unsigned pad2:6; + } ia16; + + + struct + { + int jump_count:16; /* note: signed */ + unsigned pop_count:4; + unsigned pad0:12; + } if_else; + + struct { + unsigned function:4; + unsigned int_type:1; + unsigned precision:1; + unsigned saturate:1; + unsigned data_type:1; + unsigned pad0:8; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; + } math; + + struct { + unsigned binding_table_index:8; + unsigned sampler:4; + unsigned return_format:2; + unsigned msg_type:2; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; + } sampler; + + struct brw_urb_immediate urb; + + struct { + unsigned binding_table_index:8; + unsigned msg_control:4; + unsigned msg_type:2; + unsigned target_cache:2; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; + } dp_read; + + struct { + unsigned binding_table_index:8; + unsigned msg_control:3; + unsigned pixel_scoreboard_clear:1; + unsigned msg_type:3; + unsigned send_commit_msg:1; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; + } dp_write; + + struct { + unsigned pad:16; + unsigned response_length:4; + unsigned msg_length:4; + unsigned msg_target:4; + unsigned pad1:3; + unsigned end_of_thread:1; + } generic; + + int d; + unsigned ud; + } bits3; +}; + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_surface.c b/src/gallium/drivers/i965simple/brw_surface.c new file mode 100644 index 0000000000..518845e4b2 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_surface.c @@ -0,0 +1,210 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "brw_blit.h" +#include "brw_context.h" +#include "brw_state.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/util/p_tile.h" + + +/* + * XXX note: same as code in sp_surface.c + */ +static struct pipe_surface * +brw_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice) +{ + struct brw_texture *tex = (struct brw_texture *)pt; + struct pipe_surface *ps; + unsigned offset; /* in bytes */ + + offset = tex->level_offset[level]; + + if (pt->target == PIPE_TEXTURE_CUBE) { + offset += tex->image_offset[level][face] * pt->cpp; + } + else if (pt->target == PIPE_TEXTURE_3D) { + offset += tex->image_offset[level][zslice] * pt->cpp; + } + else { + assert(face == 0); + assert(zslice == 0); + } + + ps = pipe->winsys->surface_alloc(pipe->winsys); + if (ps) { + assert(ps->format); + assert(ps->refcount); + pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer); + ps->format = pt->format; + ps->cpp = pt->cpp; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->pitch = tex->pitch; + ps->offset = offset; + } + return ps; +} + + +/* Upload data to a rectangular sub-region. Lots of choices how to do this: + * + * - memcpy by span to current destination + * - upload data as new buffer and blit + * + * Currently always memcpy. + */ +static void +brw_surface_data(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + const void *src, unsigned src_pitch, + unsigned srcx, unsigned srcy, unsigned width, unsigned height) +{ + pipe_copy_rect(pipe_surface_map(dst) + dst->offset, + dst->cpp, dst->pitch, + dstx, dsty, width, height, src, src_pitch, srcx, srcy); + + pipe_surface_unmap(dst); +} + + +/* Assumes all values are within bounds -- no checking at this level - + * do it higher up if required. + */ +static void +brw_surface_copy(struct pipe_context *pipe, + unsigned do_flip, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + struct pipe_surface *src, + unsigned srcx, unsigned srcy, unsigned width, unsigned height) +{ + assert(dst != src); + assert(dst->cpp == src->cpp); + + if (0) { + pipe_copy_rect(pipe_surface_map(dst) + dst->offset, + dst->cpp, + dst->pitch, + dstx, dsty, + width, height, + pipe_surface_map(src) + src->offset, + do_flip ? -src->pitch : src->pitch, + srcx, do_flip ? 1 - srcy - height : srcy); + + pipe_surface_unmap(src); + pipe_surface_unmap(dst); + } + else { + brw_copy_blit(brw_context(pipe), + do_flip, + dst->cpp, + (short) src->pitch, src->buffer, src->offset, FALSE, + (short) dst->pitch, dst->buffer, dst->offset, FALSE, + (short) srcx, (short) srcy, (short) dstx, (short) dsty, + (short) width, (short) height, PIPE_LOGICOP_COPY); + } +} + +/* Fill a rectangular sub-region. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +static void * +get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y) +{ + return (char *)dst_map + (y * dst->pitch + x) * dst->cpp; +} + + +static void +brw_surface_fill(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, unsigned value) +{ + if (0) { + unsigned i, j; + void *dst_map = pipe_surface_map(dst); + + switch (dst->cpp) { + case 1: { + ubyte *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + memset(row, value, width); + row += dst->pitch; + } + } + break; + case 2: { + ushort *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = (ushort) value; + row += dst->pitch; + } + } + break; + case 4: { + unsigned *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = value; + row += dst->pitch; + } + } + break; + default: + assert(0); + break; + } + + pipe_surface_unmap( dst ); + } + else { + brw_fill_blit(brw_context(pipe), + dst->cpp, + (short) dst->pitch, + dst->buffer, dst->offset, FALSE, + (short) dstx, (short) dsty, + (short) width, (short) height, + value); + } +} + +void +brw_init_surface_functions(struct brw_context *brw) +{ + brw->pipe.get_tex_surface = brw_get_tex_surface; + brw->pipe.surface_copy = brw_surface_copy; + brw->pipe.surface_fill = brw_surface_fill; +} diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.c b/src/gallium/drivers/i965simple/brw_tex_layout.c new file mode 100644 index 0000000000..90561f1307 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_tex_layout.c @@ -0,0 +1,353 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +/* Code to layout images in a mipmap tree for i965. + */ + +#include "brw_tex_layout.h" + +#include "pipe/p_state.h" +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" + +#include "brw_context.h" + +#define FILE_DEBUG_FLAG DEBUG_TEXTURE + +#if 0 +unsigned intel_compressed_alignment(unsigned internalFormat) +{ + unsigned alignment = 4; + + switch (internalFormat) { + case GL_COMPRESSED_RGB_FXT1_3DFX: + case GL_COMPRESSED_RGBA_FXT1_3DFX: + alignment = 8; + break; + + default: + break; + } + + return alignment; +} +#endif + +static unsigned minify( unsigned d ) +{ + return MAX2(1, d>>1); +} + + +static boolean brw_miptree_layout(struct pipe_context *, struct brw_texture *); + +static void intel_miptree_set_image_offset(struct brw_texture *tex, + unsigned level, + unsigned img, + unsigned x, unsigned y) +{ + struct pipe_texture *pt = &tex->base; + if (img == 0 && level == 0) + assert(x == 0 && y == 0); + assert(img < tex->nr_images[level]); + + tex->image_offset[level][img] = (x + y * tex->pitch) * pt->cpp; +} + +static void intel_miptree_set_level_info(struct brw_texture *tex, + unsigned level, + unsigned nr_images, + unsigned x, unsigned y, + unsigned w, unsigned h, unsigned d) +{ + struct pipe_texture *pt = &tex->base; + + assert(level < PIPE_MAX_TEXTURE_LEVELS); + + pt->width[level] = w; + pt->height[level] = h; + pt->depth[level] = d; + + tex->level_offset[level] = (x + y * tex->pitch) * pt->cpp; + tex->nr_images[level] = nr_images; + + /* + DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__, + level, w, h, d, x, y, tex->level_offset[level]); + */ + + /* Not sure when this would happen, but anyway: + */ + if (tex->image_offset[level]) { + FREE(tex->image_offset[level]); + tex->image_offset[level] = NULL; + } + + assert(nr_images); + assert(!tex->image_offset[level]); + + tex->image_offset[level] = (unsigned *) MALLOC(nr_images * sizeof(unsigned)); + tex->image_offset[level][0] = 0; +} + +static void i945_miptree_layout_2d(struct brw_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + unsigned align_h = 2, align_w = 4; + unsigned level; + unsigned x = 0; + unsigned y = 0; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + + tex->pitch = pt->width[0]; + +#if 0 + if (pt->compressed) { + align_w = intel_compressed_alignment(pt->internal_format); + tex->pitch = ALIGN(pt->width[0], align_w); + } +#endif + + /* May need to adjust pitch to accomodate the placement of + * the 2nd mipmap. This occurs when the alignment + * constraints of mipmap placement push the right edge of the + * 2nd mipmap out past the width of its parent. + */ + if (pt->last_level > 0) { + unsigned mip1_width; + + if (pt->compressed) { + mip1_width = align(minify(pt->width[0]), align_w) + + align(minify(minify(pt->width[0])), align_w); + } else { + mip1_width = align(minify(pt->width[0]), align_w) + + minify(minify(pt->width[0])); + } + + if (mip1_width > tex->pitch) { + tex->pitch = mip1_width; + } + } + + /* Pitch must be a whole number of dwords, even though we + * express it in texels. + */ + tex->pitch = align(tex->pitch * pt->cpp, 4) / pt->cpp; + tex->total_height = 0; + + for (level = 0; level <= pt->last_level; level++) { + unsigned img_height; + + intel_miptree_set_level_info(tex, level, 1, x, y, width, + height, 1); + + if (pt->compressed) + img_height = MAX2(1, height/4); + else + img_height = align(height, align_h); + + + /* Because the images are packed better, the final offset + * might not be the maximal one: + */ + tex->total_height = MAX2(tex->total_height, y + img_height); + + /* Layout_below: step right after second mipmap. + */ + if (level == 1) { + x += align(width, align_w); + } + else { + y += img_height; + } + + width = minify(width); + height = minify(height); + } +} + +static boolean brw_miptree_layout(struct pipe_context *pipe, struct brw_texture *tex) +{ + struct pipe_texture *pt = &tex->base; + /* XXX: these vary depending on image format: + */ +/* int align_w = 4; */ + + switch (pt->target) { + case PIPE_TEXTURE_CUBE: + case PIPE_TEXTURE_3D: { + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + unsigned pack_x_pitch, pack_x_nr; + unsigned pack_y_pitch; + unsigned level; + unsigned align_h = 2; + unsigned align_w = 4; + + tex->total_height = 0; +#if 0 + if (pt->compressed) { + align_w = intel_compressed_alignment(pt->internal_format); + pt->pitch = align(width, align_w); + pack_y_pitch = (height + 3) / 4; + } else +#endif + { + tex->pitch = align(pt->width[0] * pt->cpp, 4) / pt->cpp; + pack_y_pitch = align(pt->height[0], align_h); + } + + pack_x_pitch = tex->pitch; + pack_x_nr = 1; + + for (level = 0; level <= pt->last_level; level++) { + unsigned nr_images = pt->target == PIPE_TEXTURE_3D ? depth : 6; + int x = 0; + int y = 0; + uint q, j; + + intel_miptree_set_level_info(tex, level, nr_images, + 0, tex->total_height, + width, height, depth); + + for (q = 0; q < nr_images;) { + for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) { + intel_miptree_set_image_offset(tex, level, q, x, y); + x += pack_x_pitch; + } + + x = 0; + y += pack_y_pitch; + } + + + tex->total_height += y; + width = minify(width); + height = minify(height); + depth = minify(depth); + + if (pt->compressed) { + pack_y_pitch = (height + 3) / 4; + + if (pack_x_pitch > align(width, align_w)) { + pack_x_pitch = align(width, align_w); + pack_x_nr <<= 1; + } + } else { + if (pack_x_pitch > 4) { + pack_x_pitch >>= 1; + pack_x_nr <<= 1; + assert(pack_x_pitch * pack_x_nr <= tex->pitch); + } + + if (pack_y_pitch > 2) { + pack_y_pitch >>= 1; + pack_y_pitch = align(pack_y_pitch, align_h); + } + } + + } + break; + } + + default: + i945_miptree_layout_2d(tex); + break; + } +#if 0 + PRINT("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, + pt->pitch, + pt->total_height, + pt->cpp, + pt->pitch * pt->total_height * pt->cpp ); +#endif + + return TRUE; +} + + +struct pipe_texture * +brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat) +{ + struct brw_texture *tex = CALLOC_STRUCT(brw_texture); + + if (tex) { + tex->base = *templat; + + if (brw_miptree_layout(pipe, tex)) + tex->buffer = pipe->winsys->buffer_create(pipe->winsys, 64, + PIPE_BUFFER_USAGE_PIXEL, + tex->pitch * tex->base.cpp * + tex->total_height); + + if (!tex->buffer) { + FREE(tex); + return NULL; + } + } + + return &tex->base; +} + +void +brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt) +{ + if (!*pt) + return; + + /* + DBG("%s %p refcount will be %d\n", + __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); + */ + if (--(*pt)->refcount <= 0) { + struct brw_texture *tex = (struct brw_texture *)*pt; + uint i; + + /* + DBG("%s deleting %p\n", __FUNCTION__, (void *) tex); + */ + + pipe_buffer_reference(pipe->winsys, &tex->buffer, NULL); + + for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++) + if (tex->image_offset[i]) + free(tex->image_offset[i]); + + free(tex); + } + *pt = NULL; +} diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.h b/src/gallium/drivers/i965simple/brw_tex_layout.h new file mode 100644 index 0000000000..cfd6b1ef3a --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_tex_layout.h @@ -0,0 +1,15 @@ +#ifndef BRW_TEX_LAYOUT_H +#define BRW_TEX_LAYOUT_H + +#include "pipe/p_compiler.h" + +struct pipe_context; +struct pipe_texture; + +extern struct pipe_texture * +brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat); + +extern void +brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_urb.c b/src/gallium/drivers/i965simple/brw_urb.c new file mode 100644 index 0000000000..101a4367b9 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_urb.c @@ -0,0 +1,186 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +//#include "brw_state.h" +#include "brw_batch.h" +#include "brw_defines.h" + +#define VS 0 +#define GS 1 +#define CLP 2 +#define SF 3 +#define CS 4 + +/* XXX: Are the min_entry_size numbers useful? + * XXX: Verify min_nr_entries, esp for VS. + * XXX: Verify SF min_entry_size. + */ +static const struct { + unsigned min_nr_entries; + unsigned preferred_nr_entries; + unsigned min_entry_size; + unsigned max_entry_size; +} limits[CS+1] = { + { 8, 32, 1, 5 }, /* vs */ + { 4, 8, 1, 5 }, /* gs */ + { 6, 8, 1, 5 }, /* clp */ + { 1, 8, 1, 12 }, /* sf */ + { 1, 4, 1, 32 } /* cs */ +}; + + +static boolean check_urb_layout( struct brw_context *brw ) +{ + brw->urb.vs_start = 0; + brw->urb.gs_start = brw->urb.nr_vs_entries * brw->urb.vsize; + brw->urb.clip_start = brw->urb.gs_start + brw->urb.nr_gs_entries * brw->urb.vsize; + brw->urb.sf_start = brw->urb.clip_start + brw->urb.nr_clip_entries * brw->urb.vsize; + brw->urb.cs_start = brw->urb.sf_start + brw->urb.nr_sf_entries * brw->urb.sfsize; + + return brw->urb.cs_start + brw->urb.nr_cs_entries * brw->urb.csize <= 256; +} + +/* Most minimal update, forces re-emit of URB fence packet after GS + * unit turned on/off. + */ +static void recalculate_urb_fence( struct brw_context *brw ) +{ + unsigned csize = brw->curbe.total_size; + unsigned vsize = brw->vs.prog_data->urb_entry_size; + unsigned sfsize = brw->sf.prog_data->urb_entry_size; + + if (csize < limits[CS].min_entry_size) + csize = limits[CS].min_entry_size; + + if (vsize < limits[VS].min_entry_size) + vsize = limits[VS].min_entry_size; + + if (sfsize < limits[SF].min_entry_size) + sfsize = limits[SF].min_entry_size; + + if (brw->urb.vsize < vsize || + brw->urb.sfsize < sfsize || + brw->urb.csize < csize || + (brw->urb.constrained && (brw->urb.vsize > brw->urb.vsize || + brw->urb.sfsize > brw->urb.sfsize || + brw->urb.csize > brw->urb.csize))) { + + + brw->urb.csize = csize; + brw->urb.sfsize = sfsize; + brw->urb.vsize = vsize; + + brw->urb.nr_vs_entries = limits[VS].preferred_nr_entries; + brw->urb.nr_gs_entries = limits[GS].preferred_nr_entries; + brw->urb.nr_clip_entries = limits[CLP].preferred_nr_entries; + brw->urb.nr_sf_entries = limits[SF].preferred_nr_entries; + brw->urb.nr_cs_entries = limits[CS].preferred_nr_entries; + + if (!check_urb_layout(brw)) { + brw->urb.nr_vs_entries = limits[VS].min_nr_entries; + brw->urb.nr_gs_entries = limits[GS].min_nr_entries; + brw->urb.nr_clip_entries = limits[CLP].min_nr_entries; + brw->urb.nr_sf_entries = limits[SF].min_nr_entries; + brw->urb.nr_cs_entries = limits[CS].min_nr_entries; + + brw->urb.constrained = 1; + + if (!check_urb_layout(brw)) { + /* This is impossible, given the maximal sizes of urb + * entries and the values for minimum nr of entries + * provided above. + */ + debug_printf("couldn't calculate URB layout!\n"); + exit(1); + } + + if (BRW_DEBUG & (DEBUG_URB|DEBUG_FALLBACKS)) + debug_printf("URB CONSTRAINED\n"); + } + else + brw->urb.constrained = 0; + + if (BRW_DEBUG & DEBUG_URB) + debug_printf("URB fence: %d ..VS.. %d ..GS.. %d ..CLP.. %d ..SF.. %d ..CS.. %d\n", + brw->urb.vs_start, + brw->urb.gs_start, + brw->urb.clip_start, + brw->urb.sf_start, + brw->urb.cs_start, + 256); + + brw->state.dirty.brw |= BRW_NEW_URB_FENCE; + } +} + + +const struct brw_tracked_state brw_recalculate_urb_fence = { + .dirty = { + .brw = BRW_NEW_CURBE_OFFSETS, + .cache = (CACHE_NEW_VS_PROG | + CACHE_NEW_SF_PROG) + }, + .update = recalculate_urb_fence +}; + + + + + +void brw_upload_urb_fence(struct brw_context *brw) +{ + struct brw_urb_fence uf; + memset(&uf, 0, sizeof(uf)); + + uf.header.opcode = CMD_URB_FENCE; + uf.header.length = sizeof(uf)/4-2; + uf.header.vs_realloc = 1; + uf.header.gs_realloc = 1; + uf.header.clp_realloc = 1; + uf.header.sf_realloc = 1; + uf.header.vfe_realloc = 1; + uf.header.cs_realloc = 1; + + /* The ordering below is correct, not the layout in the + * instruction. + * + * There are 256 urb reg pairs in total. + */ + uf.bits0.vs_fence = brw->urb.gs_start; + uf.bits0.gs_fence = brw->urb.clip_start; + uf.bits0.clp_fence = brw->urb.sf_start; + uf.bits1.sf_fence = brw->urb.cs_start; + uf.bits1.cs_fence = 256; + + BRW_BATCH_STRUCT(brw, &uf); +} diff --git a/src/gallium/drivers/i965simple/brw_util.c b/src/gallium/drivers/i965simple/brw_util.c new file mode 100644 index 0000000000..42391d7c8c --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_util.c @@ -0,0 +1,104 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_util.h" +#include "brw_defines.h" + +#include "pipe/p_defines.h" + +unsigned brw_count_bits( unsigned val ) +{ + unsigned i; + for (i = 0; val ; val >>= 1) + if (val & 1) + i++; + return i; +} + + +unsigned brw_translate_blend_equation( int mode ) +{ + switch (mode) { + case PIPE_BLEND_ADD: + return BRW_BLENDFUNCTION_ADD; + case PIPE_BLEND_MIN: + return BRW_BLENDFUNCTION_MIN; + case PIPE_BLEND_MAX: + return BRW_BLENDFUNCTION_MAX; + case PIPE_BLEND_SUBTRACT: + return BRW_BLENDFUNCTION_SUBTRACT; + case PIPE_BLEND_REVERSE_SUBTRACT: + return BRW_BLENDFUNCTION_REVERSE_SUBTRACT; + default: + assert(0); + return BRW_BLENDFUNCTION_ADD; + } +} + +unsigned brw_translate_blend_factor( int factor ) +{ + switch(factor) { + case PIPE_BLENDFACTOR_ZERO: + return BRW_BLENDFACTOR_ZERO; + case PIPE_BLENDFACTOR_SRC_ALPHA: + return BRW_BLENDFACTOR_SRC_ALPHA; + case PIPE_BLENDFACTOR_ONE: + return BRW_BLENDFACTOR_ONE; + case PIPE_BLENDFACTOR_SRC_COLOR: + return BRW_BLENDFACTOR_SRC_COLOR; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + return BRW_BLENDFACTOR_INV_SRC_COLOR; + case PIPE_BLENDFACTOR_DST_COLOR: + return BRW_BLENDFACTOR_DST_COLOR; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + return BRW_BLENDFACTOR_INV_DST_COLOR; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + return BRW_BLENDFACTOR_INV_SRC_ALPHA; + case PIPE_BLENDFACTOR_DST_ALPHA: + return BRW_BLENDFACTOR_DST_ALPHA; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + return BRW_BLENDFACTOR_INV_DST_ALPHA; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + return BRW_BLENDFACTOR_SRC_ALPHA_SATURATE; + case PIPE_BLENDFACTOR_CONST_COLOR: + return BRW_BLENDFACTOR_CONST_COLOR; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + return BRW_BLENDFACTOR_INV_CONST_COLOR; + case PIPE_BLENDFACTOR_CONST_ALPHA: + return BRW_BLENDFACTOR_CONST_ALPHA; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + return BRW_BLENDFACTOR_INV_CONST_ALPHA; + default: + assert(0); + return BRW_BLENDFACTOR_ZERO; + } +} diff --git a/src/gallium/drivers/i965simple/brw_util.h b/src/gallium/drivers/i965simple/brw_util.h new file mode 100644 index 0000000000..d60e5934db --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_util.h @@ -0,0 +1,43 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_UTIL_H +#define BRW_UTIL_H + +#include "pipe/p_state.h" + +extern unsigned brw_count_bits( unsigned val ); +extern unsigned brw_translate_blend_factor( int factor ); +extern unsigned brw_translate_blend_equation( int mode ); + + +#endif diff --git a/src/gallium/drivers/i965simple/brw_vs.c b/src/gallium/drivers/i965simple/brw_vs.c new file mode 100644 index 0000000000..738c6346d5 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs.c @@ -0,0 +1,120 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_vs.h" +#include "brw_util.h" +#include "brw_state.h" + + +static void do_vs_prog( struct brw_context *brw, + const struct brw_vertex_program *vp, + struct brw_vs_prog_key *key ) +{ + unsigned program_size; + const unsigned *program; + struct brw_vs_compile c; + + memset(&c, 0, sizeof(c)); + memcpy(&c.key, key, sizeof(*key)); + + brw_init_compile(&c.func); + c.vp = vp; + + c.prog_data.outputs_written = vp->program.num_outputs; + c.prog_data.inputs_read = vp->program.num_inputs; + +#if 0 + if (c.key.copy_edgeflag) { + c.prog_data.outputs_written |= 1<<VERT_RESULT_EDGE; + c.prog_data.inputs_read |= 1<<VERT_ATTRIB_EDGEFLAG; + } +#endif + + /* Emit GEN4 code. + */ + brw_vs_emit(&c); + + /* get the program + */ + program = brw_get_program(&c.func, &program_size); + + /* + */ + brw->vs.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_VS_PROG], + &c.key, + sizeof(c.key), + program, + program_size, + &c.prog_data, + &brw->vs.prog_data); +} + + +static void brw_upload_vs_prog( struct brw_context *brw ) +{ + struct brw_vs_prog_key key; + const struct brw_vertex_program *vp = brw->attribs.VertexProgram; + + assert(vp); + + memset(&key, 0, sizeof(key)); + + /* Just upload the program verbatim for now. Always send it all + * the inputs it asks for, whether they are varying or not. + */ + key.program_string_id = vp->id; + key.nr_userclip = brw->attribs.Clip.nr; + key.copy_edgeflag = (brw->attribs.Raster->fill_cw != PIPE_POLYGON_MODE_FILL || + brw->attribs.Raster->fill_ccw != PIPE_POLYGON_MODE_FILL); + + /* Make an early check for the key. + */ + if (brw_search_cache(&brw->cache[BRW_VS_PROG], + &key, sizeof(key), + &brw->vs.prog_data, + &brw->vs.prog_gs_offset)) + return; + + do_vs_prog(brw, vp, &key); +} + + +/* See brw_vs.c: + */ +const struct brw_tracked_state brw_vs_prog = { + .dirty = { + .brw = BRW_NEW_VS, + .cache = 0 + }, + .update = brw_upload_vs_prog +}; diff --git a/src/gallium/drivers/i965simple/brw_vs.h b/src/gallium/drivers/i965simple/brw_vs.h new file mode 100644 index 0000000000..0e58f043b0 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs.h @@ -0,0 +1,82 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_VS_H +#define BRW_VS_H + + +#include "brw_context.h" +#include "brw_eu.h" + + +struct brw_vs_prog_key { + unsigned program_string_id; + unsigned nr_userclip:4; + unsigned copy_edgeflag:1; + unsigned know_w_is_one:1; + unsigned pad:26; +}; + + +struct brw_vs_compile { + struct brw_compile func; + struct brw_vs_prog_key key; + struct brw_vs_prog_data prog_data; + + struct brw_vertex_program *vp; + + unsigned nr_inputs; + + unsigned first_output; + unsigned nr_outputs; + + unsigned first_tmp; + unsigned last_tmp; + + struct brw_reg r0; + struct brw_reg r1; + struct brw_reg regs[12][128]; + struct brw_reg tmp; + struct brw_reg stack; + + struct { + boolean used_in_src; + struct brw_reg reg; + } output_regs[128]; + + struct brw_reg userplane[6]; + +}; + +void brw_vs_emit( struct brw_vs_compile *c ); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_vs_emit.c b/src/gallium/drivers/i965simple/brw_vs_emit.c new file mode 100644 index 0000000000..98915ba101 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs_emit.c @@ -0,0 +1,1332 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_context.h" +#include "brw_vs.h" + +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" + +struct brw_prog_info { + unsigned num_temps; + unsigned num_addrs; + unsigned num_consts; + + unsigned writes_psize; + + unsigned pos_idx; + unsigned result_edge_idx; + unsigned edge_flag_idx; + unsigned psize_idx; +}; + +/* Do things as simply as possible. Allocate and populate all regs + * ahead of time. + */ +static void brw_vs_alloc_regs( struct brw_vs_compile *c, + struct brw_prog_info *info ) +{ + unsigned i, reg = 0, mrf; + unsigned nr_params; + + /* r0 -- reserved as usual + */ + c->r0 = brw_vec8_grf(reg, 0); reg++; + + /* User clip planes from curbe: + */ + if (c->key.nr_userclip) { + for (i = 0; i < c->key.nr_userclip; i++) { + c->userplane[i] = stride( brw_vec4_grf(reg+3+i/2, (i%2) * 4), 0, 4, 1); + } + + /* Deal with curbe alignment: + */ + reg += ((6+c->key.nr_userclip+3)/4)*2; + } + + /* Vertex program parameters from curbe: + */ + nr_params = c->prog_data.max_const; + for (i = 0; i < nr_params; i++) { + c->regs[TGSI_FILE_CONSTANT][i] = stride(brw_vec4_grf(reg+i/2, (i%2) * 4), 0, 4, 1); + } + reg += (nr_params+1)/2; + c->prog_data.curb_read_length = reg - 1; + + + + /* Allocate input regs: + */ + c->nr_inputs = c->vp->program.num_inputs; + for (i = 0; i < c->nr_inputs; i++) { + c->regs[TGSI_FILE_INPUT][i] = brw_vec8_grf(reg, 0); + reg++; + } + + + /* Allocate outputs: TODO: could organize the non-position outputs + * to go straight into message regs. + */ + c->nr_outputs = 0; + c->first_output = reg; + mrf = 4; + for (i = 0; i < c->vp->program.num_outputs; i++) { + c->nr_outputs++; +#if 0 + if (i == VERT_RESULT_HPOS) { + c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0); + reg++; + } + else if (i == VERT_RESULT_PSIZ) { + c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0); + reg++; + mrf++; /* just a placeholder? XXX fix later stages & remove this */ + } + else { + c->regs[TGSI_FILE_OUTPUT][i] = brw_message_reg(mrf); + mrf++; + } +#else + /*treat pos differently for now */ + if (i == info->pos_idx) { + c->regs[TGSI_FILE_OUTPUT][i] = brw_vec8_grf(reg, 0); + reg++; + } else { + c->regs[TGSI_FILE_OUTPUT][i] = brw_message_reg(mrf); + mrf++; + } +#endif + } + + /* Allocate program temporaries: + */ + for (i = 0; i < info->num_temps; i++) { + c->regs[TGSI_FILE_TEMPORARY][i] = brw_vec8_grf(reg, 0); + reg++; + } + + /* Address reg(s). Don't try to use the internal address reg until + * deref time. + */ + for (i = 0; i < info->num_addrs; i++) { + c->regs[TGSI_FILE_ADDRESS][i] = brw_reg(BRW_GENERAL_REGISTER_FILE, + reg, + 0, + BRW_REGISTER_TYPE_D, + BRW_VERTICAL_STRIDE_8, + BRW_WIDTH_8, + BRW_HORIZONTAL_STRIDE_1, + BRW_SWIZZLE_XXXX, + TGSI_WRITEMASK_X); + reg++; + } + + for (i = 0; i < 128; i++) { + if (c->output_regs[i].used_in_src) { + c->output_regs[i].reg = brw_vec8_grf(reg, 0); + reg++; + } + } + + c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg, 0); + reg += 2; + + + /* Some opcodes need an internal temporary: + */ + c->first_tmp = reg; + c->last_tmp = reg; /* for allocation purposes */ + + /* Each input reg holds data from two vertices. The + * urb_read_length is the number of registers read from *each* + * vertex urb, so is half the amount: + */ + c->prog_data.urb_read_length = (c->nr_inputs+1)/2; + + c->prog_data.urb_entry_size = (c->nr_outputs+2+3)/4; + c->prog_data.total_grf = reg; +} + + +static struct brw_reg get_tmp( struct brw_vs_compile *c ) +{ + struct brw_reg tmp = brw_vec8_grf(c->last_tmp, 0); + + if (++c->last_tmp > c->prog_data.total_grf) + c->prog_data.total_grf = c->last_tmp; + + return tmp; +} + +static void release_tmp( struct brw_vs_compile *c, struct brw_reg tmp ) +{ + if (tmp.nr == c->last_tmp-1) + c->last_tmp--; +} + +static void release_tmps( struct brw_vs_compile *c ) +{ + c->last_tmp = c->first_tmp; +} + + +static void unalias1( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0, + void (*func)( struct brw_vs_compile *, + struct brw_reg, + struct brw_reg )) +{ + if (dst.file == arg0.file && dst.nr == arg0.nr) { + struct brw_compile *p = &c->func; + struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask); + func(c, tmp, arg0); + brw_MOV(p, dst, tmp); + } + else { + func(c, dst, arg0); + } +} + +static void unalias2( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1, + void (*func)( struct brw_vs_compile *, + struct brw_reg, + struct brw_reg, + struct brw_reg )) +{ + if ((dst.file == arg0.file && dst.nr == arg0.nr) || + (dst.file == arg1.file && dst.nr == arg1.nr)) { + struct brw_compile *p = &c->func; + struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask); + func(c, tmp, arg0, arg1); + brw_MOV(p, dst, tmp); + } + else { + func(c, dst, arg0, arg1); + } +} + +static void emit_sop( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1, + unsigned cond) +{ + brw_push_insn_state(p); + brw_CMP(p, brw_null_reg(), cond, arg0, arg1); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_MOV(p, dst, brw_imm_f(1.0f)); + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + brw_MOV(p, dst, brw_imm_f(0.0f)); + brw_pop_insn_state(p); +} + +static void emit_seq( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_EQ); +} + +static void emit_sne( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_NEQ); +} +static void emit_slt( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_L); +} + +static void emit_sle( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_LE); +} + +static void emit_sgt( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_G); +} + +static void emit_sge( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + emit_sop(p, dst, arg0, arg1, BRW_CONDITIONAL_GE); +} + +static void emit_max( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, arg1); + brw_SEL(p, dst, arg1, arg0); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); +} + +static void emit_min( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1 ) +{ + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, arg1); + brw_SEL(p, dst, arg0, arg1); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); +} + + +static void emit_math1( struct brw_vs_compile *c, + unsigned function, + struct brw_reg dst, + struct brw_reg arg0, + unsigned precision) +{ + /* There are various odd behaviours with SEND on the simulator. In + * addition there are documented issues with the fact that the GEN4 + * processor doesn't do dependency control properly on SEND + * results. So, on balance, this kludge to get around failures + * with writemasked math results looks like it might be necessary + * whether that turns out to be a simulator bug or not: + */ + struct brw_compile *p = &c->func; + struct brw_reg tmp = dst; + boolean need_tmp = (dst.dw1.bits.writemask != 0xf || + dst.file != BRW_GENERAL_REGISTER_FILE); + + if (need_tmp) + tmp = get_tmp(c); + + brw_math(p, + tmp, + function, + BRW_MATH_SATURATE_NONE, + 2, + arg0, + BRW_MATH_DATA_SCALAR, + precision); + + if (need_tmp) { + brw_MOV(p, dst, tmp); + release_tmp(c, tmp); + } +} + +static void emit_math2( struct brw_vs_compile *c, + unsigned function, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1, + unsigned precision) +{ + struct brw_compile *p = &c->func; + struct brw_reg tmp = dst; + boolean need_tmp = (dst.dw1.bits.writemask != 0xf || + dst.file != BRW_GENERAL_REGISTER_FILE); + + if (need_tmp) + tmp = get_tmp(c); + + brw_MOV(p, brw_message_reg(3), arg1); + + brw_math(p, + tmp, + function, + BRW_MATH_SATURATE_NONE, + 2, + arg0, + BRW_MATH_DATA_SCALAR, + precision); + + if (need_tmp) { + brw_MOV(p, dst, tmp); + release_tmp(c, tmp); + } +} + + + +static void emit_exp_noalias( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0 ) +{ + struct brw_compile *p = &c->func; + + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_X) { + struct brw_reg tmp = get_tmp(c); + struct brw_reg tmp_d = retype(tmp, BRW_REGISTER_TYPE_D); + + /* tmp_d = floor(arg0.x) */ + brw_RNDD(p, tmp_d, brw_swizzle1(arg0, 0)); + + /* result[0] = 2.0 ^ tmp */ + + /* Adjust exponent for floating point: + * exp += 127 + */ + brw_ADD(p, brw_writemask(tmp_d, TGSI_WRITEMASK_X), tmp_d, brw_imm_d(127)); + + /* Install exponent and sign. + * Excess drops off the edge: + */ + brw_SHL(p, brw_writemask(retype(dst, BRW_REGISTER_TYPE_D), TGSI_WRITEMASK_X), + tmp_d, brw_imm_d(23)); + + release_tmp(c, tmp); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Y) { + /* result[1] = arg0.x - floor(arg0.x) */ + brw_FRC(p, brw_writemask(dst, TGSI_WRITEMASK_Y), brw_swizzle1(arg0, 0)); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z) { + /* As with the LOG instruction, we might be better off just + * doing a taylor expansion here, seeing as we have to do all + * the prep work. + * + * If mathbox partial precision is too low, consider also: + * result[3] = result[0] * EXP(result[1]) + */ + emit_math1(c, + BRW_MATH_FUNCTION_EXP, + brw_writemask(dst, TGSI_WRITEMASK_Z), + brw_swizzle1(arg0, 0), + BRW_MATH_PRECISION_PARTIAL); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W) { + /* result[3] = 1.0; */ + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_W), brw_imm_f(1)); + } +} + + +static void emit_log_noalias( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0 ) +{ + struct brw_compile *p = &c->func; + struct brw_reg tmp = dst; + struct brw_reg tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD); + struct brw_reg arg0_ud = retype(arg0, BRW_REGISTER_TYPE_UD); + boolean need_tmp = (dst.dw1.bits.writemask != 0xf || + dst.file != BRW_GENERAL_REGISTER_FILE); + + if (need_tmp) { + tmp = get_tmp(c); + tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD); + } + + /* Perform mant = frexpf(fabsf(x), &exp), adjust exp and mnt + * according to spec: + * + * These almost look likey they could be joined up, but not really + * practical: + * + * result[0].f = (x.i & ((1<<31)-1) >> 23) - 127 + * result[1].i = (x.i & ((1<<23)-1) + (127<<23) + */ + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_XZ) { + brw_AND(p, + brw_writemask(tmp_ud, TGSI_WRITEMASK_X), + brw_swizzle1(arg0_ud, 0), + brw_imm_ud((1U<<31)-1)); + + brw_SHR(p, + brw_writemask(tmp_ud, TGSI_WRITEMASK_X), + tmp_ud, + brw_imm_ud(23)); + + brw_ADD(p, + brw_writemask(tmp, TGSI_WRITEMASK_X), + retype(tmp_ud, BRW_REGISTER_TYPE_D), /* does it matter? */ + brw_imm_d(-127)); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_YZ) { + brw_AND(p, + brw_writemask(tmp_ud, TGSI_WRITEMASK_Y), + brw_swizzle1(arg0_ud, 0), + brw_imm_ud((1<<23)-1)); + + brw_OR(p, + brw_writemask(tmp_ud, TGSI_WRITEMASK_Y), + tmp_ud, + brw_imm_ud(127<<23)); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z) { + /* result[2] = result[0] + LOG2(result[1]); */ + + /* Why bother? The above is just a hint how to do this with a + * taylor series. Maybe we *should* use a taylor series as by + * the time all the above has been done it's almost certainly + * quicker than calling the mathbox, even with low precision. + * + * Options are: + * - result[0] + mathbox.LOG2(result[1]) + * - mathbox.LOG2(arg0.x) + * - result[0] + inline_taylor_approx(result[1]) + */ + emit_math1(c, + BRW_MATH_FUNCTION_LOG, + brw_writemask(tmp, TGSI_WRITEMASK_Z), + brw_swizzle1(tmp, 1), + BRW_MATH_PRECISION_FULL); + + brw_ADD(p, + brw_writemask(tmp, TGSI_WRITEMASK_Z), + brw_swizzle1(tmp, 2), + brw_swizzle1(tmp, 0)); + } + + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W) { + /* result[3] = 1.0; */ + brw_MOV(p, brw_writemask(tmp, TGSI_WRITEMASK_W), brw_imm_f(1)); + } + + if (need_tmp) { + brw_MOV(p, dst, tmp); + release_tmp(c, tmp); + } +} + + + + +/* Need to unalias - consider swizzles: r0 = DST r0.xxxx r1 + */ +static void emit_dst_noalias( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0, + struct brw_reg arg1) +{ + struct brw_compile *p = &c->func; + + /* There must be a better way to do this: + */ + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_X) + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_X), brw_imm_f(1.0)); + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Y) + brw_MUL(p, brw_writemask(dst, TGSI_WRITEMASK_Y), arg0, arg1); + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_Z) + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_Z), arg0); + if (dst.dw1.bits.writemask & TGSI_WRITEMASK_W) + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_W), arg1); +} + +static void emit_xpd( struct brw_compile *p, + struct brw_reg dst, + struct brw_reg t, + struct brw_reg u) +{ + brw_MUL(p, brw_null_reg(), brw_swizzle(t, 1,2,0,3), brw_swizzle(u,2,0,1,3)); + brw_MAC(p, dst, negate(brw_swizzle(t, 2,0,1,3)), brw_swizzle(u,1,2,0,3)); +} + + + +static void emit_lit_noalias( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0 ) +{ + struct brw_compile *p = &c->func; + struct brw_instruction *if_insn; + struct brw_reg tmp = dst; + boolean need_tmp = (dst.file != BRW_GENERAL_REGISTER_FILE); + + if (need_tmp) + tmp = get_tmp(c); + + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_YZ), brw_imm_f(0)); + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_XW), brw_imm_f(1)); + + /* Need to use BRW_EXECUTE_8 and also do an 8-wide compare in order + * to get all channels active inside the IF. In the clipping code + * we run with NoMask, so it's not an option and we can use + * BRW_EXECUTE_1 for all comparisions. + */ + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,0), brw_imm_f(0)); + if_insn = brw_IF(p, BRW_EXECUTE_8); + { + brw_MOV(p, brw_writemask(dst, TGSI_WRITEMASK_Y), brw_swizzle1(arg0,0)); + + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,1), brw_imm_f(0)); + brw_MOV(p, brw_writemask(tmp, TGSI_WRITEMASK_Z), brw_swizzle1(arg0,1)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + + emit_math2(c, + BRW_MATH_FUNCTION_POW, + brw_writemask(dst, TGSI_WRITEMASK_Z), + brw_swizzle1(tmp, 2), + brw_swizzle1(arg0, 3), + BRW_MATH_PRECISION_PARTIAL); + } + + brw_ENDIF(p, if_insn); +} + + + + + +/* TODO: relative addressing! + */ +static struct brw_reg get_reg( struct brw_vs_compile *c, + unsigned file, + unsigned index ) +{ + switch (file) { + case TGSI_FILE_TEMPORARY: + case TGSI_FILE_INPUT: + case TGSI_FILE_OUTPUT: + assert(c->regs[file][index].nr != 0); + return c->regs[file][index]; + case TGSI_FILE_CONSTANT: + assert(c->regs[TGSI_FILE_CONSTANT][index + c->prog_data.num_imm].nr != 0); + return c->regs[TGSI_FILE_CONSTANT][index + c->prog_data.num_imm]; + case TGSI_FILE_IMMEDIATE: + assert(c->regs[TGSI_FILE_CONSTANT][index].nr != 0); + return c->regs[TGSI_FILE_CONSTANT][index]; + case TGSI_FILE_ADDRESS: + assert(index == 0); + return c->regs[file][index]; + + case TGSI_FILE_NULL: /* undef values */ + return brw_null_reg(); + + default: + assert(0); + return brw_null_reg(); + } +} + + + +static struct brw_reg deref( struct brw_vs_compile *c, + struct brw_reg arg, + int offset) +{ + struct brw_compile *p = &c->func; + struct brw_reg tmp = vec4(get_tmp(c)); + struct brw_reg vp_address = retype(vec1(get_reg(c, TGSI_FILE_ADDRESS, 0)), BRW_REGISTER_TYPE_UW); + unsigned byte_offset = arg.nr * 32 + arg.subnr + offset * 16; + struct brw_reg indirect = brw_vec4_indirect(0,0); + + { + brw_push_insn_state(p); + brw_set_access_mode(p, BRW_ALIGN_1); + + /* This is pretty clunky - load the address register twice and + * fetch each 4-dword value in turn. There must be a way to do + * this in a single pass, but I couldn't get it to work. + */ + brw_ADD(p, brw_address_reg(0), vp_address, brw_imm_d(byte_offset)); + brw_MOV(p, tmp, indirect); + + brw_ADD(p, brw_address_reg(0), suboffset(vp_address, 8), brw_imm_d(byte_offset)); + brw_MOV(p, suboffset(tmp, 4), indirect); + + brw_pop_insn_state(p); + } + + return vec8(tmp); +} + + +static void emit_arl( struct brw_vs_compile *c, + struct brw_reg dst, + struct brw_reg arg0 ) +{ + struct brw_compile *p = &c->func; + struct brw_reg tmp = dst; + boolean need_tmp = (dst.file != BRW_GENERAL_REGISTER_FILE); + + if (need_tmp) + tmp = get_tmp(c); + + brw_RNDD(p, tmp, arg0); + brw_MUL(p, dst, tmp, brw_imm_d(16)); + + if (need_tmp) + release_tmp(c, tmp); +} + + +/* Will return mangled results for SWZ op. The emit_swz() function + * ignores this result and recalculates taking extended swizzles into + * account. + */ +static struct brw_reg get_arg( struct brw_vs_compile *c, + struct tgsi_src_register *src ) +{ + struct brw_reg reg; + + if (src->File == TGSI_FILE_NULL) + return brw_null_reg(); + +#if 0 + if (src->RelAddr) + reg = deref(c, c->regs[PROGRAM_STATE_VAR][0], src->Index); + else +#endif + reg = get_reg(c, src->File, src->Index); + + /* Convert 3-bit swizzle to 2-bit. + */ + reg.dw1.bits.swizzle = BRW_SWIZZLE4(src->SwizzleX, + src->SwizzleY, + src->SwizzleZ, + src->SwizzleW); + + /* Note this is ok for non-swizzle instructions: + */ + reg.negate = src->Negate ? 