diff options
author | Brian Paul <brianp@vmware.com> | 2009-02-10 16:44:02 -0700 |
---|---|---|
committer | Brian Paul <brianp@vmware.com> | 2009-02-10 16:44:02 -0700 |
commit | 5340b6dff73a0a23531ce2a5f28fba8303adab6e (patch) | |
tree | b141fc3648568dd8b941c966059e6ed32a8bd0ad /src/gallium/drivers/i965simple | |
parent | 9fd26daec24f21dbe17afcb2e2ab272667ee9a69 (diff) | |
parent | ee4c921b65fb76998711f3c40330505cbc49a0e0 (diff) |
Merge commit 'origin/gallium-master-merge'
This is the big merge of the gallium-0.2 branch into master.
gallium-master-merge was just the staging area for it.
Both gallium-0.2 and gallium-master-merge are considered closed now.
Conflicts:
progs/demos/Makefile
src/mesa/main/state.c
src/mesa/main/texenvprogram.c
Diffstat (limited to 'src/gallium/drivers/i965simple')
66 files changed, 18731 insertions, 0 deletions
diff --git a/src/gallium/drivers/i965simple/Makefile b/src/gallium/drivers/i965simple/Makefile new file mode 100644 index 0000000000..e97146e57c --- /dev/null +++ b/src/gallium/drivers/i965simple/Makefile @@ -0,0 +1,54 @@ +TOP = ../../../.. +include $(TOP)/configs/current + +LIBNAME = i965simple + +C_SOURCES = \ + brw_blit.c \ + brw_flush.c \ + brw_screen.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_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 + +include ../../Makefile.template + +symlinks: diff --git a/src/gallium/drivers/i965simple/SConscript b/src/gallium/drivers/i965simple/SConscript new file mode 100644 index 0000000000..43fc2a4005 --- /dev/null +++ b/src/gallium/drivers/i965simple/SConscript @@ -0,0 +1,54 @@ +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_screen.c', + 'brw_sf.c', + 'brw_sf_emit.c', + 'brw_sf_state.c', + 'brw_state.c', + 'brw_state_batch.c', + 'brw_state_cache.c', + 'brw_state_pool.c', + 'brw_state_upload.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..4d11f8d2ab --- /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/internal/p_winsys_screen.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..3668123e2e --- /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 "util/u_math.h" +#include "util/u_memory.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].writemask; + cc.cc1.stencil_test_mask = brw->attribs.DepthStencil->stencil[0].valuemask; + + 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].writemask; + cc.cc2.bf_stencil_test_mask = brw->attribs.DepthStencil->stencil[1].valuemask; + } + + /* Not really sure about this: + */ + if (brw->attribs.DepthStencil->stencil[0].writemask || + brw->attribs.DepthStencil->stencil[1].writemask) + 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); + + cc.cc7.alpha_ref.ub[0] = float_to_ubyte(brw->attribs.DepthStencil->alpha.ref_value); + + 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..d70fc094ff --- /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_MAX_ATTRIBS]; +}; + +#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..8e78dd51be --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_clip_state.c @@ -0,0 +1,93 @@ +/* + 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 "util/u_math.h" +#include "util/u_memory.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..c74cbf8d73 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_context.c @@ -0,0 +1,114 @@ +/* + 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/internal/p_winsys_screen.h" +#include "pipe/p_context.h" +#include "util/u_memory.h" +#include "pipe/p_screen.h" + + +#ifndef BRW_DEBUG +int BRW_DEBUG = (0); +#endif + + +static void brw_destroy(struct pipe_context *pipe) +{ + struct brw_context *brw = brw_context(pipe); + + if(brw->winsys->destroy) + brw->winsys->destroy(brw->winsys); + + 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); +} + + +struct pipe_context *brw_create(struct pipe_screen *screen, + struct brw_winsys *brw_winsys, + unsigned pci_id) +{ + struct brw_context *brw; + + debug_printf("%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 = screen->winsys; + brw->pipe.screen = screen; + + brw->pipe.destroy = brw_destroy; + brw->pipe.clear = brw_clear; + + brw_init_surface_functions(brw); + brw_init_texture_functions(brw); + brw_init_state_functions(brw); + brw_init_flush_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..3079485180 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_context.h @@ -0,0 +1,684 @@ +/* + 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 "tgsi/tgsi_scan.