From 2f5f7c07732577f60666e3cee69c75c9b035c145 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 23 Oct 2009 16:55:02 +0100 Subject: i965g: re-starting from the dri driver --- src/gallium/drivers/i965/brw_eu.c | 254 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 254 insertions(+) create mode 100644 src/gallium/drivers/i965/brw_eu.c (limited to 'src/gallium/drivers/i965/brw_eu.c') diff --git a/src/gallium/drivers/i965/brw_eu.c b/src/gallium/drivers/i965/brw_eu.c new file mode 100644 index 0000000000..1df561386e --- /dev/null +++ b/src/gallium/drivers/i965/brw_eu.c @@ -0,0 +1,254 @@ +/* + 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 + */ + + +#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, GLuint 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, GLuint pc ) +{ + p->current->header.predicate_control = pc; +} + +void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional ) +{ + p->current->header.destreg__conditionalmod = conditional; +} + +void brw_set_access_mode( struct brw_compile *p, GLuint access_mode ) +{ + p->current->header.access_mode = access_mode; +} + +void brw_set_compression_control( struct brw_compile *p, GLboolean compression_control ) +{ + p->current->header.compression_control = compression_control; +} + +void brw_set_mask_control( struct brw_compile *p, GLuint value ) +{ + p->current->header.mask_control = value; +} + +void brw_set_saturate( struct brw_compile *p, GLuint 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_context *brw, struct brw_compile *p ) +{ + p->brw = brw; + 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 GLuint *brw_get_program( struct brw_compile *p, + GLuint *sz ) +{ + GLuint i; + + for (i = 0; i < 8; i++) + brw_NOP(p); + + *sz = p->nr_insn * sizeof(struct brw_instruction); + return (const GLuint *)p->store; +} + + + +/** + * Subroutine calls require special attention. + * Mesa instructions may be expanded into multiple hardware instructions + * so the prog_instruction::BranchTarget field can't be used as an index + * into the hardware instructions. + * + * The BranchTarget field isn't needed, however. Mesa's GLSL compiler + * emits CAL and BGNSUB instructions with labels that can be used to map + * subroutine calls to actual subroutine code blocks. + * + * The structures and function here implement patching of CAL instructions + * so they jump to the right subroutine code... + */ + + +/** + * For each OPCODE_BGNSUB we create one of these. + */ +struct brw_glsl_label +{ + const char *name; /**< the label string */ + GLuint position; /**< the position of the brw instruction for this label */ + struct brw_glsl_label *next; /**< next in linked list */ +}; + + +/** + * For each OPCODE_CAL we create one of these. + */ +struct brw_glsl_call +{ + GLuint call_inst_pos; /**< location of the CAL instruction */ + const char *sub_name; /**< name of subroutine to call */ + struct brw_glsl_call *next; /**< next in linked list */ +}; + + +/** + * Called for each OPCODE_BGNSUB. + */ +void +brw_save_label(struct brw_compile *c, const char *name, GLuint position) +{ + struct brw_glsl_label *label = CALLOC_STRUCT(brw_glsl_label); + label->name = name; + label->position = position; + label->next = c->first_label; + c->first_label = label; +} + + +/** + * Called for each OPCODE_CAL. + */ +void +brw_save_call(struct brw_compile *c, const char *name, GLuint call_pos) +{ + struct brw_glsl_call *call = CALLOC_STRUCT(brw_glsl_call); + call->call_inst_pos = call_pos; + call->sub_name = name; + call->next = c->first_call; + c->first_call = call; +} + + +/** + * Lookup a label, return label's position/offset. + */ +static GLuint +brw_lookup_label(struct brw_compile *c, const char *name) +{ + const struct brw_glsl_label *label; + for (label = c->first_label; label; label = label->next) { + if (strcmp(name, label->name) == 0) { + return label->position; + } + } + abort(); /* should never happen */ + return ~0; +} + + +/** + * When we're done generating code, this function is called to resolve + * subroutine calls. + */ +void +brw_resolve_cals(struct brw_compile *c) +{ + const struct brw_glsl_call *call; + + for (call = c->first_call; call; call = call->next) { + const GLuint sub_loc = brw_lookup_label(c, call->sub_name); + struct brw_instruction *brw_call_inst = &c->store[call->call_inst_pos]; + struct brw_instruction *brw_sub_inst = &c->store[sub_loc]; + GLint offset = brw_sub_inst - brw_call_inst; + + /* patch brw_inst1 to point to brw_inst2 */ + brw_set_src1(brw_call_inst, brw_imm_d(offset * 16)); + } + + /* free linked list of calls */ + { + struct brw_glsl_call *call, *next; + for (call = c->first_call; call; call = next) { + next = call->next; + _mesa_free(call); + } + c->first_call = NULL; + } + + /* free linked list of labels */ + { + struct brw_glsl_label *label, *next; + for (label = c->first_label; label; label = next) { + next = label->next; + _mesa_free(label); + } + c->first_label = NULL; + } +} -- cgit v1.2.3 From 4dd2f6640b70e2313f8771f7588aa49a861153aa Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Sun, 25 Oct 2009 00:02:16 +0100 Subject: i965g: more work on compiling, particularly the brw_draw files --- src/gallium/auxiliary/util/u_debug.c | 27 +++ src/gallium/auxiliary/util/u_prim.h | 2 + src/gallium/auxiliary/util/u_upload_mgr.h | 2 + src/gallium/drivers/i965/Makefile | 2 +- src/gallium/drivers/i965/brw_batchbuffer.c | 198 +++++++++++++++ src/gallium/drivers/i965/brw_batchbuffer.h | 14 +- src/gallium/drivers/i965/brw_cc.c | 8 +- src/gallium/drivers/i965/brw_clip.c | 4 +- src/gallium/drivers/i965/brw_clip_state.c | 4 +- src/gallium/drivers/i965/brw_context.c | 2 +- src/gallium/drivers/i965/brw_context.h | 68 ++++-- src/gallium/drivers/i965/brw_curbe.c | 13 +- src/gallium/drivers/i965/brw_draw.c | 165 +++++++------ src/gallium/drivers/i965/brw_draw.h | 3 +- src/gallium/drivers/i965/brw_draw_upload.c | 372 +++++++++++++++++------------ src/gallium/drivers/i965/brw_eu.c | 5 +- src/gallium/drivers/i965/brw_eu_debug.c | 13 +- src/gallium/drivers/i965/brw_misc_state.c | 18 +- src/gallium/drivers/i965/brw_pipe_flush.c | 3 + src/gallium/drivers/i965/brw_pipe_shader.c | 19 ++ src/gallium/drivers/i965/brw_pipe_vertex.c | 25 +- src/gallium/drivers/i965/brw_screen.h | 22 ++ src/gallium/drivers/i965/brw_sf.c | 2 +- src/gallium/drivers/i965/brw_sf_state.c | 39 +-- src/gallium/drivers/i965/brw_state.h | 6 +- src/gallium/drivers/i965/brw_state_batch.c | 4 +- src/gallium/drivers/i965/brw_swtnl.c | 6 +- src/gallium/drivers/i965/brw_winsys.h | 7 + src/gallium/drivers/i965/brw_wm.c | 2 +- src/gallium/drivers/i965/brw_wm.h | 8 +- src/gallium/drivers/i965/brw_wm_glsl.c | 28 --- src/gallium/drivers/i965/brw_wm_pass0.c | 32 +-- src/mesa/state_tracker/st_draw.c | 3 +- 33 files changed, 722 insertions(+), 404 deletions(-) create mode 100644 src/gallium/drivers/i965/brw_batchbuffer.c (limited to 'src/gallium/drivers/i965/brw_eu.c') diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c index 96d400c839..321ac59a7d 100644 --- a/src/gallium/auxiliary/util/u_debug.c +++ b/src/gallium/auxiliary/util/u_debug.c @@ -69,6 +69,7 @@ #include "util/u_stream.h" #include "util/u_math.h" #include "util/u_tile.h" +#include "util/u_prim.h" #ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY @@ -600,6 +601,32 @@ const char *pf_name( enum pipe_format format ) } + +static const struct debug_named_value pipe_prim_names[] = { +#ifdef DEBUG + DEBUG_NAMED_VALUE(PIPE_PRIM_POINTS), + DEBUG_NAMED_VALUE(PIPE_PRIM_LINES), + DEBUG_NAMED_VALUE(PIPE_PRIM_LINE_LOOP), + DEBUG_NAMED_VALUE(PIPE_PRIM_LINE_STRIP), + DEBUG_NAMED_VALUE(PIPE_PRIM_TRIANGLES), + DEBUG_NAMED_VALUE(PIPE_PRIM_TRIANGLE_STRIP), + DEBUG_NAMED_VALUE(PIPE_PRIM_TRIANGLE_FAN), + DEBUG_NAMED_VALUE(PIPE_PRIM_QUADS), + DEBUG_NAMED_VALUE(PIPE_PRIM_QUAD_STRIP), + DEBUG_NAMED_VALUE(PIPE_PRIM_POLYGON), +#endif + DEBUG_NAMED_VALUE_END +}; + + +const char *u_prim_name( unsigned prim ) +{ + return debug_dump_enum(pipe_prim_names, prim); +} + + + + #ifdef DEBUG void debug_dump_image(const char *prefix, unsigned format, unsigned cpp, diff --git a/src/gallium/auxiliary/util/u_prim.h b/src/gallium/auxiliary/util/u_prim.h index a9b533eea7..7434329962 100644 --- a/src/gallium/auxiliary/util/u_prim.h +++ b/src/gallium/auxiliary/util/u_prim.h @@ -135,4 +135,6 @@ static INLINE unsigned u_reduced_prim( unsigned pipe_prim ) } } +const char *u_prim_name( unsigned pipe_prim ); + #endif diff --git a/src/gallium/auxiliary/util/u_upload_mgr.h b/src/gallium/auxiliary/util/u_upload_mgr.h index 745b5834af..d414a1f2f6 100644 --- a/src/gallium/auxiliary/util/u_upload_mgr.h +++ b/src/gallium/auxiliary/util/u_upload_mgr.h @@ -32,6 +32,8 @@ #ifndef U_UPLOAD_MGR_H #define U_UPLOAD_MGR_H +#include "pipe/p_error.h" + struct pipe_screen; struct pipe_buffer; struct u_upload_mgr; diff --git a/src/gallium/drivers/i965/Makefile b/src/gallium/drivers/i965/Makefile index 40c8364824..40e8aa8786 100644 --- a/src/gallium/drivers/i965/Makefile +++ b/src/gallium/drivers/i965/Makefile @@ -61,7 +61,7 @@ C_SOURCES = \ brw_wm_state.c \ brw_wm_surface_state.c \ brw_bo.c \ - intel_batchbuffer.c \ + brw_batchbuffer.c \ intel_tex_layout.c include ../../Makefile.template diff --git a/src/gallium/drivers/i965/brw_batchbuffer.c b/src/gallium/drivers/i965/brw_batchbuffer.c new file mode 100644 index 0000000000..8bcac76ede --- /dev/null +++ b/src/gallium/drivers/i965/brw_batchbuffer.c @@ -0,0 +1,198 @@ +/************************************************************************** + * + * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "brw_batchbuffer.h" +#include "brw_decode.h" +#include "brw_reg.h" +#include "brw_winsys.h" + + +void +brw_batchbuffer_reset(struct brw_batchbuffer *batch) +{ + struct intel_context *intel = batch->intel; + + if (batch->buf != NULL) { + brw->sws->bo_unreference(batch->buf); + batch->buf = NULL; + } + + if (!batch->buffer && intel->ttm == GL_TRUE) + batch->buffer = malloc (intel->maxBatchSize); + + batch->buf = batch->sws->bo_alloc(batch->sws, + BRW_BUFFER_TYPE_BATCH, + intel->maxBatchSize, 4096); + if (batch->buffer) + batch->map = batch->buffer; + else { + batch->sws->bo_map(batch->buf, GL_TRUE); + batch->map = batch->buf->virtual; + } + batch->size = intel->maxBatchSize; + batch->ptr = batch->map; + batch->dirty_state = ~0; + batch->cliprect_mode = IGNORE_CLIPRECTS; +} + +struct brw_batchbuffer * +brw_batchbuffer_alloc(struct brw_winsys_screen *sws) +{ + struct brw_batchbuffer *batch = CALLOC_STRUCT(brw_batchbuffer); + + batch->sws = sws; + brw_batchbuffer_reset(batch); + + return batch; +} + +void +brw_batchbuffer_free(struct brw_batchbuffer *batch) +{ + if (batch->map) { + dri_bo_unmap(batch->buf); + batch->map = NULL; + } + + brw->sws->bo_unreference(batch->buf); + batch->buf = NULL; + FREE(batch); +} + + +void +_brw_batchbuffer_flush(struct brw_batchbuffer *batch, const char *file, + int line) +{ + struct intel_context *intel = batch->intel; + GLuint used = batch->ptr - batch->map; + + if (used == 0) + return; + + if (intel->first_post_swapbuffers_batch == NULL) { + intel->first_post_swapbuffers_batch = intel->batch->buf; + batch->sws->bo_reference(intel->first_post_swapbuffers_batch); + } + + if (intel->first_post_swapbuffers_batch == NULL) { + intel->first_post_swapbuffers_batch = intel->batch->buf; + batch->sws->bo_reference(intel->first_post_swapbuffers_batch); + } + + + if (INTEL_DEBUG & DEBUG_BATCH) + fprintf(stderr, "%s:%d: Batchbuffer flush with %db used\n", file, line, + used); + + /* Emit a flush if the bufmgr doesn't do it for us. */ + if (intel->always_flush_cache || !intel->ttm) { + *(GLuint *) (batch->ptr) = intel->vtbl.flush_cmd(); + batch->ptr += 4; + used = batch->ptr - batch->map; + } + + /* Round batchbuffer usage to 2 DWORDs. */ + + if ((used & 4) == 0) { + *(GLuint *) (batch->ptr) = 0; /* noop */ + batch->ptr += 4; + used = batch->ptr - batch->map; + } + + /* Mark the end of the buffer. */ + *(GLuint *) (batch->ptr) = MI_BATCH_BUFFER_END; /* noop */ + batch->ptr += 4; + used = batch->ptr - batch->map; + + batch->sws->bo_unmap(batch->buf); + + batch->map = NULL; + batch->ptr = NULL; + + batch->sws->bo_exec(batch->buf, used, NULL, 0, 0 ); + + if (INTEL_DEBUG & DEBUG_BATCH) { + dri_bo_map(batch->buf, GL_FALSE); + intel_decode(batch->buf->virtual, used / 4, batch->buf->offset, + brw->brw_screen->pci_id); + dri_bo_unmap(batch->buf); + } + + if (INTEL_DEBUG & DEBUG_SYNC) { + fprintf(stderr, "waiting for idle\n"); + dri_bo_map(batch->buf, GL_TRUE); + dri_bo_unmap(batch->buf); + } + + /* Reset the buffer: + */ + brw_batchbuffer_reset(batch); +} + + +/* This is the only way buffers get added to the validate list. + */ +GLboolean +brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch, + struct brw_winsys_buffer *buffer, + uint32_t read_domains, uint32_t write_domain, + uint32_t delta) +{ + int ret; + + if (batch->ptr - batch->map > batch->buf->size) + _mesa_printf ("bad relocation ptr %p map %p offset %d size %d\n", + batch->ptr, batch->map, batch->ptr - batch->map, batch->buf->size); + + ret = batch->sws->bo_emit_reloc(batch->buf, + read_domains, + write_domain, + delta, + batch->ptr - batch->map, + buffer); + + /* + * Using the old buffer offset, write in what the right data would be, in case + * the buffer doesn't move and we can short-circuit the relocation processing + * in the kernel + */ + brw_batchbuffer_emit_dword (batch, buffer->offset + delta); + + return GL_TRUE; +} + +void +brw_batchbuffer_data(struct brw_batchbuffer *batch, + const void *data, GLuint bytes, + enum cliprect_mode cliprect_mode) +{ + assert((bytes & 3) == 0); + brw_batchbuffer_require_space(batch, bytes); + __memcpy(batch->ptr, data, bytes); + batch->ptr += bytes; +} diff --git a/src/gallium/drivers/i965/brw_batchbuffer.h b/src/gallium/drivers/i965/brw_batchbuffer.h index b8492882e1..25bb9cefca 100644 --- a/src/gallium/drivers/i965/brw_batchbuffer.h +++ b/src/gallium/drivers/i965/brw_batchbuffer.h @@ -33,18 +33,16 @@ void brw_batchbuffer_reset(struct brw_batchbuffer *batch); * Consider it a convenience function wrapping multple * intel_buffer_dword() calls. */ -void brw_batchbuffer_data(struct brw_batchbuffer *batch, +int brw_batchbuffer_data(struct brw_batchbuffer *batch, const void *data, GLuint bytes, enum cliprect_mode cliprect_mode); -void brw_batchbuffer_release_space(struct brw_batchbuffer *batch, - GLuint bytes); -GLboolean brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch, - struct brw_winsys_buffer *buffer, - uint32_t read_domains, - uint32_t write_domain, - uint32_t offset); +int