1 : 0; + + return reg; +} + + +static struct brw_reg get_dst( struct brw_vs_compile *c, + const struct tgsi_dst_register *dst ) +{ + struct brw_reg reg = get_reg(c, dst->File, dst->Index); + + reg.dw1.bits.writemask = dst->WriteMask; + + return reg; +} + + + + +static void emit_swz( struct brw_vs_compile *c, + struct brw_reg dst, + struct tgsi_src_register src ) +{ + struct brw_compile *p = &c->func; + unsigned zeros_mask = 0; + unsigned ones_mask = 0; + unsigned src_mask = 0; + ubyte src_swz[4]; + boolean need_tmp = (src.Negate && + dst.file != BRW_GENERAL_REGISTER_FILE); + struct brw_reg tmp = dst; + unsigned i; + + if (need_tmp) + tmp = get_tmp(c); + + for (i = 0; i < 4; i++) { + if (dst.dw1.bits.writemask & (1<<i)) { + ubyte s = 0; + switch(i) { + case 0: + s = src.SwizzleX; + break; + s = src.SwizzleY; + case 1: + break; + s = src.SwizzleZ; + case 2: + break; + s = src.SwizzleW; + case 3: + break; + } + switch (s) { + case TGSI_SWIZZLE_X: + case TGSI_SWIZZLE_Y: + case TGSI_SWIZZLE_Z: + case TGSI_SWIZZLE_W: + src_mask |= 1<<i; + src_swz[i] = s; + break; + case TGSI_EXTSWIZZLE_ZERO: + zeros_mask |= 1<<i; + break; + case TGSI_EXTSWIZZLE_ONE: + ones_mask |= 1<<i; + break; + } + } + } + + /* Do src first, in case dst aliases src: + */ + if (src_mask) { + struct brw_reg arg0; + +#if 0 + if (src.RelAddr) + arg0 = deref(c, c->regs[PROGRAM_STATE_VAR][0], src.Index); + else +#endif + arg0 = get_reg(c, src.File, src.Index); + + arg0 = brw_swizzle(arg0, + src_swz[0], src_swz[1], + src_swz[2], src_swz[3]); + + brw_MOV(p, brw_writemask(tmp, src_mask), arg0); + } + + if (zeros_mask) + brw_MOV(p, brw_writemask(tmp, zeros_mask), brw_imm_f(0)); + + if (ones_mask) + brw_MOV(p, brw_writemask(tmp, ones_mask), brw_imm_f(1)); + + if (src.Negate) + brw_MOV(p, brw_writemask(tmp, src.Negate), negate(tmp)); + + if (need_tmp) { + brw_MOV(p, dst, tmp); + release_tmp(c, tmp); + } +} + + + +/* Post-vertex-program processing. Send the results to the URB. + */ +static void emit_vertex_write( struct brw_vs_compile *c, struct brw_prog_info *info) +{ + struct brw_compile *p = &c->func; + struct brw_reg m0 = brw_message_reg(0); + struct brw_reg pos = c->regs[TGSI_FILE_OUTPUT][info->pos_idx]; + struct brw_reg ndc; + + if (c->key.copy_edgeflag) { + brw_MOV(p, + get_reg(c, TGSI_FILE_OUTPUT, info->result_edge_idx), + get_reg(c, TGSI_FILE_INPUT, info->edge_flag_idx)); + } + + + /* Build ndc coords? TODO: Shortcircuit when w is known to be one. + */ + if (!c->key.know_w_is_one) { + ndc = get_tmp(c); + emit_math1(c, BRW_MATH_FUNCTION_INV, ndc, brw_swizzle1(pos, 3), BRW_MATH_PRECISION_FULL); + brw_MUL(p, brw_writemask(ndc, TGSI_WRITEMASK_XYZ), pos, ndc); + } + else { + ndc = pos; + } + + /* This includes the workaround for -ve rhw, so is no longer an + * optional step: + */ + if (info->writes_psize || + c->key.nr_userclip || + !c->key.know_w_is_one) + { + struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); + unsigned i; + + brw_MOV(p, header1, brw_imm_ud(0)); + + brw_set_access_mode(p, BRW_ALIGN_16); + + if (info->writes_psize) { + struct brw_reg psiz = c->regs[TGSI_FILE_OUTPUT][info->psize_idx]; + brw_MUL(p, brw_writemask(header1, TGSI_WRITEMASK_W), + brw_swizzle1(psiz, 0), brw_imm_f(1<<11)); + brw_AND(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1, + brw_imm_ud(0x7ff<<8)); + } + + + for (i = 0; i < c->key.nr_userclip; i++) { + brw_set_conditionalmod(p, BRW_CONDITIONAL_L); + brw_DP4(p, brw_null_reg(), pos, c->userplane[i]); + brw_OR(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1, brw_imm_ud(1<<i)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } + + + /* i965 clipping workaround: + * 1) Test for -ve rhw + * 2) If set, + * set ndc = (0,0,0,0) + * set ucp[6] = 1 + * + * Later, clipping will detect ucp[6] and ensure the primitive is + * clipped against all fixed planes. + */ + if (!c->key.know_w_is_one) { + brw_CMP(p, + vec8(brw_null_reg()), + BRW_CONDITIONAL_L, + brw_swizzle1(ndc, 3), + brw_imm_f(0)); + + brw_OR(p, brw_writemask(header1, TGSI_WRITEMASK_W), header1, brw_imm_ud(1<<6)); + brw_MOV(p, ndc, brw_imm_f(0)); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + } + + brw_set_access_mode(p, BRW_ALIGN_1); /* why? */ + brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), header1); + brw_set_access_mode(p, BRW_ALIGN_16); + + release_tmp(c, header1); + } + else { + brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), brw_imm_ud(0)); + } + + + /* Emit the (interleaved) headers for the two vertices - an 8-reg + * of zeros followed by two sets of NDC coordinates: + */ + brw_set_access_mode(p, BRW_ALIGN_1); + brw_MOV(p, offset(m0, 2), ndc); + brw_MOV(p, offset(m0, 3), pos); + + + brw_urb_WRITE(p, + brw_null_reg(), /* dest */ + 0, /* starting mrf reg nr */ + c->r0, /* src */ + 0, /* allocate */ + 1, /* used */ + c->nr_outputs + 3, /* msg len */ + 0, /* response len */ + 1, /* eot */ + 1, /* writes complete */ + 0, /* urb destination offset */ + BRW_URB_SWIZZLE_INTERLEAVE); + +} + +static void +post_vs_emit( struct brw_vs_compile *c, struct brw_instruction *end_inst ) +{ + struct tgsi_parse_context parse; + const struct tgsi_token *tokens = c->vp->program.tokens; + tgsi_parse_init(&parse, tokens); + while (!tgsi_parse_end_of_tokens(&parse)) { + tgsi_parse_token(&parse); + if (parse.FullToken.Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION) { +#if 0 + struct brw_instruction *brw_inst1, *brw_inst2; + const struct tgsi_full_instruction *inst1, *inst2; + int offset; + inst1 = &parse.FullToken.FullInstruction; + brw_inst1 = inst1->Data; + switch (inst1->Opcode) { + case TGSI_OPCODE_CAL: + case TGSI_OPCODE_BRA: + target_insn = inst1->BranchTarget; + inst2 = &c->vp->program.Base.Instructions[target_insn]; + brw_inst2 = inst2->Data; + offset = brw_inst2 - brw_inst1; + brw_set_src1(brw_inst1, brw_imm_d(offset*16)); + break; + case TGSI_OPCODE_END: + offset = end_inst - brw_inst1; + brw_set_src1(brw_inst1, brw_imm_d(offset*16)); + break; + default: + break; + } +#endif + } + } + tgsi_parse_free(&parse); +} + +static void process_declaration(const struct tgsi_full_declaration *decl, + struct brw_prog_info *info) +{ + int first = decl->u.DeclarationRange.First; + int last = decl->u.DeclarationRange.Last; + + assert (decl->Declaration.Declare != TGSI_DECLARE_MASK); + + switch(decl->Declaration.File) { + case TGSI_FILE_CONSTANT: + info->num_consts += last - first + 1; + break; + case TGSI_FILE_INPUT: { + } + break; + case TGSI_FILE_OUTPUT: { + assert(last == first); /* for now */ + if (decl->Declaration.Semantic) { + switch (decl->Semantic.SemanticName) { + case TGSI_SEMANTIC_POSITION: { + info->pos_idx = first; + } + break; + case TGSI_SEMANTIC_COLOR: + break; + case TGSI_SEMANTIC_BCOLOR: + break; + case TGSI_SEMANTIC_FOG: + break; + case TGSI_SEMANTIC_PSIZE: { + info->writes_psize = TRUE; + info->psize_idx = first; + } + break; + case TGSI_SEMANTIC_GENERIC: + break; + } + } + } + break; + case TGSI_FILE_TEMPORARY: { + info->num_temps += (last - first) + 1; + } + break; + case TGSI_FILE_SAMPLER: { + } + break; + case TGSI_FILE_ADDRESS: { + info->num_addrs += (last - first) + 1; + } + break; + case TGSI_FILE_IMMEDIATE: { + } + break; + case TGSI_FILE_NULL: { + } + break; + } +} + +static void process_instruction(struct brw_vs_compile *c, + struct tgsi_full_instruction *inst, + struct brw_prog_info *info) +{ + struct brw_reg args[3], dst; + struct brw_compile *p = &c->func; + struct brw_indirect stack_index = brw_indirect(0, 0); + unsigned i; + unsigned index; + unsigned file; + /*FIXME: might not be the only one*/ + const struct tgsi_dst_register *dst_reg = &inst->FullDstRegisters[0].DstRegister; + /* + struct brw_instruction *if_inst[MAX_IFSN]; + unsigned insn, if_insn = 0; + */ + + for (i = 0; i < 3; i++) { + struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i]; + index = src->SrcRegister.Index; + file = src->SrcRegister.File; + if (file == TGSI_FILE_OUTPUT && c->output_regs[index].used_in_src) + args[i] = c->output_regs[index].reg; + else + args[i] = get_arg(c, &src->SrcRegister); + } + + /* Get dest regs. Note that it is possible for a reg to be both + * dst and arg, given the static allocation of registers. So + * care needs to be taken emitting multi-operation instructions. + */ + index = dst_reg->Index; + file = dst_reg->File; + if (file == TGSI_FILE_OUTPUT && c->output_regs[index].used_in_src) + dst = c->output_regs[index].reg; + else + dst = get_dst(c, dst_reg); + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_ABS: + brw_MOV(p, dst, brw_abs(args[0])); + break; + case TGSI_OPCODE_ADD: + brw_ADD(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_DP3: + brw_DP3(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_DP4: + brw_DP4(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_DPH: + brw_DPH(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_DST: + unalias2(c, dst, args[0], args[1], emit_dst_noalias); + break; + case TGSI_OPCODE_EXP: + unalias1(c, dst, args[0], emit_exp_noalias); + break; + case TGSI_OPCODE_EX2: + emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, args[0], BRW_MATH_PRECISION_FULL); + break; + case TGSI_OPCODE_ARL: + emit_arl(c, dst, args[0]); + break; + case TGSI_OPCODE_FLR: + brw_RNDD(p, dst, args[0]); + break; + case TGSI_OPCODE_FRC: + brw_FRC(p, dst, args[0]); + break; + case TGSI_OPCODE_LOG: + unalias1(c, dst, args[0], emit_log_noalias); + break; + case TGSI_OPCODE_LG2: + emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, args[0], BRW_MATH_PRECISION_FULL); + break; + case TGSI_OPCODE_LIT: + unalias1(c, dst, args[0], emit_lit_noalias); + break; + case TGSI_OPCODE_MAD: + brw_MOV(p, brw_acc_reg(), args[2]); + brw_MAC(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_MAX: + emit_max(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_MIN: + emit_min(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_MOV: +#if 0 + case TGSI_OPCODE_SWZ: + /* The args[0] value can't be used here as it won't have + * correctly encoded the full swizzle: + */ + emit_swz(c, dst, inst->SrcReg[0] ); +#endif + brw_MOV(p, dst, args[0]); + break; + case TGSI_OPCODE_MUL: + brw_MUL(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_POW: + emit_math2(c, BRW_MATH_FUNCTION_POW, dst, args[0], args[1], BRW_MATH_PRECISION_FULL); + break; + case TGSI_OPCODE_RCP: + emit_math1(c, BRW_MATH_FUNCTION_INV, dst, args[0], BRW_MATH_PRECISION_FULL); + break; + case TGSI_OPCODE_RSQ: + emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, args[0], BRW_MATH_PRECISION_FULL); + break; + + case TGSI_OPCODE_SEQ: + emit_seq(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SNE: + emit_sne(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SGE: + emit_sge(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SGT: + emit_sgt(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SLT: + emit_slt(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SLE: + emit_sle(p, dst, args[0], args[1]); + break; + case TGSI_OPCODE_SUB: + brw_ADD(p, dst, args[0], negate(args[1])); + break; + case TGSI_OPCODE_XPD: + emit_xpd(p, dst, args[0], args[1]); + break; +#if 0 + case TGSI_OPCODE_IF: + assert(if_insn < MAX_IFSN); + if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8); + break; + case TGSI_OPCODE_ELSE: + if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]); + break; + case TGSI_OPCODE_ENDIF: + assert(if_insn > 0); + brw_ENDIF(p, if_inst[--if_insn]); + break; + case TGSI_OPCODE_BRA: + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); + brw_set_predicate_control_flag_value(p, 0xff); + break; + case TGSI_OPCODE_CAL: + brw_set_access_mode(p, BRW_ALIGN_1); + brw_ADD(p, deref_1uw(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16)); + brw_set_access_mode(p, BRW_ALIGN_16); + brw_ADD(p, get_addr_reg(stack_index), + get_addr_reg(stack_index), brw_imm_d(4)); + inst->Data = &p->store[p->nr_insn]; + brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); + break; +#endif + case TGSI_OPCODE_RET: +#if 0 + brw_ADD(p, get_addr_reg(stack_index), + get_addr_reg(stack_index), brw_imm_d(-4)); + brw_set_access_mode(p, BRW_ALIGN_1); + brw_MOV(p, brw_ip_reg(), deref_1uw(stack_index, 0)); + brw_set_access_mode(p, BRW_ALIGN_16); +#else + /*brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));*/ +#endif + break; + case TGSI_OPCODE_END: + brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); + break; + case TGSI_OPCODE_BGNSUB: + case TGSI_OPCODE_ENDSUB: + break; + default: + debug_printf("Unsupport opcode %d in vertex shader\n", inst->Instruction.Opcode); + break; + } + + if (dst_reg->File == TGSI_FILE_OUTPUT + && dst_reg->Index != info->pos_idx + && c->output_regs[dst_reg->Index].used_in_src) + brw_MOV(p, get_dst(c, dst_reg), dst); + + release_tmps(c); +} + +/* Emit the fragment program instructions here. + */ +void brw_vs_emit(struct brw_vs_compile *c) +{ +#define MAX_IFSN 32 + struct brw_compile *p = &c->func; + struct brw_instruction *end_inst; + struct tgsi_parse_context parse; + struct brw_indirect stack_index = brw_indirect(0, 0); + const struct tgsi_token *tokens = c->vp->program.tokens; + struct brw_prog_info prog_info; + unsigned allocated_registers = 0; + memset(&prog_info, 0, sizeof(struct brw_prog_info)); + + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_set_access_mode(p, BRW_ALIGN_16); + + tgsi_parse_init(&parse, tokens); + /* Message registers can't be read, so copy the output into GRF register + if they are used in source registers */ + while (!tgsi_parse_end_of_tokens(&parse)) { + tgsi_parse_token(&parse); + unsigned i; + switch (parse.FullToken.Token.Type) { + case TGSI_TOKEN_TYPE_INSTRUCTION: { + const struct tgsi_full_instruction *inst = &parse.FullToken.FullInstruction; + for (i = 0; i < 3; ++i) { + const struct tgsi_src_register *src = &inst->FullSrcRegisters[i].SrcRegister; + unsigned index = src->Index; + unsigned file = src->File; + if (file == TGSI_FILE_OUTPUT) + c->output_regs[index].used_in_src = TRUE; + } + } + break; + default: + /* nothing */ + break; + } + } + tgsi_parse_free(&parse); + + tgsi_parse_init(&parse, tokens); + + while (!tgsi_parse_end_of_tokens(&parse)) { + tgsi_parse_token(&parse); + + switch (parse.FullToken.Token.Type) { + case TGSI_TOKEN_TYPE_DECLARATION: { + struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration; + process_declaration(decl, &prog_info); + } + break; + case TGSI_TOKEN_TYPE_IMMEDIATE: { + struct tgsi_full_immediate *imm = &parse.FullToken.FullImmediate; + /*assert(imm->Immediate.Size == 4);*/ + c->prog_data.imm_buf[c->prog_data.num_imm][0] = imm->u.ImmediateFloat32[0].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][1] = imm->u.ImmediateFloat32[1].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][2] = imm->u.ImmediateFloat32[2].Float; + c->prog_data.imm_buf[c->prog_data.num_imm][3] = imm->u.ImmediateFloat32[3].Float; + c->prog_data.num_imm++; + } + break; + case TGSI_TOKEN_TYPE_INSTRUCTION: { + struct tgsi_full_instruction *inst = &parse.FullToken.FullInstruction; + if (!allocated_registers) { + /* first instruction (declerations finished). + * now that we know what vars are being used allocate + * registers for them.*/ + c->prog_data.num_consts = prog_info.num_consts; + c->prog_data.max_const = prog_info.num_consts + c->prog_data.num_imm; + brw_vs_alloc_regs(c, &prog_info); + + brw_set_access_mode(p, BRW_ALIGN_1); + brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); + brw_set_access_mode(p, BRW_ALIGN_16); + allocated_registers = 1; + } + process_instruction(c, inst, &prog_info); + } + break; + } + } + + end_inst = &p->store[p->nr_insn]; + emit_vertex_write(c, &prog_info); + post_vs_emit(c, end_inst); + tgsi_parse_free(&parse); + +} diff --git a/src/gallium/drivers/i965simple/brw_vs_state.c b/src/gallium/drivers/i965simple/brw_vs_state.c new file mode 100644 index 0000000000..c73469929c --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs_state.c @@ -0,0 +1,102 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + +#include "pipe/p_util.h" + +static void upload_vs_unit( struct brw_context *brw ) +{ + struct brw_vs_unit_state vs; + + memset(&vs, 0, sizeof(vs)); + + /* CACHE_NEW_VS_PROG */ + vs.thread0.kernel_start_pointer = brw->vs.prog_gs_offset >> 6; + vs.thread0.grf_reg_count = align(brw->vs.prog_data->total_grf, 16) / 16 - 1; + vs.thread3.urb_entry_read_length = brw->vs.prog_data->urb_read_length; + vs.thread3.const_urb_entry_read_length = brw->vs.prog_data->curb_read_length; + vs.thread3.dispatch_grf_start_reg = 1; + + + /* BRW_NEW_URB_FENCE */ + vs.thread4.nr_urb_entries = brw->urb.nr_vs_entries; + vs.thread4.urb_entry_allocation_size = brw->urb.vsize - 1; + vs.thread4.max_threads = MIN2( + MAX2(0, (brw->urb.nr_vs_entries - 6) / 2 - 1), + 15); + + + + if (BRW_DEBUG & DEBUG_SINGLE_THREAD) + vs.thread4.max_threads = 0; + + /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */ + if (0 /*brw->attribs.Clip->ClipPlanesEnabled*/) { + /* Note that we read in the userclip planes as well, hence + * clip_start: + */ + vs.thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2; + } + else { + vs.thread3.const_urb_entry_read_offset = brw->curbe.vs_start * 2; + } + + vs.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; + vs.thread3.urb_entry_read_offset = 0; + + /* No samplers for ARB_vp programs: + */ + vs.vs5.sampler_count = 0; + + if (BRW_DEBUG & DEBUG_STATS) + vs.thread4.stats_enable = 1; + + /* Vertex program always enabled: + */ + vs.vs6.vs_enable = 1; + + brw->vs.state_gs_offset = brw_cache_data( &brw->cache[BRW_VS_UNIT], &vs ); +} + + +const struct brw_tracked_state brw_vs_unit = { + .dirty = { + .brw = (BRW_NEW_CLIP | + BRW_NEW_CURBE_OFFSETS | + BRW_NEW_URB_FENCE), + .cache = CACHE_NEW_VS_PROG + }, + .update = upload_vs_unit +}; diff --git a/src/gallium/drivers/i965simple/brw_winsys.h b/src/gallium/drivers/i965simple/brw_winsys.h new file mode 100644 index 0000000000..3523a58614 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_winsys.h @@ -0,0 +1,205 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \file + * This is the interface that i965simple requires any window system + * hosting it to implement. This is the only include file in i965simple + * which is public. + * + */ + +#ifndef BRW_WINSYS_H +#define BRW_WINSYS_H + + +#include "pipe/p_defines.h" + + +/* Pipe drivers are (meant to be!) independent of both GL and the + * window system. The window system provides a buffer manager and a + * set of additional hooks for things like command buffer submission, + * etc. + * + * There clearly has to be some agreement between the window system + * driver and the hardware driver about the format of command buffers, + * etc. + */ + +struct pipe_buffer; +struct pipe_fence_handle; +struct pipe_winsys; + +/* The pipe driver currently understands the following chipsets: + */ +#define PCI_CHIP_I965_G 0x29A2 +#define PCI_CHIP_I965_Q 0x2992 +#define PCI_CHIP_I965_G_1 0x2982 +#define PCI_CHIP_I965_GM 0x2A02 +#define PCI_CHIP_I965_GME 0x2A12 + + +/* These are the names of all the state caches managed by the driver. + * + * When data is uploaded to a buffer with buffer_subdata, we use the + * special version of that function below so that information about + * what type of data this is can be passed to the winsys backend. + * That in turn allows the correct flags to be set in the aub file + * dump to allow human-readable file dumps later on. + */ + +enum brw_cache_id { + BRW_CC_VP, + BRW_CC_UNIT, + BRW_WM_PROG, + BRW_SAMPLER_DEFAULT_COLOR, + BRW_SAMPLER, + BRW_WM_UNIT, + BRW_SF_PROG, + BRW_SF_VP, + BRW_SF_UNIT, + BRW_VS_UNIT, + BRW_VS_PROG, + BRW_GS_UNIT, + BRW_GS_PROG, + BRW_CLIP_VP, + BRW_CLIP_UNIT, + BRW_CLIP_PROG, + BRW_SS_SURFACE, + BRW_SS_SURF_BIND, + + BRW_MAX_CACHE +}; + +#define BRW_CONSTANT_BUFFER BRW_MAX_CACHE + +/** + * Additional winsys interface for i965simple. + * + * It is an over-simple batchbuffer mechanism. Will want to improve the + * performance of this, perhaps based on the cmdstream stuff. It + * would be pretty impossible to implement swz on top of this + * interface. + * + * Will also need additions/changes to implement static/dynamic + * indirect state. + */ +struct brw_winsys { + + /** + * Reserve space on batch buffer. + * + * Returns a null pointer if there is insufficient space in the batch buffer + * to hold the requested number of dwords and relocations. + * + * The number of dwords should also include the number of relocations. + */ + unsigned *(*batch_start)(struct brw_winsys *sws, + unsigned dwords, + unsigned relocs); + + void (*batch_dword)(struct brw_winsys *sws, + unsigned dword); + + /** + * Emit a relocation to a buffer. + * + * Used not only when the buffer addresses are not pinned, but also to + * ensure refered buffers will not be destroyed until the current batch + * buffer execution is finished. + * + * The access flags is a combination of I915_BUFFER_ACCESS_WRITE and + * I915_BUFFER_ACCESS_READ macros. + */ + void (*batch_reloc)(struct brw_winsys *sws, + struct pipe_buffer *buf, + unsigned access_flags, + unsigned delta); + + + /* Not used yet, but really want this: + */ + void (*batch_end)( struct brw_winsys *sws ); + + /** + * Flush the batch buffer. + * + * Fence argument must point to NULL or to a previous fence, and the caller + * must call fence_reference when done with the fence. + */ + void (*batch_flush)(struct brw_winsys *sws, + struct pipe_fence_handle **fence); + + + /* A version of buffer_subdata that includes information for the + * simulator: + */ + void (*buffer_subdata_typed)(struct brw_winsys *sws, + struct pipe_buffer *buf, + unsigned long offset, + unsigned long size, + const void *data, + unsigned data_type); + + + /* A cheat so we don't have to think about relocations in a couple + * of places yet: + */ + unsigned (*get_buffer_offset)( struct brw_winsys *sws, + struct pipe_buffer *buf, + unsigned flags ); + +}; + +#define BRW_BUFFER_ACCESS_WRITE 0x1 +#define BRW_BUFFER_ACCESS_READ 0x2 + +#define BRW_BUFFER_USAGE_LIT_VERTEX (PIPE_BUFFER_USAGE_CUSTOM << 0) + + +struct pipe_context *brw_create(struct pipe_winsys *, + struct brw_winsys *, + unsigned pci_id); + +static inline boolean brw_batchbuffer_data(struct brw_winsys *winsys, + const void *data, + unsigned bytes) +{ + static const unsigned incr = sizeof(unsigned); + uint i; + const unsigned *udata = (const unsigned*)(data); + unsigned size = bytes/incr; + + winsys->batch_start(winsys, size, 0); + for (i = 0; i < size; ++i) { + winsys->batch_dword(winsys, udata[i]); + } + winsys->batch_end(winsys); + + return (i == size); +} +#endif diff --git a/src/gallium/drivers/i965simple/brw_wm.c b/src/gallium/drivers/i965simple/brw_wm.c new file mode 100644 index 0000000000..539b170744 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm.c @@ -0,0 +1,210 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_util.h" +#include "brw_wm.h" +#include "brw_eu.h" +#include "brw_state.h" +#include "pipe/p_util.h" + + + +static void do_wm_prog( struct brw_context *brw, + struct brw_fragment_program *fp, + struct brw_wm_prog_key *key) +{ + struct brw_wm_compile *c = CALLOC_STRUCT(brw_wm_compile); + const unsigned *program; + unsigned program_size; + + c->key = *key; + c->fp = fp; + + c->delta_xy[0] = brw_null_reg(); + c->delta_xy[1] = brw_null_reg(); + c->pixel_xy[0] = brw_null_reg(); + c->pixel_xy[1] = brw_null_reg(); + c->pixel_w = brw_null_reg(); + + + debug_printf("XXXXXXXX FP\n"); + + brw_wm_glsl_emit(c); + + /* get the program + */ + program = brw_get_program(&c->func, &program_size); + + /* + */ + brw->wm.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_WM_PROG], + &c->key, + sizeof(c->key), + program, + program_size, + &c->prog_data, + &brw->wm.prog_data ); + + FREE(c); +} + + + +static void brw_wm_populate_key( struct brw_context *brw, + struct brw_wm_prog_key *key ) +{ + /* BRW_NEW_FRAGMENT_PROGRAM */ + struct brw_fragment_program *fp = + (struct brw_fragment_program *)brw->attribs.FragmentProgram; + unsigned lookup = 0; + unsigned line_aa; + + memset(key, 0, sizeof(*key)); + + /* Build the index for table lookup + */ + /* BRW_NEW_DEPTH_STENCIL */ + if (fp->UsesKill || + brw->attribs.DepthStencil->alpha.enabled) + lookup |= IZ_PS_KILL_ALPHATEST_BIT; + + if (fp->ComputesDepth) + lookup |= IZ_PS_COMPUTES_DEPTH_BIT; + + if (brw->attribs.DepthStencil->depth.enabled) + lookup |= IZ_DEPTH_TEST_ENABLE_BIT; + + if (brw->attribs.DepthStencil->depth.enabled && + brw->attribs.DepthStencil->depth.writemask) /* ?? */ + lookup |= IZ_DEPTH_WRITE_ENABLE_BIT; + + if (brw->attribs.DepthStencil->stencil[0].enabled) { + lookup |= IZ_STENCIL_TEST_ENABLE_BIT; + + if (brw->attribs.DepthStencil->stencil[0].write_mask || + brw->attribs.DepthStencil->stencil[1].write_mask) + lookup |= IZ_STENCIL_WRITE_ENABLE_BIT; + } + + /* XXX: when should this be disabled? + */ + if (1) + lookup |= IZ_EARLY_DEPTH_TEST_BIT; + + + line_aa = AA_NEVER; + + /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */ + if (brw->attribs.Raster->line_smooth) { + if (brw->reduced_primitive == PIPE_PRIM_LINES) { + line_aa = AA_ALWAYS; + } + else if (brw->reduced_primitive == PIPE_PRIM_TRIANGLES) { + if (brw->attribs.Raster->fill_ccw == PIPE_POLYGON_MODE_LINE) { + line_aa = AA_SOMETIMES; + + if (brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_LINE || + (brw->attribs.Raster->cull_mode == PIPE_WINDING_CW)) + line_aa = AA_ALWAYS; + } + else if (brw->attribs.Raster->fill_cw == PIPE_POLYGON_MODE_LINE) { + line_aa = AA_SOMETIMES; + + if (brw->attribs.Raster->cull_mode == PIPE_WINDING_CCW) + line_aa = AA_ALWAYS; + } + } + } + + brw_wm_lookup_iz(line_aa, + lookup, + key); + + +#if 0 + /* BRW_NEW_SAMPLER + * + * Not doing any of this at the moment: + */ + for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { + const struct pipe_sampler_state *unit = brw->attribs.Samplers[i]; + + if (unit) { + + if (unit->compare && + unit->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + key->shadowtex_mask |= 1<<i; + } + if (t->Image[0][t->BaseLevel]->InternalFormat == GL_YCBCR_MESA) + key->yuvtex_mask |= 1<<i; + } + } +#endif + + + /* Extra info: + */ + key->program_string_id = fp->id; + +} + + +static void brw_upload_wm_prog( struct brw_context *brw ) +{ + struct brw_wm_prog_key key; + struct brw_fragment_program *fp = (struct brw_fragment_program *) + brw->attribs.FragmentProgram; + + brw_wm_populate_key(brw, &key); + + /* Make an early check for the key. + */ + if (brw_search_cache(&brw->cache[BRW_WM_PROG], + &key, sizeof(key), + &brw->wm.prog_data, + &brw->wm.prog_gs_offset)) + return; + + do_wm_prog(brw, fp, &key); +} + + +const struct brw_tracked_state brw_wm_prog = { + .dirty = { + .brw = (BRW_NEW_FS | + BRW_NEW_REDUCED_PRIMITIVE), + .cache = 0 + }, + .update = brw_upload_wm_prog +}; + diff --git a/src/gallium/drivers/i965simple/brw_wm.h b/src/gallium/drivers/i965simple/brw_wm.h new file mode 100644 index 0000000000..a1ac0f504a --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm.h @@ -0,0 +1,142 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#ifndef BRW_WM_H +#define BRW_WM_H + + +#include "brw_context.h" +#include "brw_eu.h" + +/* A big lookup table is used to figure out which and how many + * additional regs will inserted before the main payload in the WM + * program execution. These mainly relate to depth and stencil + * processing and the early-depth-test optimization. + */ +#define IZ_PS_KILL_ALPHATEST_BIT 0x1 +#define IZ_PS_COMPUTES_DEPTH_BIT 0x2 +#define IZ_DEPTH_WRITE_ENABLE_BIT 0x4 +#define IZ_DEPTH_TEST_ENABLE_BIT 0x8 +#define IZ_STENCIL_WRITE_ENABLE_BIT 0x10 +#define IZ_STENCIL_TEST_ENABLE_BIT 0x20 +#define IZ_EARLY_DEPTH_TEST_BIT 0x40 +#define IZ_BIT_MAX 0x80 + +#define AA_NEVER 0 +#define AA_SOMETIMES 1 +#define AA_ALWAYS 2 + +struct brw_wm_prog_key { + unsigned source_depth_reg:3; + unsigned aa_dest_stencil_reg:3; + unsigned dest_depth_reg:3; + unsigned nr_depth_regs:3; + unsigned shadowtex_mask:8; + unsigned computes_depth:1; /* could be derived from program string */ + unsigned source_depth_to_render_target:1; + unsigned runtime_check_aads_emit:1; + + unsigned yuvtex_mask:8; + + unsigned program_string_id; +}; + + + + + +#define PROGRAM_INTERNAL_PARAM +#define MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS 1024 /* 72 for GL_ARB_f_p */ +#define BRW_WM_MAX_INSN (MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS*3 + PIPE_ATTRIB_MAX + 3) +#define BRW_WM_MAX_GRF 128 /* hardware limit */ +#define BRW_WM_MAX_VREG (BRW_WM_MAX_INSN * 4) +#define BRW_WM_MAX_REF (BRW_WM_MAX_INSN * 12) +#define BRW_WM_MAX_PARAM 256 +#define BRW_WM_MAX_CONST 256 +#define BRW_WM_MAX_KILLS MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS + +#define PAYLOAD_DEPTH (PIPE_ATTRIB_MAX) + +#define MAX_IFSN 32 +#define MAX_LOOP_DEPTH 32 + +struct brw_wm_compile { + struct brw_compile func; + struct brw_wm_prog_key key; + struct brw_wm_prog_data prog_data; /* result */ + + struct brw_fragment_program *fp; + + unsigned grf_limit; + unsigned max_wm_grf; + + + struct brw_reg pixel_xy[2]; + struct brw_reg delta_xy[2]; + struct brw_reg pixel_w; + + + struct brw_reg wm_regs[8][32][4]; + + struct brw_reg payload_depth[4]; + struct brw_reg payload_coef[16]; + + struct brw_reg emit_mask_reg; + + struct brw_instruction *if_inst[MAX_IFSN]; + int if_insn; + + struct brw_instruction *loop_inst[MAX_LOOP_DEPTH]; + int loop_insn; + + struct brw_instruction *inst0; + struct brw_instruction *inst1; + + struct brw_reg stack; + struct brw_indirect stack_index; + + unsigned reg_index; + + unsigned tmp_start; + unsigned tmp_index; +}; + + + +void brw_wm_lookup_iz( unsigned line_aa, + unsigned lookup, + struct brw_wm_prog_key *key ); + +void brw_wm_glsl_emit(struct brw_wm_compile *c); +void brw_wm_emit_decls(struct brw_wm_compile *c); + +#endif diff --git a/src/gallium/drivers/i965simple/brw_wm_decl.c b/src/gallium/drivers/i965simple/brw_wm_decl.c new file mode 100644 index 0000000000..b45a333a2e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_decl.c @@ -0,0 +1,383 @@ + +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_wm.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" + +static struct brw_reg alloc_tmp(struct brw_wm_compile *c) +{ + c->tmp_index++; + c->reg_index = MAX2(c->reg_index, c->tmp_start + c->tmp_index); + return brw_vec8_grf(c->tmp_start + c->tmp_index, 0); +} + +static void release_tmps(struct brw_wm_compile *c) +{ + c->tmp_index = 0; +} + + + +static int is_null( struct brw_reg reg ) +{ + return (reg.file == BRW_ARCHITECTURE_REGISTER_FILE && + reg.nr == BRW_ARF_NULL); +} + +static void emit_pixel_xy( struct brw_wm_compile *c ) +{ + if (is_null(c->pixel_xy[0])) { + + struct brw_compile *p = &c->func; + struct brw_reg r1_uw = retype(brw_vec1_grf(1, 0), BRW_REGISTER_TYPE_UW); + + c->pixel_xy[0] = vec8(retype(alloc_tmp(c), BRW_REGISTER_TYPE_UW)); + c->pixel_xy[1] = vec8(retype(alloc_tmp(c), BRW_REGISTER_TYPE_UW)); + + /* Calculate pixel centers by adding 1 or 0 to each of the + * micro-tile coordinates passed in r1. + */ + brw_ADD(p, + c->pixel_xy[0], + stride(suboffset(r1_uw, 4), 2, 4, 0), + brw_imm_v(0x10101010)); + + brw_ADD(p, + c->pixel_xy[1], + stride(suboffset(r1_uw, 5), 2, 4, 0), + brw_imm_v(0x11001100)); + } +} + + + + + + +static void emit_delta_xy( struct brw_wm_compile *c ) +{ + if (is_null(c->delta_xy[0])) { + struct brw_compile *p = &c->func; + struct brw_reg r1 = brw_vec1_grf(1, 0); + + emit_pixel_xy(c); + + c->delta_xy[0] = alloc_tmp(c); + c->delta_xy[1] = alloc_tmp(c); + + /* Calc delta X,Y by subtracting origin in r1 from the pixel + * centers. + */ + brw_ADD(p, + c->delta_xy[0], + retype(c->pixel_xy[0], BRW_REGISTER_TYPE_UW), + negate(r1)); + + brw_ADD(p, + c->delta_xy[1], + retype(c->pixel_xy[1], BRW_REGISTER_TYPE_UW), + negate(suboffset(r1,1))); + } +} + + + +#if 0 +static void emit_pixel_w( struct brw_wm_compile *c ) +{ + if (is_null(c->pixel_w)) { + struct brw_compile *p = &c->func; + + struct brw_reg interp_wpos = c->coef_wpos; + + c->pixel_w = alloc_tmp(c); + + emit_delta_xy(c); + + /* Calc 1/w - just linterp wpos[3] optimized by putting the + * result straight into a message reg. + */ + struct brw_reg interp3 = brw_vec1_grf(interp_wpos.nr+1, 4); + brw_LINE(p, brw_null_reg(), interp3, c->delta_xy[0]); + brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), c->delta_xy[1]); + + /* Calc w */ + brw_math_16( p, + c->pixel_w, + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 2, + brw_null_reg(), + BRW_MATH_PRECISION_FULL); + } +} +#endif + + +static void emit_cinterp(struct brw_wm_compile *c, + int idx, + int mask ) +{ + struct brw_compile *p = &c->func; + struct brw_reg interp[4]; + struct brw_reg coef = c->payload_coef[idx]; + int i; + + interp[0] = brw_vec1_grf(coef.nr, 0); + interp[1] = brw_vec1_grf(coef.nr, 4); + interp[2] = brw_vec1_grf(coef.nr+1, 0); + interp[3] = brw_vec1_grf(coef.nr+1, 4); + + for(i = 0; i < 4; i++ ) { + if (mask & (1<<i)) { + struct brw_reg dst = c->wm_regs[TGSI_FILE_INPUT][idx][i]; + brw_MOV(p, dst, suboffset(interp[i],3)); + } + } +} + +static void emit_linterp(struct brw_wm_compile *c, + int idx, + int mask ) +{ + struct brw_compile *p = &c->func; + struct brw_reg interp[4]; + struct brw_reg coef = c->payload_coef[idx]; + int i; + + emit_delta_xy(c); + + interp[0] = brw_vec1_grf(coef.nr, 0); + interp[1] = brw_vec1_grf(coef.nr, 4); + interp[2] = brw_vec1_grf(coef.nr+1, 0); + interp[3] = brw_vec1_grf(coef.nr+1, 4); + + for(i = 0; i < 4; i++ ) { + if (mask & (1<<i)) { + struct brw_reg dst = c->wm_regs[TGSI_FILE_INPUT][idx][i]; + brw_LINE(p, brw_null_reg(), interp[i], c->delta_xy[0]); + brw_MAC(p, dst, suboffset(interp[i],1), c->delta_xy[1]); + } + } +} + +#if 0 +static void emit_pinterp(struct brw_wm_compile *c, + int idx, + int mask ) +{ + struct brw_compile *p = &c->func; + struct brw_reg interp[4]; + struct brw_reg coef = c->payload_coef[idx]; + int i; + + get_delta_xy(c); + get_pixel_w(c); + + interp[0] = brw_vec1_grf(coef.nr, 0); + interp[1] = brw_vec1_grf(coef.nr, 4); + interp[2] = brw_vec1_grf(coef.nr+1, 0); + interp[3] = brw_vec1_grf(coef.nr+1, 4); + + for(i = 0; i < 4; i++ ) { + if (mask & (1<<i)) { + struct brw_reg dst = allocate_reg(c, TGSI_FILE_INPUT, idx, i); + brw_LINE(p, brw_null_reg(), interp[i], c->delta_xy[0]); + brw_MAC(p, dst, suboffset(interp[i],1), c->delta_xy[1]); + brw_MUL(p, dst, dst, c->pixel_w); + } + } +} +#endif + + + +#if 0 +static void emit_wpos( ) +{ + struct prog_dst_register dst = dst_reg(PROGRAM_INPUT, idx); + struct tgsi_full_src_register interp = src_reg(PROGRAM_PAYLOAD, idx); + struct tgsi_full_src_register deltas = get_delta_xy(c); + struct tgsi_full_src_register arg2; + unsigned opcode; + + opcode = WM_LINTERP; + arg2 = src_undef(); + + /* Have to treat wpos.xy specially: + */ + emit_op(c, + WM_WPOSXY, + dst_mask(dst, WRITEMASK_XY), + 0, 0, 0, + get_pixel_xy(c), + src_undef(), + src_undef()); + + dst = dst_mask(dst, WRITEMASK_ZW); + + /* PROGRAM_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw + */ + emit_op(c, + WM_LINTERP, + dst, + 0, 0, 0, + interp, + deltas, + arg2); +} +#endif + + + + +/* Perform register allocation: + * + * -- r0??? + * -- passthrough depth regs (and stencil/aa??) + * -- curbe ?? + * -- inputs (coefficients) + * + * Use a totally static register allocation. This will perform poorly + * but is an easy way to get started (again). + */ +static void prealloc_reg(struct brw_wm_compile *c) +{ + int i, j; + int nr_curbe_regs = 0; + + /* R0, then some depth related regs: + */ + for (i = 0; i < c->key.nr_depth_regs; i++) { + c->payload_depth[i] = brw_vec8_grf(i*2, 0); + c->reg_index += 2; + } + + + /* Then a copy of our part of the CURBE entry: + */ + { + int nr_constants = c->fp->info.nr_regs[TGSI_FILE_CONSTANT]; + int index = 0; + + c->prog_data.max_const = 4*nr_constants; + for (i = 0; i < nr_constants; i++) { + for (j = 0; j < 4; j++, index++) + c->wm_regs[TGSI_FILE_CONSTANT][i][j] = brw_vec1_grf(c->reg_index + index/8, + index%8); + } + + nr_curbe_regs = 2*((4*nr_constants+15)/16); + c->reg_index += nr_curbe_regs; + } + + /* Adjust for parameter coefficients for position, which are + * currently always provided. + */ +// c->position_coef[i] = brw_vec8_grf(c->reg_index, 0); + c->reg_index += 2; + + /* Next we receive the plane coefficients for parameter + * interpolation: + */ + for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_INPUT]; i++) { + c->payload_coef[i] = brw_vec8_grf(c->reg_index, 0); + c->reg_index += 2; + } + + c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2; + c->prog_data.urb_read_length = (c->fp->program.num_inputs + 1) * 2; + c->prog_data.curb_read_length = nr_curbe_regs; + + /* That's the end of the payload, now we can start allocating registers. + */ + c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); + c->reg_index++; + + c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); + c->reg_index += 2; + + /* Now allocate room for the interpolated inputs and staging + * registers for the outputs: + */ + for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_INPUT]; i++) + for (j = 0; j < 4; j++) + c->wm_regs[TGSI_FILE_INPUT][i][j] = brw_vec8_grf( c->reg_index++, 0 ); + + for (i = 0; i < c->fp->info.nr_regs[TGSI_FILE_OUTPUT]; i++) + for (j = 0; j < 4; j++) + c->wm_regs[TGSI_FILE_OUTPUT][i][j] = brw_vec8_grf( c->reg_index++, 0 ); + + /* Beyond this we should only need registers for internal temporaries: + */ + c->tmp_start = c->reg_index; +} + + + + + +/* Need to interpolate fragment program inputs in as a preamble to the + * shader. A more sophisticated compiler would do this on demand, but + * we'll do it up front: + */ +void brw_wm_emit_decls(struct brw_wm_compile *c) +{ + struct tgsi_parse_context parse; + int done = 0; + + prealloc_reg(c); + + tgsi_parse_init( &parse, c->fp->program.tokens ); + + while( !done && + !tgsi_parse_end_of_tokens( &parse ) ) + { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + { + const struct tgsi_full_declaration *decl = &parse.FullToken.FullDeclaration; + unsigned first = decl->u.DeclarationRange.First; + unsigned last = decl->u.DeclarationRange.Last; + unsigned mask = decl->Declaration.UsageMask; /* ? */ + unsigned i; + + if (decl->Declaration.File != TGSI_FILE_INPUT) + break; + + assert(decl->Declaration.Interpolate); + + for( i = first; i <= last; i++ ) { + switch (decl->Interpolation.Interpolate) { + case TGSI_INTERPOLATE_CONSTANT: + emit_cinterp(c, i, mask); + break; + + case TGSI_INTERPOLATE_LINEAR: + emit_linterp(c, i, mask); + break; + + case TGSI_INTERPOLATE_PERSPECTIVE: + //emit_pinterp(c, i, mask); + emit_linterp(c, i, mask); + break; + } + } + break; + } + case TGSI_TOKEN_TYPE_IMMEDIATE: + case TGSI_TOKEN_TYPE_INSTRUCTION: + default: + done = 1; + break; + } + } + + tgsi_parse_free (&parse); + + release_tmps(c); +} diff --git a/src/gallium/drivers/i965simple/brw_wm_glsl.c b/src/gallium/drivers/i965simple/brw_wm_glsl.c new file mode 100644 index 0000000000..d95645d108 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_glsl.c @@ -0,0 +1,1079 @@ + +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_wm.