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) \ + debug_printf(__VA_ARGS__); \ +} while(0) + +#define PRINT(...) do { \ + debug_printf(__VA_ARGS__); \ +} while(0) + +struct brw_state_flags { + unsigned cache; + unsigned brw; +}; + + +struct brw_vertex_program { + struct pipe_shader_state program; + struct tgsi_shader_info info; + int id; +}; + + +struct brw_fragment_program { + struct pipe_shader_state program; + struct tgsi_shader_info info; + + boolean UsesDepth; /* XXX add this to tgsi_shader_info? */ + 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 stride; + unsigned depth_pitch; /* per-image on i945? */ + unsigned total_nblocksy; + + 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_MAX_ATTRIBS > 32. Wouldn't life + * be easier if C allowed arrays of packed elements? + */ +#define ATTRIB_BIT_DWORDS ((PIPE_MAX_ATTRIBS+31)/32) + + + + +struct brw_vertex_info { + unsigned varying; /* varying:1[PIPE_MAX_ATTRIBS] */ + unsigned sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[PIPE_MAX_ATTRIBS] */ +}; + + + + + +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; + + unsigned num_samplers; + unsigned num_textures; + + 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: + */ + const struct pipe_vertex_buffer *vbo_array[PIPE_MAX_ATTRIBS]; + + struct brw_vertex_element_state inputs[PIPE_MAX_ATTRIBS]; + +#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..904cde8e30 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_curbe.c @@ -0,0 +1,369 @@ +/* + 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/internal/p_winsys_screen.h" +#include "util/u_math.h" +#include "util/u_memory.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->buffer->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->buffer->size); + ws->buffer_unmap(ws, cbuffer->buffer); + offset += cbuffer->buffer->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..648aaa0da5 --- /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/internal/p_winsys_screen.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..2d9ca3f2ea --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_draw_upload.c @@ -0,0 +1,300 @@ +/************************************************************************** + * + * 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]->stride; + 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->info.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..e6001c30d9 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_flush.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: + * Keith Whitwell <keith@tungstengraphics.com> + */ + + +#include "pipe/p_defines.h" +#include "brw_context.h" +#include "brw_defines.h" +#include "brw_batch.h" + + +static void brw_flush( struct pipe_context *pipe, + unsigned flags, + struct pipe_fence_handle **fence ) +{ + struct brw_context *brw = brw_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)) { + 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 ); +} + + + +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..5b8016b2e9 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_gs_state.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 "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" +#include "util/u_math.h" +#include "util/u_memory.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..99ff4403a5 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_misc_state.c @@ -0,0 +1,488 @@ +/* + 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; + struct brw_texture *tex = (struct brw_texture *)depth_surface->texture; + assert(depth_surface->block.width == 1); + assert(depth_surface->block.height == 1); + switch (depth_surface->block.size) { + 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->stride - 1) | + (format << 18) | + (BRW_TILEWALK_YMAJOR << 26) | +// (depth_surface->region->tiled << 27) | + (BRW_SURFACE_2D << 29)); + OUT_RELOC(tex->buffer, + PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE, 0); + OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) | + ((depth_surface->stride/depth_surface->block.size - 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_screen.c b/src/gallium/drivers/i965simple/brw_screen.c new file mode 100644 index 0000000000..b22e105f10 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_screen.c @@ -0,0 +1,246 @@ +/************************************************************************** + * + * 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 "util/u_memory.h" +#include "pipe/internal/p_winsys_screen.h" +#include "util/u_string.h" +#include "util/u_simple_screen.h" + +#include "brw_context.h" +#include "brw_screen.h" +#include "brw_tex_layout.h" + + +static const char * +brw_get_vendor( struct pipe_screen *screen ) +{ + return "Tungsten Graphics, Inc."; +} + + +static const char * +brw_get_name( struct pipe_screen *screen ) +{ + static char buffer[128]; + const char *chipset; + + switch (brw_screen(screen)->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; + } + + util_snprintf(buffer, sizeof(buffer), "i965 (chipset: %s)", chipset); + return buffer; +} + + +static int +brw_get_param(struct pipe_screen *screen, 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_screen *screen, 