brw_batchbuffer_emit_reloc(struct brw_batchbuffer *batch, + struct brw_winsys_buffer *buffer, + uint32_t read_domains, + uint32_t write_domain, + uint32_t offset); /* Inline functions - might actually be better off with these * non-inlined. Certainly better off switching all command packets to diff --git a/src/gallium/drivers/i965/brw_cc.c b/src/gallium/drivers/i965/brw_cc.c index 76759304eb..ca10bc73f6 100644 --- a/src/gallium/drivers/i965/brw_cc.c +++ b/src/gallium/drivers/i965/brw_cc.c @@ -57,7 +57,7 @@ static void calc_sane_viewport( const struct pipe_viewport_state *vp, svp->far = 1; } -static void prepare_cc_vp( struct brw_context *brw ) +static int prepare_cc_vp( struct brw_context *brw ) { struct brw_cc_viewport ccv; struct sane_viewport svp; @@ -72,6 +72,8 @@ static void prepare_cc_vp( struct brw_context *brw ) brw->sws->bo_unreference(brw->cc.vp_bo); brw->cc.vp_bo = brw_cache_data( &brw->cache, BRW_CC_VP, &ccv, NULL, 0 ); + + return 0; } const struct brw_tracked_state brw_cc_vp = { @@ -158,7 +160,7 @@ cc_unit_create_from_key(struct brw_context *brw, struct brw_cc_unit_key *key) return bo; } -static void prepare_cc_unit( struct brw_context *brw ) +static int prepare_cc_unit( struct brw_context *brw ) { struct brw_cc_unit_key key; @@ -172,6 +174,8 @@ static void prepare_cc_unit( struct brw_context *brw ) if (brw->cc.state_bo == NULL) brw->cc.state_bo = cc_unit_create_from_key(brw, &key); + + return 0; } const struct brw_tracked_state brw_cc_unit = { diff --git a/src/gallium/drivers/i965/brw_clip.c b/src/gallium/drivers/i965/brw_clip.c index 622d9dba96..1a52fa771b 100644 --- a/src/gallium/drivers/i965/brw_clip.c +++ b/src/gallium/drivers/i965/brw_clip.c @@ -146,7 +146,7 @@ static void compile_clip_prog( struct brw_context *brw, /* Calculate interpolants for triangle and line rasterization. */ -static void upload_clip_prog(struct brw_context *brw) +static int upload_clip_prog(struct brw_context *brw) { struct brw_clip_prog_key key; @@ -173,6 +173,8 @@ static void upload_clip_prog(struct brw_context *brw) &brw->clip.prog_data); if (brw->clip.prog_bo == NULL) compile_clip_prog( brw, &key ); + + return 0; } diff --git a/src/gallium/drivers/i965/brw_clip_state.c b/src/gallium/drivers/i965/brw_clip_state.c index 25b8c6372f..bf4e6f5103 100644 --- a/src/gallium/drivers/i965/brw_clip_state.c +++ b/src/gallium/drivers/i965/brw_clip_state.c @@ -159,7 +159,7 @@ clip_unit_create_from_key(struct brw_context *brw, return bo; } -static void upload_clip_unit( struct brw_context *brw ) +static int upload_clip_unit( struct brw_context *brw ) { struct brw_clip_unit_key key; @@ -173,6 +173,8 @@ static void upload_clip_unit( struct brw_context *brw ) if (brw->clip.state_bo == NULL) { brw->clip.state_bo = clip_unit_create_from_key(brw, &key); } + + return 0; } const struct brw_tracked_state brw_clip_unit = { diff --git a/src/gallium/drivers/i965/brw_context.c b/src/gallium/drivers/i965/brw_context.c index e9605bafe6..e10b7d8bf5 100644 --- a/src/gallium/drivers/i965/brw_context.c +++ b/src/gallium/drivers/i965/brw_context.c @@ -105,7 +105,7 @@ struct pipe_context *brw_create_context(struct pipe_screen *screen) brw->state.dirty.mesa = ~0; brw->state.dirty.brw = ~0; - brw->emit_state_always = 0; + brw->flags.always_emit_state = 0; make_empty_list(&brw->query.active_head); diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h index dd782fdba9..7ead641811 100644 --- a/src/gallium/drivers/i965/brw_context.h +++ b/src/gallium/drivers/i965/brw_context.h @@ -182,6 +182,8 @@ struct brw_fragment_shader { #define PIPE_NEW_FRAGMENT_CONSTANTS 0x2 #define PIPE_NEW_VERTEX_CONSTANTS 0x2 #define