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/tgsi/util/tgsi_parse.h" + + + +static int get_scalar_dst_index(struct tgsi_full_instruction *inst) +{ + struct tgsi_dst_register dst = inst->FullDstRegisters[0].DstRegister; + int i; + for (i = 0; i < 4; i++) + if (dst.WriteMask & (1<<i)) + break; + return i; +} + +static struct brw_reg alloc_tmp(struct brw_wm_compile *c) +{ + c->tmp_index++; + c->reg_index = MAX2(c->reg_index, c->tmp_index); + return brw_vec8_grf(c->tmp_start + c->tmp_index, 0); +} + +static void release_tmps(struct brw_wm_compile *c) +{ + c->tmp_index = 0; +} + + +static struct brw_reg +get_reg(struct brw_wm_compile *c, int file, int index, int component ) +{ + switch (file) { + case TGSI_FILE_NULL: + return brw_null_reg(); + + case TGSI_FILE_SAMPLER: + /* Should never get here: + */ + assert (0); + return brw_null_reg(); + + case TGSI_FILE_IMMEDIATE: + /* These need a different path: + */ + assert(0); + return brw_null_reg(); + + + case TGSI_FILE_CONSTANT: + case TGSI_FILE_INPUT: + case TGSI_FILE_OUTPUT: + case TGSI_FILE_TEMPORARY: + case TGSI_FILE_ADDRESS: + return c->wm_regs[file][index][component]; + + default: + assert(0); + return brw_null_reg(); + } +} + + +static struct brw_reg get_dst_reg(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst, + int component) +{ + return get_reg(c, + inst->FullDstRegisters[0].DstRegister.File, + inst->FullDstRegisters[0].DstRegister.Index, + component); +} + +static int get_swz( struct tgsi_src_register src, int index ) +{ + switch (index & 3) { + case 0: return src.SwizzleX; + case 1: return src.SwizzleY; + case 2: return src.SwizzleZ; + case 3: return src.SwizzleW; + default: return 0; + } +} + +static int get_ext_swz( struct tgsi_src_register_ext_swz src, int index ) +{ + switch (index & 3) { + case 0: return src.ExtSwizzleX; + case 1: return src.ExtSwizzleY; + case 2: return src.ExtSwizzleZ; + case 3: return src.ExtSwizzleW; + default: return 0; + } +} + +static struct brw_reg get_src_reg(struct brw_wm_compile *c, + struct tgsi_full_src_register *src, + int index) +{ + struct brw_reg reg; + int component = index; + int neg = 0; + int abs = 0; + + if (src->SrcRegister.Negate) + neg = 1; + + component = get_swz(src->SrcRegister, component); + + /* Yes, there are multiple negates: + */ + switch (component & 3) { + case 0: neg ^= src->SrcRegisterExtSwz.NegateX; break; + case 1: neg ^= src->SrcRegisterExtSwz.NegateY; break; + case 2: neg ^= src->SrcRegisterExtSwz.NegateZ; break; + case 3: neg ^= src->SrcRegisterExtSwz.NegateW; break; + } + + /* And multiple swizzles, fun isn't it: + */ + component = get_ext_swz(src->SrcRegisterExtSwz, component); + + /* Can't handle this, don't know if we need to: + */ + assert(src->SrcRegisterExtSwz.ExtDivide == TGSI_EXTSWIZZLE_ONE); + + /* Not handling indirect lookups yet: + */ + assert(src->SrcRegister.Indirect == 0); + + /* Don't know what dimension means: + */ + assert(src->SrcRegister.Dimension == 0); + + /* Will never handle any of this stuff: + */ + assert(src->SrcRegisterExtMod.Complement == 0); + assert(src->SrcRegisterExtMod.Bias == 0); + assert(src->SrcRegisterExtMod.Scale2X == 0); + + if (src->SrcRegisterExtMod.Absolute) + abs = 1; + + /* Another negate! This is a post-absolute negate, which we + * can't do. Need to clean the crap out of tgsi somehow. + */ + assert(src->SrcRegisterExtMod.Negate == 0); + + switch( component ) { + case TGSI_EXTSWIZZLE_X: + case TGSI_EXTSWIZZLE_Y: + case TGSI_EXTSWIZZLE_Z: + case TGSI_EXTSWIZZLE_W: + reg = get_reg(c, + src->SrcRegister.File, + src->SrcRegister.Index, + component ); + + if (neg) + reg = negate(reg); + + if (abs) + reg = brw_abs(reg); + + break; + + /* XXX: this won't really work in the general case, but we know + * that the extended swizzle is only allowed in the SWZ + * instruction (right??), in which case using an immediate + * directly will work. + */ + case TGSI_EXTSWIZZLE_ZERO: + reg = brw_imm_f(0); + break; + + case TGSI_EXTSWIZZLE_ONE: + if (neg && !abs) + reg = brw_imm_f(-1.0); + else + reg = brw_imm_f(1.0); + break; + + default: + assert(0); + break; + } + + + return reg; +} + +static void emit_abs( struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + + int i; + struct brw_compile *p = &c->func; + brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE); + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + struct brw_reg src, dst; + dst = get_dst_reg(c, inst, i); + src = get_src_reg(c, &inst->FullSrcRegisters[0], i); + brw_MOV(p, dst, brw_abs(src)); /* NOTE */ + } + } + brw_set_saturate(p, 0); +} + + +static void emit_xpd(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + int i; + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + for (i = 0; i < 4; i++) { + unsigned i2 = (i+2)%3; + unsigned i1 = (i+1)%3; + if (mask & (1<<i)) { + struct brw_reg src0, src1, dst; + dst = get_dst_reg(c, inst, i); + src0 = negate(get_src_reg(c, &inst->FullSrcRegisters[0], i2)); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i1); + brw_MUL(p, brw_null_reg(), src0, src1); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i1); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i2); + brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE); + brw_MAC(p, dst, src0, src1); + brw_set_saturate(p, 0); + } + } + brw_set_saturate(p, 0); +} + +static void emit_dp3(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_reg src0[3], src1[3], dst; + int i; + struct brw_compile *p = &c->func; + for (i = 0; i < 3; i++) { + src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i); + } + + dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + brw_MUL(p, brw_null_reg(), src0[0], src1[0]); + brw_MAC(p, brw_null_reg(), src0[1], src1[1]); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_MAC(p, dst, src0[2], src1[2]); + brw_set_saturate(p, 0); +} + +static void emit_dp4(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_reg src0[4], src1[4], dst; + int i; + struct brw_compile *p = &c->func; + for (i = 0; i < 4; i++) { + src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i); + } + dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + brw_MUL(p, brw_null_reg(), src0[0], src1[0]); + brw_MAC(p, brw_null_reg(), src0[1], src1[1]); + brw_MAC(p, brw_null_reg(), src0[2], src1[2]); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_MAC(p, dst, src0[3], src1[3]); + brw_set_saturate(p, 0); +} + +static void emit_dph(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_reg src0[4], src1[4], dst; + int i; + struct brw_compile *p = &c->func; + for (i = 0; i < 4; i++) { + src0[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1[i] = get_src_reg(c, &inst->FullSrcRegisters[1], i); + } + dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + brw_MUL(p, brw_null_reg(), src0[0], src1[0]); + brw_MAC(p, brw_null_reg(), src0[1], src1[1]); + brw_MAC(p, dst, src0[2], src1[2]); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_ADD(p, dst, src0[3], src1[3]); + brw_set_saturate(p, 0); +} + +static void emit_math1(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst, unsigned func) +{ + struct brw_compile *p = &c->func; + struct brw_reg src0, dst; + + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0); + dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + brw_MOV(p, brw_message_reg(2), src0); + brw_math(p, + dst, + func, + ((inst->Instruction.Saturate != TGSI_SAT_NONE) + ? BRW_MATH_SATURATE_SATURATE + : BRW_MATH_SATURATE_NONE), + 2, + brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); +} + + +static void emit_alu2(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst, + unsigned opcode) +{ + struct brw_compile *p = &c->func; + struct brw_reg src0, src1, dst; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + int i; + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + for (i = 0 ; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + brw_alu2(p, opcode, dst, src0, src1); + } + } + brw_set_saturate(p, 0); +} + + +static void emit_alu1(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst, + unsigned opcode) +{ + struct brw_compile *p = &c->func; + struct brw_reg src0, dst; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + int i; + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + for (i = 0 ; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + brw_alu1(p, opcode, dst, src0); + } + } + if (inst->Instruction.Saturate != TGSI_SAT_NONE) + brw_set_saturate(p, 0); +} + + +static void emit_max(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg src0, src1, dst; + int i; + brw_push_insn_state(p); + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_MOV(p, dst, src0); + brw_set_saturate(p, 0); + + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src0, src1); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + brw_MOV(p, dst, src1); + brw_set_saturate(p, 0); + brw_set_predicate_control_flag_value(p, 0xff); + } + } + brw_pop_insn_state(p); +} + +static void emit_min(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg src0, src1, dst; + int i; + brw_push_insn_state(p); + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_MOV(p, dst, src0); + brw_set_saturate(p, 0); + + brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src1, src0); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + brw_MOV(p, dst, src1); + brw_set_saturate(p, 0); + brw_set_predicate_control_flag_value(p, 0xff); + } + } + brw_pop_insn_state(p); +} + +static void emit_pow(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + struct brw_reg dst, src0, src1; + dst = get_dst_reg(c, inst, get_scalar_dst_index(inst)); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], 0); + + brw_MOV(p, brw_message_reg(2), src0); + brw_MOV(p, brw_message_reg(3), src1); + + brw_math(p, + dst, + BRW_MATH_FUNCTION_POW, + (inst->Instruction.Saturate != TGSI_SAT_NONE + ? BRW_MATH_SATURATE_SATURATE + : BRW_MATH_SATURATE_NONE), + 2, + brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); +} + +static void emit_lrp(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg dst, tmp1, tmp2, src0, src1, src2; + int i; + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + + if (src1.nr == dst.nr) { + tmp1 = alloc_tmp(c); + brw_MOV(p, tmp1, src1); + } else + tmp1 = src1; + + src2 = get_src_reg(c, &inst->FullSrcRegisters[2], i); + if (src2.nr == dst.nr) { + tmp2 = alloc_tmp(c); + brw_MOV(p, tmp2, src2); + } else + tmp2 = src2; + + brw_ADD(p, dst, negate(src0), brw_imm_f(1.0)); + brw_MUL(p, brw_null_reg(), dst, tmp2); + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_MAC(p, dst, src0, tmp1); + brw_set_saturate(p, 0); + } + release_tmps(c); + } +} + +static void emit_kil(struct brw_wm_compile *c) +{ + struct brw_compile *p = &c->func; + struct brw_reg depth = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW); + brw_push_insn_state(p); + brw_set_mask_control(p, BRW_MASK_DISABLE); + brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); //IMASK + brw_AND(p, depth, c->emit_mask_reg, depth); + brw_pop_insn_state(p); +} + +static void emit_mad(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg dst, src0, src1, src2; + int i; + + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + src2 = get_src_reg(c, &inst->FullSrcRegisters[2], i); + brw_MUL(p, dst, src0, src1); + + brw_set_saturate(p, (inst->Instruction.Saturate != TGSI_SAT_NONE) ? 1 : 0); + brw_ADD(p, dst, dst, src2); + brw_set_saturate(p, 0); + } + } +} + +static void emit_sop(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst, unsigned cond) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg dst, src0, src1; + int i; + + brw_push_insn_state(p); + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], i); + src1 = get_src_reg(c, &inst->FullSrcRegisters[1], i); + brw_CMP(p, brw_null_reg(), cond, src0, src1); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + brw_MOV(p, dst, brw_imm_f(0.0)); + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + brw_MOV(p, dst, brw_imm_f(1.0)); + } + } + brw_pop_insn_state(p); +} + + +static void emit_ddx(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg interp[4]; + struct brw_reg dst; + struct brw_reg src0, w; + unsigned nr, i; + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0); + w = get_src_reg(c, &inst->FullSrcRegisters[1], 3); + nr = src0.nr; + interp[0] = brw_vec1_grf(nr, 0); + interp[1] = brw_vec1_grf(nr, 4); + interp[2] = brw_vec1_grf(nr+1, 0); + interp[3] = brw_vec1_grf(nr+1, 4); + brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE); + for(i = 0; i < 4; i++ ) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + brw_MOV(p, dst, interp[i]); + brw_MUL(p, dst, dst, w); + } + } + brw_set_saturate(p, 0); +} + +static void emit_ddy(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + unsigned mask = inst->FullDstRegisters[0].DstRegister.WriteMask; + struct brw_reg interp[4]; + struct brw_reg dst; + struct brw_reg src0, w; + unsigned nr, i; + + src0 = get_src_reg(c, &inst->FullSrcRegisters[0], 0); + nr = src0.nr; + w = get_src_reg(c, &inst->FullSrcRegisters[1], 3); + interp[0] = brw_vec1_grf(nr, 0); + interp[1] = brw_vec1_grf(nr, 4); + interp[2] = brw_vec1_grf(nr+1, 0); + interp[3] = brw_vec1_grf(nr+1, 4); + brw_set_saturate(p, inst->Instruction.Saturate != TGSI_SAT_NONE); + for(i = 0; i < 4; i++ ) { + if (mask & (1<<i)) { + dst = get_dst_reg(c, inst, i); + brw_MOV(p, dst, suboffset(interp[i], 1)); + brw_MUL(p, dst, dst, w); + } + } + brw_set_saturate(p, 0); +} + +/* TODO + BIAS on SIMD8 not workind yet... +*/ +static void emit_txb(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ +#if 0 + struct brw_compile *p = &c->func; + struct brw_reg payload_reg = c->payload_depth[0]; + struct brw_reg dst[4], src[4]; + unsigned i; + for (i = 0; i < 4; i++) + dst[i] = get_dst_reg(c, inst, i); + for (i = 0; i < 4; i++) + src[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i); + +#if 0 + switch (inst->TexSrcTarget) { + case TEXTURE_1D_INDEX: + brw_MOV(p, brw_message_reg(2), src[0]); + brw_MOV(p, brw_message_reg(3), brw_imm_f(0)); + brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); + break; + case TEXTURE_2D_INDEX: + case TEXTURE_RECT_INDEX: + brw_MOV(p, brw_message_reg(2), src[0]); + brw_MOV(p, brw_message_reg(3), src[1]); + brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); + break; + default: + brw_MOV(p, brw_message_reg(2), src[0]); + brw_MOV(p, brw_message_reg(3), src[1]); + brw_MOV(p, brw_message_reg(4), src[2]); + break; + } +#else + brw_MOV(p, brw_message_reg(2), src[0]); + brw_MOV(p, brw_message_reg(3), src[1]); + brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); +#endif + + brw_MOV(p, brw_message_reg(5), src[3]); + brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); + brw_SAMPLE(p, + retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), + 1, + retype(payload_reg, BRW_REGISTER_TYPE_UW), + inst->TexSrcUnit + 1, /* surface */ + inst->TexSrcUnit, /* sampler */ + inst->FullDstRegisters[0].DstRegister.WriteMask, + BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS, + 4, + 4, + 0); +#endif +} + +static void emit_tex(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ +#if 0 + struct brw_compile *p = &c->func; + struct brw_reg payload_reg = c->payload_depth[0]; + struct brw_reg dst[4], src[4]; + unsigned msg_len; + unsigned i, nr; + unsigned emit; + boolean shadow = (c->key.shadowtex_mask & (1<<inst->TexSrcUnit)) ? 1 : 0; + + for (i = 0; i < 4; i++) + dst[i] = get_dst_reg(c, inst, i); + for (i = 0; i < 4; i++) + src[i] = get_src_reg(c, &inst->FullSrcRegisters[0], i); + +#if 0 + switch (inst->TexSrcTarget) { + case TEXTURE_1D_INDEX: + emit = WRITEMASK_X; + nr = 1; + break; + case TEXTURE_2D_INDEX: + case TEXTURE_RECT_INDEX: + emit = WRITEMASK_XY; + nr = 2; + break; + default: + emit = WRITEMASK_XYZ; + nr = 3; + break; + } +#else + emit = WRITEMASK_XY; + nr = 2; +#endif + + msg_len = 1; + + for (i = 0; i < nr; i++) { + static const unsigned swz[4] = {0,1,2,2}; + if (emit & (1<<i)) + brw_MOV(p, brw_message_reg(msg_len+1), src[swz[i]]); + else + brw_MOV(p, brw_message_reg(msg_len+1), brw_imm_f(0)); + msg_len += 1; + } + + if (shadow) { + brw_MOV(p, brw_message_reg(5), brw_imm_f(0)); + brw_MOV(p, brw_message_reg(6), src[2]); + } + + brw_SAMPLE(p, + retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), + 1, + retype(payload_reg, BRW_REGISTER_TYPE_UW), + inst->TexSrcUnit + 1, /* surface */ + inst->TexSrcUnit, /* sampler */ + inst->FullDstRegisters[0].DstRegister.WriteMask, + BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE, + 4, + shadow ? 6 : 4, + 0); + + if (shadow) + brw_MOV(p, dst[3], brw_imm_f(1.0)); +#endif +} + + + + + + + + +static void emit_fb_write(struct brw_wm_compile *c, + struct tgsi_full_instruction *inst) +{ + struct brw_compile *p = &c->func; + int nr = 2; + int channel; + int base_reg = 0; + + // src0 = output color + // src1 = payload_depth[0] + // src2 = output depth + // dst = ??? + + + + /* Reserve a space for AA - may not be needed: + */ + if (c->key.aa_dest_stencil_reg) + nr += 1; + + { + brw_push_insn_state(p); + for (channel = 0; channel < 4; channel++) { + struct brw_reg src0 = c->wm_regs[TGSI_FILE_OUTPUT][0][channel]; + + /* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */ + /* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */ + brw_MOV(p, brw_message_reg(nr + channel), src0); + } + /* skip over the regs populated above: */ + nr += 8; + brw_pop_insn_state(p); + } + + + /* Pass through control information: + */ + /* mov (8) m1.0<1>:ud r1.0<8;8,1>:ud { Align1 NoMask } */ + { + brw_push_insn_state(p); + brw_set_mask_control(p, BRW_MASK_DISABLE); /* ? */ + brw_MOV(p, + brw_message_reg(base_reg + 1), + brw_vec8_grf(1, 0)); + brw_pop_insn_state(p); + } + + /* Send framebuffer write message: */ + brw_fb_WRITE(p, + retype(vec8(brw_null_reg()), BRW_REGISTER_TYPE_UW), + base_reg, + retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW), + 0, /* render surface always 0 */ + nr, + 0, + 1); + +} + + +static void brw_wm_emit_instruction( struct brw_wm_compile *c, + struct tgsi_full_instruction *inst ) +{ + struct brw_compile *p = &c->func; + +#if 0 + if (inst->CondUpdate) + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + else + brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); +#else + brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); +#endif + + switch (inst->Instruction.Opcode) { + case TGSI_OPCODE_ABS: + emit_abs(c, inst); + break; + case TGSI_OPCODE_ADD: + emit_alu2(c, inst, BRW_OPCODE_ADD); + break; + case TGSI_OPCODE_SUB: + assert(0); +// emit_alu2(c, inst, BRW_OPCODE_SUB); + break; + case TGSI_OPCODE_FRC: + emit_alu1(c, inst, BRW_OPCODE_FRC); + break; + case TGSI_OPCODE_FLR: + assert(0); +// emit_alu1(c, inst, BRW_OPCODE_FLR); + break; + case TGSI_OPCODE_LRP: + emit_lrp(c, inst); + break; + case TGSI_OPCODE_INT: + emit_alu1(c, inst, BRW_OPCODE_RNDD); + break; + case TGSI_OPCODE_MOV: + emit_alu1(c, inst, BRW_OPCODE_MOV); + break; + case TGSI_OPCODE_DP3: + emit_dp3(c, inst); + break; + case TGSI_OPCODE_DP4: + emit_dp4(c, inst); + break; + case TGSI_OPCODE_XPD: + emit_xpd(c, inst); + break; + case TGSI_OPCODE_DPH: + emit_dph(c, inst); + break; + case TGSI_OPCODE_RCP: + emit_math1(c, inst, BRW_MATH_FUNCTION_INV); + break; + case TGSI_OPCODE_RSQ: + emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ); + break; + case TGSI_OPCODE_SIN: + emit_math1(c, inst, BRW_MATH_FUNCTION_SIN); + break; + case TGSI_OPCODE_COS: + emit_math1(c, inst, BRW_MATH_FUNCTION_COS); + break; + case TGSI_OPCODE_EX2: + emit_math1(c, inst, BRW_MATH_FUNCTION_EXP); + break; + case TGSI_OPCODE_LG2: + emit_math1(c, inst, BRW_MATH_FUNCTION_LOG); + break; + case TGSI_OPCODE_MAX: + emit_max(c, inst); + break; + case TGSI_OPCODE_MIN: + emit_min(c, inst); + break; + case TGSI_OPCODE_DDX: + emit_ddx(c, inst); + break; + case TGSI_OPCODE_DDY: + emit_ddy(c, inst); + break; + case TGSI_OPCODE_SLT: + emit_sop(c, inst, BRW_CONDITIONAL_L); + break; + case TGSI_OPCODE_SLE: + emit_sop(c, inst, BRW_CONDITIONAL_LE); + break; + case TGSI_OPCODE_SGT: + emit_sop(c, inst, BRW_CONDITIONAL_G); + break; + case TGSI_OPCODE_SGE: + emit_sop(c, inst, BRW_CONDITIONAL_GE); + break; + case TGSI_OPCODE_SEQ: + emit_sop(c, inst, BRW_CONDITIONAL_EQ); + break; + case TGSI_OPCODE_SNE: + emit_sop(c, inst, BRW_CONDITIONAL_NEQ); + break; + case TGSI_OPCODE_MUL: + emit_alu2(c, inst, BRW_OPCODE_MUL); + break; + case TGSI_OPCODE_POW: + emit_pow(c, inst); + break; + case TGSI_OPCODE_MAD: + emit_mad(c, inst); + break; + case TGSI_OPCODE_TEX: + emit_tex(c, inst); + break; + case TGSI_OPCODE_TXB: + emit_txb(c, inst); + break; + case TGSI_OPCODE_TEXKILL: + emit_kil(c); + break; + case TGSI_OPCODE_IF: + assert(c->if_insn < MAX_IFSN); + c->if_inst[c->if_insn++] = brw_IF(p, BRW_EXECUTE_8); + break; + case TGSI_OPCODE_ELSE: + c->if_inst[c->if_insn-1] = brw_ELSE(p, c->if_inst[c->if_insn-1]); + break; + case TGSI_OPCODE_ENDIF: + assert(c->if_insn > 0); + brw_ENDIF(p, c->if_inst[--c->if_insn]); + break; + case TGSI_OPCODE_BGNSUB: + case TGSI_OPCODE_ENDSUB: + break; + case TGSI_OPCODE_CAL: + brw_push_insn_state(p); + brw_set_mask_control(p, BRW_MASK_DISABLE); + brw_set_access_mode(p, BRW_ALIGN_1); + brw_ADD(p, deref_1ud(c->stack_index, 0), brw_ip_reg(), brw_imm_d(3*16)); + brw_set_access_mode(p, BRW_ALIGN_16); + brw_ADD(p, + get_addr_reg(c->stack_index), + get_addr_reg(c->stack_index), brw_imm_d(4)); +// orig_inst = inst->Data; +// orig_inst->Data = &p->store[p->nr_insn]; + assert(0); + brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); + brw_pop_insn_state(p); + break; + + case TGSI_OPCODE_RET: +#if 0 + brw_push_insn_state(p); + brw_set_mask_control(p, BRW_MASK_DISABLE); + brw_ADD(p, + get_addr_reg(c->stack_index), + get_addr_reg(c->stack_index), brw_imm_d(-4)); + brw_set_access_mode(p, BRW_ALIGN_1); + brw_MOV(p, brw_ip_reg(), deref_1ud(c->stack_index, 0)); + brw_set_access_mode(p, BRW_ALIGN_16); + brw_pop_insn_state(p); +#else + emit_fb_write(c, inst); +#endif + + break; + case TGSI_OPCODE_LOOP: + c->loop_inst[c->loop_insn++] = brw_DO(p, BRW_EXECUTE_8); + break; + case TGSI_OPCODE_BRK: + brw_BREAK(p); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + break; + case TGSI_OPCODE_CONT: + brw_CONT(p); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + break; + case TGSI_OPCODE_ENDLOOP: + c->loop_insn--; + c->inst0 = c->inst1 = brw_WHILE(p, c->loop_inst[c->loop_insn]); + /* patch all the BREAK instructions from + last BEGINLOOP */ + while (c->inst0 > c->loop_inst[c->loop_insn]) { + c->inst0--; + if (c->inst0->header.opcode == BRW_OPCODE_BREAK) { + c->inst0->bits3.if_else.jump_count = c->inst1 - c->inst0 + 1; + c->inst0->bits3.if_else.pop_count = 0; + } else if (c->inst0->header.opcode == BRW_OPCODE_CONTINUE) { + c->inst0->bits3.if_else.jump_count = c->inst1 - c->inst0; + c->inst0->bits3.if_else.pop_count = 0; + } + } + break; + case TGSI_OPCODE_END: + emit_fb_write(c, inst); + break; + + default: + debug_printf("unsupported IR in fragment shader %d\n", + inst->Instruction.Opcode); + } +#if 0 + if (inst->CondUpdate) + brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); + else + brw_set_predicate_control(p, BRW_PREDICATE_NONE); +#endif +} + + + + + + +void brw_wm_glsl_emit(struct brw_wm_compile *c) +{ + struct tgsi_parse_context parse; + struct brw_compile *p = &c->func; + + brw_init_compile(&c->func); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + + c->reg_index = 0; + c->if_insn = 0; + c->loop_insn = 0; + c->stack_index = brw_indirect(0,0); + + /* Do static register allocation and parameter interpolation: + */ + brw_wm_emit_decls( c ); + + /* Emit the actual program. All done with very direct translation, + * hopefully we can improve on this shortly... + */ + brw_MOV(p, get_addr_reg(c->stack_index), brw_address(c->stack)); + + tgsi_parse_init( &parse, c->fp->program.tokens ); + + while( !tgsi_parse_end_of_tokens( &parse ) ) + { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + /* already done */ + break; + + case TGSI_TOKEN_TYPE_IMMEDIATE: + /* not handled yet */ + assert(0); + break; + + case TGSI_TOKEN_TYPE_INSTRUCTION: + brw_wm_emit_instruction(c, &parse.FullToken.FullInstruction); + break; + + default: + assert( 0 ); + } + } + + tgsi_parse_free (&parse); + + /* Fix up call targets: + */ +#if 0 + { + unsigned nr_insns = c->fp->program.Base.NumInstructions; + unsigned insn, target_insn; + struct tgsi_full_instruction *inst1, *inst2; + struct brw_instruction *brw_inst1, *brw_inst2; + int offset; + for (insn = 0; insn < nr_insns; insn++) { + inst1 = &c->fp->program.Base.Instructions[insn]; + brw_inst1 = inst1->Data; + switch (inst1->Opcode) { + case TGSI_OPCODE_CAL: + target_insn = inst1->BranchTarget; + inst2 = &c->fp->program.Base.Instructions[target_insn]; + brw_inst2 = inst2->Data; + offset = brw_inst2 - brw_inst1; + brw_set_src1(brw_inst1, brw_imm_d(offset*16)); + break; + default: + break; + } + } + } +#endif + + c->prog_data.total_grf = c->reg_index; + c->prog_data.total_scratch = 0; +} diff --git a/src/gallium/drivers/i965simple/brw_wm_iz.c b/src/gallium/drivers/i965simple/brw_wm_iz.c new file mode 100644 index 0000000000..6c5f25bf39 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_iz.c @@ -0,0 +1,214 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_wm.h" + + +#undef P /* prompted depth */ +#undef C /* computed */ +#undef N /* non-promoted? */ + +#define P 0 +#define C 1 +#define N 2 + +const struct { + unsigned mode:2; + unsigned sd_present:1; + unsigned sd_to_rt:1; + unsigned dd_present:1; + unsigned ds_present:1; +} wm_iz_table[IZ_BIT_MAX] = +{ + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 1, 1, 0, 0 }, + { C, 1, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 1, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 1, 1, 0, 0 }, + { C, 1, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 0, 1, 0, 0 }, + { C, 1, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 1, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 0, 0, 1 }, + { C, 0, 0, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 1, 1, 0, 1 }, + { C, 1, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 1, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 0, 0, 1 }, + { C, 0, 0, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 1, 1, 0, 1 }, + { C, 1, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 1, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 0 }, + { N, 0, 1, 0, 0 }, + { N, 0, 1, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 0 }, + { N, 0, 1, 0, 0 }, + { N, 0, 1, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 0 }, + { C, 0, 1, 1, 0 }, + { C, 0, 1, 1, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 1 }, + { N, 0, 1, 0, 1 }, + { N, 0, 1, 0, 1 }, + { P, 0, 0, 0, 0 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { P, 0, 0, 0, 0 }, + { N, 1, 1, 0, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { P, 0, 0, 0, 0 }, + { C, 0, 0, 0, 1 }, + { P, 0, 0, 0, 0 }, + { C, 0, 1, 0, 1 }, + { P, 0, 0, 0, 0 }, + { C, 1, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { C, 0, 1, 0, 1 }, + { P, 0, 0, 0, 0 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { P, 0, 0, 0, 0 }, + { C, 1, 1, 1, 1 }, + { C, 0, 1, 1, 1 }, + { C, 0, 1, 1, 1 } +}; + +void brw_wm_lookup_iz( unsigned line_aa, + unsigned lookup, + struct brw_wm_prog_key *key ) +{ + unsigned reg = 2; + + assert (lookup < IZ_BIT_MAX); + + if (lookup & IZ_PS_COMPUTES_DEPTH_BIT) + key->computes_depth = 1; + + if (wm_iz_table[lookup].sd_present) { + key->source_depth_reg = reg; + reg += 2; + } + + if (wm_iz_table[lookup].sd_to_rt) + key->source_depth_to_render_target = 1; + + if (wm_iz_table[lookup].ds_present || line_aa != AA_NEVER) { + key->aa_dest_stencil_reg = reg; + key->runtime_check_aads_emit = (!wm_iz_table[lookup].ds_present && + line_aa == AA_SOMETIMES); + reg++; + } + + if (wm_iz_table[lookup].dd_present) { + key->dest_depth_reg = reg; + reg+=2; + } + + key->nr_depth_regs = (reg+1)/2; +} + diff --git a/src/gallium/drivers/i965simple/brw_wm_sampler_state.c b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c new file mode 100644 index 0000000000..de42ffc5b1 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c @@ -0,0 +1,273 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + +#include "pipe/p_util.h" + + +#define COMPAREFUNC_ALWAYS 0 +#define COMPAREFUNC_NEVER 0x1 +#define COMPAREFUNC_LESS 0x2 +#define COMPAREFUNC_EQUAL 0x3 +#define COMPAREFUNC_LEQUAL 0x4 +#define COMPAREFUNC_GREATER 0x5 +#define COMPAREFUNC_NOTEQUAL 0x6 +#define COMPAREFUNC_GEQUAL 0x7 + +/* Samplers aren't strictly wm state from the hardware's perspective, + * but that is the only situation in which we use them in this driver. + */ + +static int intel_translate_shadow_compare_func(unsigned func) +{ + switch(func) { + case PIPE_FUNC_NEVER: + return COMPAREFUNC_ALWAYS; + case PIPE_FUNC_LESS: + return COMPAREFUNC_LEQUAL; + case PIPE_FUNC_LEQUAL: + return COMPAREFUNC_LESS; + case PIPE_FUNC_GREATER: + return COMPAREFUNC_GEQUAL; + case PIPE_FUNC_GEQUAL: + return COMPAREFUNC_GREATER; + case PIPE_FUNC_NOTEQUAL: + return COMPAREFUNC_EQUAL; + case PIPE_FUNC_EQUAL: + return COMPAREFUNC_NOTEQUAL; + case PIPE_FUNC_ALWAYS: + return COMPAREFUNC_NEVER; + } + + debug_printf("Unknown value in %s: %x\n", __FUNCTION__, func); + return COMPAREFUNC_NEVER; +} + +/* The brw (and related graphics cores) do not support GL_CLAMP. The + * Intel drivers for "other operating systems" implement GL_CLAMP as + * GL_CLAMP_TO_EDGE, so the same is done here. + */ +static unsigned translate_wrap_mode( int wrap ) +{ + switch( wrap ) { + case PIPE_TEX_WRAP_REPEAT: + return BRW_TEXCOORDMODE_WRAP; + case PIPE_TEX_WRAP_CLAMP: + return BRW_TEXCOORDMODE_CLAMP; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + return BRW_TEXCOORDMODE_CLAMP; /* conform likes it this way */ + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + return BRW_TEXCOORDMODE_CLAMP_BORDER; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + return BRW_TEXCOORDMODE_MIRROR; + default: + return BRW_TEXCOORDMODE_WRAP; + } +} + + +static unsigned U_FIXED(float value, unsigned frac_bits) +{ + value *= (1<<frac_bits); + return value < 0 ? 0 : value; +} + +static int S_FIXED(float value, unsigned frac_bits) +{ + return value * (1<<frac_bits); +} + + +static unsigned upload_default_color( struct brw_context *brw, + const float *color ) +{ + struct brw_sampler_default_color sdc; + + COPY_4V(sdc.color, color); + + return brw_cache_data( &brw->cache[BRW_SAMPLER_DEFAULT_COLOR], &sdc ); +} + + +/* + */ +static void brw_update_sampler_state( const struct pipe_sampler_state *pipe_sampler, + unsigned sdc_gs_offset, + struct brw_sampler_state *sampler) +{ + memset(sampler, 0, sizeof(*sampler)); + + switch (pipe_sampler->min_mip_filter) { + case PIPE_TEX_FILTER_NEAREST: + sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST; + break; + case PIPE_TEX_FILTER_LINEAR: + sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR; + break; + default: + break; + } + + switch (pipe_sampler->min_mip_filter) { + case PIPE_TEX_MIPFILTER_NEAREST: + sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST; + break; + case PIPE_TEX_MIPFILTER_LINEAR: + sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR; + break; + case PIPE_TEX_MIPFILTER_NONE: + sampler->ss0.mip_filter = BRW_MIPFILTER_NONE; + break; + default: + break; + } + /* Set Anisotropy: + */ + if (pipe_sampler->max_anisotropy > 1.0) { + sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC; + sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC; + + if (pipe_sampler->max_anisotropy > 2.0) { + sampler->ss3.max_aniso = MAX2((pipe_sampler->max_anisotropy - 2) / 2, + BRW_ANISORATIO_16); + } + } + else { + switch (pipe_sampler->mag_img_filter) { + case PIPE_TEX_FILTER_NEAREST: + sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST; + break; + case PIPE_TEX_FILTER_LINEAR: + sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR; + break; + default: + break; + } + } + + sampler->ss1.s_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_s); + sampler->ss1.r_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_r); + sampler->ss1.t_wrap_mode = translate_wrap_mode(pipe_sampler->wrap_t); + + /* Fulsim complains if I don't do this. Hardware doesn't mind: + */ +#if 0 + if (texObj->Target == GL_TEXTURE_CUBE_MAP_ARB) { + sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; + } +#endif + + /* Set shadow function: + */ + if (pipe_sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + /* Shadowing is "enabled" by emitting a particular sampler + * message (sample_c). So need to recompile WM program when + * shadow comparison is enabled on each/any texture unit. + */ + sampler->ss0.shadow_function = intel_translate_shadow_compare_func(pipe_sampler->compare_func); + } + + /* Set LOD bias: + */ + sampler->ss0.lod_bias = S_FIXED(CLAMP(pipe_sampler->lod_bias, -16, 15), 6); + + sampler->ss0.lod_preclamp = 1; /* OpenGL mode */ + sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */ + + /* Set BaseMipLevel, MaxLOD, MinLOD: + * + * XXX: I don't think that using firstLevel, lastLevel works, + * because we always setup the surface state as if firstLevel == + * level zero. Probably have to subtract firstLevel from each of + * these: + */ + sampler->ss0.base_level = U_FIXED(0, 1); + + sampler->ss1.max_lod = U_FIXED(MIN2(MAX2(pipe_sampler->max_lod, 0), 13), 6); + sampler->ss1.min_lod = U_FIXED(MIN2(MAX2(pipe_sampler->min_lod, 0), 13), 6); + + sampler->ss2.default_color_pointer = sdc_gs_offset >> 5; +} + + + +/* All samplers must be uploaded in a single contiguous array, which + * complicates various things. However, this is still too confusing - + * FIXME: simplify all the different new texture state flags. + */ +static void upload_wm_samplers(struct brw_context *brw) +{ + unsigned unit; + unsigned sampler_count = 0; + + /* BRW_NEW_SAMPLER */ + for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) { + /* determine unit enable/disable by looking for a bound texture */ + if (brw->attribs.Texture[unit]) { + const struct pipe_sampler_state *sampler = brw->attribs.Samplers[unit]; + unsigned sdc_gs_offset = upload_default_color(brw, sampler->border_color); + + brw_update_sampler_state(sampler, + sdc_gs_offset, + &brw->wm.sampler[unit]); + + sampler_count = unit + 1; + } + } + + if (brw->wm.sampler_count != sampler_count) { + brw->wm.sampler_count = sampler_count; + brw->state.dirty.cache |= CACHE_NEW_SAMPLER; + } + + brw->wm.sampler_gs_offset = 0; + + if (brw->wm.sampler_count) + brw->wm.sampler_gs_offset = + brw_cache_data_sz(&brw->cache[BRW_SAMPLER], + brw->wm.sampler, + sizeof(struct brw_sampler_state) * brw->wm.sampler_count); +} + +const struct brw_tracked_state brw_wm_samplers = { + .dirty = { + .brw = BRW_NEW_SAMPLER, + .cache = 0 + }, + .update = upload_wm_samplers +}; + diff --git a/src/gallium/drivers/i965simple/brw_wm_state.c b/src/gallium/drivers/i965simple/brw_wm_state.c new file mode 100644 index 0000000000..5ccd488842 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_state.c @@ -0,0 +1,194 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "brw_wm.h" +#include "pipe/p_util.h" + +/*********************************************************************** + * WM unit - fragment programs and rasterization + */ +static void upload_wm_unit(struct brw_context *brw ) +{ + struct brw_wm_unit_state wm; + unsigned max_threads; + unsigned per_thread; + + if (BRW_DEBUG & DEBUG_SINGLE_THREAD) + max_threads = 0; + else + max_threads = 31; + + + memset(&wm, 0, sizeof(wm)); + + /* CACHE_NEW_WM_PROG */ + wm.thread0.grf_reg_count = align(brw->wm.prog_data->total_grf, 16) / 16 - 1; + wm.thread0.kernel_start_pointer = brw->wm.prog_gs_offset >> 6; + wm.thread3.dispatch_grf_start_reg = brw->wm.prog_data->first_curbe_grf; + wm.thread3.urb_entry_read_length = brw->wm.prog_data->urb_read_length; + wm.thread3.const_urb_entry_read_length = brw->wm.prog_data->curb_read_length; + + wm.wm5.max_threads = max_threads; + + per_thread = align(brw->wm.prog_data->total_scratch, 1024); + assert(per_thread <= 12 * 1024); + +#if 0 + if (brw->wm.prog_data->total_scratch) { + unsigned total = per_thread * (max_threads + 1); + + /* Scratch space -- just have to make sure there is sufficient + * allocated for the active program and current number of threads. + */ + brw->wm.scratch_buffer_size = total; + if (brw->wm.scratch_buffer && + brw->wm.scratch_buffer_size > brw->wm.scratch_buffer->size) { + dri_bo_unreference(brw->wm.scratch_buffer); + brw->wm.scratch_buffer = NULL; + } + if (!brw->wm.scratch_buffer) { + brw->wm.scratch_buffer = dri_bo_alloc(intel->intelScreen->bufmgr, + "wm scratch", + brw->wm.scratch_buffer_size, + 4096, DRM_BO_FLAG_MEM_TT); + } + } + /* XXX: Scratch buffers are not implemented correectly. + * + * The scratch offset to be programmed into wm is relative to the general + * state base address. However, using dri_bo_alloc/dri_bo_emit_reloc (or + * the previous bmGenBuffers scheme), we get an offset relative to the + * start of framebuffer. Even before then, it was broken in other ways, + * so just fail for now if we hit that path. + */ + assert(brw->wm.prog_data->total_scratch == 0); +#endif + + /* CACHE_NEW_SURFACE */ + wm.thread1.binding_table_entry_count = brw->wm.nr_surfaces; + + /* BRW_NEW_CURBE_OFFSETS */ + wm.thread3.const_urb_entry_read_offset = brw->curbe.wm_start * 2; + + wm.thread3.urb_entry_read_offset = 0; + wm.thread1.depth_coef_urb_read_offset = 1; + wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; + + /* CACHE_NEW_SAMPLER */ + wm.wm4.sampler_count = (brw->wm.sampler_count + 1) / 4; + wm.wm4.sampler_state_pointer = brw->wm.sampler_gs_offset >> 5; + + /* BRW_NEW_FRAGMENT_PROGRAM */ + { + const struct brw_fragment_program *fp = brw->attribs.FragmentProgram; + + if (fp->UsesDepth) + wm.wm5.program_uses_depth = 1; /* as far as we can tell */ + + if (fp->ComputesDepth) + wm.wm5.program_computes_depth = 1; + + /* BRW_NEW_ALPHA_TEST */ + if (fp->UsesKill || + brw->attribs.DepthStencil->alpha.enabled) + wm.wm5.program_uses_killpixel = 1; + + wm.wm5.enable_8_pix = 1; + } + + wm.wm5.thread_dispatch_enable = 1; /* AKA: color_write */ + wm.wm5.legacy_line_rast = 0; + wm.wm5.legacy_global_depth_bias = 0; + wm.wm5.early_depth_test = 1; /* never need to disable */ + wm.wm5.line_aa_region_width = 0; + wm.wm5.line_endcap_aa_region_width = 1; + + /* BRW_NEW_RASTERIZER */ + if (brw->attribs.Raster->poly_stipple_enable) + wm.wm5.polygon_stipple = 1; + +#if 0 + if (brw->attribs.Polygon->OffsetFill) { + wm.wm5.depth_offset = 1; + /* Something wierd going on with legacy_global_depth_bias, + * offset_constant, scaling and MRD. This value passes glean + * but gives some odd results elsewere (eg. the + * quad-offset-units test). + */ + wm.global_depth_offset_constant = brw->attribs.Polygon->OffsetUnits * 2; + + /* This is the only value that passes glean: + */ + wm.global_depth_offset_scale = brw->attribs.Polygon->OffsetFactor; + } +#endif + + if (brw->attribs.Raster->line_stipple_enable) { + wm.wm5.line_stipple = 1; + } + + if (BRW_DEBUG & DEBUG_STATS) + wm.wm4.stats_enable = 1; + + brw->wm.state_gs_offset = brw_cache_data( &brw->cache[BRW_WM_UNIT], &wm ); + + if (brw->wm.prog_data->total_scratch) { + /* + dri_emit_reloc(brw->cache[BRW_WM_UNIT].pool->buffer, + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + (per_thread / 1024) - 1, + brw->wm.state_gs_offset + + ((char *)&wm.thread2 - (char *)&wm), + brw->wm.scratch_buffer); + */ + } else { + wm.thread2.scratch_space_base_pointer = 0; + } +} + +const struct brw_tracked_state brw_wm_unit = { + .dirty = { + .brw = (BRW_NEW_RASTERIZER | + BRW_NEW_ALPHA_TEST | + BRW_NEW_FS | + BRW_NEW_CURBE_OFFSETS), + + .cache = (CACHE_NEW_SURFACE | + CACHE_NEW_WM_PROG | + CACHE_NEW_SAMPLER) + }, + .update = upload_wm_unit +}; + diff --git a/src/gallium/drivers/i965simple/brw_wm_surface_state.c b/src/gallium/drivers/i965simple/brw_wm_surface_state.c new file mode 100644 index 0000000000..d16d919bce --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_surface_state.c @@ -0,0 +1,304 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + +static unsigned translate_tex_target( enum pipe_texture_target target ) +{ + switch (target) { + case PIPE_TEXTURE_1D: + return BRW_SURFACE_1D; + + case PIPE_TEXTURE_2D: + return BRW_SURFACE_2D; + + case PIPE_TEXTURE_3D: + return BRW_SURFACE_3D; + + case PIPE_TEXTURE_CUBE: + return BRW_SURFACE_CUBE; + + default: + assert(0); + return 0; + } +} + +static unsigned translate_tex_format( enum pipe_format pipe_format ) +{ + switch( pipe_format ) { + case PIPE_FORMAT_U_L8: + return BRW_SURFACEFORMAT_L8_UNORM; + + case PIPE_FORMAT_U_I8: + return BRW_SURFACEFORMAT_I8_UNORM; + + case PIPE_FORMAT_U_A8: + return BRW_SURFACEFORMAT_A8_UNORM; + + case PIPE_FORMAT_U_A8_L8: + return BRW_SURFACEFORMAT_L8A8_UNORM; + + case PIPE_FORMAT_R8G8B8_UNORM: + assert(0); /* not supported for sampling */ + return BRW_SURFACEFORMAT_R8G8B8_UNORM; + + case PIPE_FORMAT_B8G8R8A8_UNORM: + return BRW_SURFACEFORMAT_B8G8R8A8_UNORM; + + case PIPE_FORMAT_R8G8B8A8_UNORM: + return BRW_SURFACEFORMAT_R8G8B8A8_UNORM; + + case PIPE_FORMAT_R5G6B5_UNORM: + return BRW_SURFACEFORMAT_B5G6R5_UNORM; + + case PIPE_FORMAT_A1R5G5B5_UNORM: + return BRW_SURFACEFORMAT_B5G5R5A1_UNORM; + + case PIPE_FORMAT_A4R4G4B4_UNORM: + return BRW_SURFACEFORMAT_B4G4R4A4_UNORM; + + case PIPE_FORMAT_YCBCR_REV: + return BRW_SURFACEFORMAT_YCRCB_NORMAL; + + case PIPE_FORMAT_YCBCR: + return BRW_SURFACEFORMAT_YCRCB_SWAPUVY; +#if 0 + case PIPE_FORMAT_RGB_FXT1: + case PIPE_FORMAT_RGBA_FXT1: + return BRW_SURFACEFORMAT_FXT1; +#endif + + case PIPE_FORMAT_Z16_UNORM: + return BRW_SURFACEFORMAT_I16_UNORM; +#if 0 + case PIPE_FORMAT_RGB_DXT1: + return BRW_SURFACEFORMAT_DXT1_RGB; + + case PIPE_FORMAT_RGBA_DXT1: + return BRW_SURFACEFORMAT_BC1_UNORM; + + case PIPE_FORMAT_RGBA_DXT3: + return BRW_SURFACEFORMAT_BC2_UNORM; + + case PIPE_FORMAT_RGBA_DXT5: + return BRW_SURFACEFORMAT_BC3_UNORM; + + case PIPE_FORMAT_SRGBA8: + return BRW_SURFACEFORMAT_R8G8B8A8_UNORM_SRGB; + case PIPE_FORMAT_SRGB_DXT1: + return BRW_SURFACEFORMAT_BC1_UNORM_SRGB; +#endif + + default: + assert(0); + return 0; + } +} + +static unsigned brw_buffer_offset(struct brw_context *brw, + struct pipe_buffer *buffer) +{ + return brw->winsys->get_buffer_offset(brw->winsys, + buffer, + 0); +} + +static +void brw_update_texture_surface( struct brw_context *brw, + unsigned unit ) +{ + const struct brw_texture *tObj = brw->attribs.Texture[unit]; + struct brw_surface_state surf; + + memset(&surf, 0, sizeof(surf)); + + surf.