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_screen *screen, + enum pipe_format format, + enum pipe_texture_target target, + unsigned tex_usage, + unsigned geom_flags ) +{ +#if 0 + /* XXX: This is broken -- rewrite if still needed. */ + static const unsigned tex_supported[] = { + PIPE_FORMAT_R8G8B8A8_UNORM, + PIPE_FORMAT_A8R8G8B8_UNORM, + PIPE_FORMAT_R5G6B5_UNORM, + PIPE_FORMAT_L8_UNORM, + PIPE_FORMAT_A8_UNORM, + PIPE_FORMAT_I8_UNORM, + PIPE_FORMAT_L8A8_UNORM, + 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_A8R8G8B8_UNORM, + PIPE_FORMAT_R5G6B5_UNORM, + }; + + /* + */ + static const unsigned z_stencil_supported[] = { + PIPE_FORMAT_Z16_UNORM, + PIPE_FORMAT_Z32_UNORM, + PIPE_FORMAT_S8Z24_UNORM, + }; + + 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 +} + + +static void +brw_destroy_screen( struct pipe_screen *screen ) +{ + struct pipe_winsys *winsys = screen->winsys; + + if(winsys->destroy) + winsys->destroy(winsys); + + FREE(screen); +} + + +/** + * Create a new brw_screen object + */ +struct pipe_screen * +brw_create_screen(struct pipe_winsys *winsys, uint pci_id) +{ + struct brw_screen *brwscreen = CALLOC_STRUCT(brw_screen); + + if (!brwscreen) + return NULL; + + brwscreen->pci_id = pci_id; + + brwscreen->screen.winsys = winsys; + + brwscreen->screen.destroy = brw_destroy_screen; + + brwscreen->screen.get_name = brw_get_name; + brwscreen->screen.get_vendor = brw_get_vendor; + brwscreen->screen.get_param = brw_get_param; + brwscreen->screen.get_paramf = brw_get_paramf; + brwscreen->screen.is_format_supported = brw_is_format_supported; + + brw_init_screen_texture_funcs(&brwscreen->screen); + u_simple_screen_init(&brwscreen->screen); + + return &brwscreen->screen; +} diff --git a/src/gallium/drivers/i965simple/brw_screen.h b/src/gallium/drivers/i965simple/brw_screen.h new file mode 100644 index 0000000000..d3c70387e6 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_screen.h @@ -0,0 +1,68 @@ +/************************************************************************** + * + * 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 BRW_SCREEN_H +#define BRW_SCREEN_H + + +#include "pipe/p_screen.h" + + +#ifdef __cplusplus +extern "C" { +#endif + + +/** + * Subclass of pipe_screen + */ +struct brw_screen +{ + struct pipe_screen screen; + + uint pci_id; +}; + + +/** cast wrapper */ +static INLINE struct brw_screen * +brw_screen(struct pipe_screen *pscreen) +{ + return (struct brw_screen *) pscreen; +} + + +extern struct pipe_screen * +brw_create_screen(struct pipe_winsys *winsys, uint pci_id); + + +#ifdef __cplusplus +} +#endif + +#endif /* BRW_SCREEN_H */ diff --git a/src/gallium/drivers/i965simple/brw_sf.c b/src/gallium/drivers/i965simple/brw_sf.c new file mode 100644 index 0000000000..b82a2e143b --- /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/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.file_max[TGSI_FILE_INPUT] + 1; + + + /* 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.DeclarationRange.First; + int last = parse.FullToken.FullDeclaration.DeclarationRange.Last; + int interp_mode = parse.FullToken.FullDeclaration.Declaration.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.DeclarationRange.First; + int last = parse.FullToken.FullDeclaration.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..2a5de61c21 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_sf_state.c @@ -0,0 +1,181 @@ +/* + 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 "util/u_math.h" +#include "util/u_memory.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..86d877d7ef --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_shader_info.c @@ -0,0 +1,48 @@ + +#include "brw_context.h" +#include "brw_state.h" +#include "util/u_memory.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/tgsi_parse.h" + + +/** + * XXX this obsolete new and no longer compiled. + */ +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->DeclarationRange.Last; + + // 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..b47f5373f3 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state.c @@ -0,0 +1,469 @@ +/************************************************************************** + * + * 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/internal/p_winsys_screen.h" +#include "util/u_memory.h" +#include "pipe/p_inlines.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_parse.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_states(struct pipe_context *pipe, + unsigned num, void **sampler) +{ + struct brw_context *brw = brw_context(pipe); + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == brw->num_samplers && + !memcmp(brw->attribs.Samplers, sampler, num * sizeof(void *))) + return; + + memcpy(brw->attribs.Samplers, sampler, num * sizeof(void *)); + memset(&brw->attribs.Samplers[num], 0, (PIPE_MAX_SAMPLERS - num) * + sizeof(void *)); + + brw->num_samplers = num; + + 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); + + brw_fp->program.tokens = tgsi_dup_tokens(shader->tokens); + brw_fp->id = brw_context(pipe)->program_id++; + + tgsi_scan_shader(shader->tokens, &brw_fp->info); + +#if 0 + brw_shader_info(shader->tokens, + &brw_fp->info2); +#endif + + 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) +{ + struct brw_fragment_program *brw_fp = (struct brw_fragment_program *) shader; + + FREE((void *) brw_fp->program.tokens); + FREE(brw_fp); +} + + +/************************************************************************ + * 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); + + brw_vp->program.tokens = tgsi_dup_tokens(shader->tokens); + brw_vp->id = brw_context(pipe)->program_id++; + + tgsi_scan_shader(shader->tokens, &brw_vp->info); + +#if 0 + brw_shader_info(shader->tokens, + &brw_vp->info2); +#endif + 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) +{ + struct brw_vertex_program *brw_vp = (struct brw_vertex_program *) shader; + + FREE((void *) brw_vp->program.tokens); + FREE(brw_vp); +} + + +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_buffers(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_buffer *buffers) +{ + struct brw_context *brw = brw_context(pipe); + memcpy(brw->vb.vbo_array, buffers, count * sizeof(buffers[0])); +} + +static void brw_set_vertex_elements(struct pipe_context *pipe, + unsigned count, + const struct pipe_vertex_element *elements) +{ + /* flush ? */ + struct brw_context *brw = brw_context(pipe); + uint i; + + assert(count <= PIPE_MAX_ATTRIBS); + + for (i = 0; i < count; i++) { + struct brw_vertex_element_state el; + memset(&el, 0, sizeof(el)); + + el.ve0.src_offset = elements[i].src_offset; + el.ve0.src_format = brw_translate_surface_format(elements[i].src_format); + el.ve0.valid = 1; + el.ve0.vertex_buffer_index = elements[i].vertex_buffer_index; + + el.ve1.dst_offset = i * 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 (elements[i].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[i] = 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_textures(struct pipe_context *pipe, + unsigned num, + struct pipe_texture **texture) +{ + struct brw_context *brw = brw_context(pipe); + uint i; + + assert(num <= PIPE_MAX_SAMPLERS); + + /* Check for no-op */ + if (num == brw->num_textures && + !memcmp(brw->attribs.Texture, texture, num * + sizeof(struct pipe_texture *))) + return; + + for (i = 0; i < num; i++) + pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i], + texture[i]); + + for (i = num; i < brw->num_textures; i++) + pipe_texture_reference((struct pipe_texture **) &brw->attribs.Texture[i], + NULL); + + brw->num_textures = num; + + 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_states = brw_bind_sampler_states; + 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_textures = brw_set_sampler_textures; + brw->pipe.set_viewport_state = brw_set_viewport_state; + brw->pipe.set_vertex_buffers = brw_set_vertex_buffers; + brw->pipe.set_vertex_elements = brw_set_vertex_elements; +} diff --git a/src/gallium/drivers/i965simple/brw_state.h b/src/gallium/drivers/i965simple/brw_state.h new file mode 100644 index 0000000000..de0a6371b8 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state.h @@ -0,0 +1,151 @@ +/* + 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 ); + + +#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..43a1c89fc4 --- /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 "util/u_memory.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..094248fa69 --- /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 "util/u_memory.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..e91263cb1f --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_state_pool.c @@ -0,0 +1,138 @@ +/* + 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/internal/p_winsys_screen.h" +#include "util/u_math.h" +#include "util/u_memory.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 = pipe_buffer_create(brw->pipe.screen, + 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.screen, + &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..bac9161b5f --- /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 "util/u_memory.