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW; + surf.ss0.surface_type = translate_tex_target(tObj->base.target); + surf.ss0.surface_format = translate_tex_format(tObj->base.format); + + /* This is ok for all textures with channel width 8bit or less: + */ +/* surf.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */ + + /* Updated in emit_reloc */ + surf.ss1.base_addr = brw_buffer_offset( brw, tObj->buffer ); + + surf.ss2.mip_count = tObj->base.last_level; + surf.ss2.width = tObj->base.width[0] - 1; + surf.ss2.height = tObj->base.height[0] - 1; + + surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR; + surf.ss3.tiled_surface = 0; /* always zero */ + surf.ss3.pitch = tObj->pitch - 1; + surf.ss3.depth = tObj->base.depth[0] - 1; + + surf.ss4.min_lod = 0; + + if (tObj->base.target == PIPE_TEXTURE_CUBE) { + surf.ss0.cube_pos_x = 1; + surf.ss0.cube_pos_y = 1; + surf.ss0.cube_pos_z = 1; + surf.ss0.cube_neg_x = 1; + surf.ss0.cube_neg_y = 1; + surf.ss0.cube_neg_z = 1; + } + + brw->wm.bind.surf_ss_offset[unit + 1] = + brw_cache_data( &brw->cache[BRW_SS_SURFACE], &surf ); +} + + + +#define OFFSET(TYPE, FIELD) ( (unsigned)&(((TYPE *)0)->FIELD) ) + + +static void upload_wm_surfaces(struct brw_context *brw ) +{ + unsigned i; + + { + struct brw_surface_state surf; + + /* BRW_NEW_FRAMEBUFFER + */ + struct pipe_surface *pipe_surface = brw->attribs.FrameBuffer.cbufs[0];/*fixme*/ + + memset(&surf, 0, sizeof(surf)); + + if (pipe_surface != NULL) { + if (pipe_surface->cpp == 4) + surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; + else + surf.ss0.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM; + + surf.ss0.surface_type = BRW_SURFACE_2D; + + surf.ss1.base_addr = brw_buffer_offset( brw, pipe_surface->buffer ); + + surf.ss2.width = pipe_surface->width - 1; + surf.ss2.height = pipe_surface->height - 1; + surf.ss3.tile_walk = BRW_TILEWALK_XMAJOR; + surf.ss3.tiled_surface = 0; + surf.ss3.pitch = (pipe_surface->pitch * pipe_surface->cpp) - 1; + } else { + surf.ss0.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; + surf.ss0.surface_type = BRW_SURFACE_NULL; + } + + /* BRW_NEW_BLEND */ + surf.ss0.color_blend = (!brw->attribs.Blend->logicop_enable && + brw->attribs.Blend->blend_enable); + + + surf.ss0.writedisable_red = !(brw->attribs.Blend->colormask & PIPE_MASK_R); + surf.ss0.writedisable_green = !(brw->attribs.Blend->colormask & PIPE_MASK_G); + surf.ss0.writedisable_blue = !(brw->attribs.Blend->colormask & PIPE_MASK_B); + surf.ss0.writedisable_alpha = !(brw->attribs.Blend->colormask & PIPE_MASK_A); + + + + + brw->wm.bind.surf_ss_offset[0] = brw_cache_data( &brw->cache[BRW_SS_SURFACE], &surf ); + + brw->wm.nr_surfaces = 1; + } + + + /* BRW_NEW_TEXTURE + */ + for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { + const struct brw_texture *texUnit = brw->attribs.Texture[i]; + + if (texUnit && + texUnit->base.refcount/*(texUnit->refcount > 0) == really used */) { + + brw_update_texture_surface(brw, i); + + brw->wm.nr_surfaces = i+2; + } + else { + brw->wm.bind.surf_ss_offset[i+1] = 0; + } + } + + brw->wm.bind_ss_offset = brw_cache_data( &brw->cache[BRW_SS_SURF_BIND], + &brw->wm.bind ); +} + + +/* KW: Will find a different way to acheive this, see for example the + * state caches with relocs in the i915 swz driver. + */ +#if 0 +static void emit_reloc_wm_surfaces(struct brw_context *brw) +{ + int unit; + + if (brw->state.draw_region != NULL) { + /* Emit framebuffer relocation */ + dri_emit_reloc(brw_cache_buffer(brw, BRW_SS_SURFACE), + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, + 0, + brw->wm.bind.surf_ss_offset[0] + + offsetof(struct brw_surface_state, ss1), + brw->state.draw_region->buffer); + } + + /* Emit relocations for texture buffers */ + for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) { + struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[unit]; + struct gl_texture_object *tObj = texUnit->_Current; + struct intel_texture_object *intelObj = intel_texture_object(tObj); + + if (texUnit->_ReallyEnabled && intelObj->mt != NULL) { + dri_emit_reloc(brw_cache_buffer(brw, BRW_SS_SURFACE), + DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, + 0, + brw->wm.bind.surf_ss_offset[unit + 1] + + offsetof(struct brw_surface_state, ss1), + intelObj->mt->region->buffer); + } + } +} +#endif + +const struct brw_tracked_state brw_wm_surfaces = { + .dirty = { + .brw = (BRW_NEW_FRAMEBUFFER | + BRW_NEW_BLEND | + BRW_NEW_TEXTURE), + .cache = 0 + }, + .update = upload_wm_surfaces, +}; diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile new file mode 100644 index 0000000000..31438a882e --- /dev/null +++ b/src/gallium/drivers/softpipe/Makefile @@ -0,0 +1,50 @@ + +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = softpipe + +DRIVER_SOURCES = \ + sp_clear.c \ + sp_flush.c \ + sp_query.c \ + sp_context.c \ + sp_draw_arrays.c \ + sp_prim_setup.c \ + sp_prim_vbuf.c \ + sp_quad.c \ + sp_quad_alpha_test.c \ + sp_quad_blend.c \ + sp_quad_bufloop.c \ + sp_quad_colormask.c \ + sp_quad_coverage.c \ + sp_quad_depth_test.c \ + sp_quad_earlyz.c \ + sp_quad_fs.c \ + sp_quad_occlusion.c \ + sp_quad_output.c \ + sp_quad_stencil.c \ + sp_quad_stipple.c \ + sp_state_blend.c \ + sp_state_clip.c \ + sp_state_derived.c \ + sp_state_fs.c \ + sp_state_sampler.c \ + sp_state_rasterizer.c \ + sp_state_surface.c \ + sp_state_vertex.c \ + sp_texture.c \ + sp_tex_sample.c \ + sp_tile_cache.c \ + sp_surface.c + +C_SOURCES = \ + $(COMMON_SOURCES) \ + $(DRIVER_SOURCES) + +ASM_SOURCES = + +include ../Makefile.template + +symlinks: + diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript new file mode 100644 index 0000000000..d581ee8d3c --- /dev/null +++ b/src/gallium/drivers/softpipe/SConscript @@ -0,0 +1,42 @@ +Import('*') + +env = env.Clone() + +softpipe = env.ConvenienceLibrary( + target = 'softpipe', + source = [ + 'sp_clear.c', + 'sp_context.c', + 'sp_draw_arrays.c', + 'sp_flush.c', + 'sp_prim_setup.c', + 'sp_prim_vbuf.c', + 'sp_quad_alpha_test.c', + 'sp_quad_blend.c', + 'sp_quad_bufloop.c', + 'sp_quad.c', + 'sp_quad_colormask.c', + 'sp_quad_coverage.c', + 'sp_quad_depth_test.c', + 'sp_quad_earlyz.c', + 'sp_quad_fs.c', + 'sp_quad_occlusion.c', + 'sp_quad_output.c', + 'sp_quad_stencil.c', + 'sp_quad_stipple.c', + 'sp_query.c', + 'sp_state_blend.c', + 'sp_state_clip.c', + 'sp_state_derived.c', + 'sp_state_fs.c', + 'sp_state_rasterizer.c', + 'sp_state_sampler.c', + 'sp_state_surface.c', + 'sp_state_vertex.c', + 'sp_surface.c', + 'sp_tex_sample.c', + 'sp_texture.c', + 'sp_tile_cache.c', + ]) + +Export('softpipe')
\ No newline at end of file diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c new file mode 100644 index 0000000000..8d295a30ca --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_clear.c @@ -0,0 +1,73 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + */ + + +#include "pipe/p_defines.h" +#include "sp_clear.h" +#include "sp_context.h" +#include "sp_surface.h" +#include "sp_state.h" +#include "sp_tile_cache.h" + + +/** + * Clear the given surface to the specified value. + * No masking, no scissor (clear entire buffer). + */ +void +softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + uint i; + +#if 0 + softpipe_update_derived(softpipe); /* not needed?? */ +#endif + + if (ps == sp_tile_cache_get_surface(softpipe->zsbuf_cache)) { + sp_tile_cache_clear(softpipe->zsbuf_cache, clearValue); +#if TILE_CLEAR_OPTIMIZATION + return; +#endif + } + + for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) { + if (ps == sp_tile_cache_get_surface(softpipe->cbuf_cache[i])) { + sp_tile_cache_clear(softpipe->cbuf_cache[i], clearValue); + } + } + +#if !TILE_CLEAR_OPTIMIZATION + /* non-cached surface */ + pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, clearValue); +#endif +} diff --git a/src/gallium/drivers/softpipe/sp_clear.h b/src/gallium/drivers/softpipe/sp_clear.h new file mode 100644 index 0000000000..a8ed1c4ecc --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_clear.h @@ -0,0 +1,43 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + */ + +#ifndef SP_CLEAR_H +#define SP_CLEAR_H + +#include "pipe/p_state.h" +struct pipe_context; + +extern void +softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps, + unsigned clearValue); + + +#endif /* SP_CLEAR_H */ diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c new file mode 100644 index 0000000000..cea6b90104 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -0,0 +1,333 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "pipe/draw/draw_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "sp_clear.h" +#include "sp_context.h" +#include "sp_flush.h" +#include "sp_prim_setup.h" +#include "sp_prim_vbuf.h" +#include "sp_state.h" +#include "sp_surface.h" +#include "sp_tile_cache.h" +#include "sp_texture.h" +#include "sp_winsys.h" +#include "sp_query.h" + + + +/** + * Query format support for creating a texture, drawing surface, etc. + * \param format the format to test + * \param type one of PIPE_TEXTURE, PIPE_SURFACE + */ +static boolean +softpipe_is_format_supported( struct pipe_context *pipe, + enum pipe_format format, uint type ) +{ + switch (type) { + case PIPE_TEXTURE: + /* softpipe supports all texture formats */ + return TRUE; + case PIPE_SURFACE: + /* softpipe supports all (off-screen) surface formats */ + return TRUE; + default: + assert(0); + return FALSE; + } +} + + +/** + * Map any drawing surfaces which aren't already mapped + */ +void +softpipe_map_surfaces(struct softpipe_context *sp) +{ + unsigned i; + + for (i = 0; i < sp->framebuffer.num_cbufs; i++) { + sp_tile_cache_map_surfaces(sp->cbuf_cache[i]); + } + + sp_tile_cache_map_surfaces(sp->zsbuf_cache); +} + + +/** + * Unmap any mapped drawing surfaces + */ +void +softpipe_unmap_surfaces(struct softpipe_context *sp) +{ + uint i; + + for (i = 0; i < sp->framebuffer.num_cbufs; i++) + sp_flush_tile_cache(sp, sp->cbuf_cache[i]); + sp_flush_tile_cache(sp, sp->zsbuf_cache); + + for (i = 0; i < sp->framebuffer.num_cbufs; i++) { + sp_tile_cache_unmap_surfaces(sp->cbuf_cache[i]); + } + sp_tile_cache_unmap_surfaces(sp->zsbuf_cache); +} + + +static void softpipe_destroy( struct pipe_context *pipe ) +{ + struct softpipe_context *softpipe = softpipe_context( pipe ); + struct pipe_winsys *ws = pipe->winsys; + uint i; + + draw_destroy( softpipe->draw ); + + softpipe->quad.polygon_stipple->destroy( softpipe->quad.polygon_stipple ); + softpipe->quad.earlyz->destroy( softpipe->quad.earlyz ); + softpipe->quad.shade->destroy( softpipe->quad.shade ); + softpipe->quad.alpha_test->destroy( softpipe->quad.alpha_test ); + softpipe->quad.depth_test->destroy( softpipe->quad.depth_test ); + softpipe->quad.stencil_test->destroy( softpipe->quad.stencil_test ); + softpipe->quad.occlusion->destroy( softpipe->quad.occlusion ); + softpipe->quad.coverage->destroy( softpipe->quad.coverage ); + softpipe->quad.bufloop->destroy( softpipe->quad.bufloop ); + softpipe->quad.blend->destroy( softpipe->quad.blend ); + softpipe->quad.colormask->destroy( softpipe->quad.colormask ); + softpipe->quad.output->destroy( softpipe->quad.output ); + + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) + sp_destroy_tile_cache(softpipe->cbuf_cache[i]); + sp_destroy_tile_cache(softpipe->zsbuf_cache); + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) + sp_destroy_tile_cache(softpipe->tex_cache[i]); + + for (i = 0; i < Elements(softpipe->constants); i++) { + if (softpipe->constants[i].buffer) { + pipe_buffer_reference(ws, &softpipe->constants[i].buffer, NULL); + } + } + + FREE( softpipe ); +} + + +static const char *softpipe_get_name( struct pipe_context *pipe ) +{ + return "softpipe"; +} + +static const char *softpipe_get_vendor( struct pipe_context *pipe ) +{ + return "Tungsten Graphics, Inc."; +} + +static int softpipe_get_param(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + return 8; + case PIPE_CAP_NPOT_TEXTURES: + return 1; + case PIPE_CAP_TWO_SIDED_STENCIL: + return 1; + case PIPE_CAP_GLSL: + return 1; + case PIPE_CAP_S3TC: + return 0; + case PIPE_CAP_ANISOTROPIC_FILTER: + return 0; + case PIPE_CAP_POINT_SPRITE: + return 1; + case PIPE_CAP_MAX_RENDER_TARGETS: + return 1; + case PIPE_CAP_OCCLUSION_QUERY: + return 1; + case PIPE_CAP_TEXTURE_SHADOW_MAP: + return 1; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + return 12; /* max 2Kx2K */ + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + return 8; /* max 128x128x128 */ + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + return 12; /* max 2Kx2K */ + default: + return 0; + } +} + +static float softpipe_get_paramf(struct pipe_context *pipe, int param) +{ + switch (param) { + case PIPE_CAP_MAX_LINE_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_LINE_WIDTH_AA: + return 255.0; /* arbitrary */ + + case PIPE_CAP_MAX_POINT_WIDTH: + /* fall-through */ + case PIPE_CAP_MAX_POINT_WIDTH_AA: + return 255.0; /* arbitrary */ + + case PIPE_CAP_MAX_TEXTURE_ANISOTROPY: + return 0.0; + + case PIPE_CAP_MAX_TEXTURE_LOD_BIAS: + return 16.0; /* arbitrary */ + + default: + return 0; + } +} + +struct pipe_context *softpipe_create( struct pipe_winsys *pipe_winsys, + struct softpipe_winsys *softpipe_winsys ) +{ + struct softpipe_context *softpipe = CALLOC_STRUCT(softpipe_context); + uint i; + +#if defined(__i386__) || defined(__386__) + softpipe->use_sse = GETENV( "GALLIUM_NOSSE" ) == NULL; +#else + softpipe->use_sse = FALSE; +#endif + + softpipe->dump_fs = GETENV( "GALLIUM_DUMP_FS" ) != NULL; + + softpipe->pipe.winsys = pipe_winsys; + softpipe->pipe.destroy = softpipe_destroy; + + /* queries */ + softpipe->pipe.is_format_supported = softpipe_is_format_supported; + softpipe->pipe.get_name = softpipe_get_name; + softpipe->pipe.get_vendor = softpipe_get_vendor; + softpipe->pipe.get_param = softpipe_get_param; + softpipe->pipe.get_paramf = softpipe_get_paramf; + + /* state setters */ + softpipe->pipe.create_blend_state = softpipe_create_blend_state; + softpipe->pipe.bind_blend_state = softpipe_bind_blend_state; + softpipe->pipe.delete_blend_state = softpipe_delete_blend_state; + + softpipe->pipe.create_sampler_state = softpipe_create_sampler_state; + softpipe->pipe.bind_sampler_state = softpipe_bind_sampler_state; + softpipe->pipe.delete_sampler_state = softpipe_delete_sampler_state; + + softpipe->pipe.create_depth_stencil_alpha_state = softpipe_create_depth_stencil_state; + softpipe->pipe.bind_depth_stencil_alpha_state = softpipe_bind_depth_stencil_state; + softpipe->pipe.delete_depth_stencil_alpha_state = softpipe_delete_depth_stencil_state; + + softpipe->pipe.create_rasterizer_state = softpipe_create_rasterizer_state; + softpipe->pipe.bind_rasterizer_state = softpipe_bind_rasterizer_state; + softpipe->pipe.delete_rasterizer_state = softpipe_delete_rasterizer_state; + + softpipe->pipe.create_fs_state = softpipe_create_fs_state; + softpipe->pipe.bind_fs_state = softpipe_bind_fs_state; + softpipe->pipe.delete_fs_state = softpipe_delete_fs_state; + + softpipe->pipe.create_vs_state = softpipe_create_vs_state; + softpipe->pipe.bind_vs_state = softpipe_bind_vs_state; + softpipe->pipe.delete_vs_state = softpipe_delete_vs_state; + + softpipe->pipe.set_blend_color = softpipe_set_blend_color; + softpipe->pipe.set_clip_state = softpipe_set_clip_state; + softpipe->pipe.set_constant_buffer = softpipe_set_constant_buffer; + softpipe->pipe.set_framebuffer_state = softpipe_set_framebuffer_state; + softpipe->pipe.set_polygon_stipple = softpipe_set_polygon_stipple; + softpipe->pipe.set_scissor_state = softpipe_set_scissor_state; + softpipe->pipe.set_sampler_texture = softpipe_set_sampler_texture; + softpipe->pipe.set_viewport_state = softpipe_set_viewport_state; + + softpipe->pipe.set_vertex_buffer = softpipe_set_vertex_buffer; + softpipe->pipe.set_vertex_element = softpipe_set_vertex_element; + + softpipe->pipe.draw_arrays = softpipe_draw_arrays; + softpipe->pipe.draw_elements = softpipe_draw_elements; + + softpipe->pipe.clear = softpipe_clear; + softpipe->pipe.flush = softpipe_flush; + + softpipe_init_query_funcs( softpipe ); + + /* textures */ + softpipe->pipe.texture_create = softpipe_texture_create; + softpipe->pipe.texture_release = softpipe_texture_release; + softpipe->pipe.get_tex_surface = softpipe_get_tex_surface; + + /* + * Alloc caches for accessing drawing surfaces and textures. + * Must be before quad stage setup! + */ + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) + softpipe->cbuf_cache[i] = sp_create_tile_cache(); + softpipe->zsbuf_cache = sp_create_tile_cache(); + + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) + softpipe->tex_cache[i] = sp_create_tile_cache(); + + + /* setup quad rendering stages */ + softpipe->quad.polygon_stipple = sp_quad_polygon_stipple_stage(softpipe); + softpipe->quad.earlyz = sp_quad_earlyz_stage(softpipe); + softpipe->quad.shade = sp_quad_shade_stage(softpipe); + softpipe->quad.alpha_test = sp_quad_alpha_test_stage(softpipe); + softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe); + softpipe->quad.stencil_test = sp_quad_stencil_test_stage(softpipe); + softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe); + softpipe->quad.coverage = sp_quad_coverage_stage(softpipe); + softpipe->quad.bufloop = sp_quad_bufloop_stage(softpipe); + softpipe->quad.blend = sp_quad_blend_stage(softpipe); + softpipe->quad.colormask = sp_quad_colormask_stage(softpipe); + softpipe->quad.output = sp_quad_output_stage(softpipe); + + softpipe->winsys = softpipe_winsys; + + /* + * Create drawing context and plug our rendering stage into it. + */ + softpipe->draw = draw_create(); + assert(softpipe->draw); + softpipe->setup = sp_draw_render_stage(softpipe); + + if (GETENV( "SP_VBUF" ) != NULL) { + sp_init_vbuf(softpipe); + } + else { + draw_set_rasterize_stage(softpipe->draw, softpipe->setup); + } + + sp_init_surface_functions(softpipe); + + return &softpipe->pipe; +} diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h new file mode 100644 index 0000000000..aff8c2cc5d --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -0,0 +1,152 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_CONTEXT_H +#define SP_CONTEXT_H + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" + +#include "pipe/draw/draw_vertex.h" + +#include "sp_quad.h" + + +struct softpipe_winsys; +struct softpipe_vbuf_render; +struct draw_context; +struct draw_stage; +struct softpipe_tile_cache; +struct sp_fragment_shader_state; +struct sp_vertex_shader_state; + + +struct softpipe_context { + struct pipe_context pipe; /**< base class */ + struct softpipe_winsys *winsys; /**< window system interface */ + + + /* The most recent drawing state as set by the driver: + */ + const struct pipe_blend_state *blend; + const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS]; + const struct pipe_depth_stencil_alpha_state *depth_stencil; + const struct pipe_rasterizer_state *rasterizer; + const struct sp_fragment_shader_state *fs; + const struct sp_vertex_shader_state *vs; + + struct pipe_blend_color blend_color; + struct pipe_clip_state clip; + struct pipe_constant_buffer constants[2]; + struct pipe_framebuffer_state framebuffer; + struct pipe_poly_stipple poly_stipple; + struct pipe_scissor_state scissor; + struct softpipe_texture *texture[PIPE_MAX_SAMPLERS]; + struct pipe_viewport_state viewport; + struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX]; + struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX]; + unsigned dirty; + + /* Counter for occlusion queries. Note this supports overlapping + * queries. + */ + uint64 occlusion_count; + + /* + * Mapped vertex buffers + */ + ubyte *mapped_vbuffer[PIPE_ATTRIB_MAX]; + + /** Mapped constant buffers */ + void *mapped_constants[PIPE_SHADER_TYPES]; + + /** Vertex format */ + struct vertex_info vertex_info; + struct vertex_info vertex_info_vbuf; + + int psize_slot; + +#if 0 + /* Stipple derived state: + */ + ubyte stipple_masks[16][16]; +#endif + + /** Derived from scissor and surface bounds: */ + struct pipe_scissor_state cliprect; + + unsigned line_stipple_counter; + + /** Software quad rendering pipeline */ + struct { + struct quad_stage *polygon_stipple; + struct quad_stage *earlyz; + struct quad_stage *shade; + struct quad_stage *alpha_test; + struct quad_stage *stencil_test; + struct quad_stage *depth_test; + struct quad_stage *occlusion; + struct quad_stage *coverage; + struct quad_stage *bufloop; + struct quad_stage *blend; + struct quad_stage *colormask; + struct quad_stage *output; + + struct quad_stage *first; /**< points to one of the above stages */ + } quad; + + /** The primitive drawing context */ + struct draw_context *draw; + struct draw_stage *setup; + struct draw_stage *vbuf; + struct softpipe_vbuf_render *vbuf_render; + + uint current_cbuf; /**< current color buffer being written to */ + + struct softpipe_tile_cache *cbuf_cache[PIPE_MAX_COLOR_BUFS]; + struct softpipe_tile_cache *zsbuf_cache; + + struct softpipe_tile_cache *tex_cache[PIPE_MAX_SAMPLERS]; + + int use_sse : 1; + int dump_fs : 1; +}; + + + + +static INLINE struct softpipe_context * +softpipe_context( struct pipe_context *pipe ) +{ + return (struct softpipe_context *)pipe; +} + + +#endif /* SP_CONTEXT_H */ diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c new file mode 100644 index 0000000000..71a303a8b5 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c @@ -0,0 +1,164 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Brian Paul + * Keith Whitwell + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_context.h" +#include "pipe/p_winsys.h" + +#include "sp_context.h" +#include "sp_state.h" + +#include "pipe/draw/draw_context.h" + + + +static void +softpipe_map_constant_buffers(struct softpipe_context *sp) +{ + struct pipe_winsys *ws = sp->pipe.winsys; + uint i; + for (i = 0; i < 2; i++) { + if (sp->constants[i].size) + sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + } + + draw_set_mapped_constant_buffer(sp->draw, + sp->mapped_constants[PIPE_SHADER_VERTEX]); +} + +static void +softpipe_unmap_constant_buffers(struct softpipe_context *sp) +{ + struct pipe_winsys *ws = sp->pipe.winsys; + uint i; + for (i = 0; i < 2; i++) { + if (sp->constants[i].size) + ws->buffer_unmap(ws, sp->constants[i].buffer); + sp->mapped_constants[i] = NULL; + } +} + + +boolean +softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count) +{ + return softpipe_draw_elements(pipe, NULL, 0, mode, start, count); +} + + + +/** + * Draw vertex arrays, with optional indexing. + * Basically, map the vertex buffers (and drawing surfaces), then hand off + * the drawing to the 'draw' module. + * + * XXX should the element buffer be specified/bound with a separate function? + */ +boolean +softpipe_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count) +{ + struct softpipe_context *sp = softpipe_context(pipe); + struct draw_context *draw = sp->draw; + unsigned i; + + /* first, check that the primitive is not malformed. It is the + * state tracker's responsibility to do send only correctly formed + * primitives down. It currently isn't doing that though... + */ +#if 1 + count = draw_trim_prim( mode, count ); +#else + if (!draw_validate_prim( mode, count )) + assert(0); +#endif + + + if (sp->dirty) + softpipe_update_derived( sp ); + + softpipe_map_surfaces(sp); + softpipe_map_constant_buffers(sp); + + /* + * Map vertex buffers + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (sp->vertex_buffer[i].buffer) { + void *buf + = pipe->winsys->buffer_map(pipe->winsys, + sp->vertex_buffer[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_vertex_buffer(draw, i, buf); + } + } + /* Map index buffer, if present */ + if (indexBuffer) { + void *mapped_indexes + = pipe->winsys->buffer_map(pipe->winsys, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes); + } + else { + /* no index/element buffer */ + draw_set_mapped_element_buffer(draw, 0, NULL); + } + + + /* draw! */ + draw_arrays(draw, mode, start, count); + + /* + * unmap vertex/index buffers - will cause draw module to flush + */ + for (i = 0; i < PIPE_ATTRIB_MAX; i++) { + if (sp->vertex_buffer[i].buffer) { + draw_set_mapped_vertex_buffer(draw, i, NULL); + pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer); + } + } + if (indexBuffer) { + draw_set_mapped_element_buffer(draw, 0, NULL); + pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer); + } + + + /* Note: leave drawing surfaces mapped */ + softpipe_unmap_constant_buffers(sp); + + return TRUE; +} diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c new file mode 100644 index 0000000000..ced0d5d098 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_flush.c @@ -0,0 +1,76 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_defines.h" +#include "pipe/draw/draw_context.h" +#include "sp_flush.h" +#include "sp_context.h" +#include "sp_surface.h" +#include "sp_state.h" +#include "sp_tile_cache.h" +#include "sp_winsys.h" + + +/* There will be actual work to do here. In future we may want a + * fence-like interface instead of finish, and perhaps flush will take + * flags to indicate what type of flush is required. + */ +void +softpipe_flush( struct pipe_context *pipe, + unsigned flags ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + uint i; + + draw_flush(softpipe->draw); + + /* - flush the quad pipeline + * - flush the texture cache + * - flush the render cache + */ + + for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) + if (softpipe->cbuf_cache[i]) + sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]); + + if (softpipe->zsbuf_cache) + sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache); + + /* Need this call for hardware buffers before swapbuffers. + * + * there should probably be another/different flush-type function + * that's called before swapbuffers because we don't always want + * to unmap surfaces when flushing. + */ + softpipe_unmap_surfaces(softpipe); +} + diff --git a/src/gallium/drivers/softpipe/sp_flush.h b/src/gallium/drivers/softpipe/sp_flush.h new file mode 100644 index 0000000000..34ec617866 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_flush.h @@ -0,0 +1,35 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SP_FLUSH_H +#define SP_FLUSH_H + +struct pipe_context; + +void softpipe_flush(struct pipe_context *pipe, unsigned flags ); + +#endif diff --git a/src/gallium/drivers/softpipe/sp_headers.h b/src/gallium/drivers/softpipe/sp_headers.h new file mode 100644 index 0000000000..0ae31d8796 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_headers.h @@ -0,0 +1,82 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_HEADERS_H +#define SP_HEADERS_H + +#include "pipe/tgsi/exec/tgsi_exec.h" + +#define PRIM_POINT 1 +#define PRIM_LINE 2 +#define PRIM_TRI 3 + + +/* The rasterizer generates 2x2 quads of fragment and feeds them to + * the current fp_machine (see below). + * Remember that Y=0=top with Y increasing down the window. + */ +#define QUAD_TOP_LEFT 0 +#define QUAD_TOP_RIGHT 1 +#define QUAD_BOTTOM_LEFT 2 +#define QUAD_BOTTOM_RIGHT 3 + +#define MASK_TOP_LEFT (1 << QUAD_TOP_LEFT) +#define MASK_TOP_RIGHT (1 << QUAD_TOP_RIGHT) +#define MASK_BOTTOM_LEFT (1 << QUAD_BOTTOM_LEFT) +#define MASK_BOTTOM_RIGHT (1 << QUAD_BOTTOM_RIGHT) +#define MASK_ALL 0xf + + +/** + * Encodes everything we need to know about a 2x2 pixel block. Uses + * "Channel-Serial" or "SoA" layout. + */ +struct quad_header { + int x0; + int y0; + unsigned mask:4; + unsigned facing:1; /**< Front (0) or back (1) facing? */ + unsigned prim:2; /**< PRIM_POINT, LINE, TRI */ + + struct { + float color[NUM_CHANNELS][QUAD_SIZE]; /* rrrr, gggg, bbbb, aaaa */ + float depth[QUAD_SIZE]; + } outputs; + + float coverage[QUAD_SIZE]; /** fragment coverage for antialiasing */ + + const struct tgsi_interp_coef *coef; + const struct tgsi_interp_coef *posCoef; + + unsigned nr_attrs; +}; + + +#endif /* SP_HEADERS_H */ diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c new file mode 100644 index 0000000000..2772048661 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_prim_setup.c @@ -0,0 +1,1247 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief Primitive rasterization/rendering (points, lines, triangles) + * + * \author Keith Whitwell <keith@tungstengraphics.com> + * \author Brian Paul + */ + + +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_quad.h" +#include "sp_state.h" +#include "sp_prim_setup.h" +#include "pipe/draw/draw_private.h" +#include "pipe/draw/draw_vertex.h" +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" + +#define DEBUG_VERTS 0 + +/** + * Triangle edge info + */ +struct edge { + float dx; /**< X(v1) - X(v0), used only during setup */ + float dy; /**< Y(v1) - Y(v0), used only during setup */ + float dxdy; /**< dx/dy */ + float sx, sy; /**< first sample point coord */ + int lines; /**< number of lines on this edge */ +}; + + +/** + * Triangle setup info (derived from draw_stage). + * Also used for line drawing (taking some liberties). + */ +struct setup_stage { + struct draw_stage stage; /**< This must be first (base class) */ + + struct softpipe_context *softpipe; + + /* Vertices are just an array of floats making up each attribute in + * turn. Currently fixed at 4 floats, but should change in time. + * Codegen will help cope with this. + */ + const struct vertex_header *vmax; + const struct vertex_header *vmid; + const struct vertex_header *vmin; + const struct vertex_header *vprovoke; + + struct edge ebot; + struct edge etop; + struct edge emaj; + + float oneoverarea; + + struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; + struct tgsi_interp_coef posCoef; /* For Z, W */ + struct quad_header quad; + + struct { + int left[2]; /**< [0] = row0, [1] = row1 */ + int right[2]; + int y; + unsigned y_flags; + unsigned mask; /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */ + } span; +}; + + + +/** + * Basically a cast wrapper. + */ +static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) +{ + return (struct setup_stage *)stage; +} + + +/** + * Clip setup->quad against the scissor/surface bounds. + */ +static INLINE void +quad_clip(struct setup_stage *setup) +{ + const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect; + const int minx = (int) cliprect->minx; + const int maxx = (int) cliprect->maxx; + const int miny = (int) cliprect->miny; + const int maxy = (int) cliprect->maxy; + + if (setup->quad.x0 >= maxx || + setup->quad.y0 >= maxy || + setup->quad.x0 + 1 < minx || + setup->quad.y0 + 1 < miny) { + /* totally clipped */ + setup->quad.mask = 0x0; + return; + } + if (setup->quad.x0 < minx) + setup->quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); + if (setup->quad.y0 < miny) + setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); + if (setup->quad.x0 == maxx - 1) + setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); + if (setup->quad.y0 == maxy - 1) + setup->quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); +} + + +/** + * Emit a quad (pass to next stage) with clipping. + */ +static INLINE void +clip_emit_quad(struct setup_stage *setup) +{ + quad_clip(setup); + if (setup->quad.mask) { + struct softpipe_context *sp = setup->softpipe; + sp->quad.first->run(sp->quad.first, &setup->quad); + } +} + + +/** + * Emit a quad (pass to next stage). No clipping is done. + */ +static INLINE void +emit_quad( struct setup_stage *setup, int x, int y, unsigned mask ) +{ + struct softpipe_context *sp = setup->softpipe; + setup->quad.x0 = x; + setup->quad.y0 = y; + setup->quad.mask = mask; + sp->quad.first->run(sp->quad.first, &setup->quad); +} + + +/** + * Given an X or Y coordinate, return the block/quad coordinate that it + * belongs to. + */ +static INLINE int block( int x ) +{ + return x & ~1; +} + + +/** + * Compute mask which indicates which pixels in the 2x2 quad are actually inside + * the triangle's bounds. + * + * this is pretty nasty... may need to rework flush_spans again to + * fix it, if possible. + */ +static unsigned calculate_mask( struct setup_stage *setup, int x ) +{ + unsigned mask = 0x0; + + if (x >= setup->span.left[0] && x < setup->span.right[0]) + mask |= MASK_TOP_LEFT; + + if (x >= setup->span.left[1] && x < setup->span.right[1]) + mask |= MASK_BOTTOM_LEFT; + + if (x+1 >= setup->span.left[0] && x+1 < setup->span.right[0]) + mask |= MASK_TOP_RIGHT; + + if (x+1 >= setup->span.left[1] && x+1 < setup->span.right[1]) + mask |= MASK_BOTTOM_RIGHT; + + return mask; +} + + +/** + * Render a horizontal span of quads + */ +static void flush_spans( struct setup_stage *setup ) +{ + int minleft, maxright; + int x; + + switch (setup->span.y_flags) { + case 0x3: + /* both odd and even lines written (both quad rows) */ + minleft = MIN2(setup->span.left[0], setup->span.left[1]); + maxright = MAX2(setup->span.right[0], setup->span.right[1]); + break; + + case 0x1: + /* only even line written (quad top row) */ + minleft = setup->span.left[0]; + maxright = setup->span.right[0]; + break; + + case 0x2: + /* only odd line written (quad bottom row) */ + minleft = setup->span.left[1]; + maxright = setup->span.right[1]; + break; + + default: + return; + } + + /* XXX this loop could be moved into the above switch cases and + * calculate_mask() could be simplified a bit... + */ + for (x = block(minleft); x <= block(maxright); x += 2) { + emit_quad( setup, x, setup->span.y, + calculate_mask( setup, x ) ); + } + + setup->span.y = 0; + setup->span.y_flags = 0; + setup->span.right[0] = 0; + setup->span.right[1] = 0; +} + +#if DEBUG_VERTS +static void print_vertex(const struct setup_stage *setup, + const struct vertex_header *v) +{ + int i; + debug_printf("Vertex: (%p)\n", v); + for (i = 0; i < setup->quad.nr_attrs; i++) { + debug_printf(" %d: %f %f %f %f\n", i, + v->data[i][0], v->data[i][1], v->data[i][2], v->data[i][3]); + } +} +#endif + +static boolean setup_sort_vertices( struct setup_stage *setup, + const struct prim_header *prim ) +{ + const struct vertex_header *v0 = prim->v[0]; + const struct vertex_header *v1 = prim->v[1]; + const struct vertex_header *v2 = prim->v[2]; + +#if DEBUG_VERTS + debug_printf("Triangle:\n"); + print_vertex(setup, v0); + print_vertex(setup, v1); + print_vertex(setup, v2); +#endif + + setup->vprovoke = v2; + + /* determine bottom to top order of vertices */ + { + float y0 = v0->data[0][1]; + float y1 = v1->data[0][1]; + float y2 = v2->data[0][1]; + if (y0 <= y1) { + if (y1 <= y2) { + /* y0<=y1<=y2 */ + setup->vmin = v0; + setup->vmid = v1; + setup->vmax = v2; + } + else if (y2 <= y0) { + /* y2<=y0<=y1 */ + setup->vmin = v2; + setup->vmid = v0; + setup->vmax = v1; + } + else { + /* y0<=y2<=y1 */ + setup->vmin = v0; + setup->vmid = v2; + setup->vmax = v1; + } + } + else { + if (y0 <= y2) { + /* y1<=y0<=y2 */ + setup->vmin = v1; + setup->vmid = v0; + setup->vmax = v2; + } + else if (y2 <= y1) { + /* y2<=y1<=y0 */ + setup->vmin = v2; + setup->vmid = v1; + setup->vmax = v0; + } + else { + /* y1<=y2<=y0 */ + setup->vmin = v1; + setup->vmid = v2; + setup->vmax = v0; + } + } + } + + setup->ebot.dx = setup->vmid->data[0][0] - setup->vmin->data[0][0]; + setup->ebot.dy = setup->vmid->data[0][1] - setup->vmin->data[0][1]; + setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0]; + setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1]; + setup->etop.dx = setup->vmax->data[0][0] - setup->vmid->data[0][0]; + setup->etop.dy = setup->vmax->data[0][1] - setup->vmid->data[0][1]; + + /* + * Compute triangle's area. Use 1/area to compute partial + * derivatives of attributes later. + * + * The area will be the same as prim->det, but the sign may be + * different depending on how the vertices get sorted above. + * + * To determine whether the primitive is front or back facing we + * use the prim->det value because its sign is correct. + */ + { + const float area = (setup->emaj.dx * setup->ebot.dy - + setup->ebot.dx * setup->emaj.dy); + + setup->oneoverarea = 1.0f / area; + /* + debug_printf("%s one-over-area %f area %f det %f\n", + __FUNCTION__, setup->oneoverarea, area, prim->det ); + */ + } + + /* We need to know if this is a front or back-facing triangle for: + * - the GLSL gl_FrontFacing fragment attribute (bool) + * - two-sided stencil test + */ + setup->quad.facing = (prim->det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW); + + return TRUE; +} + + +/** + * Compute a0 for a constant-valued coefficient (GL_FLAT shading). + * The value value comes from vertex->data[slot][i]. + * The result will be put into setup->coef[slot].a0[i]. + * \param slot which attribute slot + * \param i which component of the slot (0..3) + */ +static void const_coeff( struct setup_stage *setup, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + assert(i <= 3); + + coef->dadx[i] = 0; + coef->dady[i] = 0; + + /* need provoking vertex info! + */ + coef->a0[i] = setup->vprovoke->data[vertSlot][i]; +} + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a triangle. + */ +static void tri_linear_coeff( struct setup_stage *setup, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + float botda = setup->vmid->data[vertSlot][i] - setup->vmin->data[vertSlot][i]; + float majda = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i]; + float a = setup->ebot.dy * majda - botda * setup->emaj.dy; + float b = setup->emaj.dx * botda - majda * setup->ebot.dx; + float dadx = a * setup->oneoverarea; + float dady = b * setup->oneoverarea; + + assert(i <= 3); + + coef->dadx[i] = dadx; + coef->dady[i] = dady; + + /* calculate a0 as the value which would be sampled for the + * fragment at (0,0), taking into account that we want to sample at + * pixel centers, in other words (0.5, 0.5). + * + * this is neat but unfortunately not a good way to do things for + * triangles with very large values of dadx or dady as it will + * result in the subtraction and re-addition from a0 of a very + * large number, which means we'll end up loosing a lot of the + * fractional bits and precision from a0. the way to fix this is + * to define a0 as the sample at a pixel center somewhere near vmin + * instead - i'll switch to this later. + */ + coef->a0[i] = (setup->vmin->data[vertSlot][i] - + (dadx * (setup->vmin->data[0][0] - 0.5f) + + dady * (setup->vmin->data[0][1] - 0.5f))); + + /* + debug_printf("attr[%d].