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_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..0a95dce194 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_surface.c @@ -0,0 +1,127 @@ +/************************************************************************** + * + * 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_inlines.h" +#include "pipe/internal/p_winsys_screen.h" +#include "util/u_tile.h" +#include "util/u_rect.h" + + + +/* 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, + boolean 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->block.size == src->block.size ); + assert( dst->block.width == src->block.height ); + assert( dst->block.height == src->block.height ); + + if (0) { + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + + const void *src_map = pipe->screen->surface_map( pipe->screen, + src, + PIPE_BUFFER_USAGE_CPU_READ ); + + pipe_copy_rect(dst_map, + &dst->block, + dst->stride, + dstx, dsty, + width, height, + src_map, + do_flip ? -(int) src->stride : src->stride, + srcx, do_flip ? height - 1 - srcy : srcy); + + pipe->screen->surface_unmap(pipe->screen, src); + pipe->screen->surface_unmap(pipe->screen, dst); + } + else { + struct brw_texture *dst_tex = (struct brw_texture *)dst->texture; + struct brw_texture *src_tex = (struct brw_texture *)src->texture; + assert(dst->block.width == 1); + assert(dst->block.height == 1); + brw_copy_blit(brw_context(pipe), + do_flip, + dst->block.size, + (short) src->stride/src->block.size, src_tex->buffer, src->offset, FALSE, + (short) dst->stride/dst->block.size, dst_tex->buffer, dst->offset, FALSE, + (short) srcx, (short) srcy, (short) dstx, (short) dsty, + (short) width, (short) height, PIPE_LOGICOP_COPY); + } +} + + +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) { + void *dst_map = pipe->screen->surface_map( pipe->screen, + dst, + PIPE_BUFFER_USAGE_CPU_WRITE ); + + pipe_fill_rect(dst_map, &dst->block, dst->stride, dstx, dsty, width, height, value); + + pipe->screen->surface_unmap(pipe->screen, dst); + } + else { + struct brw_texture *tex = (struct brw_texture *)dst->texture; + assert(dst->block.width == 1); + assert(dst->block.height == 1); + brw_fill_blit(brw_context(pipe), + dst->block.size, + (short) dst->stride/dst->block.size, + tex->buffer, dst->offset, FALSE, + (short) dstx, (short) dsty, + (short) width, (short) height, + value); + } +} + + +void +brw_init_surface_functions(struct brw_context *brw) +{ + 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..448229ed4e --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_tex_layout.c @@ -0,0 +1,398 @@ +/* + 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 "pipe/p_state.h" +#include "pipe/p_context.h" +#include "pipe/p_defines.h" +#include "pipe/p_inlines.h" +#include "pipe/internal/p_winsys_screen.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "brw_context.h" +#include "brw_tex_layout.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 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] = y * tex->stride + x * pt->block.size; +} + +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; + + pt->nblocksx[level] = pf_get_nblocksx(&pt->block, w); + pt->nblocksy[level] = pf_get_nblocksy(&pt->block, h); + + tex->level_offset[level] = y * tex->stride + x * tex->base.block.size; + 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; + const int align_x = 2, align_y = 4; + unsigned level; + unsigned x = 0; + unsigned y = 0; + unsigned width = pt->width[0]; + unsigned height = pt->height[0]; + unsigned nblocksx = pt->nblocksx[0]; + unsigned nblocksy = pt->nblocksy[0]; + + tex->stride = align(pt->nblocksx[0] * pt->block.size, 4); + + /* 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_nblocksx + = align(pf_get_nblocksx(&pt->block, minify(width)), align_x) + + pf_get_nblocksx(&pt->block, minify(minify(width))); + + if (mip1_nblocksx > nblocksx) + tex->stride = mip1_nblocksx * pt->block.size; + } + + /* Pitch must be a whole number of dwords + */ + tex->stride = align(tex->stride, 64); + tex->total_nblocksy = 0; + + for (level = 0; level <= pt->last_level; level++) { + intel_miptree_set_level_info(tex, level, 1, x, y, width, + height, 1); + + nblocksy = align(nblocksy, align_y); + + /* Because the images are packed better, the final offset + * might not be the maximal one: + */ + tex->total_nblocksy = MAX2(tex->total_nblocksy, y + nblocksy); + + /* Layout_below: step right after second mipmap level. + */ + if (level == 1) { + x += align(nblocksx, align_x); + } + else { + y += nblocksy; + } + + width = minify(width); + height = minify(height); + nblocksx = pf_get_nblocksx(&pt->block, width); + nblocksy = pf_get_nblocksy(&pt->block, height); + } +} + +static boolean brw_miptree_layout(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 nblocksx = pt->nblocksx[0]; + unsigned nblocksy = pt->nblocksy[0]; + unsigned pack_x_pitch, pack_x_nr; + unsigned pack_y_pitch; + unsigned level; + unsigned align_h = 2; + unsigned align_w = 4; + + tex->total_nblocksy = 0; + + tex->stride = align(pt->nblocksx[0], 4); + pack_y_pitch = align(pt->nblocksy[0], align_h); + + pack_x_pitch = tex->stride / pt->block.size; + 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_nblocksy, + 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_nblocksy += y; + width = minify(width); + height = minify(height); + depth = minify(depth); + nblocksx = pf_get_nblocksx(&pt->block, width); + nblocksy = pf_get_nblocksy(&pt->block, height); + + 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 * pt->block.size <= tex->stride); + } + + 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_nblocksy, + pt->block.size, + pt->stride * pt->total_nblocksy ); +#endif + + return TRUE; +} + + +static struct pipe_texture * +brw_texture_create_screen(struct pipe_screen *screen, + const struct pipe_texture *templat) +{ + struct pipe_winsys *ws = screen->winsys; + struct brw_texture *tex = CALLOC_STRUCT(brw_texture); + + if (tex) { + tex->base = *templat; + tex->base.refcount = 1; + + tex->base.nblocksx[0] = pf_get_nblocksx(&tex->base.block, tex->base.width[0]); + tex->base.nblocksy[0] = pf_get_nblocksy(&tex->base.block, tex->base.height[0]); + + if (brw_miptree_layout(tex)) + tex->buffer = ws->buffer_create(ws, 64, + PIPE_BUFFER_USAGE_PIXEL, + tex->stride * + tex->total_nblocksy); + + if (!tex->buffer) { + FREE(tex); + return NULL; + } + } + + return &tex->base; +} + + +static void +brw_texture_release_screen(struct pipe_screen *screen, + 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(screen, &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; +} + + +static struct pipe_surface * +brw_get_tex_surface_screen(struct pipe_screen *screen, + 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]; + } + else if (pt->target == PIPE_TEXTURE_3D) { + offset += tex->image_offset[level][zslice]; + } + else { + assert(face == 0); + assert(zslice == 0); + } + + ps = CALLOC_STRUCT(pipe_surface); + if (ps) { + ps->refcount = 1; + pipe_texture_reference(&ps->texture, pt); + ps->format = pt->format; + ps->width = pt->width[level]; + ps->height = pt->height[level]; + ps->block = pt->block; + ps->nblocksx = pt->nblocksx[level]; + ps->nblocksy = pt->nblocksy[level]; + ps->stride = tex->stride; + ps->offset = offset; + ps->status = PIPE_SURFACE_STATUS_DEFINED; + } + return ps; +} + + +void +brw_init_texture_functions(struct brw_context *brw) +{ +// brw->pipe.texture_update = brw_texture_update; +} + + +void +brw_init_screen_texture_funcs(struct pipe_screen *screen) +{ + screen->texture_create = brw_texture_create_screen; + screen->texture_release = brw_texture_release_screen; + screen->get_tex_surface = brw_get_tex_surface_screen; +} + 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..a6b6ba8146 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_tex_layout.h @@ -0,0 +1,44 @@ +/* + 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. + + **********************************************************************/ + + +#ifndef BRW_TEX_LAYOUT_H +#define BRW_TEX_LAYOUT_H + + +struct brw_context; +struct pipe_screen; + + +extern void +brw_init_texture_functions(struct brw_context *brw); + +extern void +brw_init_screen_texture_funcs(struct pipe_screen *screen); + + +#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..92327e896d --- /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->info.num_outputs; + c.prog_data.inputs_read = vp->info.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..070f9dfcae --- /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; + + const 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..e03d653482 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs_emit.c @@ -0,0 +1,1330 @@ +/* + 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 "tgsi/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->info.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->info.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->DeclarationRange.First; + int last = decl->DeclarationRange.Last; + + 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: + case TGSI_OPCODE_SWZ: +#if 0 + /* 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.NrTokens == 4 + 1); + 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..1eaff87892 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_vs_state.c @@ -0,0 +1,103 @@ +/* + 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 "util/u_math.h" +#include "util/u_memory.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..ec1e400418 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_winsys.h @@ -0,0 +1,209 @@ +/************************************************************************** + * + * 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; +struct pipe_screen; + + +/* 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 { + + void (*destroy)(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_screen *, + 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..10161f2d2f --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm.c @@ -0,0 +1,209 @@ +/* + 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 "util/u_memory.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->info.uses_kill || + brw->attribs.DepthStencil->alpha.enabled) + lookup |= IZ_PS_KILL_ALPHATEST_BIT; + + if (fp->info.writes_z) + 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].writemask || + brw->attribs.DepthStencil->stencil[1].writemask) + 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_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..b29c4393f0 --- /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_MAX_ATTRIBS + 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_MAX_ATTRIBS) + +#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..d50e66f613 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_decl.c @@ -0,0 +1,392 @@ + +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_wm.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/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.file_max[TGSI_FILE_CONSTANT] + 1; + int index = 0; + + /* XXX number of constants, or highest numbered constant? */ + assert(nr_constants == c->fp->info.file_count[TGSI_FILE_CONSTANT]); + + 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: + */ + assert(c->fp->info.file_max[TGSI_FILE_INPUT] == c->fp->info.num_inputs); + for (i = 0; i < c->fp->info.file_max[TGSI_FILE_INPUT] + 1; 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->info.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: + */ + /* XXX do we want to loop over the _number_ of inputs/outputs or loop + * to the highest input/output index that's used? + * Probably the same, actually. + */ + assert(c->fp->info.file_max[TGSI_FILE_INPUT] + 1 == c->fp->info.num_inputs); + assert(c->fp->info.file_max[TGSI_FILE_OUTPUT] + 1 == c->fp->info.num_outputs); + for (i = 0; i < c->fp->info.file_max[TGSI_FILE_INPUT] + 1; 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.file_max[TGSI_FILE_OUTPUT] + 1; 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->DeclarationRange.First; + unsigned last = decl->DeclarationRange.Last; + unsigned mask = decl->Declaration.UsageMask; /* ? */ + unsigned i; + + if (decl->Declaration.File != TGSI_FILE_INPUT) + break; + + for( i = first; i <= last; i++ ) { + switch (decl->Declaration.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..ab6410aa60 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_glsl.c @@ -0,0 +1,1076 @@ + +#include "brw_context.h" +#include "brw_eu.h" +#include "brw_wm.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "pipe/p_shader_tokens.h" +#include "tgsi/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); + + /* 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..52b2909a65 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_sampler_state.c @@ -0,0 +1,275 @@ +/* + 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 "util/u_math.h" +#include "util/u_memory.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; + case PIPE_TEX_FILTER_ANISO: + sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC; + 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: + */ + 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; + case PIPE_TEX_FILTER_ANISO: + sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR; + break; + default: + break; + } + + if (pipe_sampler->max_anisotropy > 2.0) { + sampler->ss3.max_aniso = MAX2((pipe_sampler->max_anisotropy - 2) / 2, + BRW_ANISORATIO_16); + } + + 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->num_textures && unit < brw->num_samplers; + 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..37a9bf919c --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_state.c @@ -0,0 +1,195 @@ +/* + 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 "util/u_math.h" +#include "util/u_memory.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->info.writes_z) + wm.wm5.program_computes_depth = 1; + + /* BRW_NEW_ALPHA_TEST */ + if (fp->info.uses_kill || + 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..1bab5bfdb3 --- /dev/null +++ b/src/gallium/drivers/i965simple/brw_wm_surface_state.c @@ -0,0 +1,305 @@ +/* + 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_L8_UNORM: + return BRW_SURFACEFORMAT_L8_UNORM; + + case PIPE_FORMAT_I8_UNORM: + return BRW_SURFACEFORMAT_I8_UNORM; + + case PIPE_FORMAT_A8_UNORM: + return BRW_SURFACEFORMAT_A8_UNORM; + + case PIPE_FORMAT_A8L8_UNORM: + 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->stride - 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*/ + struct brw_texture *tex = (struct brw_texture *)pipe_surface->texture; + + memset(&surf, 0, sizeof(surf)); + + if (pipe_surface != NULL) { + if (pipe_surface->block.size == 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, tex->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->stride - 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->num_textures && i < brw->num_samplers; 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, +}; |