%c: %f dx:%f dy:%f\n", + slot, "xyzw"[i], + setup->coef[slot].a0[i], + setup->coef[slot].dadx[i], + setup->coef[slot].dady[i]); + */ +} + + +/** + * Compute a0, dadx and dady for a perspective-corrected interpolant, + * for a triangle. + * We basically multiply the vertex value by 1/w before computing + * the plane coefficients (a0, dadx, dady). + * Later, when we compute the value at a particular fragment position we'll + * divide the interpolated value by the interpolated W at that fragment. + */ +static void tri_persp_coeff( struct setup_stage *setup, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + /* premultiply by 1/w (v->data[0][3] is always W): + */ + float mina = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3]; + float mida = setup->vmid->data[vertSlot][i] * setup->vmid->data[0][3]; + float maxa = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3]; + float botda = mida - mina; + float majda = maxa - mina; + float a = setup->ebot.dy * majda - botda * setup->emaj.dy; + float b = setup->emaj.dx * botda - majda * setup->ebot.dx; + float dadx = a * setup->oneoverarea; + float dady = b * setup->oneoverarea; + + /* + debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i, + setup->vmin->data[vertSlot][i], + setup->vmid->data[vertSlot][i], + setup->vmax->data[vertSlot][i] + ); + */ + assert(i <= 3); + + coef->dadx[i] = dadx; + coef->dady[i] = dady; + coef->a0[i] = (mina - + (dadx * (setup->vmin->data[0][0] - 0.5f) + + dady * (setup->vmin->data[0][1] - 0.5f))); +} + + +/** + * Special coefficient setup for gl_FragCoord. + * X and Y are trivial, though Y has to be inverted for OpenGL. + * Z and W are copied from posCoef which should have already been computed. + * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask. + */ +static void +setup_fragcoord_coeff(struct setup_stage *setup) +{ + /*X*/ + setup->coef[0].a0[0] = 0; + setup->coef[0].dadx[0] = 1.0; + setup->coef[0].dady[0] = 0.0; + /*Y*/ + if (setup->softpipe->rasterizer->origin_lower_left) { + /* y=0=bottom */ + const int winHeight = setup->softpipe->framebuffer.cbufs[0]->height; + setup->coef[0].a0[1] = (float) (winHeight - 1); + setup->coef[0].dady[1] = -1.0; + } + else { + /* y=0=top */ + setup->coef[0].a0[1] = 0.0; + setup->coef[0].dady[1] = 1.0; + } + setup->coef[0].dadx[1] = 0.0; + /*Z*/ + setup->coef[0].a0[2] = setup->posCoef.a0[2]; + setup->coef[0].dadx[2] = setup->posCoef.dadx[2]; + setup->coef[0].dady[2] = setup->posCoef.dady[2]; + /*W*/ + setup->coef[0].a0[3] = setup->posCoef.a0[3]; + setup->coef[0].dadx[3] = setup->posCoef.dadx[3]; + setup->coef[0].dady[3] = setup->posCoef.dady[3]; +} + + + +/** + * Compute the setup->coef[] array dadx, dady, a0 values. + * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized. + */ +static void setup_tri_coefficients( struct setup_stage *setup ) +{ + struct softpipe_context *softpipe = setup->softpipe; + const struct pipe_shader_state *fs = &softpipe->fs->shader; + const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe); + uint fragSlot; + + /* z and w are done by linear interpolation: + */ + tri_linear_coeff(setup, &setup->posCoef, 0, 2); + tri_linear_coeff(setup, &setup->posCoef, 0, 3); + + /* setup interpolation for all the remaining attributes: + */ + for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) { + const uint vertSlot = vinfo->src_index[fragSlot]; + uint j; + + switch (vinfo->interp_mode[fragSlot]) { + case INTERP_CONSTANT: + for (j = 0; j < NUM_CHANNELS; j++) + const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_LINEAR: + for (j = 0; j < NUM_CHANNELS; j++) + tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_PERSPECTIVE: + for (j = 0; j < NUM_CHANNELS; j++) + tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_POS: + assert(fragSlot == 0); + setup_fragcoord_coeff(setup); + break; + default: + assert(0); + } + + if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) { + /* FOG.y = front/back facing XXX fix this */ + setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing; + setup->coef[fragSlot].dadx[1] = 0.0; + setup->coef[fragSlot].dady[1] = 0.0; + } + } +} + + + +static void setup_tri_edges( struct setup_stage *setup ) +{ + float vmin_x = setup->vmin->data[0][0] + 0.5f; + float vmid_x = setup->vmid->data[0][0] + 0.5f; + + float vmin_y = setup->vmin->data[0][1] - 0.5f; + float vmid_y = setup->vmid->data[0][1] - 0.5f; + float vmax_y = setup->vmax->data[0][1] - 0.5f; + + setup->emaj.sy = CEILF(vmin_y); + setup->emaj.lines = (int) CEILF(vmax_y - setup->emaj.sy); + setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy; + setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy; + + setup->etop.sy = CEILF(vmid_y); + setup->etop.lines = (int) CEILF(vmax_y - setup->etop.sy); + setup->etop.dxdy = setup->etop.dx / setup->etop.dy; + setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy; + + setup->ebot.sy = CEILF(vmin_y); + setup->ebot.lines = (int) CEILF(vmid_y - setup->ebot.sy); + setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy; + setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy; +} + + +/** + * Render the upper or lower half of a triangle. + * Scissoring/cliprect is applied here too. + */ +static void subtriangle( struct setup_stage *setup, + struct edge *eleft, + struct edge *eright, + unsigned lines ) +{ + const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect; + const int minx = (int) cliprect->minx; + const int maxx = (int) cliprect->maxx; + const int miny = (int) cliprect->miny; + const int maxy = (int) cliprect->maxy; + int y, start_y, finish_y; + int sy = (int)eleft->sy; + + assert((int)eleft->sy == (int) eright->sy); + + /* clip top/bottom */ + start_y = sy; + finish_y = sy + lines; + + if (start_y < miny) + start_y = miny; + + if (finish_y > maxy) + finish_y = maxy; + + start_y -= sy; + finish_y -= sy; + + /* + debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y); + */ + + for (y = start_y; y < finish_y; y++) { + + /* avoid accumulating adds as floats don't have the precision to + * accurately iterate large triangle edges that way. luckily we + * can just multiply these days. + * + * this is all drowned out by the attribute interpolation anyway. + */ + int left = (int)(eleft->sx + y * eleft->dxdy); + int right = (int)(eright->sx + y * eright->dxdy); + + /* clip left/right */ + if (left < minx) + left = minx; + if (right > maxx) + right = maxx; + + if (left < right) { + int _y = sy + y; + if (block(_y) != setup->span.y) { + flush_spans(setup); + setup->span.y = block(_y); + } + + setup->span.left[_y&1] = left; + setup->span.right[_y&1] = right; + setup->span.y_flags |= 1<<(_y&1); + } + } + + + /* save the values so that emaj can be restarted: + */ + eleft->sx += lines * eleft->dxdy; + eright->sx += lines * eright->dxdy; + eleft->sy += lines; + eright->sy += lines; +} + + +/** + * Do setup for triangle rasterization, then render the triangle. + */ +static void setup_tri( struct draw_stage *stage, + struct prim_header *prim ) +{ + struct setup_stage *setup = setup_stage( stage ); + + /* + debug_printf("%s\n", __FUNCTION__ ); + */ + + setup_sort_vertices( setup, prim ); + setup_tri_coefficients( setup ); + setup_tri_edges( setup ); + + setup->quad.prim = PRIM_TRI; + + setup->span.y = 0; + setup->span.y_flags = 0; + setup->span.right[0] = 0; + setup->span.right[1] = 0; + /* setup->span.z_mode = tri_z_mode( setup->ctx ); */ + + /* init_constant_attribs( setup ); */ + + if (setup->oneoverarea < 0.0) { + /* emaj on left: + */ + subtriangle( setup, &setup->emaj, &setup->ebot, setup->ebot.lines ); + subtriangle( setup, &setup->emaj, &setup->etop, setup->etop.lines ); + } + else { + /* emaj on right: + */ + subtriangle( setup, &setup->ebot, &setup->emaj, setup->ebot.lines ); + subtriangle( setup, &setup->etop, &setup->emaj, setup->etop.lines ); + } + + flush_spans( setup ); +} + + + +/** + * Compute a0, dadx and dady for a linearly interpolated coefficient, + * for a line. + */ +static void +line_linear_coeff(struct setup_stage *setup, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + const float da = setup->vmax->data[vertSlot][i] - setup->vmin->data[vertSlot][i]; + const float dadx = da * setup->emaj.dx * setup->oneoverarea; + const float dady = da * setup->emaj.dy * setup->oneoverarea; + coef->dadx[i] = dadx; + coef->dady[i] = dady; + coef->a0[i] = (setup->vmin->data[vertSlot][i] - + (dadx * (setup->vmin->data[0][0] - 0.5f) + + dady * (setup->vmin->data[0][1] - 0.5f))); +} + + +/** + * Compute a0, dadx and dady for a perspective-corrected interpolant, + * for a line. + */ +static void +line_persp_coeff(struct setup_stage *setup, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + /* XXX double-check/verify this arithmetic */ + const float a0 = setup->vmin->data[vertSlot][i] * setup->vmin->data[0][3]; + const float a1 = setup->vmax->data[vertSlot][i] * setup->vmax->data[0][3]; + const float da = a1 - a0; + const float dadx = da * setup->emaj.dx * setup->oneoverarea; + const float dady = da * setup->emaj.dy * setup->oneoverarea; + coef->dadx[i] = dadx; + coef->dady[i] = dady; + coef->a0[i] = (setup->vmin->data[vertSlot][i] - + (dadx * (setup->vmin->data[0][0] - 0.5f) + + dady * (setup->vmin->data[0][1] - 0.5f))); +} + + +/** + * Compute the setup->coef[] array dadx, dady, a0 values. + * Must be called after setup->vmin,vmax are initialized. + */ +static INLINE void +setup_line_coefficients(struct setup_stage *setup, struct prim_header *prim) +{ + struct softpipe_context *softpipe = setup->softpipe; + const struct pipe_shader_state *fs = &setup->softpipe->fs->shader; + const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe); + uint fragSlot; + + /* use setup->vmin, vmax to point to vertices */ + setup->vprovoke = prim->v[1]; + setup->vmin = prim->v[0]; + setup->vmax = prim->v[1]; + + setup->emaj.dx = setup->vmax->data[0][0] - setup->vmin->data[0][0]; + setup->emaj.dy = setup->vmax->data[0][1] - setup->vmin->data[0][1]; + /* NOTE: this is not really 1/area */ + setup->oneoverarea = 1.0f / (setup->emaj.dx * setup->emaj.dx + + setup->emaj.dy * setup->emaj.dy); + + /* z and w are done by linear interpolation: + */ + line_linear_coeff(setup, &setup->posCoef, 0, 2); + line_linear_coeff(setup, &setup->posCoef, 0, 3); + + /* setup interpolation for all the remaining attributes: + */ + for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) { + const uint vertSlot = vinfo->src_index[fragSlot]; + uint j; + + switch (vinfo->interp_mode[fragSlot]) { + case INTERP_CONSTANT: + for (j = 0; j < NUM_CHANNELS; j++) + const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_LINEAR: + for (j = 0; j < NUM_CHANNELS; j++) + line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_PERSPECTIVE: + for (j = 0; j < NUM_CHANNELS; j++) + line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_POS: + assert(fragSlot == 0); + assert(0); /* XXX fix this: */ + setup_fragcoord_coeff(setup); + break; + default: + assert(0); + } + + if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) { + /* FOG.y = front/back facing XXX fix this */ + setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing; + setup->coef[fragSlot].dadx[1] = 0.0; + setup->coef[fragSlot].dady[1] = 0.0; + } + } +} + + +/** + * Plot a pixel in a line segment. + */ +static INLINE void +plot(struct setup_stage *setup, int x, int y) +{ + const int iy = y & 1; + const int ix = x & 1; + const int quadX = x - ix; + const int quadY = y - iy; + const int mask = (1 << ix) << (2 * iy); + + if (quadX != setup->quad.x0 || + quadY != setup->quad.y0) + { + /* flush prev quad, start new quad */ + + if (setup->quad.x0 != -1) + clip_emit_quad(setup); + + setup->quad.x0 = quadX; + setup->quad.y0 = quadY; + setup->quad.mask = 0x0; + } + + setup->quad.mask |= mask; +} + + +/** + * Do setup for line rasterization, then render the line. + * Single-pixel width, no stipple, etc. We rely on the 'draw' module + * to handle stippling and wide lines. + */ +static void +setup_line(struct draw_stage *stage, struct prim_header *prim) +{ + const struct vertex_header *v0 = prim->v[0]; + const struct vertex_header *v1 = prim->v[1]; + struct setup_stage *setup = setup_stage( stage ); + int x0 = (int) v0->data[0][0]; + int x1 = (int) v1->data[0][0]; + int y0 = (int) v0->data[0][1]; + int y1 = (int) v1->data[0][1]; + int dx = x1 - x0; + int dy = y1 - y0; + int xstep, ystep; + + if (dx == 0 && dy == 0) + return; + + setup_line_coefficients(setup, prim); + + if (dx < 0) { + dx = -dx; /* make positive */ + xstep = -1; + } + else { + xstep = 1; + } + + if (dy < 0) { + dy = -dy; /* make positive */ + ystep = -1; + } + else { + ystep = 1; + } + + assert(dx >= 0); + assert(dy >= 0); + + setup->quad.x0 = setup->quad.y0 = -1; + setup->quad.mask = 0x0; + setup->quad.prim = PRIM_LINE; + /* XXX temporary: set coverage to 1.0 so the line appears + * if AA mode happens to be enabled. + */ + setup->quad.coverage[0] = + setup->quad.coverage[1] = + setup->quad.coverage[2] = + setup->quad.coverage[3] = 1.0; + + if (dx > dy) { + /*** X-major line ***/ + int i; + const int errorInc = dy + dy; + int error = errorInc - dx; + const int errorDec = error - dx; + + for (i = 0; i < dx; i++) { + plot(setup, x0, y0); + + x0 += xstep; + if (error < 0) { + error += errorInc; + } + else { + error += errorDec; + y0 += ystep; + } + } + } + else { + /*** Y-major line ***/ + int i; + const int errorInc = dx + dx; + int error = errorInc - dy; + const int errorDec = error - dy; + + for (i = 0; i < dy; i++) { + plot(setup, x0, y0); + + y0 += ystep; + if (error < 0) { + error += errorInc; + } + else { + error += errorDec; + x0 += xstep; + } + } + } + + /* draw final quad */ + if (setup->quad.mask) { + clip_emit_quad(setup); + } +} + + +static void +point_persp_coeff(struct setup_stage *setup, + const struct vertex_header *vert, + struct tgsi_interp_coef *coef, + uint vertSlot, uint i) +{ + assert(i <= 3); + coef->dadx[i] = 0.0F; + coef->dady[i] = 0.0F; + coef->a0[i] = vert->data[vertSlot][i] * vert->data[0][3]; +} + + +/** + * Do setup for point rasterization, then render the point. + * Round or square points... + * XXX could optimize a lot for 1-pixel points. + */ +static void +setup_point(struct draw_stage *stage, struct prim_header *prim) +{ + struct setup_stage *setup = setup_stage( stage ); + struct softpipe_context *softpipe = setup->softpipe; + const struct pipe_shader_state *fs = &softpipe->fs->shader; + const struct vertex_header *v0 = prim->v[0]; + const int sizeAttr = setup->softpipe->psize_slot; + const float size + = sizeAttr > 0 ? v0->data[sizeAttr][0] + : setup->softpipe->rasterizer->point_size; + const float halfSize = 0.5F * size; + const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth; + const float x = v0->data[0][0]; /* Note: data[0] is always position */ + const float y = v0->data[0][1]; + const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe); + uint fragSlot; + + /* For points, all interpolants are constant-valued. + * However, for point sprites, we'll need to setup texcoords appropriately. + * XXX: which coefficients are the texcoords??? + * We may do point sprites as textured quads... + * + * KW: We don't know which coefficients are texcoords - ultimately + * the choice of what interpolation mode to use for each attribute + * should be determined by the fragment program, using + * per-attribute declaration statements that include interpolation + * mode as a parameter. So either the fragment program will have + * to be adjusted for pointsprite vs normal point behaviour, or + * otherwise a special interpolation mode will have to be defined + * which matches the required behaviour for point sprites. But - + * the latter is not a feature of normal hardware, and as such + * probably should be ruled out on that basis. + */ + setup->vprovoke = prim->v[0]; + + /* setup Z, W */ + const_coeff(setup, &setup->posCoef, 0, 2); + const_coeff(setup, &setup->posCoef, 0, 3); + + for (fragSlot = 0; fragSlot < fs->num_inputs; fragSlot++) { + const uint vertSlot = vinfo->src_index[fragSlot]; + uint j; + + switch (vinfo->interp_mode[fragSlot]) { + case INTERP_CONSTANT: + /* fall-through */ + case INTERP_LINEAR: + for (j = 0; j < NUM_CHANNELS; j++) + const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_PERSPECTIVE: + for (j = 0; j < NUM_CHANNELS; j++) + point_persp_coeff(setup, setup->vprovoke, + &setup->coef[fragSlot], vertSlot, j); + break; + case INTERP_POS: + assert(fragSlot == 0); + assert(0); /* XXX fix this: */ + setup_fragcoord_coeff(setup); + break; + default: + assert(0); + } + + if (fs->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) { + /* FOG.y = front/back facing XXX fix this */ + setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing; + setup->coef[fragSlot].dadx[1] = 0.0; + setup->coef[fragSlot].dady[1] = 0.0; + } + } + + setup->quad.prim = PRIM_POINT; + + if (halfSize <= 0.5 && !round) { + /* special case for 1-pixel points */ + const int ix = ((int) x) & 1; + const int iy = ((int) y) & 1; + setup->quad.x0 = (int) x - ix; + setup->quad.y0 = (int) y - iy; + setup->quad.mask = (1 << ix) << (2 * iy); + clip_emit_quad(setup); + } + else { + if (round) { + /* rounded points */ + const int ixmin = block((int) (x - halfSize)); + const int ixmax = block((int) (x + halfSize)); + const int iymin = block((int) (y - halfSize)); + const int iymax = block((int) (y + halfSize)); + const float rmin = halfSize - 0.7071F; /* 0.7071 = sqrt(2)/2 */ + const float rmax = halfSize + 0.7071F; + const float rmin2 = MAX2(0.0F, rmin * rmin); + const float rmax2 = rmax * rmax; + const float cscale = 1.0F / (rmax2 - rmin2); + int ix, iy; + + for (iy = iymin; iy <= iymax; iy += 2) { + for (ix = ixmin; ix <= ixmax; ix += 2) { + float dx, dy, dist2, cover; + + setup->quad.mask = 0x0; + + dx = (ix + 0.5f) - x; + dy = (iy + 0.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f); + setup->quad.mask |= MASK_TOP_LEFT; + } + + dx = (ix + 1.5f) - x; + dy = (iy + 0.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f); + setup->quad.mask |= MASK_TOP_RIGHT; + } + + dx = (ix + 0.5f) - x; + dy = (iy + 1.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f); + setup->quad.mask |= MASK_BOTTOM_LEFT; + } + + dx = (ix + 1.5f) - x; + dy = (iy + 1.5f) - y; + dist2 = dx * dx + dy * dy; + if (dist2 <= rmax2) { + cover = 1.0F - (dist2 - rmin2) * cscale; + setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f); + setup->quad.mask |= MASK_BOTTOM_RIGHT; + } + + if (setup->quad.mask) { + setup->quad.x0 = ix; + setup->quad.y0 = iy; + clip_emit_quad(setup); + } + } + } + } + else { + /* square points */ + const int xmin = (int) (x + 0.75 - halfSize); + const int ymin = (int) (y + 0.25 - halfSize); + const int xmax = xmin + (int) size; + const int ymax = ymin + (int) size; + /* XXX could apply scissor to xmin,ymin,xmax,ymax now */ + const int ixmin = block(xmin); + const int ixmax = block(xmax - 1); + const int iymin = block(ymin); + const int iymax = block(ymax - 1); + int ix, iy; + + /* + debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax); + */ + for (iy = iymin; iy <= iymax; iy += 2) { + uint rowMask = 0xf; + if (iy < ymin) { + /* above the top edge */ + rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); + } + if (iy + 1 >= ymax) { + /* below the bottom edge */ + rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); + } + + for (ix = ixmin; ix <= ixmax; ix += 2) { + uint mask = rowMask; + + if (ix < xmin) { + /* fragment is past left edge of point, turn off left bits */ + mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); + } + if (ix + 1 >= xmax) { + /* past the right edge */ + mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); + } + + setup->quad.mask = mask; + setup->quad.x0 = ix; + setup->quad.y0 = iy; + clip_emit_quad(setup); + } + } + } + } +} + + + +static void setup_begin( struct draw_stage *stage ) +{ + struct setup_stage *setup = setup_stage(stage); + struct softpipe_context *sp = setup->softpipe; + const struct pipe_shader_state *fs = &setup->softpipe->fs->shader; + + setup->quad.nr_attrs = fs->num_inputs; + + sp->quad.first->begin(sp->quad.first); + + stage->point = setup_point; + stage->line = setup_line; + stage->tri = setup_tri; +} + + +static void setup_first_point( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->point( stage, header ); +} + +static void setup_first_line( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->line( stage, header ); +} + + +static void setup_first_tri( struct draw_stage *stage, + struct prim_header *header ) +{ + setup_begin(stage); + stage->tri( stage, header ); +} + + + +static void setup_flush( struct draw_stage *stage, + unsigned flags ) +{ + stage->point = setup_first_point; + stage->line = setup_first_line; + stage->tri = setup_first_tri; +} + + +static void reset_stipple_counter( struct draw_stage *stage ) +{ +} + + +static void render_destroy( struct draw_stage *stage ) +{ + FREE( stage ); +} + + +/** + * Create a new primitive setup/render stage. + */ +struct draw_stage *sp_draw_render_stage( struct softpipe_context *softpipe ) +{ + struct setup_stage *setup = CALLOC_STRUCT(setup_stage); + + setup->softpipe = softpipe; + setup->stage.draw = softpipe->draw; + setup->stage.point = setup_first_point; + setup->stage.line = setup_first_line; + setup->stage.tri = setup_first_tri; + setup->stage.flush = setup_flush; + setup->stage.reset_stipple_counter = reset_stipple_counter; + setup->stage.destroy = render_destroy; + + setup->quad.coef = setup->coef; + setup->quad.posCoef = &setup->posCoef; + + return &setup->stage; +} diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.h b/src/gallium/drivers/softpipe/sp_prim_setup.h new file mode 100644 index 0000000000..f3e8a79dd9 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_prim_setup.h @@ -0,0 +1,79 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#ifndef SP_PRIM_SETUP_H +#define SP_PRIM_SETUP_H + + +/** + * vbuf is a special stage to gather the stream of triangles, lines, points + * together and reconstruct vertex buffers for hardware upload. + * + * First attempt, work in progress. + * + * TODO: + * - separate out vertex buffer building and primitive emit, ie >1 draw per vb. + * - tell vbuf stage how to build hw vertices directly + * - pass vbuf stage a buffer pointer for direct emit to agp/vram. + * + * + * + * Vertices are just an array of floats, with all the attributes + * packed. We currently assume a layout like: + * + * attr[0][0..3] - window position + * attr[1..n][0..3] - remaining attributes. + * + * Attributes are assumed to be 4 floats wide but are packed so that + * all the enabled attributes run contiguously. + */ + + +struct draw_stage; +struct softpipe_context; + + +typedef void (*vbuf_draw_func)( struct pipe_context *pipe, + unsigned prim, + const ushort *elements, + unsigned nr_elements, + const void *vertex_buffer, + unsigned nr_vertices ); + + +extern struct draw_stage * +sp_draw_render_stage( struct softpipe_context *softpipe ); + + +extern struct draw_stage * +sp_draw_vbuf_stage( struct draw_context *draw_context, + struct pipe_context *pipe, + vbuf_draw_func draw ); + + +#endif /* SP_PRIM_SETUP_H */ diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c new file mode 100644 index 0000000000..7f71fdb6a9 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c @@ -0,0 +1,221 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Post-transform vertex buffering. This is an optional part of the + * softpipe rendering pipeline. + * Probably not desired in general, but useful for testing/debuggin. + * Enabled/Disabled with SP_VBUF env var. + * + * Authors + * Brian Paul + */ + + +#include "sp_context.h" +#include "sp_state.h" +#include "sp_prim_vbuf.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_private.h" +#include "pipe/draw/draw_vbuf.h" + + +#define SP_MAX_VBUF_INDEXES 1024 +#define SP_MAX_VBUF_SIZE 4096 + + +/** + * Subclass of vbuf_render. + */ +struct softpipe_vbuf_render +{ + struct vbuf_render base; + struct softpipe_context *softpipe; + uint prim; + uint vertex_size; + void *vertex_buffer; +}; + + +/** cast wrapper */ +static struct softpipe_vbuf_render * +softpipe_vbuf_render(struct vbuf_render *vbr) +{ + return (struct softpipe_vbuf_render *) vbr; +} + + + +static const struct vertex_info * +sp_vbuf_get_vertex_info(struct vbuf_render *vbr) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + return softpipe_get_vbuf_vertex_info(cvbr->softpipe); +} + + +static void * +sp_vbuf_allocate_vertices(struct vbuf_render *vbr, + ushort vertex_size, ushort nr_vertices) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + assert(!cvbr->vertex_buffer); + cvbr->vertex_buffer = align_malloc(vertex_size * nr_vertices, 16); + cvbr->vertex_size = vertex_size; + return cvbr->vertex_buffer; +} + + +static void +sp_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices, + unsigned vertex_size, unsigned vertices_used) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + align_free(vertices); + assert(vertices == cvbr->vertex_buffer); + cvbr->vertex_buffer = NULL; +} + + +static void +sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + cvbr->prim = prim; +} + + +/** + * Recalculate prim's determinant. + * XXX is this needed? + */ +static void +calc_det(struct prim_header *header) +{ + /* Window coords: */ + const float *v0 = header->v[0]->data[0]; + const float *v1 = header->v[1]->data[0]; + const float *v2 = header->v[2]->data[0]; + + /* edge vectors e = v0 - v2, f = v1 - v2 */ + const float ex = v0[0] - v2[0]; + const float ey = v0[1] - v2[1]; + const float fx = v1[0] - v2[0]; + const float fy = v1[1] - v2[1]; + + /* det = cross(e,f).z */ + header->det = ex * fy - ey * fx; +} + + +static void +sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr_indices) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + struct softpipe_context *softpipe = cvbr->softpipe; + struct draw_stage *setup = softpipe->setup; + struct prim_header prim; + unsigned vertex_size = softpipe->vertex_info_vbuf.size * sizeof(float); + unsigned i, j; + void *vertex_buffer = cvbr->vertex_buffer; + + prim.det = 0; + prim.reset_line_stipple = 0; + prim.edgeflags = 0; + prim.pad = 0; + + switch (cvbr->prim) { + case PIPE_PRIM_TRIANGLES: + for (i = 0; i < nr_indices; i += 3) { + for (j = 0; j < 3; j++) + prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + + indices[i+j] * vertex_size); + + calc_det(&prim); + setup->tri( setup, &prim ); + } + break; + + case PIPE_PRIM_LINES: + for (i = 0; i < nr_indices; i += 2) { + for (j = 0; j < 2; j++) + prim.v[j] = (struct vertex_header *)((char *)vertex_buffer + + indices[i+j] * vertex_size); + + setup->line( setup, &prim ); + } + break; + + case PIPE_PRIM_POINTS: + for (i = 0; i < nr_indices; i++) { + prim.v[0] = (struct vertex_header *)((char *)vertex_buffer + + indices[i] * vertex_size); + setup->point( setup, &prim ); + } + break; + } + + setup->flush( setup, 0 ); +} + + +static void +sp_vbuf_destroy(struct vbuf_render *vbr) +{ + struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr); + cvbr->softpipe->vbuf_render = NULL; + FREE(cvbr); +} + + +/** + * Initialize the post-transform vertex buffer information for the given + * context. + */ +void +sp_init_vbuf(struct softpipe_context *sp) +{ + assert(sp->draw); + + sp->vbuf_render = CALLOC_STRUCT(softpipe_vbuf_render); + + sp->vbuf_render->base.max_indices = SP_MAX_VBUF_INDEXES; + sp->vbuf_render->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE; + + sp->vbuf_render->base.get_vertex_info = sp_vbuf_get_vertex_info; + sp->vbuf_render->base.allocate_vertices = sp_vbuf_allocate_vertices; + sp->vbuf_render->base.set_primitive = sp_vbuf_set_primitive; + sp->vbuf_render->base.draw = sp_vbuf_draw; + sp->vbuf_render->base.release_vertices = sp_vbuf_release_vertices; + sp->vbuf_render->base.destroy = sp_vbuf_destroy; + + sp->vbuf_render->softpipe = sp; + + sp->vbuf = draw_vbuf_stage(sp->draw, &sp->vbuf_render->base); + + draw_set_rasterize_stage(sp->draw, sp->vbuf); +} diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.h b/src/gallium/drivers/softpipe/sp_prim_vbuf.h new file mode 100644 index 0000000000..1de9cc2a89 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.h @@ -0,0 +1,38 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SP_VBUF_H +#define SP_VBUF_H + + +struct softpipe_context; + +extern void +sp_init_vbuf(struct softpipe_context *softpipe); + + +#endif /* SP_VBUF_H */ diff --git a/src/gallium/drivers/softpipe/sp_quad.c b/src/gallium/drivers/softpipe/sp_quad.c new file mode 100644 index 0000000000..6bd468a51c --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad.c @@ -0,0 +1,118 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +#include "sp_context.h" +#include "sp_state.h" +#include "pipe/p_shader_tokens.h" + +static void +sp_push_quad_first( + struct softpipe_context *sp, + struct quad_stage *quad ) +{ + quad->next = sp->quad.first; + sp->quad.first = quad; +} + +static void +sp_build_depth_stencil( + struct softpipe_context *sp ) +{ + if (sp->depth_stencil->stencil[0].enabled || + sp->depth_stencil->stencil[1].enabled) { + sp_push_quad_first( sp, sp->quad.stencil_test ); + } + else if (sp->depth_stencil->depth.enabled && + sp->framebuffer.zsbuf) { + sp_push_quad_first( sp, sp->quad.depth_test ); + } +} + +void +sp_build_quad_pipeline(struct softpipe_context *sp) +{ + boolean early_depth_test = + sp->depth_stencil->depth.enabled && + sp->framebuffer.zsbuf && + !sp->depth_stencil->alpha.enabled && + sp->fs->shader.output_semantic_name[0] != TGSI_SEMANTIC_POSITION; + + /* build up the pipeline in reverse order... */ + + sp->quad.first = sp->quad.output; + + if (sp->blend->colormask != 0xf) { + sp_push_quad_first( sp, sp->quad.colormask ); + } + + if (sp->blend->blend_enable || + sp->blend->logicop_enable) { + sp_push_quad_first( sp, sp->quad.blend ); + } + + if (sp->framebuffer.num_cbufs == 1) { + /* the usual case: write to exactly one colorbuf */ + sp->current_cbuf = 0; + } + else { + /* insert bufloop stage */ + sp_push_quad_first( sp, sp->quad.bufloop ); + } + + if (sp->depth_stencil->depth.occlusion_count) { + sp_push_quad_first( sp, sp->quad.occlusion ); + } + + if (sp->rasterizer->poly_smooth || + sp->rasterizer->line_smooth || + sp->rasterizer->point_smooth) { + sp_push_quad_first( sp, sp->quad.coverage ); + } + + if (!early_depth_test) { + sp_build_depth_stencil( sp ); + } + + if (sp->depth_stencil->alpha.enabled) { + sp_push_quad_first( sp, sp->quad.alpha_test ); + } + + /* XXX always enable shader? */ + if (1) { + sp_push_quad_first( sp, sp->quad.shade ); + } + + if (early_depth_test) { + sp_build_depth_stencil( sp ); + sp_push_quad_first( sp, sp->quad.earlyz ); + } + + if (sp->rasterizer->poly_stipple_enable) { + sp_push_quad_first( sp, sp->quad.polygon_stipple ); + } +} diff --git a/src/gallium/drivers/softpipe/sp_quad.h b/src/gallium/drivers/softpipe/sp_quad.h new file mode 100644 index 0000000000..f1e0281764 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad.h @@ -0,0 +1,70 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_QUAD_H +#define SP_QUAD_H + + +struct softpipe_context; +struct quad_header; + + +struct quad_stage { + struct softpipe_context *softpipe; + + struct quad_stage *next; + + void (*begin)(struct quad_stage *qs); + + /** the stage action */ + void (*run)(struct quad_stage *qs, struct quad_header *quad); + + void (*destroy)(struct quad_stage *qs); +}; + + +struct quad_stage *sp_quad_polygon_stipple_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_earlyz_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_alpha_test_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe ); +struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe ); + +void sp_build_quad_pipeline(struct softpipe_context *sp); + +void sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad); + +#endif /* SP_QUAD_H */ diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c new file mode 100644 index 0000000000..4ffeac35e1 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c @@ -0,0 +1,108 @@ + +/** + * quad alpha test + */ + +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_quad.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" + + +static void +alpha_test_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + const float ref = softpipe->depth_stencil->alpha.ref; + unsigned passMask = 0x0, j; + const float *aaaa = quad->outputs.color[3]; + + switch (softpipe->depth_stencil->alpha.func) { + case PIPE_FUNC_NEVER: + quad->mask = 0x0; + break; + case PIPE_FUNC_LESS: + /* + * If mask were an array [4] we could do this SIMD-style: + * passMask = (quad->outputs.color[3] <= vec4(ref)); + */ + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] < ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_EQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] == ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_LEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] <= ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GREATER: + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] > ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_NOTEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] != ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (aaaa[j] >= ref) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_ALWAYS: + passMask = MASK_ALL; + break; + default: + abort(); + } + + quad->mask &= passMask; + + if (quad->mask) + qs->next->run(qs->next, quad); +} + + +static void alpha_test_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void alpha_test_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage * +sp_quad_alpha_test_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = alpha_test_begin; + stage->run = alpha_test_quad; + stage->destroy = alpha_test_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c new file mode 100644 index 0000000000..17f3ecd0b8 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_blend.c @@ -0,0 +1,749 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * quad blending + * \author Brian Paul + */ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_tile_cache.h" +#include "sp_quad.h" + + +#define VEC4_COPY(DST, SRC) \ +do { \ + DST[0] = SRC[0]; \ + DST[1] = SRC[1]; \ + DST[2] = SRC[2]; \ + DST[3] = SRC[3]; \ +} while(0) + +#define VEC4_SCALAR(DST, SRC) \ +do { \ + DST[0] = SRC; \ + DST[1] = SRC; \ + DST[2] = SRC; \ + DST[3] = SRC; \ +} while(0) + +#define VEC4_ADD(R, A, B) \ +do { \ + R[0] = A[0] + B[0]; \ + R[1] = A[1] + B[1]; \ + R[2] = A[2] + B[2]; \ + R[3] = A[3] + B[3]; \ +} while (0) + +#define VEC4_SUB(R, A, B) \ +do { \ + R[0] = A[0] - B[0]; \ + R[1] = A[1] - B[1]; \ + R[2] = A[2] - B[2]; \ + R[3] = A[3] - B[3]; \ +} while (0) + +#define VEC4_MUL(R, A, B) \ +do { \ + R[0] = A[0] * B[0]; \ + R[1] = A[1] * B[1]; \ + R[2] = A[2] * B[2]; \ + R[3] = A[3] * B[3]; \ +} while (0) + +#define VEC4_MIN(R, A, B) \ +do { \ + R[0] = (A[0] < B[0]) ? A[0] : B[0]; \ + R[1] = (A[1] < B[1]) ? A[1] : B[1]; \ + R[2] = (A[2] < B[2]) ? A[2] : B[2]; \ + R[3] = (A[3] < B[3]) ? A[3] : B[3]; \ +} while (0) + +#define VEC4_MAX(R, A, B) \ +do { \ + R[0] = (A[0] > B[0]) ? A[0] : B[0]; \ + R[1] = (A[1] > B[1]) ? A[1] : B[1]; \ + R[2] = (A[2] > B[2]) ? A[2] : B[2]; \ + R[3] = (A[3] > B[3]) ? A[3] : B[3]; \ +} while (0) + + + +static void +logicop_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + float dest[4][QUAD_SIZE]; + ubyte src[4][4], dst[4][4], res[4][4]; + uint *src4 = (uint *) src; + uint *dst4 = (uint *) dst; + uint *res4 = (uint *) res; + struct softpipe_cached_tile * + tile = sp_get_cached_tile(softpipe, + softpipe->cbuf_cache[softpipe->current_cbuf], + quad->x0, quad->y0); + float (*quadColor)[4] = quad->outputs.color; + uint i, j; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } + + /* convert to ubyte */ + for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */ + UNCLAMPED_FLOAT_TO_UBYTE(dst[j][0], dest[j][0]); /* P0 */ + UNCLAMPED_FLOAT_TO_UBYTE(dst[j][1], dest[j][1]); /* P1 */ + UNCLAMPED_FLOAT_TO_UBYTE(dst[j][2], dest[j][2]); /* P2 */ + UNCLAMPED_FLOAT_TO_UBYTE(dst[j][3], dest[j][3]); /* P3 */ + + UNCLAMPED_FLOAT_TO_UBYTE(src[j][0], quadColor[j][0]); /* P0 */ + UNCLAMPED_FLOAT_TO_UBYTE(src[j][1], quadColor[j][1]); /* P1 */ + UNCLAMPED_FLOAT_TO_UBYTE(src[j][2], quadColor[j][2]); /* P2 */ + UNCLAMPED_FLOAT_TO_UBYTE(src[j][3], quadColor[j][3]); /* P3 */ + } + + switch (softpipe->blend->logicop_func) { + case PIPE_LOGICOP_CLEAR: + for (j = 0; j < 4; j++) + res4[j] = 0; + break; + case PIPE_LOGICOP_NOR: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] | dst4[j]); + break; + case PIPE_LOGICOP_AND_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_COPY_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j]; + break; + case PIPE_LOGICOP_AND_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & ~dst4[j]; + break; + case PIPE_LOGICOP_INVERT: + for (j = 0; j < 4; j++) + res4[j] = ~dst4[j]; + break; + case PIPE_LOGICOP_XOR: + for (j = 0; j < 4; j++) + res4[j] = dst4[j] ^ src4[j]; + break; + case PIPE_LOGICOP_NAND: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] & dst4[j]); + break; + case PIPE_LOGICOP_AND: + for (j = 0; j < 4; j++) + res4[j] = src4[j] & dst4[j]; + break; + case PIPE_LOGICOP_EQUIV: + for (j = 0; j < 4; j++) + res4[j] = ~(src4[j] ^ dst4[j]); + break; + case PIPE_LOGICOP_NOOP: + for (j = 0; j < 4; j++) + res4[j] = dst4[j]; + break; + case PIPE_LOGICOP_OR_INVERTED: + for (j = 0; j < 4; j++) + res4[j] = ~src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_COPY: + for (j = 0; j < 4; j++) + res4[j] = src4[j]; + break; + case PIPE_LOGICOP_OR_REVERSE: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | ~dst4[j]; + break; + case PIPE_LOGICOP_OR: + for (j = 0; j < 4; j++) + res4[j] = src4[j] | dst4[j]; + break; + case PIPE_LOGICOP_SET: + for (j = 0; j < 4; j++) + res4[j] = ~0; + break; + default: + assert(0); + } + + for (j = 0; j < 4; j++) { + quadColor[j][0] = UBYTE_TO_FLOAT(res[j][0]); + quadColor[j][1] = UBYTE_TO_FLOAT(res[j][1]); + quadColor[j][2] = UBYTE_TO_FLOAT(res[j][2]); + quadColor[j][3] = UBYTE_TO_FLOAT(res[j][3]); + } + + /* pass quad to next stage */ + qs->next->run(qs->next, quad); +} + + + + +static void +blend_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + static const float zero[4] = { 0, 0, 0, 0 }; + static const float one[4] = { 1, 1, 1, 1 }; + float source[4][QUAD_SIZE], dest[4][QUAD_SIZE]; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe, + softpipe->cbuf_cache[softpipe->current_cbuf], + quad->x0, quad->y0); + float (*quadColor)[4] = quad->outputs.color; + uint i, j; + + if (softpipe->blend->logicop_enable) { + logicop_quad(qs, quad); + return; + } + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } + + /* + * Compute src/first term RGB + */ + switch (softpipe->blend->rgb_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[0], quadColor[0]); /* R */ + VEC4_COPY(source[1], quadColor[1]); /* G */ + VEC4_COPY(source[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const float *alpha = quadColor[3]; + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */ + VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */ + VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + { + const float *alpha = dest[3]; + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + { + const float *alpha = quadColor[3]; + float diff[4]; + VEC4_SUB(diff, one, dest[3]); + VEC4_MIN(source[0], alpha, diff); /* R */ + VEC4_MIN(source[1], alpha, diff); /* G */ + VEC4_MIN(source[2], alpha, diff); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], comp); /* R */ + VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], comp); /* G */ + VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float alpha[4]; + VEC4_SCALAR(alpha, softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quadColor[0], alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[0], zero); /* R */ + VEC4_COPY(source[1], zero); /* G */ + VEC4_COPY(source[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, quadColor[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, dest[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[0]); /* R */ + VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, dest[1]); /* G */ + VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, dest[2]); /* B */ + VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + { + float inv_comp[4]; + /* R */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); + VEC4_MUL(source[0], quadColor[0], inv_comp); + /* G */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); + VEC4_MUL(source[1], quadColor[1], inv_comp); + /* B */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); + VEC4_MUL(source[2], quadColor[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_alpha[4]; + VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + assert(0); /* to do */ + break; + default: + abort(); + } + + /* + * Compute src/first term A + */ + switch (softpipe->blend->alpha_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[3], quadColor[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const float *alpha = quadColor[3]; + VEC4_MUL(source[3], quadColor[3], alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + { + const float *alpha = quadColor[3]; + float diff[4]; + VEC4_SUB(diff, one, dest[3]); + VEC4_MIN(source[3], alpha, diff); /* A */ + } + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(source[3], quadColor[3], comp); /* A */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, quadColor[3]); + VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_alpha[4]; + VEC4_SUB(inv_alpha, one, dest[3]); + VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + /* A */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(source[3], quadColor[3], inv_comp); + } + break; + default: + abort(); + } + + + /* + * Compute dest/second term RGB + */ + switch (softpipe->blend->rgb_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */ + VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */ + VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */ + VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */ + VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(dest[0], dest[0], dest[0]); /* R */ + VEC4_MUL(dest[1], dest[1], dest[1]); /* G */ + VEC4_MUL(dest[2], dest[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ + VEC4_MUL(dest[0], dest[0], comp); /* R */ + VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ + VEC4_MUL(dest[1], dest[1], comp); /* G */ + VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ + VEC4_MUL(dest[2], dest[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(dest[0], dest[0], comp); /* R */ + VEC4_MUL(dest[1], dest[1], comp); /* G */ + VEC4_MUL(dest[2], dest[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[0], zero); /* R */ + VEC4_COPY(dest[1], zero); /* G */ + VEC4_COPY(dest[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + case PIPE_BLENDFACTOR_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, quadColor[0]); /* R */ + VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ + VEC4_SUB(inv_comp, one, quadColor[1]); /* G */ + VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ + VEC4_SUB(inv_comp, one, quadColor[2]); /* B */ + VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quadColor[3]); + VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ + VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ + VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, quadColor[3]); /* A */ + VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */ + VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */ + VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[0]); /* R */ + VEC4_MUL(dest[0], dest[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, dest[1]); /* G */ + VEC4_MUL(dest[1], dest[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, dest[2]); /* B */ + VEC4_MUL(dest[2], dest[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + { + float inv_comp[4]; + /* R */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]); + VEC4_MUL(dest[0], dest[0], inv_comp); + /* G */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]); + VEC4_MUL(dest[1], dest[1], inv_comp); + /* B */ + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]); + VEC4_MUL(dest[2], dest[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(dest[0], dest[0], inv_comp); + VEC4_MUL(dest[1], dest[1], inv_comp); + VEC4_MUL(dest[2], dest[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + /* XXX what are these? */ + assert(0); + break; + default: + assert(0); + } + + /* + * Compute dest/second term A + */ + switch (softpipe->blend->alpha_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_SRC_ALPHA: + VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */ + break; + case PIPE_BLENDFACTOR_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_DST_ALPHA: + VEC4_MUL(dest[3], dest[3], dest[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* illegal */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + float comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */ + VEC4_MUL(dest[3], dest[3], comp); /* A */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[3], zero); /* A */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + float one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quadColor[3]); + VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + float inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[3]); /* A */ + VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + /* fall-through */ + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + float inv_comp[4]; + VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]); + VEC4_MUL(dest[3], dest[3], inv_comp); + } + break; + default: + assert(0); + } + + /* + * Combine RGB terms + */ + switch (softpipe->blend->rgb_func) { + case PIPE_BLEND_ADD: + VEC4_ADD(quadColor[0], source[0], dest[0]); /* R */ + VEC4_ADD(quadColor[1], source[1], dest[1]); /* G */ + VEC4_ADD(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB(quadColor[0], source[0], dest[0]); /* R */ + VEC4_SUB(quadColor[1], source[1], dest[1]); /* G */ + VEC4_SUB(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB(quadColor[0], dest[0], source[0]); /* R */ + VEC4_SUB(quadColor[1], dest[1], source[1]); /* G */ + VEC4_SUB(quadColor[2], dest[2], source[2]); /* B */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */ + VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */ + VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */ + break; + default: + assert(0); + } + + /* + * Combine A terms + */ + switch (softpipe->blend->alpha_func) { + case PIPE_BLEND_ADD: + VEC4_ADD(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB(quadColor[3], dest[3], source[3]); /* A */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */ + break; + default: + abort(); + } + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quad); +} + + +static void blend_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void blend_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = blend_begin; + stage->run = blend_quad; + stage->destroy = blend_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_bufloop.c b/src/gallium/drivers/softpipe/sp_quad_bufloop.c new file mode 100644 index 0000000000..2ae4e22a7d --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_bufloop.c @@ -0,0 +1,72 @@ + +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_quad.h" + + +/** + * Loop over colorbuffers, passing quad to next stage each time. + */ +static void +cbuf_loop_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + float tmp[4][QUAD_SIZE]; + unsigned i; + + assert(sizeof(quad->outputs.color) == sizeof(tmp)); + assert(softpipe->framebuffer.num_cbufs <= PIPE_MAX_COLOR_BUFS); + + /* make copy of original colors since they can get modified + * by blending and masking. + * XXX we won't have to do this if the fragment program actually emits + * N separate colors and we're drawing to N color buffers (MRT). + * But if we emitted one color and glDrawBuffer(GL_FRONT_AND_BACK) is + * in effect, we need to save/restore colors like this. + */ + memcpy(tmp, quad->outputs.color, sizeof(tmp)); + + for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) { + /* set current cbuffer */ + softpipe->current_cbuf = i; + + /* pass blended quad to next stage */ + qs->next->run(qs->next, quad); + + /* restore quad's colors for next buffer */ + memcpy(quad->outputs.color, tmp, sizeof(tmp)); + } +} + + +static void cbuf_loop_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void cbuf_loop_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +/** + * Create the colorbuffer loop stage. + * This is used to implement multiple render targets and GL_FRONT_AND_BACK + * rendering. + */ +struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = cbuf_loop_begin; + stage->run = cbuf_loop_quad; + stage->destroy = cbuf_loop_destroy; + + return stage; +} + diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c new file mode 100644 index 0000000000..1f09d900ca --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_colormask.c @@ -0,0 +1,110 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief quad colormask stage + * \author Brian Paul + */ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_quad.h" +#include "sp_tile_cache.h" + + + +/** + * XXX colormask could be rolled into blending... + */ +static void +colormask_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + float dest[4][QUAD_SIZE]; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe, + softpipe->cbuf_cache[softpipe->current_cbuf], + quad->x0, quad->y0); + float (*quadColor)[4] = quad->outputs.color; + uint i, j; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1); + int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1); + for (i = 0; i < 4; i++) { + dest[i][j] = tile->data.color[y][x][i]; + } + } + + /* R */ + if (!(softpipe->blend->colormask & PIPE_MASK_R)) + COPY_4V(quadColor[0], dest[0]); + + /* G */ + if (!(softpipe->blend->colormask & PIPE_MASK_G)) + COPY_4V(quadColor[1], dest[1]); + + /* B */ + if (!(softpipe->blend->colormask & PIPE_MASK_B)) + COPY_4V(quadColor[2], dest[2]); + + /* A */ + if (!(softpipe->blend->colormask & PIPE_MASK_A)) + COPY_4V(quadColor[3], dest[3]); + + /* pass quad to next stage */ + qs->next->run(qs->next, quad); +} + + +static void colormask_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void colormask_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = colormask_begin; + stage->run = colormask_quad; + stage->destroy = colormask_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c new file mode 100644 index 0000000000..b3d3fae22f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c @@ -0,0 +1,88 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * \brief Apply AA coverage to quad alpha valus + * \author Brian Paul + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_quad.h" + + +/** + * Multiply quad's alpha values by the fragment coverage. + */ +static void +coverage_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + + if ((softpipe->rasterizer->poly_smooth && quad->prim == PRIM_TRI) || + (softpipe->rasterizer->line_smooth && quad->prim == PRIM_LINE) || + (softpipe->rasterizer->point_smooth && quad->prim == PRIM_POINT)) { + float (*quadColor)[4] = quad->outputs.color; + unsigned j; + for (j = 0; j < QUAD_SIZE; j++) { + assert(quad->coverage[j] >= 0.0); + assert(quad->coverage[j] <= 1.0); + quadColor[3][j] *= quad->coverage[j]; + } + } + + qs->next->run(qs->next, quad); +} + + +static void coverage_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void coverage_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = coverage_begin; + stage->run = coverage_quad; + stage->destroy = coverage_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c new file mode 100644 index 0000000000..a9a0754f27 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c @@ -0,0 +1,276 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief Quad depth testing + */ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_quad.h" +#include "sp_tile_cache.h" + + +/** + * Do depth testing for a quad. + * Not static since it's used by the stencil code. + */ + +/* + * To increase efficiency, we should probably have multiple versions + * of this function that are specifically for Z16, Z32 and FP Z buffers. + * Try to effectively do that with codegen... + */ + +void +sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + struct pipe_surface *ps = softpipe->framebuffer.zsbuf; + const enum pipe_format format = ps->format; + unsigned bzzzz[QUAD_SIZE]; /**< Z values fetched from depth buffer */ + unsigned qzzzz[QUAD_SIZE]; /**< Z values from the quad */ + unsigned zmask = 0; + unsigned j; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0); + + assert(ps); /* shouldn't get here if there's no zbuffer */ + + /* + * Convert quad's float depth values to int depth values (qzzzz). + * If the Z buffer stores integer values, we _have_ to do the depth + * compares with integers (not floats). Otherwise, the float->int->float + * conversion of Z values (which isn't an identity function) will cause + * Z-fighting errors. + * + * Also, get the zbuffer values (bzzzz) from the cached tile. + */ + switch (format) { + case PIPE_FORMAT_Z16_UNORM: + { + float scale = 65535.0; + + for (j = 0; j < QUAD_SIZE; j++) { + qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale); + } + + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + bzzzz[j] = tile->data.depth16[y][x]; + } + } + break; + case PIPE_FORMAT_Z32_UNORM: + { + double scale = (double) (uint) ~0UL; + + for (j = 0; j < QUAD_SIZE; j++) { + qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale); + } + + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + bzzzz[j] = tile->data.depth32[y][x]; + } + } + break; + case PIPE_FORMAT_S8Z24_UNORM: + { + float scale = (float) ((1 << 24) - 1); + + for (j = 0; j < QUAD_SIZE; j++) { + qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale); + } + + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + bzzzz[j] = tile->data.depth32[y][x] & 0xffffff; + } + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + { + float scale = (float) ((1 << 24) - 1); + + for (j = 0; j < QUAD_SIZE; j++) { + qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale); + } + + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + bzzzz[j] = tile->data.depth32[y][x] >> 8; + } + } + break; + default: + assert(0); + } + + switch (softpipe->depth_stencil->depth.func) { + case PIPE_FUNC_NEVER: + /* zmask = 0 */ + break; + case PIPE_FUNC_LESS: + /* Note this is pretty much a single sse or cell instruction. + * Like this: quad->mask &= (quad->outputs.depth < zzzz); + */ + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] < bzzzz[j]) + zmask |= 1 << j; + } + break; + case PIPE_FUNC_EQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] == bzzzz[j]) + zmask |= 1 << j; + } + break; + case PIPE_FUNC_LEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] <= bzzzz[j]) + zmask |= (1 << j); + } + break; + case PIPE_FUNC_GREATER: + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] > bzzzz[j]) + zmask |= (1 << j); + } + break; + case PIPE_FUNC_NOTEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] != bzzzz[j]) + zmask |= (1 << j); + } + break; + case PIPE_FUNC_GEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (qzzzz[j] >= bzzzz[j]) + zmask |= (1 << j); + } + break; + case PIPE_FUNC_ALWAYS: + zmask = MASK_ALL; + break; + default: + abort(); + } + + quad->mask &= zmask; + + if (softpipe->depth_stencil->depth.writemask) { + + /* This is also efficient with sse / spe instructions: + */ + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->mask & (1 << j)) { + bzzzz[j] = qzzzz[j]; + } + } + + /* put updated Z values back into cached tile */ + switch (format) { + case PIPE_FORMAT_Z16_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + tile->data.depth16[y][x] = (ushort) bzzzz[j]; + } + break; + case PIPE_FORMAT_Z32_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + tile->data.depth32[y][x] = bzzzz[j]; + } + break; + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + uint s8z24 = tile->data.depth32[y][x]; + s8z24 = (s8z24 & 0xff000000) | bzzzz[j]; + tile->data.depth32[y][x] = s8z24; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + uint z24s8 = tile->data.depth32[y][x]; + z24s8 = (z24s8 & 0xff) | (bzzzz[j] << 24); + tile->data.depth32[y][x] = z24s8; + } + break; + default: + assert(0); + } + } +} + + +static void +depth_test_quad(struct quad_stage *qs, struct quad_header *quad) +{ + sp_depth_test_quad(qs, quad); + + if (quad->mask) + qs->next->run(qs->next, quad); +} + + +static void depth_test_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void depth_test_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = depth_test_begin; + stage->run = depth_test_quad; + stage->destroy = depth_test_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c new file mode 100644 index 0000000000..22ea99049f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_earlyz.c @@ -0,0 +1,88 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * \brief Quad early-z testing + */ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_headers.h" +#include "sp_quad.h" + + +/** + * All this stage does is compute the quad's Z values (which is normally + * done by the shading stage). + * The next stage will do the actual depth test. + */ +static void +earlyz_quad( + struct quad_stage *qs, + struct quad_header *quad ) +{ + const float fx = (float) quad->x0; + const float fy = (float) quad->y0; + const float dzdx = quad->posCoef->dadx[2]; + const float dzdy = quad->posCoef->dady[2]; + const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy; + + quad->outputs.depth[0] = z0; + quad->outputs.depth[1] = z0 + dzdx; + quad->outputs.depth[2] = z0 + dzdy; + quad->outputs.depth[3] = z0 + dzdx + dzdy; + + qs->next->run( qs->next, quad ); +} + +static void +earlyz_begin( + struct quad_stage *qs ) +{ + qs->next->begin( qs->next ); +} + +static void +earlyz_destroy( + struct quad_stage *qs ) +{ + FREE( qs ); +} + +struct quad_stage * +sp_quad_earlyz_stage( + struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT( quad_stage ); + + stage->softpipe = softpipe; + stage->begin = earlyz_begin; + stage->run = earlyz_quad; + stage->destroy = earlyz_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c new file mode 100644 index 0000000000..3316858413 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_fs.c @@ -0,0 +1,390 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Vertices are just an array of floats, with all the attributes + * packed. We currently assume a layout like: + * + * attr[0][0..3] - window position + * attr[1..n][0..3] - remaining attributes. + * + * Attributes are assumed to be 4 floats wide but are packed so that + * all the enabled attributes run contiguously. + */ + +#include "pipe/p_util.h" +#include "pipe/p_defines.h" +#include "pipe/p_shader_tokens.h" + +#include "x86/rtasm/x86sse.h" + +#ifdef MESA_LLVM +#include "pipe/llvm/gallivm.h" +#endif + +#include "sp_context.h" +#include "sp_state.h" +#include "sp_headers.h" +#include "sp_quad.h" +#include "sp_texture.h" +#include "sp_tex_sample.h" + + +struct quad_shade_stage +{ + struct quad_stage stage; + struct tgsi_sampler samplers[PIPE_MAX_SAMPLERS]; + struct tgsi_exec_machine machine; + struct tgsi_exec_vector *inputs, *outputs; + int colorOutSlot, depthOutSlot; +#ifdef MESA_LLVM + struct gallivm_prog *llvm_prog; +#endif +}; + + +/** cast wrapper */ +static INLINE struct quad_shade_stage * +quad_shade_stage(struct quad_stage *qs) +{ + return (struct quad_shade_stage *) qs; +} + + +/** + * Compute quad X,Y,Z,W for the four fragments in a quad. + * Note that we only need to "compute" X and Y for the upper-left fragment. + * We could do less work if we're not depth testing, or there's no + * perspective-corrected attributes, but that's seldom. + */ +static void +setup_pos_vector(const struct tgsi_interp_coef *coef, + float x, float y, + struct tgsi_exec_vector *quadpos) +{ + uint chan; + /* do X */ + quadpos->xyzw[0].f[0] = x; + /* do Y */ + quadpos->xyzw[1].f[0] = y; + /* do Z and W for all fragments in the quad */ + for (chan = 2; chan < 4; chan++) { + const float dadx = coef->dadx[chan]; + const float dady = coef->dady[chan]; + const float a0 = coef->a0[chan] + dadx * x + dady * y; + quadpos->xyzw[chan].f[0] = a0; + quadpos->xyzw[chan].f[1] = a0 + dadx; + quadpos->xyzw[chan].f[2] = a0 + dady; + quadpos->xyzw[chan].f[3] = a0 + dadx + dady; + } +} + + +typedef void (XSTDCALL *codegen_function)( + const struct tgsi_exec_vector *input, + struct tgsi_exec_vector *output, + float (*constant)[4], + struct tgsi_exec_vector *temporary, + const struct tgsi_interp_coef *coef +#if 0 + ,const struct tgsi_exec_vector *quadPos +#endif + ); + + +/** + * Execute fragment shader for the four fragments in the quad. + */ +static void +shade_quad( + struct quad_stage *qs, + struct quad_header *quad ) +{ + struct quad_shade_stage *qss = quad_shade_stage( qs ); + struct softpipe_context *softpipe = qs->softpipe; + struct tgsi_exec_machine *machine = &qss->machine; + + /* Consts do not require 16 byte alignment. */ + machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT]; + + machine->InterpCoefs = quad->coef; + + /* Compute X, Y, Z, W vals for this quad */ + setup_pos_vector(quad->posCoef, (float) quad->x0, (float) quad->y0, &machine->QuadPos); + + /* run shader */ +#if defined(__i386__) || defined(__386__) + if( softpipe->use_sse ) { + codegen_function func = (codegen_function) x86_get_func( &softpipe->fs->sse2_program ); + func( + machine->Inputs, + machine->Outputs, + machine->Consts, + machine->Temps, + machine->InterpCoefs +#if 0 + ,machine->QuadPos +#endif + ); + quad->mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]); + } + else +#endif + { + quad->mask &= tgsi_exec_machine_run( machine ); + } + + /* store result color */ + if (qss->colorOutSlot >= 0) { + /* XXX need to handle multiple color outputs someday */ + assert(qss->stage.softpipe->fs->shader.output_semantic_name[qss->colorOutSlot] + == TGSI_SEMANTIC_COLOR); + memcpy( + quad->outputs.color, + &machine->Outputs[qss->colorOutSlot].xyzw[0].f[0], + sizeof( quad->outputs.color ) ); + } + + /* + * XXX the following code for updating quad->outputs.depth + * isn't really needed if we did early z testing. + */ + + /* store result Z */ + if (qss->depthOutSlot >= 0) { + /* output[slot] is new Z */ + uint i; + for (i = 0; i < 4; i++) { + quad->outputs.depth[i] = machine->Outputs[0].xyzw[2].f[i]; + } + } + else { + /* copy input Z (which was interpolated by the executor) to output Z */ + uint i; + for (i = 0; i < 4; i++) { + quad->outputs.depth[i] = machine->Inputs[0].xyzw[2].f[i]; + /* XXX not sure the above line is always correct. The following + * might be better: + quad->outputs.depth[i] = machine->QuadPos.xyzw[2].f[i]; + */ + } + } + + /* shader may cull fragments */ + if( quad->mask ) { + qs->next->run( qs->next, quad ); + } +} + +#if 0 +#ifdef MESA_LLVM +#define DLLVM 0 +static void +shade_quad_llvm(struct quad_stage *qs, + struct quad_header *quad) +{ + struct quad_shade_stage *qss = quad_shade_stage(qs); + struct softpipe_context *softpipe = qs->softpipe; + float dests[4][16][4] ALIGN16_ATTRIB; + float inputs[4][16][4] ALIGN16_ATTRIB; + const float fx = (float) quad->x0; + const float fy = (float) quad->y0; + struct gallivm_prog *llvm = qss->llvm_prog; + + inputs[0][0][0] = fx; + inputs[1][0][0] = fx + 1.0f; + inputs[2][0][0] = fx; + inputs[3][0][0] = fx + 1.0f; + + inputs[0][0][1] = fy; + inputs[1][0][1] = fy; + inputs[2][0][1] = fy + 1.0f; + inputs[3][0][1] = fy + 1.0f; +#if DLLVM + debug_printf("MASK = %d\n", quad->mask); +#endif + gallivm_prog_inputs_interpolate(llvm, inputs, quad->coef); +#if DLLVM + for (int i = 0; i < 4; ++i) { + for (int j = 0; j < 2; ++j) { + debug_printf("IN(%d,%d) [%f %f %f %f]\n", i, j, + inputs[i][j][0], inputs[i][j][1], inputs[i][j][2], inputs[i][j][3]); + } + } +#endif + + quad->mask &= + gallivm_fragment_shader_exec(llvm, fx, fy, dests, inputs, + softpipe->mapped_constants[PIPE_SHADER_FRAGMENT], + qss->samplers); +#if DLLVM + debug_printf("OUT LLVM = 1[%f %f %f %f], 2[%f %f %f %f]\n", + dests[0][0][0], dests[0][0][1], dests[0][0][2], dests[0][0][3], + dests[0][1][0], dests[0][1][1], dests[0][1][2], dests[0][1][3]); +#endif + + /* store result color */ + if (qss->colorOutSlot >= 0) { + unsigned i; + /* XXX need to handle multiple color outputs someday */ + assert(qss->stage.softpipe->fs->shader.output_semantic_name[qss->colorOutSlot] + == TGSI_SEMANTIC_COLOR); + for (i = 0; i < QUAD_SIZE; ++i) { + quad->outputs.color[0][i] = dests[i][qss->colorOutSlot][0]; + quad->outputs.color[1][i] = dests[i][qss->colorOutSlot][1]; + quad->outputs.color[2][i] = dests[i][qss->colorOutSlot][2]; + quad->outputs.color[3][i] = dests[i][qss->colorOutSlot][3]; + } + } +#if DLLVM + for (int i = 0; i < QUAD_SIZE; ++i) { + debug_printf("QLLVM%d(%d) [%f, %f, %f, %f]\n", i, qss->colorOutSlot, + quad->outputs.color[0][i], + quad->outputs.color[1][i], + quad->outputs.color[2][i], + quad->outputs.color[3][i]); + } +#endif + + /* store result Z */ + if (qss->depthOutSlot >= 0) { + /* output[slot] is new Z */ + uint i; + for (i = 0; i < 4; i++) { + quad->outputs.depth[i] = dests[i][0][2]; + } + } + else { + /* copy input Z (which was interpolated by the executor) to output Z */ + uint i; + for (i = 0; i < 4; i++) { + quad->outputs.depth[i] = inputs[i][0][2]; + } + } +#if DLLVM + debug_printf("D [%f, %f, %f, %f] mask = %d\n", + quad->outputs.depth[0], + quad->outputs.depth[1], + quad->outputs.depth[2], + quad->outputs.depth[3], quad->mask); +#endif + + /* shader may cull fragments */ + if( quad->mask ) { + qs->next->run( qs->next, quad ); + } +} +#endif /*MESA_LLVM*/ +#endif + +/** + * Per-primitive (or per-begin?) setup + */ +static void shade_begin(struct quad_stage *qs) +{ + struct quad_shade_stage *qss = quad_shade_stage(qs); + struct softpipe_context *softpipe = qs->softpipe; + unsigned i; + + /* set TGSI sampler state that varies */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + qss->samplers[i].state = softpipe->sampler[i]; + qss->samplers[i].texture = &softpipe->texture[i]->base; + } + +#ifdef MESA_LLVM + qss->llvm_prog = softpipe->fs->llvm_prog; +#endif + /* XXX only do this if the fragment shader changes... */ + tgsi_exec_machine_init(&qss->machine, + softpipe->fs->shader.tokens, + PIPE_MAX_SAMPLERS, + qss->samplers ); + + /* find output slots for depth, color */ + qss->colorOutSlot = -1; + qss->depthOutSlot = -1; + for (i = 0; i < qss->stage.softpipe->fs->shader.num_outputs; i++) { + switch (qss->stage.softpipe->fs->shader.output_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + qss->depthOutSlot = i; + break; + case TGSI_SEMANTIC_COLOR: + qss->colorOutSlot = i; + break; + } + } + + qs->next->begin(qs->next); +} + + +static void shade_destroy(struct quad_stage *qs) +{ + struct quad_shade_stage *qss = (struct quad_shade_stage *) qs; + + tgsi_exec_machine_free_data(&qss->machine); + FREE( qss->inputs ); + FREE( qss->outputs ); + FREE( qs ); +} + + +struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe ) +{ + struct quad_shade_stage *qss = CALLOC_STRUCT(quad_shade_stage); + uint i; + + /* allocate storage for program inputs/outputs, aligned to 16 bytes */ + qss->inputs = MALLOC(PIPE_ATTRIB_MAX * sizeof(*qss->inputs) + 16); + qss->outputs = MALLOC(PIPE_ATTRIB_MAX * sizeof(*qss->outputs) + 16); + qss->machine.Inputs = align16(qss->inputs); + qss->machine.Outputs = align16(qss->outputs); + + qss->stage.softpipe = softpipe; + qss->stage.begin = shade_begin; +#ifdef MESA_LLVM + /* disable until ported to accept + * x/y and soa layout + qss->stage.run = shade_quad_llvm; + */ + softpipe->use_sse = FALSE; + qss->stage.run = shade_quad; +#else + qss->stage.run = shade_quad; +#endif + qss->stage.destroy = shade_destroy; + + /* set TGSI sampler state that's constant */ + for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { + assert(softpipe->tex_cache[i]); + qss->samplers[i].get_samples = sp_get_samples; + qss->samplers[i].pipe = &softpipe->pipe; + qss->samplers[i].cache = softpipe->tex_cache[i]; + } + + return &qss->stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/drivers/softpipe/sp_quad_occlusion.c new file mode 100644 index 0000000000..54254df1f1 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_occlusion.c @@ -0,0 +1,85 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + + +/** + * \brief Quad occlusion counter stage + * \author Brian Paul + */ + + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_quad.h" + +static unsigned count_bits( unsigned val ) +{ + unsigned i; + + for (i = 0; val ; val >>= 1) + i += (val & 1); + + return i; +} + +static void +occlusion_count_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + + softpipe->occlusion_count += count_bits(quad->mask); + + qs->next->run(qs->next, quad); +} + + +static void occlusion_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void occlusion_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = occlusion_begin; + stage->run = occlusion_count_quad; + stage->destroy = occlusion_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c new file mode 100644 index 0000000000..cfe8f11808 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_output.c @@ -0,0 +1,90 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_quad.h" +#include "sp_tile_cache.h" + + +/** + * Write quad to framebuffer, taking mask into account. + * + * Note that surfaces support only full quad reads and writes. + */ +static void +output_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe, + softpipe->cbuf_cache[softpipe->current_cbuf], + quad->x0, quad->y0); + /* in-tile pos: */ + const int itx = quad->x0 % TILE_SIZE; + const int ity = quad->y0 % TILE_SIZE; + float (*quadColor)[4] = quad->outputs.color; + int i, j; + + /* get/swizzle dest colors */ + for (j = 0; j < QUAD_SIZE; j++) { + if (quad->mask & (1 << j)) { + int x = itx + (j & 1); + int y = ity + (j >> 1); + for (i = 0; i < 4; i++) { /* loop over color chans */ + tile->data.color[y][x][i] = quadColor[i][j]; + } + } + } +} + + +static void output_begin(struct quad_stage *qs) +{ + assert(qs->next == NULL); +} + + +static void output_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = output_begin; + stage->run = output_quad; + stage->destroy = output_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c new file mode 100644 index 0000000000..92a0da0083 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c @@ -0,0 +1,352 @@ + +/** + * \brief Quad stencil testing + */ + + +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_tile_cache.h" +#include "sp_quad.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" + + +/** Only 8-bit stencil supported */ +#define STENCIL_MAX 0xff + + +/** + * Do the basic stencil test (compare stencil buffer values against the + * reference value. + * + * \param stencilVals the stencil values from the stencil buffer + * \param func the stencil func (PIPE_FUNC_x) + * \param ref the stencil reference value + * \param valMask the stencil value mask indicating which bits of the stencil + * values and ref value are to be used. + * \return mask indicating which pixels passed the stencil test + */ +static unsigned +do_stencil_test(const ubyte stencilVals[QUAD_SIZE], unsigned func, + unsigned ref, unsigned valMask) +{ + unsigned passMask = 0x0; + unsigned j; + + ref &= valMask; + + switch (func) { + case PIPE_FUNC_NEVER: + /* passMask = 0x0 */ + break; + case PIPE_FUNC_LESS: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref < (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_EQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref == (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_LEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref <= (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GREATER: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref > (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_NOTEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref != (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_GEQUAL: + for (j = 0; j < QUAD_SIZE; j++) { + if (ref >= (stencilVals[j] & valMask)) { + passMask |= (1 << j); + } + } + break; + case PIPE_FUNC_ALWAYS: + passMask = MASK_ALL; + break; + default: + assert(0); + } + + return passMask; +} + + +/** + * Apply the stencil operator to stencil values. + * + * \param stencilVals the stencil buffer values (read and written) + * \param mask indicates which pixels to update + * \param op the stencil operator (PIPE_STENCIL_OP_x) + * \param ref the stencil reference value + * \param wrtMask writemask controlling which bits are changed in the + * stencil values + */ +static void +apply_stencil_op(ubyte stencilVals[QUAD_SIZE], + unsigned mask, unsigned op, ubyte ref, ubyte wrtMask) +{ + unsigned j; + ubyte newstencil[QUAD_SIZE]; + + for (j = 0; j < QUAD_SIZE; j++) { + newstencil[j] = stencilVals[j]; + } + + switch (op) { + case PIPE_STENCIL_OP_KEEP: + /* no-op */ + break; + case PIPE_STENCIL_OP_ZERO: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = 0; + } + } + break; + case PIPE_STENCIL_OP_REPLACE: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = ref; + } + } + break; + case PIPE_STENCIL_OP_INCR: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + if (stencilVals[j] < STENCIL_MAX) { + newstencil[j] = stencilVals[j] + 1; + } + } + } + break; + case PIPE_STENCIL_OP_DECR: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + if (stencilVals[j] > 0) { + newstencil[j] = stencilVals[j] - 1; + } + } + } + break; + case PIPE_STENCIL_OP_INCR_WRAP: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = stencilVals[j] + 1; + } + } + break; + case PIPE_STENCIL_OP_DECR_WRAP: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = stencilVals[j] - 1; + } + } + break; + case PIPE_STENCIL_OP_INVERT: + for (j = 0; j < QUAD_SIZE; j++) { + if (mask & (1 << j)) { + newstencil[j] = ~stencilVals[j]; + } + } + break; + default: + assert(0); + } + + /* + * update the stencil values + */ + if (wrtMask != STENCIL_MAX) { + /* apply bit-wise stencil buffer writemask */ + for (j = 0; j < QUAD_SIZE; j++) { + stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & stencilVals[j]); + } + } + else { + for (j = 0; j < QUAD_SIZE; j++) { + stencilVals[j] = newstencil[j]; + } + } +} + + +/** + * Do stencil (and depth) testing. Stenciling depends on the outcome of + * depth testing. + */ +static void +stencil_test_quad(struct quad_stage *qs, struct quad_header *quad) +{ + struct softpipe_context *softpipe = qs->softpipe; + struct pipe_surface *ps = softpipe->framebuffer.zsbuf; + unsigned func, zFailOp, zPassOp, failOp; + ubyte ref, wrtMask, valMask; + ubyte stencilVals[QUAD_SIZE]; + struct softpipe_cached_tile *tile + = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0); + uint j; + uint face = quad->facing; + + if (!softpipe->depth_stencil->stencil[1].enabled) { + /* single-sided stencil test, use front (face=0) state */ + face = 0; + } + + /* choose front or back face function, operator, etc */ + /* XXX we could do these initializations once per primitive */ + func = softpipe->depth_stencil->stencil[face].func; + failOp = softpipe->depth_stencil->stencil[face].fail_op; + zFailOp = softpipe->depth_stencil->stencil[face].zfail_op; + zPassOp = softpipe->depth_stencil->stencil[face].zpass_op; + ref = softpipe->depth_stencil->stencil[face].ref_value; + wrtMask = softpipe->depth_stencil->stencil[face].write_mask; + valMask = softpipe->depth_stencil->stencil[face].value_mask; + + assert(ps); /* shouldn't get here if there's no stencil buffer */ + + /* get stencil values from cached tile */ + switch (ps->format) { + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.depth32[y][x] >> 24; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.depth32[y][x] & 0xff; + } + break; + case PIPE_FORMAT_U_S8: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + stencilVals[j] = tile->data.stencil8[y][x]; + } + break; + default: + assert(0); + } + + /* do the stencil test first */ + { + unsigned passMask, failMask; + passMask = do_stencil_test(stencilVals, func, ref, valMask); + failMask = quad->mask & ~passMask; + quad->mask &= passMask; + + if (failOp != PIPE_STENCIL_OP_KEEP) { + apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask); + } + } + + if (quad->mask) { + /* now the pixels that passed the stencil test are depth tested */ + if (softpipe->depth_stencil->depth.enabled) { + const unsigned origMask = quad->mask; + + sp_depth_test_quad(qs, quad); /* quad->mask is updated */ + + /* update stencil buffer values according to z pass/fail result */ + if (zFailOp != PIPE_STENCIL_OP_KEEP) { + const unsigned failMask = origMask & ~quad->mask; + apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask); + } + + if (zPassOp != PIPE_STENCIL_OP_KEEP) { + const unsigned passMask = origMask & quad->mask; + apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask); + } + } + else { + /* no depth test, apply Zpass operator to stencil buffer values */ + apply_stencil_op(stencilVals, quad->mask, zPassOp, ref, wrtMask); + } + + } + + /* put new stencil values into cached tile */ + switch (ps->format) { + case PIPE_FORMAT_S8Z24_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + uint s8z24 = tile->data.depth32[y][x]; + s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff); + tile->data.depth32[y][x] = s8z24; + } + break; + case PIPE_FORMAT_Z24S8_UNORM: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + uint z24s8 = tile->data.depth32[y][x]; + z24s8 = (z24s8 & 0xffffff00) | stencilVals[j]; + tile->data.depth32[y][x] = z24s8; + } + break; + case PIPE_FORMAT_U_S8: + for (j = 0; j < QUAD_SIZE; j++) { + int x = quad->x0 % TILE_SIZE + (j & 1); + int y = quad->y0 % TILE_SIZE + (j >> 1); + tile->data.stencil8[y][x] = stencilVals[j]; + } + break; + default: + assert(0); + } + + if (quad->mask) + qs->next->run(qs->next, quad); +} + + +static void stencil_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void stencil_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = stencil_begin; + stage->run = stencil_test_quad; + stage->destroy = stencil_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c new file mode 100644 index 0000000000..8660432259 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c @@ -0,0 +1,94 @@ + +/** + * quad polygon stipple stage + */ + +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_quad.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" + + +/** + * Apply polygon stipple to quads produced by triangle rasterization + */ +static void +stipple_quad(struct quad_stage *qs, struct quad_header *quad) +{ + static const uint bit31 = 1 << 31; + static const uint bit30 = 1 << 30; + + if (quad->prim == PRIM_TRI) { + struct softpipe_context *softpipe = qs->softpipe; + /* need to invert Y to index into OpenGL's stipple pattern */ + int y0, y1; + uint stipple0, stipple1; + if (softpipe->rasterizer->origin_lower_left) { + y0 = softpipe->framebuffer.cbufs[0]->height - 1 - quad->y0; + y1 = y0 - 1; + } + else { + y0 = quad->y0; + y1 = y0 + 1; + } + stipple0 = softpipe->poly_stipple.stipple[y0 % 32]; + stipple1 = softpipe->poly_stipple.stipple[y1 % 32]; + +#if 1 + { + const int col0 = quad->x0 % 32; + if ((stipple0 & (bit31 >> col0)) == 0) + quad->mask &= ~MASK_TOP_LEFT; + + if ((stipple0 & (bit30 >> col0)) == 0) + quad->mask &= ~MASK_TOP_RIGHT; + + if ((stipple1 & (bit31 >> col0)) == 0) + quad->mask &= ~MASK_BOTTOM_LEFT; + + if ((stipple1 & (bit30 >> col0)) == 0) + quad->mask &= ~MASK_BOTTOM_RIGHT; + } +#else + /* We'd like to use this code, but we'd need to redefine + * MASK_TOP_LEFT to be (1 << 1) and MASK_TOP_RIGHT to be (1 << 0), + * and similarly for the BOTTOM bits. But that may have undesirable + * side effects elsewhere. + */ + const int col0 = 30 - (quad->x0 % 32); + quad->mask &= (((stipple0 >> col0) & 0x3) | + (((stipple1 >> col0) & 0x3) << 2)); +#endif + if (!quad->mask) + return; + } + + qs->next->run(qs->next, quad); +} + + +static void stipple_begin(struct quad_stage *qs) +{ + qs->next->begin(qs->next); +} + + +static void stipple_destroy(struct quad_stage *qs) +{ + FREE( qs ); +} + + +struct quad_stage * +sp_quad_polygon_stipple_stage( struct softpipe_context *softpipe ) +{ + struct quad_stage *stage = CALLOC_STRUCT(quad_stage); + + stage->softpipe = softpipe; + stage->begin = stipple_begin; + stage->run = stipple_quad; + stage->destroy = stipple_destroy; + + return stage; +} diff --git a/src/gallium/drivers/softpipe/sp_query.c b/src/gallium/drivers/softpipe/sp_query.c new file mode 100644 index 0000000000..6a8a43aeda --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_query.c @@ -0,0 +1,107 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "pipe/draw/draw_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_query.h" + +struct softpipe_query { + uint64 start; + uint64 end; +}; + + +static struct softpipe_query *softpipe_query( struct pipe_query *p ) +{ + return (struct softpipe_query *)p; +} + +static struct pipe_query * +softpipe_create_query(struct pipe_context *pipe, + unsigned type) +{ + assert(type == PIPE_QUERY_OCCLUSION_COUNTER); + return (struct pipe_query *)CALLOC_STRUCT( softpipe_query ); +} + + +static void +softpipe_destroy_query(struct pipe_context *pipe, struct pipe_query *q) +{ + FREE(q); +} + + +static void +softpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q) +{ + struct softpipe_context *softpipe = softpipe_context( pipe ); + struct softpipe_query *sq = softpipe_query(q); + + sq->start = softpipe->occlusion_count; +} + + +static void +softpipe_end_query(struct pipe_context *pipe, struct pipe_query *q) +{ + struct softpipe_context *softpipe = softpipe_context( pipe ); + struct softpipe_query *sq = softpipe_query(q); + + sq->end = softpipe->occlusion_count; +} + + +static boolean +softpipe_get_query_result(struct pipe_context *pipe, + struct pipe_query *q, + boolean wait, + uint64 *result ) +{ + struct softpipe_query *sq = softpipe_query(q); + *result = sq->end - sq->start; + return TRUE; +} + + +void softpipe_init_query_funcs(struct softpipe_context *softpipe ) +{ + softpipe->pipe.create_query = softpipe_create_query; + softpipe->pipe.destroy_query = softpipe_destroy_query; + softpipe->pipe.begin_query = softpipe_begin_query; + softpipe->pipe.end_query = softpipe_end_query; + softpipe->pipe.get_query_result = softpipe_get_query_result; +} + + diff --git a/src/gallium/drivers/softpipe/sp_query.h b/src/gallium/drivers/softpipe/sp_query.h new file mode 100644 index 0000000000..05060a4575 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_query.h @@ -0,0 +1,39 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Author: + * Keith Whitwell + */ + +#ifndef SP_QUERY_H +#define SP_QUERY_H + +struct softpipe_context; +extern void softpipe_init_query_funcs(struct softpipe_context * ); + + +#endif /* SP_QUERY_H */ diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h new file mode 100644 index 0000000000..b79db0d1f1 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state.h @@ -0,0 +1,187 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_STATE_H +#define SP_STATE_H + +#include "pipe/p_state.h" + +#include "x86/rtasm/x86sse.h" + + +#define SP_NEW_VIEWPORT 0x1 +#define SP_NEW_RASTERIZER 0x2 +#define SP_NEW_FS 0x4 +#define SP_NEW_BLEND 0x8 +#define SP_NEW_CLIP 0x10 +#define SP_NEW_SCISSOR 0x20 +#define SP_NEW_STIPPLE 0x40 +#define SP_NEW_FRAMEBUFFER 0x80 +#define SP_NEW_DEPTH_STENCIL_ALPHA 0x100 +#define SP_NEW_CONSTANTS 0x200 +#define SP_NEW_SAMPLER 0x400 +#define SP_NEW_TEXTURE 0x800 +#define SP_NEW_VERTEX 0x1000 +#define SP_NEW_VS 0x2000 +#define SP_NEW_QUERY 0x4000 + + + +#ifdef MESA_LLVM +struct gallivm_prog; +#endif + + +struct vertex_info; + + +/** Subclass of pipe_shader_state */ +struct sp_fragment_shader_state { + struct pipe_shader_state shader; +#if defined(__i386__) || defined(__386__) + struct x86_function sse2_program; +#endif +#ifdef MESA_LLVM + struct gallivm_prog *llvm_prog; +#endif +}; + + +/** Subclass of pipe_shader_state */ +struct sp_vertex_shader_state { + struct pipe_shader_state shader; + struct draw_vertex_shader *draw_data; +}; + + + +void * +softpipe_create_blend_state(struct pipe_context *, + const struct pipe_blend_state *); +void softpipe_bind_blend_state(struct pipe_context *, + void *); +void softpipe_delete_blend_state(struct pipe_context *, + void *); + +void * +softpipe_create_sampler_state(struct pipe_context *, + const struct pipe_sampler_state *); +void softpipe_bind_sampler_state(struct pipe_context *, unsigned, void *); +void softpipe_delete_sampler_state(struct pipe_context *, void *); + +void * +softpipe_create_depth_stencil_state(struct pipe_context *, + const struct pipe_depth_stencil_alpha_state *); +void softpipe_bind_depth_stencil_state(struct pipe_context *, void *); +void softpipe_delete_depth_stencil_state(struct pipe_context *, void *); + +void * +softpipe_create_rasterizer_state(struct pipe_context *, + const struct pipe_rasterizer_state *); +void softpipe_bind_rasterizer_state(struct pipe_context *, void *); +void softpipe_delete_rasterizer_state(struct pipe_context *, void *); + +void softpipe_set_framebuffer_state( struct pipe_context *, + const struct pipe_framebuffer_state * ); + +void softpipe_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ); + +void softpipe_set_clip_state( struct pipe_context *, + const struct pipe_clip_state * ); + +void softpipe_set_constant_buffer(struct pipe_context *, + uint shader, uint index, + const struct pipe_constant_buffer *buf); + +void *softpipe_create_fs_state(struct pipe_context *, + const struct pipe_shader_state *); +void softpipe_bind_fs_state(struct pipe_context *, void *); +void softpipe_delete_fs_state(struct pipe_context *, void *); +void *softpipe_create_vs_state(struct pipe_context *, + const struct pipe_shader_state *); +void softpipe_bind_vs_state(struct pipe_context *, void *); +void softpipe_delete_vs_state(struct pipe_context *, void *); + +void softpipe_set_polygon_stipple( struct pipe_context *, + const struct pipe_poly_stipple * ); + +void softpipe_set_scissor_state( struct pipe_context *, + const struct pipe_scissor_state * ); + +void softpipe_set_sampler_texture( struct pipe_context *, + unsigned unit, + struct pipe_texture * ); + +void softpipe_set_viewport_state( struct pipe_context *, + const struct pipe_viewport_state * ); + +void softpipe_set_vertex_element(struct pipe_context *, + unsigned index, + const struct pipe_vertex_element *); + +void softpipe_set_vertex_buffer(struct pipe_context *, + unsigned index, + const struct pipe_vertex_buffer *); + + +void softpipe_update_derived( struct softpipe_context *softpipe ); + + +boolean softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, + unsigned start, unsigned count); + +boolean softpipe_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count); + + +void +softpipe_map_surfaces(struct softpipe_context *sp); + +void +softpipe_unmap_surfaces(struct softpipe_context *sp); + +void +softpipe_map_texture_surfaces(struct softpipe_context *sp); + +void +softpipe_unmap_texture_surfaces(struct softpipe_context *sp); + + +struct vertex_info * +softpipe_get_vertex_info(struct softpipe_context *softpipe); + +struct vertex_info * +softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe); + + +#endif diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c new file mode 100644 index 0000000000..2d40d6bd8f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_blend.c @@ -0,0 +1,98 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_state.h" + + +void * +softpipe_create_blend_state(struct pipe_context *pipe, + const struct pipe_blend_state *blend) +{ + return mem_dup(blend, sizeof(*blend)); +} + +void softpipe_bind_blend_state( struct pipe_context *pipe, + void *blend ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->blend = (const struct pipe_blend_state *)blend; + + softpipe->dirty |= SP_NEW_BLEND; +} + +void softpipe_delete_blend_state(struct pipe_context *pipe, + void *blend) +{ + FREE( blend ); +} + + +void softpipe_set_blend_color( struct pipe_context *pipe, + const struct pipe_blend_color *blend_color ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->blend_color = *blend_color; + + softpipe->dirty |= SP_NEW_BLEND; +} + + +/** XXX move someday? Or consolidate all these simple state setters + * into one file. + */ + + +void * +softpipe_create_depth_stencil_state(struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *depth_stencil) +{ + return mem_dup(depth_stencil, sizeof(*depth_stencil)); +} + +void +softpipe_bind_depth_stencil_state(struct pipe_context *pipe, + void *depth_stencil) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil; + + softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA; +} + +void +softpipe_delete_depth_stencil_state(struct pipe_context *pipe, void *depth) +{ + FREE( depth ); +} diff --git a/src/gallium/drivers/softpipe/sp_state_clip.c b/src/gallium/drivers/softpipe/sp_state_clip.c new file mode 100644 index 0000000000..08c5f06d05 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_clip.c @@ -0,0 +1,83 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ +#include "sp_context.h" +#include "sp_state.h" +#include "pipe/draw/draw_context.h" + + +void softpipe_set_clip_state( struct pipe_context *pipe, + const struct pipe_clip_state *clip ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + /* pass the clip state to the draw module */ + draw_set_clip_state(softpipe->draw, clip); +} + + + +/* Called when driver state tracker notices changes to the viewport + * matrix: + */ +void softpipe_set_viewport_state( struct pipe_context *pipe, + const struct pipe_viewport_state *viewport ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->viewport = *viewport; /* struct copy */ + softpipe->dirty |= SP_NEW_VIEWPORT; + + /* pass the viewport info to the draw module */ + draw_set_viewport_state(softpipe->draw, viewport); + + /* Using tnl/ and vf/ modules is temporary while getting started. + * Full pipe will have vertex shader, vertex fetch of its own. + */ +} + + +void softpipe_set_scissor_state( struct pipe_context *pipe, + const struct pipe_scissor_state *scissor ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + memcpy( &softpipe->scissor, scissor, sizeof(*scissor) ); + softpipe->dirty |= SP_NEW_SCISSOR; +} + + +void softpipe_set_polygon_stipple( struct pipe_context *pipe, + const struct pipe_poly_stipple *stipple ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + memcpy( &softpipe->poly_stipple, stipple, sizeof(*stipple) ); + softpipe->dirty |= SP_NEW_STIPPLE; +} diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c new file mode 100644 index 0000000000..372597869f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_derived.c @@ -0,0 +1,235 @@ +/************************************************************************** + * + * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_util.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/draw/draw_context.h" +#include "pipe/draw/draw_vertex.h" +#include "pipe/draw/draw_private.h" +#include "sp_context.h" +#include "sp_state.h" + + +/** + * Search vertex program's outputs to find a match for the given + * semantic name/index. Return the index of the output slot. + * + * Return 0 if not found. This will cause the fragment program to use + * vertex attrib 0 (position) in the cases where the fragment program + * attempts to use a missing vertex program output. This is an undefined + * condition that users shouldn't hit anyway. + */ +static int +find_vs_output(const struct pipe_shader_state *vs, + uint semantic_name, + uint semantic_index) +{ + uint i; + for (i = 0; i < vs->num_outputs; i++) { + if (vs->output_semantic_name[i] == semantic_name && + vs->output_semantic_index[i] == semantic_index) + return i; + } + return 0; +} + + +/** + * Mark the current vertex layout as "invalid". + * We'll validate the vertex layout later, when we start to actually + * render a point or line or tri. + */ +static void +invalidate_vertex_layout(struct softpipe_context *softpipe) +{ + softpipe->vertex_info.num_attribs = 0; +} + + +/** + * The vertex info describes how to convert the post-transformed vertices + * (simple float[][4]) used by the 'draw' module into vertices for + * rasterization. + * + * This function validates the vertex layout and returns a pointer to a + * vertex_info object. + */ +struct vertex_info * +softpipe_get_vertex_info(struct softpipe_context *softpipe) +{ + struct vertex_info *vinfo = &softpipe->vertex_info; + + if (vinfo->num_attribs == 0) { + /* compute vertex layout now */ + const struct pipe_shader_state *vs = &softpipe->vs->shader; + const struct pipe_shader_state *fs = &softpipe->fs->shader; + const enum interp_mode colorInterp + = softpipe->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR; + uint i; + + if (softpipe->vbuf) { + /* if using the post-transform vertex buffer, tell draw_vbuf to + * simply emit the whole post-xform vertex as-is: + */ + struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf; + vinfo_vbuf->num_attribs = 0; + draw_emit_vertex_attr(vinfo_vbuf, EMIT_ALL, INTERP_NONE, 0); + vinfo_vbuf->size = 4 * vs->num_outputs + + sizeof(struct vertex_header) / 4; + } + + /* + * Loop over fragment shader inputs, searching for the matching output + * from the vertex shader. + */ + vinfo->num_attribs = 0; + for (i = 0; i < fs->num_inputs; i++) { + int src; + switch (fs->input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src); + break; + + case TGSI_SEMANTIC_COLOR: + src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, + fs->input_semantic_index[i]); + draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src); + break; + + case TGSI_SEMANTIC_FOG: + src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src); + break; + + case TGSI_SEMANTIC_GENERIC: + /* this includes texcoords and varying vars */ + src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC, + fs->input_semantic_index[i]); + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src); + break; + + default: + assert(0); + } + } + + softpipe->psize_slot = find_vs_output(vs, TGSI_SEMANTIC_PSIZE, 0); + if (softpipe->psize_slot > 0) { + draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_CONSTANT, + softpipe->psize_slot); + } + + draw_compute_vertex_size(vinfo); + } + + return vinfo; +} + + +/** + * Called from vbuf module. + * + * Note that there's actually two different vertex layouts in softpipe. + * + * The normal one is computed in softpipe_get_vertex_info() above and is + * used by the point/line/tri "setup" code. + * + * The other one (this one) is only used by the vbuf module (which is + * not normally used by default but used in testing). For the vbuf module, + * we basically want to pass-through the draw module's vertex layout as-is. + * When the softpipe vbuf code begins drawing, the normal vertex layout + * will come into play again. + */ +struct vertex_info * +softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe) +{ + (void) softpipe_get_vertex_info(softpipe); + return &softpipe->vertex_info_vbuf; +} + + +/** + * Recompute cliprect from scissor bounds, scissor enable and surface size. + */ +static void +compute_cliprect(struct softpipe_context *sp) +{ + unsigned surfWidth, surfHeight; + + if (sp->framebuffer.num_cbufs > 0) { + surfWidth = sp->framebuffer.cbufs[0]->width; + surfHeight = sp->framebuffer.cbufs[0]->height; + } + else { + /* no surface? */ + surfWidth = sp->scissor.maxx; + surfHeight = sp->scissor.maxy; + } + + if (sp->rasterizer->scissor) { + /* clip to scissor rect */ + sp->cliprect.minx = MAX2(sp->scissor.minx, 0); + sp->cliprect.miny = MAX2(sp->scissor.miny, 0); + sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth); + sp->cliprect.maxy = MIN2(sp->scissor.maxy, surfHeight); + } + else { + /* clip to surface bounds */ + sp->cliprect.minx = 0; + sp->cliprect.miny = 0; + sp->cliprect.maxx = surfWidth; + sp->cliprect.maxy = surfHeight; + } +} + + +/* Hopefully this will remain quite simple, otherwise need to pull in + * something like the state tracker mechanism. + */ +void softpipe_update_derived( struct softpipe_context *softpipe ) +{ + if (softpipe->dirty & (SP_NEW_RASTERIZER | + SP_NEW_FS | + SP_NEW_VS)) + invalidate_vertex_layout( softpipe ); + + if (softpipe->dirty & (SP_NEW_SCISSOR | + SP_NEW_DEPTH_STENCIL_ALPHA | + SP_NEW_FRAMEBUFFER)) + compute_cliprect(softpipe); + + if (softpipe->dirty & (SP_NEW_BLEND | + SP_NEW_DEPTH_STENCIL_ALPHA | + SP_NEW_FRAMEBUFFER | + SP_NEW_RASTERIZER | + SP_NEW_FS | + SP_NEW_QUERY)) + sp_build_quad_pipeline(softpipe); + + softpipe->dirty = 0; +} diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c new file mode 100644 index 0000000000..0b814fc284 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -0,0 +1,179 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "sp_context.h" +#include "sp_state.h" + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/draw/draw_context.h" +#include "pipe/p_shader_tokens.h" +#include "pipe/llvm/gallivm.h" +#include "pipe/tgsi/util/tgsi_dump.h" +#include "pipe/tgsi/exec/tgsi_sse2.h" + + +void * +softpipe_create_fs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + struct sp_fragment_shader_state *state; + + /* Decide whether we'll be codegenerating this shader and if so do + * that now. + */ + + state = CALLOC_STRUCT(sp_fragment_shader_state); + if (!state) + return NULL; + + state->shader = *templ; + + if (softpipe->dump_fs) { + tgsi_dump(state->shader.tokens, 0); + } + +#ifdef MESA_LLVM + state->llvm_prog = 0; + +#if 0 + if (!gallivm_global_cpu_engine()) { + gallivm_cpu_engine_create(state->llvm_prog); + } + else + gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog); +#endif + +#elif defined(__i386__) || defined(__386__) + if (softpipe->use_sse) { + x86_init_func( &state->sse2_program ); + tgsi_emit_sse2_fs( state->shader.tokens, &state->sse2_program ); + } +#endif + + return state; +} + + +void +softpipe_bind_fs_state(struct pipe_context *pipe, void *fs) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->fs = (struct sp_fragment_shader_state *) fs; + + softpipe->dirty |= SP_NEW_FS; +} + + +void +softpipe_delete_fs_state(struct pipe_context *pipe, void *fs) +{ + struct sp_fragment_shader_state *state = fs; + +#if defined(__i386__) || defined(__386__) + x86_release_func( &state->sse2_program ); +#endif + + FREE( state ); +} + + +void * +softpipe_create_vs_state(struct pipe_context *pipe, + const struct pipe_shader_state *templ) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + struct sp_vertex_shader_state *state; + + state = CALLOC_STRUCT(sp_vertex_shader_state); + if (state == NULL ) { + return NULL; + } + + state->shader = *templ; + + state->draw_data = draw_create_vertex_shader(softpipe->draw, + &state->shader); + if (state->draw_data == NULL) { + FREE( state ); + return NULL; + } + + return state; +} + + +void +softpipe_bind_vs_state(struct pipe_context *pipe, void *vs) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + softpipe->vs = (const struct sp_vertex_shader_state *)vs; + + draw_bind_vertex_shader(softpipe->draw, softpipe->vs->draw_data); + + softpipe->dirty |= SP_NEW_VS; +} + + +void +softpipe_delete_vs_state(struct pipe_context *pipe, void *vs) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + struct sp_vertex_shader_state *state = + (struct sp_vertex_shader_state *)vs; + + draw_delete_vertex_shader(softpipe->draw, state->draw_data); + FREE( state ); +} + + + +void +softpipe_set_constant_buffer(struct pipe_context *pipe, + uint shader, uint index, + const struct pipe_constant_buffer *buf) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + struct pipe_winsys *ws = pipe->winsys; + + assert(shader < PIPE_SHADER_TYPES); + assert(index == 0); + + /* note: reference counting */ + pipe_buffer_reference(ws, + &softpipe->constants[shader].buffer, + buf->buffer); + softpipe->constants[shader].size = buf->size; + + softpipe->dirty |= SP_NEW_CONSTANTS; +} diff --git a/src/gallium/drivers/softpipe/sp_state_rasterizer.c b/src/gallium/drivers/softpipe/sp_state_rasterizer.c new file mode 100644 index 0000000000..53755099dd --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_rasterizer.c @@ -0,0 +1,62 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "sp_context.h" +#include "sp_state.h" +#include "pipe/draw/draw_context.h" + + + +void * +softpipe_create_rasterizer_state(struct pipe_context *pipe, + const struct pipe_rasterizer_state *rast) +{ + return mem_dup(rast, sizeof(*rast)); +} + +void softpipe_bind_rasterizer_state(struct pipe_context *pipe, + void *setup) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + /* pass-through to draw module */ + draw_set_rasterizer_state(softpipe->draw, setup); + + softpipe->rasterizer = (struct pipe_rasterizer_state *)setup; + + softpipe->dirty |= SP_NEW_RASTERIZER; +} + +void softpipe_delete_rasterizer_state(struct pipe_context *pipe, + void *rasterizer) +{ + FREE( rasterizer ); +} + + diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c new file mode 100644 index 0000000000..ea348c7e95 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_sampler.c @@ -0,0 +1,93 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: + * Brian Paul + */ + +#include "pipe/p_util.h" + +#include "pipe/draw/draw_context.h" + +#include "sp_context.h" +#include "sp_context.h" +#include "sp_state.h" +#include "sp_texture.h" +#include "sp_tile_cache.h" +#include "pipe/draw/draw_context.h" + + + +void * +softpipe_create_sampler_state(struct pipe_context *pipe, + const struct pipe_sampler_state *sampler) +{ + return mem_dup(sampler, sizeof(*sampler)); +} + +void +softpipe_bind_sampler_state(struct pipe_context *pipe, + unsigned unit, void *sampler) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + draw_flush(softpipe->draw); + + assert(unit < PIPE_MAX_SAMPLERS); + softpipe->sampler[unit] = (struct pipe_sampler_state *)sampler; + + softpipe->dirty |= SP_NEW_SAMPLER; +} + + +void +softpipe_delete_sampler_state(struct pipe_context *pipe, + void *sampler) +{ + FREE( sampler ); +} + + +void +softpipe_set_sampler_texture(struct pipe_context *pipe, + unsigned unit, + struct pipe_texture *texture) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + + draw_flush(softpipe->draw); + + assert(unit < PIPE_MAX_SAMPLERS); + softpipe->texture[unit] = softpipe_texture(texture); /* ptr, not struct */ + + sp_tile_cache_set_texture(softpipe->tex_cache[unit], texture); + + softpipe->dirty |= SP_NEW_TEXTURE; +} + + + diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c new file mode 100644 index 0000000000..e2c6893e9f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_surface.c @@ -0,0 +1,109 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ +#include "p_inlines.h" + +#include "sp_context.h" +#include "sp_state.h" +#include "sp_surface.h" +#include "sp_tile_cache.h" + + +/** + * XXX this might get moved someday + * Set the framebuffer surface info: color buffers, zbuffer, stencil buffer. + * Here, we flush the old surfaces and update the tile cache to point to the new + * surfaces. + */ +void +softpipe_set_framebuffer_state(struct pipe_context *pipe, + const struct pipe_framebuffer_state *fb) +{ + struct softpipe_context *sp = softpipe_context(pipe); + uint i; + + for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) { + /* check if changing cbuf */ + if (sp->framebuffer.cbufs[i] != fb->cbufs[i]) { + /* flush old */ + sp_flush_tile_cache(sp, sp->cbuf_cache[i]); + + /* assign new */ + sp->framebuffer.cbufs[i] = fb->cbufs[i]; + + /* update cache */ + sp_tile_cache_set_surface(sp->cbuf_cache[i], fb->cbufs[i]); + } + } + + sp->framebuffer.num_cbufs = fb->num_cbufs; + + /* zbuf changing? */ + if (sp->framebuffer.zsbuf != fb->zsbuf) { + /* flush old */ + sp_flush_tile_cache(sp, sp->zsbuf_cache); + + /* assign new */ + sp->framebuffer.zsbuf = fb->zsbuf; + + /* update cache */ + sp_tile_cache_set_surface(sp->zsbuf_cache, fb->zsbuf); + } + +#if 0 + /* XXX combined depth/stencil here */ + + /* sbuf changing? */ + if (sp->framebuffer.sbuf != fb->sbuf) { + /* flush old */ + sp_flush_tile_cache(sp, sp->sbuf_cache_sep); + + /* assign new */ + sp->framebuffer.sbuf = fb->sbuf; + + /* update cache */ + if (fb->sbuf != fb->zbuf) { + /* separate stencil buf */ + sp->sbuf_cache = sp->sbuf_cache_sep; + sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf); + } + else { + /* combined depth/stencil */ + sp->sbuf_cache = sp->zbuf_cache; + sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf); + } + } +#endif + + sp->dirty |= SP_NEW_FRAMEBUFFER; +} + + + + diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c new file mode 100644 index 0000000000..09ff540ccf --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_state_vertex.c @@ -0,0 +1,64 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "sp_context.h" +#include "sp_state.h" +#include "sp_surface.h" + +#include "pipe/draw/draw_context.h" + + +void +softpipe_set_vertex_element(struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_element *attrib) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + assert(index < PIPE_ATTRIB_MAX); + softpipe->vertex_element[index] = *attrib; /* struct copy */ + softpipe->dirty |= SP_NEW_VERTEX; + + draw_set_vertex_element(softpipe->draw, index, attrib); +} + + +void +softpipe_set_vertex_buffer(struct pipe_context *pipe, + unsigned index, + const struct pipe_vertex_buffer *buffer) +{ + struct softpipe_context *softpipe = softpipe_context(pipe); + assert(index < PIPE_ATTRIB_MAX); + softpipe->vertex_buffer[index] = *buffer; /* struct copy */ + softpipe->dirty |= SP_NEW_VERTEX; + + draw_set_vertex_buffer(softpipe->draw, index, buffer); +} diff --git a/src/gallium/drivers/softpipe/sp_surface.c b/src/gallium/drivers/softpipe/sp_surface.c new file mode 100644 index 0000000000..8802ced187 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_surface.c @@ -0,0 +1,159 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/p_winsys.h" +#include "pipe/util/p_tile.h" +#include "sp_context.h" +#include "sp_surface.h" + + + +/* Assumes all values are within bounds -- no checking at this level - + * do it higher up if required. + */ +static void +sp_surface_copy(struct pipe_context *pipe, + unsigned do_flip, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + struct pipe_surface *src, + unsigned srcx, unsigned srcy, unsigned width, unsigned height) +{ + assert( dst->cpp == src->cpp ); + + pipe_copy_rect(pipe_surface_map(dst), + dst->cpp, + dst->pitch, + dstx, dsty, + width, height, + pipe_surface_map(src), + do_flip ? -(int) src->pitch : src->pitch, + srcx, do_flip ? 1 - srcy - height : srcy); + + pipe_surface_unmap(src); + pipe_surface_unmap(dst); +} + + +static void * +get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y) +{ + return (char *)dst_map + (y * dst->pitch + x) * dst->cpp; +} + + +#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8)) + + +/** + * Fill a rectangular sub-region. Need better logic about when to + * push buffers into AGP - will currently do so whenever possible. + */ +static void +sp_surface_fill(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, unsigned value) +{ + unsigned i, j; + void *dst_map = pipe_surface_map(dst); + + assert(dst->pitch > 0); + assert(width <= dst->pitch); + + + switch (dst->cpp) { + case 1: + { + ubyte *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + memset(row, value, width); + row += dst->pitch; + } + } + break; + case 2: + { + ushort *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = (ushort) value; + row += dst->pitch; + } + } + break; + case 4: + { + unsigned *row = get_pointer(dst, dst_map, dstx, dsty); + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) + row[j] = value; + row += dst->pitch; + } + } + break; + case 8: + { + /* expand the 4-byte clear value to an 8-byte value */ + ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty); + ushort val0 = UBYTE_TO_USHORT((value >> 0) & 0xff); + ushort val1 = UBYTE_TO_USHORT((value >> 8) & 0xff); + ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff); + ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff); + val0 = (val0 << 8) | val0; + val1 = (val1 << 8) | val1; + val2 = (val2 << 8) | val2; + val3 = (val3 << 8) | val3; + for (i = 0; i < height; i++) { + for (j = 0; j < width; j++) { + row[j*4+0] = val0; + row[j*4+1] = val1; + row[j*4+2] = val2; + row[j*4+3] = val3; + } + row += dst->pitch * 4; + } + } + break; + default: + assert(0); + break; + } + + pipe_surface_unmap( dst ); +} + + +void +sp_init_surface_functions(struct softpipe_context *sp) +{ + sp->pipe.surface_copy = sp_surface_copy; + sp->pipe.surface_fill = sp_surface_fill; +} diff --git a/src/gallium/drivers/softpipe/sp_surface.h b/src/gallium/drivers/softpipe/sp_surface.h new file mode 100644 index 0000000000..22de3ba43f --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_surface.h @@ -0,0 +1,42 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* Authors: Keith Whitwell <keith@tungstengraphics.com> + */ + +#ifndef SP_SURFACE_H +#define SP_SURFACE_H + + +struct softpipe_context; + + +extern void +sp_init_surface_functions(struct softpipe_context *sp); + + +#endif /* SP_SURFACE_H */ diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c new file mode 100644 index 0000000000..325bdb86da --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tex_sample.c @@ -0,0 +1,916 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Texture sampling + * + * Authors: + * Brian Paul + */ + +#include "sp_context.h" +#include "sp_headers.h" +#include "sp_surface.h" +#include "sp_tex_sample.h" +#include "sp_tile_cache.h" +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_util.h" +#include "pipe/tgsi/exec/tgsi_exec.h" + + +/* + * Note, the FRAC macro has to work perfectly. Otherwise you'll sometimes + * see 1-pixel bands of improperly weighted linear-filtered textures. + * The tests/texwrap.c demo is a good test. + * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0. + * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x). + */ +#define FRAC(f) ((f) - ifloor(f)) + + +/** + * Linear interpolation macro + */ +#define LERP(T, A, B) ( (A) + (T) * ((B) - (A)) ) + + +/** + * Do 2D/biliner interpolation of float values. + * v00, v10, v01 and v11 are typically four texture samples in a square/box. + * a and b are the horizontal and vertical interpolants. + * It's important that this function is inlined when compiled with + * optimization! If we find that's not true on some systems, convert + * to a macro. + */ +static INLINE float +lerp_2d(float a, float b, + float v00, float v10, float v01, float v11) +{ + const float temp0 = LERP(a, v00, v10); + const float temp1 = LERP(a, v01, v11); + return LERP(b, temp0, temp1); +} + + +/** + * If A is a signed integer, A % B doesn't give the right value for A < 0 + * (in terms of texture repeat). Just casting to unsigned fixes that. + */ +#define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B)) + + +/** + * Apply texture coord wrapping mode and return integer texture index. + * \param wrapMode PIPE_TEX_WRAP_x + * \param s the texcoord + * \param size the texture image size + * \return integer texture index + */ +static INLINE int +nearest_texcoord(unsigned wrapMode, float s, unsigned size) +{ + int i; + switch (wrapMode) { + case PIPE_TEX_WRAP_REPEAT: + /* s limited to [0,1) */ + /* i limited to [0,size-1] */ + i = ifloor(s * size); + i = REMAINDER(i, size); + return i; + case PIPE_TEX_WRAP_CLAMP: + /* s limited to [0,1] */ + /* i limited to [0,size-1] */ + if (s <= 0.0F) + i = 0; + else if (s >= 1.0F) + i = size - 1; + else + i = ifloor(s * size); + return i; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + if (s < min) + i = 0; + else if (s > max) + i = size - 1; + else + i = ifloor(s * size); + } + return i; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + { + /* s limited to [min,max] */ + /* i limited to [-1, size] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + if (s <= min) + i = -1; + else if (s >= max) + i = size; + else + i = ifloor(s * size); + } + return i; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + { + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + const int flr = ifloor(s); + float u; + if (flr & 1) + u = 1.0F - (s - (float) flr); + else + u = s - (float) flr; + if (u < min) + i = 0; + else if (u > max) + i = size - 1; + else + i = ifloor(u * size); + } + return i; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + { + /* s limited to [0,1] */ + /* i limited to [0,size-1] */ + const float u = FABSF(s); + if (u <= 0.0F) + i = 0; + else if (u >= 1.0F) + i = size - 1; + else + i = ifloor(u * size); + } + return i; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = 1.0F / (2.0F * size); + const float max = 1.0F - min; + const float u = FABSF(s); + if (u < min) + i = 0; + else if (u > max) + i = size - 1; + else + i = ifloor(u * size); + } + return i; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + { + /* s limited to [min,max] */ + /* i limited to [0, size-1] */ + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + const float u = FABSF(s); + if (u < min) + i = -1; + else if (u > max) + i = size; + else + i = ifloor(u * size); + } + return i; + default: + assert(0); + return 0; + } +} + + +/** + * Used to compute texel locations for linear sampling. + * \param wrapMode PIPE_TEX_WRAP_x + * \param s the texcoord + * \param size the texture image size + * \param i0 returns first texture index + * \param i1 returns second texture index (usually *i0 + 1) + * \param a returns blend factor/weight between texture indexes + */ +static INLINE void +linear_texcoord(unsigned wrapMode, float s, unsigned size, + int *i0, int *i1, float *a) +{ + float u; + switch (wrapMode) { + case PIPE_TEX_WRAP_REPEAT: + u = s * size - 0.5F; + *i0 = REMAINDER(ifloor(u), size); + *i1 = REMAINDER(*i0 + 1, size); + break; + case PIPE_TEX_WRAP_CLAMP: + if (s <= 0.0F) + u = 0.0F; + else if (s >= 1.0F) + u = (float) size; + else + u = s * size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + break; + case PIPE_TEX_WRAP_CLAMP_TO_EDGE: + if (s <= 0.0F) + u = 0.0F; + else if (s >= 1.0F) + u = (float) size; + else + u = s * size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + if (*i0 < 0) + *i0 = 0; + if (*i1 >= (int) size) + *i1 = size - 1; + break; + case PIPE_TEX_WRAP_CLAMP_TO_BORDER: + { + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + if (s <= min) + u = min * size; + else if (s >= max) + u = max * size; + else + u = s * size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + } + break; + case PIPE_TEX_WRAP_MIRROR_REPEAT: + { + const int flr = ifloor(s); + if (flr & 1) + u = 1.0F - (s - (float) flr); + else + u = s - (float) flr; + u = (u * size) - 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + if (*i0 < 0) + *i0 = 0; + if (*i1 >= (int) size) + *i1 = size - 1; + } + break; + case PIPE_TEX_WRAP_MIRROR_CLAMP: + u = FABSF(s); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + break; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + u = FABSF(s); + if (u >= 1.0F) + u = (float) size; + else + u *= size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + if (*i0 < 0) + *i0 = 0; + if (*i1 >= (int) size) + *i1 = size - 1; + break; + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + { + const float min = -1.0F / (2.0F * size); + const float max = 1.0F - min; + u = FABSF(s); + if (u <= min) + u = min * size; + else if (u >= max) + u = max * size; + else + u *= size; + u -= 0.5F; + *i0 = ifloor(u); + *i1 = *i0 + 1; + } + break; + default: + assert(0); + } + *a = FRAC(u); +} + + +static unsigned +choose_cube_face(float rx, float ry, float rz, float *newS, float *newT) +{ + /* + major axis + direction target sc tc ma + ---------- ------------------------------- --- --- --- + +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx + -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx + +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry + -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry + +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz + -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz + */ + const float arx = FABSF(rx), ary = FABSF(ry), arz = FABSF(rz); + unsigned face; + float sc, tc, ma; + + if (arx > ary && arx > arz) { + if (rx >= 0.0F) { + face = PIPE_TEX_FACE_POS_X; + sc = -rz; + tc = -ry; + ma = arx; + } + else { + face = PIPE_TEX_FACE_NEG_X; + sc = rz; + tc = -ry; + ma = arx; + } + } + else if (ary > arx && ary > arz) { + if (ry >= 0.0F) { + face = PIPE_TEX_FACE_POS_Y; + sc = rx; + tc = rz; + ma = ary; + } + else { + face = PIPE_TEX_FACE_NEG_Y; + sc = rx; + tc = -rz; + ma = ary; + } + } + else { + if (rz > 0.0F) { + face = PIPE_TEX_FACE_POS_Z; + sc = rx; + tc = -ry; + ma = arz; + } + else { + face = PIPE_TEX_FACE_NEG_Z; + sc = -rx; + tc = -ry; + ma = arz; + } + } + + *newS = ( sc / ma + 1.0F ) * 0.5F; + *newT = ( tc / ma + 1.0F ) * 0.5F; + + return face; +} + + +/** + * Examine the quad's texture coordinates to compute the partial + * derivatives w.r.t X and Y, then compute lambda (level of detail). + * + * This is only done for fragment shaders, not vertex shaders. + */ +static float +compute_lambda(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias) +{ + float rho, lambda; + + assert(s); + { + float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]; + float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]; + dsdx = FABSF(dsdx); + dsdy = FABSF(dsdy); + rho = MAX2(dsdx, dsdy); + if (sampler->state->normalized_coords) + rho *= sampler->texture->width[0]; + } + if (t) { + float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]; + float dtdy = t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]; + float max; + dtdx = FABSF(dtdx); + dtdy = FABSF(dtdy); + max = MAX2(dtdx, dtdy); + if (sampler->state->normalized_coords) + max *= sampler->texture->height[0]; + rho = MAX2(rho, max); + } + if (p) { + float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]; + float dpdy = p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]; + float max; + dpdx = FABSF(dpdx); + dpdy = FABSF(dpdy); + max = MAX2(dpdx, dpdy); + if (sampler->state->normalized_coords) + max *= sampler->texture->depth[0]; + rho = MAX2(rho, max); + } + + lambda = LOG2(rho); + lambda += lodbias + sampler->state->lod_bias; + lambda = CLAMP(lambda, sampler->state->min_lod, sampler->state->max_lod); + + return lambda; +} + + +/** + * Do several things here: + * 1. Compute lambda from the texcoords, if needed + * 2. Determine if we're minifying or magnifying + * 3. If minifying, choose mipmap levels + * 4. Return image filter to use within mipmap images + */ +static void +choose_mipmap_levels(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + unsigned *level0, unsigned *level1, float *levelBlend, + unsigned *imgFilter) +{ + if (sampler->state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) { + /* no mipmap selection needed */ + *imgFilter = sampler->state->mag_img_filter; + *level0 = *level1 = (int) sampler->state->min_lod; + } + else { + float lambda; + + if (1) + /* fragment shader */ + lambda = compute_lambda(sampler, s, t, p, lodbias); + else + /* vertex shader */ + lambda = lodbias; /* not really a bias, but absolute LOD */ + + if (lambda < 0.0) { /* XXX threshold depends on the filter */ + /* magnifying */ + *imgFilter = sampler->state->mag_img_filter; + *level0 = *level1 = 0; + } + else { + /* minifying */ + *imgFilter = sampler->state->min_img_filter; + + /* choose mipmap level(s) and compute the blend factor between them */ + if (sampler->state->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) { + /* Nearest mipmap level */ + const int lvl = (int) (lambda + 0.5); + *level0 = + *level1 = CLAMP(lvl, 0, (int) sampler->texture->last_level); + } + else { + /* Linear interpolation between mipmap levels */ + const int lvl = (int) lambda; + *level0 = CLAMP(lvl, 0, (int) sampler->texture->last_level); + *level1 = CLAMP(lvl + 1, 0, (int) sampler->texture->last_level); + *levelBlend = FRAC(lambda); /* blending weight between levels */ + } + } + } +} + + +/** + * Get a texel from a texture, using the texture tile cache. + * + * \param face the cube face in 0..5 + * \param level the mipmap level + * \param x the x coord of texel within 2D image + * \param y the y coord of texel within 2D image + * \param z which slice of a 3D texture + * \param rgba the quad to put the texel/color into + * \param j which element of the rgba quad to write to + * + * XXX maybe move this into sp_tile_cache.c and merge with the + * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1... + */ +static void +get_texel(struct tgsi_sampler *sampler, + unsigned face, unsigned level, int x, int y, int z, + float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j) +{ + const int tx = x % TILE_SIZE; + const int ty = y % TILE_SIZE; + const struct softpipe_cached_tile *tile + = sp_get_cached_tile_tex(sampler->pipe, sampler->cache, + x, y, z, face, level); + rgba[0][j] = tile->data.color[ty][tx][0]; + rgba[1][j] = tile->data.color[ty][tx][1]; + rgba[2][j] = tile->data.color[ty][tx][2]; + rgba[3][j] = tile->data.color[ty][tx][3]; +} + + +/** + * Compare texcoord 'p' (aka R) against texture value 'rgba[0]' + * When we sampled the depth texture, the depth value was put into all + * RGBA channels. We look at the red channel here. + */ +static INLINE void +shadow_compare(uint compare_func, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const float p[QUAD_SIZE], + uint j) +{ + int k; + switch (compare_func) { + case PIPE_FUNC_LESS: + k = p[j] < rgba[0][j]; + break; + case PIPE_FUNC_LEQUAL: + k = p[j] <= rgba[0][j]; + break; + case PIPE_FUNC_GREATER: + k = p[j] > rgba[0][j]; + break; + case PIPE_FUNC_GEQUAL: + k = p[j] >= rgba[0][j]; + break; + case PIPE_FUNC_EQUAL: + k = p[j] == rgba[0][j]; + break; + case PIPE_FUNC_NOTEQUAL: + k = p[j] != rgba[0][j]; + break; + case PIPE_FUNC_ALWAYS: + k = 1; + break; + case PIPE_FUNC_NEVER: + k = 0; + break; + default: + assert(0); + } + + rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k; +} + + +/** + * Common code for sampling 1D/2D/cube textures. + * Could probably extend for 3D... + */ +static void +sp_get_samples_2d_common(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE], + const unsigned faces[4]) +{ + const uint compare_func = sampler->state->compare_func; + unsigned level0, level1, j, imgFilter; + int width, height; + float levelBlend; + + choose_mipmap_levels(sampler, s, t, p, lodbias, + &level0, &level1, &levelBlend, &imgFilter); + + if (sampler->state->normalized_coords) { + width = sampler->texture->width[level0]; + height = sampler->texture->height[level0]; + } + else { + width = height = 1; + } + + assert(width > 0); + + switch (imgFilter) { + case PIPE_TEX_FILTER_NEAREST: + for (j = 0; j < QUAD_SIZE; j++) { + int x = nearest_texcoord(sampler->state->wrap_s, s[j], width); + int y = nearest_texcoord(sampler->state->wrap_t, t[j], height); + get_texel(sampler, faces[j], level0, x, y, 0, rgba, j); + if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare(compare_func, rgba, p, j); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + unsigned c; + x = x / 2; + y = y / 2; + get_texel(sampler, faces[j], level1, x, y, 0, rgba2, j); + if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ + shadow_compare(compare_func, rgba2, p, j); + } + + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + break; + case PIPE_TEX_FILTER_LINEAR: + for (j = 0; j < QUAD_SIZE; j++) { + float tx[4][4], a, b; + int x0, y0, x1, y1, c; + linear_texcoord(sampler->state->wrap_s, s[j], width, &x0, &x1, &a); + linear_texcoord(sampler->state->wrap_t, t[j], height, &y0, &y1, &b); + get_texel(sampler, faces[j], level0, x0, y0, 0, tx, 0); + get_texel(sampler, faces[j], level0, x1, y0, 0, tx, 1); + get_texel(sampler, faces[j], level0, x0, y1, 0, tx, 2); + get_texel(sampler, faces[j], level0, x1, y1, 0, tx, 3); + if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + shadow_compare(compare_func, tx, p, 0); + shadow_compare(compare_func, tx, p, 1); + shadow_compare(compare_func, tx, p, 2); + shadow_compare(compare_func, tx, p, 3); + } + + for (c = 0; c < 4; c++) { + rgba[c][j] = lerp_2d(a, b, tx[c][0], tx[c][1], tx[c][2], tx[c][3]); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + x0 = x0 / 2; + y0 = y0 / 2; + x1 = x1 / 2; + y1 = y1 / 2; + get_texel(sampler, faces[j], level1, x0, y0, 0, tx, 0); + get_texel(sampler, faces[j], level1, x1, y0, 0, tx, 1); + get_texel(sampler, faces[j], level1, x0, y1, 0, tx, 2); + get_texel(sampler, faces[j], level1, x1, y1, 0, tx, 3); + if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){ + shadow_compare(compare_func, tx, p, 0); + shadow_compare(compare_func, tx, p, 1); + shadow_compare(compare_func, tx, p, 2); + shadow_compare(compare_func, tx, p, 3); + } + + for (c = 0; c < 4; c++) { + rgba2[c][j] = lerp_2d(a, b, + tx[c][0], tx[c][1], tx[c][2], tx[c][3]); + } + + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + break; + default: + assert(0); + } +} + + +static void +sp_get_samples_1d(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + static const unsigned faces[4] = {0, 0, 0, 0}; + static const float tzero[4] = {0, 0, 0, 0}; + sp_get_samples_2d_common(sampler, s, tzero, NULL, lodbias, rgba, faces); +} + + +static void +sp_get_samples_2d(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + static const unsigned faces[4] = {0, 0, 0, 0}; + sp_get_samples_2d_common(sampler, s, t, p, lodbias, rgba, faces); +} + + +static void +sp_get_samples_3d(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + /* get/map pipe_surfaces corresponding to 3D tex slices */ + unsigned level0, level1, j, imgFilter; + int width, height, depth; + float levelBlend; + const uint face = 0; + + choose_mipmap_levels(sampler, s, t, p, lodbias, + &level0, &level1, &levelBlend, &imgFilter); + + if (sampler->state->normalized_coords) { + width = sampler->texture->width[level0]; + height = sampler->texture->height[level0]; + depth = sampler->texture->depth[level0]; + } + else { + width = height = depth = 1; + } + + assert(width > 0); + assert(height > 0); + assert(depth > 0); + + switch (imgFilter) { + case PIPE_TEX_FILTER_NEAREST: + for (j = 0; j < QUAD_SIZE; j++) { + int x = nearest_texcoord(sampler->state->wrap_s, s[j], width); + int y = nearest_texcoord(sampler->state->wrap_t, t[j], height); + int z = nearest_texcoord(sampler->state->wrap_r, p[j], depth); + get_texel(sampler, face, level0, x, y, z, rgba, j); + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + unsigned c; + x /= 2; + y /= 2; + z /= 2; + get_texel(sampler, face, level1, x, y, z, rgba2, j); + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = LERP(levelBlend, rgba2[c][j], rgba[c][j]); + } + } + } + break; + case PIPE_TEX_FILTER_LINEAR: + for (j = 0; j < QUAD_SIZE; j++) { + float texel0[4][4], texel1[4][4]; + float xw, yw, zw; /* interpolation weights */ + int x0, x1, y0, y1, z0, z1, c; + linear_texcoord(sampler->state->wrap_s, s[j], width, &x0, &x1, &xw); + linear_texcoord(sampler->state->wrap_t, t[j], height, &y0, &y1, &yw); + linear_texcoord(sampler->state->wrap_r, p[j], depth, &z0, &z1, &zw); + get_texel(sampler, face, level0, x0, y0, z0, texel0, 0); + get_texel(sampler, face, level0, x1, y0, z0, texel0, 1); + get_texel(sampler, face, level0, x0, y1, z0, texel0, 2); + get_texel(sampler, face, level0, x1, y1, z0, texel0, 3); + get_texel(sampler, face, level0, x0, y0, z1, texel1, 0); + get_texel(sampler, face, level0, x1, y0, z1, texel1, 1); + get_texel(sampler, face, level0, x0, y1, z1, texel1, 2); + get_texel(sampler, face, level0, x1, y1, z1, texel1, 3); + + /* 3D lerp */ + for (c = 0; c < 4; c++) { + float ctemp0[4][4], ctemp1[4][4]; + ctemp0[c][j] = lerp_2d(xw, yw, + texel0[c][0], texel0[c][1], + texel0[c][2], texel0[c][3]); + ctemp1[c][j] = lerp_2d(xw, yw, + texel1[c][0], texel1[c][1], + texel1[c][2], texel1[c][3]); + rgba[c][j] = LERP(zw, ctemp0[c][j], ctemp1[c][j]); + } + + if (level0 != level1) { + /* get texels from second mipmap level and blend */ + float rgba2[4][4]; + x0 /= 2; + y0 /= 2; + z0 /= 2; + x1 /= 2; + y1 /= 2; + z1 /= 2; + get_texel(sampler, face, level1, x0, y0, z0, texel0, 0); + get_texel(sampler, face, level1, x1, y0, z0, texel0, 1); + get_texel(sampler, face, level1, x0, y1, z0, texel0, 2); + get_texel(sampler, face, level1, x1, y1, z0, texel0, 3); + get_texel(sampler, face, level1, x0, y0, z1, texel1, 0); + get_texel(sampler, face, level1, x1, y0, z1, texel1, 1); + get_texel(sampler, face, level1, x0, y1, z1, texel1, 2); + get_texel(sampler, face, level1, x1, y1, z1, texel1, 3); + + /* 3D lerp */ + for (c = 0; c < 4; c++) { + float ctemp0[4][4], ctemp1[4][4]; + ctemp0[c][j] = lerp_2d(xw, yw, + texel0[c][0], texel0[c][1], + texel0[c][2], texel0[c][3]); + ctemp1[c][j] = lerp_2d(xw, yw, + texel1[c][0], texel1[c][1], + texel1[c][2], texel1[c][3]); + rgba2[c][j] = LERP(zw, ctemp0[c][j], ctemp1[c][j]); + } + + /* blend mipmap levels */ + for (c = 0; c < NUM_CHANNELS; c++) { + rgba[c][j] = LERP(levelBlend, rgba[c][j], rgba2[c][j]); + } + } + } + break; + default: + assert(0); + } +} + + +static void +sp_get_samples_cube(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + unsigned faces[QUAD_SIZE], j; + float ssss[4], tttt[4]; + for (j = 0; j < QUAD_SIZE; j++) { + faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j); + } + sp_get_samples_2d_common(sampler, ssss, tttt, NULL, lodbias, rgba, faces); +} + + +/** + * Called via tgsi_sampler::get_samples() + * Use the sampler's state setting to get a filtered RGBA value + * from the sampler's texture. + * + * XXX we can implement many versions of this function, each + * tightly coded for a specific combination of sampler state + * (nearest + repeat), (bilinear mipmap + clamp), etc. + * + * The update_samplers() function in st_atom_sampler.c could create + * a new tgsi_sampler object for each state combo it finds.... + */ +void +sp_get_samples(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]) +{ + if (!sampler->texture) + return; + + switch (sampler->texture->target) { + case PIPE_TEXTURE_1D: + sp_get_samples_1d(sampler, s, t, p, lodbias, rgba); + break; + case PIPE_TEXTURE_2D: + sp_get_samples_2d(sampler, s, t, p, lodbias, rgba); + break; + case PIPE_TEXTURE_3D: + sp_get_samples_3d(sampler, s, t, p, lodbias, rgba); + break; + case PIPE_TEXTURE_CUBE: + sp_get_samples_cube(sampler, s, t, p, lodbias, rgba); + break; + default: + assert(0); + } +} + diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h new file mode 100644 index 0000000000..404bfd0c36 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tex_sample.h @@ -0,0 +1,17 @@ +#ifndef SP_TEX_SAMPLE_H +#define SP_TEX_SAMPLE_H + + +struct tgsi_sampler; + + +extern void +sp_get_samples(struct tgsi_sampler *sampler, + const float s[QUAD_SIZE], + const float t[QUAD_SIZE], + const float p[QUAD_SIZE], + float lodbias, + float rgba[NUM_CHANNELS][QUAD_SIZE]); + + +#endif /* SP_TEX_SAMPLE_H */ diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c new file mode 100644 index 0000000000..6de7a9b543 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_texture.c @@ -0,0 +1,166 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + * Michel Dänzer <michel@tungstengraphics.com> + */ + +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/p_util.h" +#include "pipe/p_winsys.h" + +#include "sp_context.h" +#include "sp_state.h" +#include "sp_texture.h" + + +/* Simple, maximally packed layout. + */ + +static unsigned minify( unsigned d ) +{ + return MAX2(1, d>>1); +} + + +static void +softpipe_texture_layout(struct softpipe_texture * spt) +{ + struct pipe_texture *pt = &spt->base; + unsigned level; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned depth = pt->depth[0]; + + spt->buffer_size = 0; + + for (level = 0; level <= pt->last_level; level++) { + pt->width[level] = width; + pt->height[level] = height; + pt->depth[level] = depth; + + spt->level_offset[level] = spt->buffer_size; + + spt->buffer_size += ((pt->compressed) ? MAX2(1, height/4) : height) * + ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) * + width * pt->cpp; + + width = minify(width); + height = minify(height); + depth = minify(depth); + } +} + + +struct pipe_texture * +softpipe_texture_create(struct pipe_context *pipe, + const struct pipe_texture *templat) +{ + struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture); + if (!spt) + return NULL; + + spt->base = *templat; + + softpipe_texture_layout(spt); + + spt->buffer = pipe->winsys->buffer_create(pipe->winsys, 32, + PIPE_BUFFER_USAGE_PIXEL, + spt->buffer_size); + if (!spt->buffer) { + FREE(spt); + return NULL; + } + + return &spt->base; +} + + +void +softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt) +{ + if (!*pt) + return; + + /* + DBG("%s %p refcount will be %d\n", + __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); + */ + if (--(*pt)->refcount <= 0) { + struct softpipe_texture *spt = softpipe_texture(*pt); + + /* + DBG("%s deleting %p\n", __FUNCTION__, (void *) spt); + */ + + pipe_buffer_reference(pipe->winsys, &spt->buffer, NULL); + + FREE(spt); + } + *pt = NULL; +} + + +/** + * Called via pipe->get_tex_surface() + */ +struct pipe_surface * +softpipe_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice) +{ + struct softpipe_texture *spt = softpipe_texture(pt); + struct pipe_surface *ps; + + assert(level <= pt->last_level); + + ps = pipe->winsys->surface_alloc(pipe->winsys); + if (ps) { + assert(ps->refcount); + assert(ps->winsys); + pipe_buffer_reference(pipe->winsys, &ps->buffer, spt->buffer); + ps->format = pt->format; + ps->cpp = pt->cpp; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->pitch = ps->width; + ps->offset = spt->level_offset[level]; + + if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) { + ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) * + (pt->compressed ? ps->height/4 : ps->height) * + ps->width * ps->cpp; + } else { + assert(face == 0); + assert(zslice == 0); + } + } + return ps; +} diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h new file mode 100644 index 0000000000..fa646c0de9 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_texture.h @@ -0,0 +1,71 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SP_TEXTURE_H +#define SP_TEXTURE_H + + +struct pipe_context; +struct pipe_texture; + + +struct softpipe_texture +{ + struct pipe_texture base; + + unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS]; + + /* The data is held here: + */ + struct pipe_buffer *buffer; + unsigned long buffer_size; +}; + + +/** cast wrapper */ +static INLINE struct softpipe_texture * +softpipe_texture(struct pipe_texture *pt) +{ + return (struct softpipe_texture *) pt; +} + + + +extern struct pipe_texture * +softpipe_texture_create(struct pipe_context *pipe, + const struct pipe_texture *templat); + +extern void +softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt); + +extern struct pipe_surface * +softpipe_get_tex_surface(struct pipe_context *pipe, + struct pipe_texture *pt, + unsigned face, unsigned level, unsigned zslice); + + +#endif /* SP_TEXTURE */ diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c new file mode 100644 index 0000000000..1597361b82 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tile_cache.c @@ -0,0 +1,585 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Framebuffer/surface tile caching. + * + * Author: + * Brian Paul + */ + +#include "pipe/p_util.h" +#include "pipe/p_inlines.h" +#include "pipe/util/p_tile.h" +#include "sp_context.h" +#include "sp_surface.h" +#include "sp_tile_cache.h" + +#define NUM_ENTRIES 30 + + +/** XXX move these */ +#define MAX_WIDTH 2048 +#define MAX_HEIGHT 2048 + + +struct softpipe_tile_cache +{ + struct pipe_surface *surface; /**< the surface we're caching */ + void *surface_map; + struct pipe_texture *texture; /**< if caching a texture */ + struct softpipe_cached_tile entries[NUM_ENTRIES]; + uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32]; + float clear_color[4]; + uint clear_val; + boolean depth_stencil; /** Is the surface a depth/stencil format? */ + + struct pipe_surface *tex_surf; + void *tex_surf_map; + int tex_face, tex_level, tex_z; + + struct softpipe_cached_tile tile; /**< scratch tile for clears */ +}; + + +/** + * Return the position in the cache for the tile that contains win pos (x,y). + * We currently use a direct mapped cache so this is like a hack key. + * At some point we should investige something more sophisticated, like + * a LRU replacement policy. + */ +#define CACHE_POS(x, y) \ + (((x) / TILE_SIZE + ((y) / TILE_SIZE) * 5) % NUM_ENTRIES) + + + +/** + * Is the tile at (x,y) in cleared state? + */ +static INLINE uint +is_clear_flag_set(const uint *bitvec, int x, int y) +{ + int pos, bit; + x /= TILE_SIZE; + y /= TILE_SIZE; + pos = y * (MAX_WIDTH / TILE_SIZE) + x; + assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32); + bit = bitvec[pos / 32] & (1 << (pos & 31)); + return bit; +} + + +/** + * Mark the tile at (x,y) as not cleared. + */ +static INLINE void +clear_clear_flag(uint *bitvec, int x, int y) +{ + int pos; + x /= TILE_SIZE; + y /= TILE_SIZE; + pos = y * (MAX_WIDTH / TILE_SIZE) + x; + assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32); + bitvec[pos / 32] &= ~(1 << (pos & 31)); +} + + +struct softpipe_tile_cache * +sp_create_tile_cache(void) +{ + struct softpipe_tile_cache *tc; + uint pos; + + tc = CALLOC_STRUCT( softpipe_tile_cache ); + if (tc) { + for (pos = 0; pos < NUM_ENTRIES; pos++) { + tc->entries[pos].x = + tc->entries[pos].y = -1; + } + } + return tc; +} + + +void +sp_destroy_tile_cache(struct softpipe_tile_cache *tc) +{ + uint pos; + + for (pos = 0; pos < NUM_ENTRIES; pos++) { + //assert(tc->entries[pos].x < 0); + } + if (tc->surface) { + pipe_surface_reference(&tc->surface, NULL); + } + if (tc->tex_surf) { + pipe_surface_reference(&tc->tex_surf, NULL); + } + + FREE( tc ); +} + + +/** + * Specify the surface to cache. + */ +void +sp_tile_cache_set_surface(struct softpipe_tile_cache *tc, + struct pipe_surface *ps) +{ + assert(!tc->texture); + + if (tc->surface_map) { + /*assert(tc->surface != ps);*/ + pipe_surface_unmap(tc->surface); + } + + pipe_surface_reference(&tc->surface, ps); + + if (ps) { + if (tc->surface_map) + tc->surface_map = pipe_surface_map(ps); + + tc->depth_stencil = (ps->format == PIPE_FORMAT_S8Z24_UNORM || + ps->format == PIPE_FORMAT_Z16_UNORM || + ps->format == PIPE_FORMAT_Z32_UNORM || + ps->format == PIPE_FORMAT_U_S8); + } +} + + +/** + * Return the surface being cached. + */ +struct pipe_surface * +sp_tile_cache_get_surface(struct softpipe_tile_cache *tc) +{ + return tc->surface; +} + + +void +sp_tile_cache_map_surfaces(struct softpipe_tile_cache *tc) +{ + if (tc->surface && !tc->surface_map) + tc->surface_map = pipe_surface_map(tc->surface); + + if (tc->tex_surf && !tc->tex_surf_map) + tc->tex_surf_map = pipe_surface_map(tc->tex_surf); +} + + +void +sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc) +{ + if (tc->surface_map) { + pipe_surface_unmap(tc->surface); + tc->surface_map = NULL; + } + + if (tc->tex_surf_map) { + pipe_surface_unmap(tc->tex_surf); + tc->tex_surf_map = NULL; + } +} + + +/** + * Specify the texture to cache. + */ +void +sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, + struct pipe_texture *texture) +{ + uint i; + + assert(!tc->surface); + + tc->texture = texture; + + if (tc->tex_surf_map) { + pipe_surface_unmap(tc->tex_surf); + tc->tex_surf_map = NULL; + } + pipe_surface_reference(&tc->tex_surf, NULL); + + /* mark as entries as invalid/empty */ + /* XXX we should try to avoid this when the teximage hasn't changed */ + for (i = 0; i < NUM_ENTRIES; i++) { + tc->entries[i].x = -1; + } + + tc->tex_face = -1; /* any invalid value here */ +} + + +/** + * Set pixels in a tile to the given clear color/value, float. + */ +static void +clear_tile_rgba(struct softpipe_cached_tile *tile, + enum pipe_format format, + const float clear_value[4]) +{ + if (clear_value[0] == 0.0 && + clear_value[1] == 0.0 && + clear_value[2] == 0.0 && + clear_value[3] == 0.0) { + memset(tile->data.color, 0, sizeof(tile->data.color)); + } + else { + uint i, j; + for (i = 0; i < TILE_SIZE; i++) { + for (j = 0; j < TILE_SIZE; j++) { + tile->data.color[i][j][0] = clear_value[0]; + tile->data.color[i][j][1] = clear_value[1]; + tile->data.color[i][j][2] = clear_value[2]; + tile->data.color[i][j][3] = clear_value[3]; + } + } + } +} + + +/** + * Set a tile to a solid value/color. + */ +static void +clear_tile(struct softpipe_cached_tile *tile, + enum pipe_format format, + uint clear_value) +{ + uint i, j; + + switch (pf_get_size(format)) { + case 1: + memset(tile->data.any, 0, TILE_SIZE * TILE_SIZE); + break; + case 2: + if (clear_value == 0) { + memset(tile->data.any, 0, 2 * TILE_SIZE * TILE_SIZE); + } + else { + for (i = 0; i < TILE_SIZE; i++) { + for (j = 0; j < TILE_SIZE; j++) { + tile->data.depth16[i][j] = (ushort) clear_value; + } + } + } + break; + case 4: + if (clear_value == 0) { + memset(tile->data.any, 0, 4 * TILE_SIZE * TILE_SIZE); + } + else { + for (i = 0; i < TILE_SIZE; i++) { + for (j = 0; j < TILE_SIZE; j++) { + tile->data.color32[i][j] = clear_value; + } + } + } + break; + default: + assert(0); + } +} + + +/** + * Actually clear the tiles which were flagged as being in a clear state. + */ +static void +sp_tile_cache_flush_clear(struct pipe_context *pipe, + struct softpipe_tile_cache *tc) +{ + struct pipe_surface *ps = tc->surface; + const uint w = tc->surface->width; + const uint h = tc->surface->height; + uint x, y; + uint numCleared = 0; + + /* clear the scratch tile to the clear value */ + clear_tile(&tc->tile, ps->format, tc->clear_val); + + /* push the tile to all positions marked as clear */ + for (y = 0; y < h; y += TILE_SIZE) { + for (x = 0; x < w; x += TILE_SIZE) { + if (is_clear_flag_set(tc->clear_flags, x, y)) { + pipe_put_tile_raw(pipe, ps, + x, y, TILE_SIZE, TILE_SIZE, + tc->tile.data.color32, 0/*STRIDE*/); + + /* do this? */ + clear_clear_flag(tc->clear_flags, x, y); + + numCleared++; + } + } + } +#if 0 + debug_printf("num cleared: %u\n", numCleared); +#endif +} + + +/** + * Flush the tile cache: write all dirty tiles back to the surface. + * any tiles "flagged" as cleared will be "really" cleared. + */ +void +sp_flush_tile_cache(struct softpipe_context *softpipe, + struct softpipe_tile_cache *tc) +{ + struct pipe_context *pipe = &softpipe->pipe; + struct pipe_surface *ps = tc->surface; + int inuse = 0, pos; + + if (!ps || !ps->buffer) + return; + + for (pos = 0; pos < NUM_ENTRIES; pos++) { + struct softpipe_cached_tile *tile = tc->entries + pos; + if (tile->x >= 0) { + if (tc->depth_stencil) { + pipe_put_tile_raw(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->data.depth32, 0/*STRIDE*/); + } + else { + pipe_put_tile_rgba(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + (float *) tile->data.color); + } + tile->x = tile->y = -1; /* mark as empty */ + inuse++; + } + } + +#if TILE_CLEAR_OPTIMIZATION + sp_tile_cache_flush_clear(&softpipe->pipe, tc); +#endif + +#if 0 + debug_printf("flushed tiles in use: %d\n", inuse); +#endif +} + + +/** + * Get a tile from the cache. + * \param x, y position of tile, in pixels + */ +struct softpipe_cached_tile * +sp_get_cached_tile(struct softpipe_context *softpipe, + struct softpipe_tile_cache *tc, int x, int y) +{ + struct pipe_context *pipe = &softpipe->pipe; + struct pipe_surface *ps = tc->surface; + + /* tile pos in framebuffer: */ + const int tile_x = x & ~(TILE_SIZE - 1); + const int tile_y = y & ~(TILE_SIZE - 1); + + /* cache pos/entry: */ + const int pos = CACHE_POS(x, y); + struct softpipe_cached_tile *tile = tc->entries + pos; + + if (tile_x != tile->x || + tile_y != tile->y) { + + if (tile->x != -1) { + /* put dirty tile back in framebuffer */ + if (tc->depth_stencil) { + pipe_put_tile_raw(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->data.depth32, 0/*STRIDE*/); + } + else { + pipe_put_tile_rgba(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + (float *) tile->data.color); + } + } + + tile->x = tile_x; + tile->y = tile_y; + + if (is_clear_flag_set(tc->clear_flags, x, y)) { + /* don't get tile from framebuffer, just clear it */ + if (tc->depth_stencil) { + clear_tile(tile, ps->format, tc->clear_val); + } + else { + clear_tile_rgba(tile, ps->format, tc->clear_color); + } + clear_clear_flag(tc->clear_flags, x, y); + } + else { + /* get new tile data from surface */ + if (tc->depth_stencil) { + pipe_get_tile_raw(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + tile->data.depth32, 0/*STRIDE*/); + } + else { + pipe_get_tile_rgba(pipe, ps, + tile->x, tile->y, TILE_SIZE, TILE_SIZE, + (float *) tile->data.color); + } + } + } + + return tile; +} + + +/** + * Given the texture face, level, zslice, x and y values, compute + * the cache entry position/index where we'd hope to find the + * cached texture tile. + * This is basically a direct-map cache. + * XXX There's probably lots of ways in which we can improve this. + */ +static INLINE uint +tex_cache_pos(int x, int y, int z, int face, int level) +{ + uint entry = x + y * 5 + z * 4 + face + level; + return entry % NUM_ENTRIES; +} + + +/** + * Similar to sp_get_cached_tile() but for textures. + * Tiles are read-only and indexed with more params. + */ +const struct softpipe_cached_tile * +sp_get_cached_tile_tex(struct pipe_context *pipe, + struct softpipe_tile_cache *tc, int x, int y, int z, + int face, int level) +{ + /* tile pos in framebuffer: */ + const int tile_x = x & ~(TILE_SIZE - 1); + const int tile_y = y & ~(TILE_SIZE - 1); + /* cache pos/entry: */ + const uint pos = tex_cache_pos(x / TILE_SIZE, y / TILE_SIZE, z, + face, level); + struct softpipe_cached_tile *tile = tc->entries + pos; + + if (tile_x != tile->x || + tile_y != tile->y || + z != tile->z || + face != tile->face || + level != tile->level) { + /* cache miss */ + + /* check if we need to get a new surface */ + if (!tc->tex_surf || + tc->tex_face != face || + tc->tex_level != level || + tc->tex_z != z) { + /* get new surface (view into texture) */ + + if (tc->tex_surf_map) + pipe_surface_unmap(tc->tex_surf); + + tc->tex_surf = pipe->get_tex_surface(pipe, tc->texture, face, level, z); + tc->tex_surf_map = pipe_surface_map(tc->tex_surf); + + tc->tex_face = face; + tc->tex_level = level; + tc->tex_z = z; + } + + /* get tile from the surface (view into texture) */ + pipe_get_tile_rgba(pipe, tc->tex_surf, + tile_x, tile_y, TILE_SIZE, TILE_SIZE, + (float *) tile->data.color); + tile->x = tile_x; + tile->y = tile_y; + tile->z = z; + tile->face = face; + tile->level = level; + } + + return tile; +} + + +/** + * When a whole surface is being cleared to a value we can avoid + * fetching tiles above. + * Save the color and set a 'clearflag' for each tile of the screen. + */ +void +sp_tile_cache_clear(struct softpipe_tile_cache *tc, uint clearValue) +{ + uint r, g, b, a; + uint pos; + + tc->clear_val = clearValue; + + switch (tc->surface->format) { + case PIPE_FORMAT_R8G8B8A8_UNORM: + r = (clearValue >> 24) & 0xff; + g = (clearValue >> 16) & 0xff; + b = (clearValue >> 8) & 0xff; + a = (clearValue ) & 0xff; + break; + case PIPE_FORMAT_A8R8G8B8_UNORM: + r = (clearValue >> 16) & 0xff; + g = (clearValue >> 8) & 0xff; + b = (clearValue ) & 0xff; + a = (clearValue >> 24) & 0xff; + break; + case PIPE_FORMAT_B8G8R8A8_UNORM: + r = (clearValue >> 8) & 0xff; + g = (clearValue >> 16) & 0xff; + b = (clearValue >> 24) & 0xff; + a = (clearValue ) & 0xff; + break; + default: + r = g = b = a = 0; + } + + tc->clear_color[0] = r / 255.0f; + tc->clear_color[1] = g / 255.0f; + tc->clear_color[2] = b / 255.0f; + tc->clear_color[3] = a / 255.0f; + +#if TILE_CLEAR_OPTIMIZATION + /* set flags to indicate all the tiles are cleared */ + memset(tc->clear_flags, 255, sizeof(tc->clear_flags)); +#else + /* disable the optimization */ + memset(tc->clear_flags, 0, sizeof(tc->clear_flags)); +#endif + + for (pos = 0; pos < NUM_ENTRIES; pos++) { + struct softpipe_cached_tile *tile = tc->entries + pos; + tile->x = tile->y = -1; + } +} diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h new file mode 100644 index 0000000000..7fd1081286 --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_tile_cache.h @@ -0,0 +1,104 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef SP_TILE_CACHE_H +#define SP_TILE_CACHE_H + +#define TILE_CLEAR_OPTIMIZATION 1 + + +#include "pipe/p_compiler.h" + + +struct softpipe_context; +struct softpipe_tile_cache; + + +/** + * Cache tile size (width and height). This needs to be a power of two. + */ +#define TILE_SIZE 64 + + + +struct softpipe_cached_tile +{ + int x, y; /**< pos of tile in window coords */ + int z, face, level; /**< Extra texture indexes */ + union { + float color[TILE_SIZE][TILE_SIZE][4]; + uint color32[TILE_SIZE][TILE_SIZE]; + uint depth32[TILE_SIZE][TILE_SIZE]; + ushort depth16[TILE_SIZE][TILE_SIZE]; + ubyte stencil8[TILE_SIZE][TILE_SIZE]; + ubyte any[1]; + } data; +}; + + +extern struct softpipe_tile_cache * +sp_create_tile_cache(void); + +extern void +sp_destroy_tile_cache(struct softpipe_tile_cache *tc); + +extern void +sp_tile_cache_set_surface(struct softpipe_tile_cache *tc, + struct pipe_surface *sps); + +extern struct pipe_surface * +sp_tile_cache_get_surface(struct softpipe_tile_cache *tc); + +extern void +sp_tile_cache_map_surfaces(struct softpipe_tile_cache *tc); + +extern void +sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc); + +extern void +sp_tile_cache_set_texture(struct softpipe_tile_cache *tc, + struct pipe_texture *texture); + +extern void +sp_flush_tile_cache(struct softpipe_context *softpipe, + struct softpipe_tile_cache *tc); + +extern void +sp_tile_cache_clear(struct softpipe_tile_cache *tc, uint clearValue); + +extern struct softpipe_cached_tile * +sp_get_cached_tile(struct softpipe_context *softpipe, + struct softpipe_tile_cache *tc, int x, int y); + +extern const struct softpipe_cached_tile * +sp_get_cached_tile_tex(struct pipe_context *pipe, + struct softpipe_tile_cache *tc, int x, int y, int z, + int face, int level); + + +#endif /* SP_TILE_CACHE_H */ + diff --git a/src/gallium/drivers/softpipe/sp_winsys.h b/src/gallium/drivers/softpipe/sp_winsys.h new file mode 100644 index 0000000000..d6b379f58c --- /dev/null +++ b/src/gallium/drivers/softpipe/sp_winsys.h @@ -0,0 +1,57 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* This is the interface that softpipe requires any window system + * hosting it to implement. This is the only include file in softpipe + * which is public. + */ + + +#ifndef SP_WINSYS_H +#define SP_WINSYS_H + + +#include "pipe/p_compiler.h" /* for boolean */ + +enum pipe_format; + +struct softpipe_winsys { + /** test if the given format is supported for front/back color bufs */ + boolean (*is_format_supported)( struct softpipe_winsys *sws, + enum pipe_format format ); + +}; + +struct pipe_winsys; +struct pipe_context; + + +struct pipe_context *softpipe_create( struct pipe_winsys *, + struct softpipe_winsys * ); + + +#endif /* SP_WINSYS_H */ |