PIPE_NEW_CLIP 0x2 +#define PIPE_NEW_INDEX_BUFFER 0x2 +#define PIPE_NEW_INDEX_RANGE 0x2 #define BRW_NEW_URB_FENCE 0x1 @@ -387,8 +389,8 @@ struct brw_cache { */ struct brw_tracked_state { struct brw_state_flags dirty; - void (*prepare)( struct brw_context *brw ); - void (*emit)( struct brw_context *brw ); + int (*prepare)( struct brw_context *brw ); + int (*emit)( struct brw_context *brw ); }; /* Flags for brw->state.cache. @@ -465,9 +467,7 @@ struct brw_context GLuint primitive; GLuint reduced_primitive; - GLboolean emit_state_always; - - /* Active vertex program: + /* Active state from the state tracker: */ struct { const struct brw_vertex_shader *vertex_shader; @@ -475,11 +475,31 @@ struct brw_context const struct brw_blend_state *blend; const struct brw_rasterizer_state *rast; const struct brw_depth_stencil_alpha_state *zstencil; + + struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; + struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; + unsigned num_vertex_elements; + unsigned num_vertex_buffers; + struct pipe_framebuffer_state fb; struct pipe_viewport_state vp; struct pipe_clip_state ucp; struct pipe_buffer *vertex_constants; struct pipe_buffer *fragment_constants; + + /** + * Index buffer for this draw_prims call. + * + * Updates are signaled by PIPE_NEW_INDEX_BUFFER. + */ + struct pipe_buffer *index_buffer; + unsigned index_size; + + /* Updates are signalled by PIPE_NEW_INDEX_RANGE: + */ + unsigned min_index; + unsigned max_index; + } curr; struct { @@ -504,30 +524,26 @@ struct brw_context struct brw_cached_batch_item *cached_batch_items; struct { - struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS]; - struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; - unsigned num_vertex_element; - unsigned num_vertex_buffer; - struct u_upload_mgr *upload_vertex; struct u_upload_mgr *upload_index; - - /* Summary of size and varying of active arrays, so we can check - * for changes to this state: + /* Information on uploaded vertex buffers: */ - struct brw_vertex_info info; - unsigned int min_index, max_index; + struct { + unsigned stride; /* in bytes between successive vertices */ + unsigned offset; /* in bytes, of first vertex in bo */ + unsigned vertex_count; /* count of valid vertices which may be accessed */ + struct brw_winsys_buffer *bo; + } vb[PIPE_MAX_ATTRIBS]; + + struct { + } ve[PIPE_MAX_ATTRIBS]; + + unsigned nr_vb; /* currently the same as curr.num_vertex_buffers */ + unsigned nr_ve; /* currently the same as curr.num_vertex_elements */ } vb; struct { - /** - * Index buffer for this draw_prims call. - * - * Updates are signaled by BRW_NEW_INDICES. - */ - const struct _mesa_index_buffer *ib; - /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */ struct brw_winsys_buffer *bo; unsigned int offset; @@ -668,6 +684,14 @@ struct brw_context int index; GLboolean active; } query; + + struct { + unsigned always_emit_state:1; + unsigned always_flush_batch:1; + unsigned force_swtnl:1; + unsigned no_swtnl:1; + } flags; + /* Used to give every program string a unique id */ GLuint program_id; diff --git a/src/gallium/drivers/i965/brw_curbe.c b/src/gallium/drivers/i965/brw_curbe.c index edc39ff223..278ffa4ca2 100644 --- a/src/gallium/drivers/i965/brw_curbe.c +++ b/src/gallium/drivers/i965/brw_curbe.c @@ -48,7 +48,7 @@ * constants. That greatly reduces the demand for space in the CURBE. * Some of the comments within are dated... */ -static void calculate_curbe_offsets( struct brw_context *brw ) +static int calculate_curbe_offsets( struct brw_context *brw ) { /* CACHE_NEW_WM_PROG */ const GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16; @@ -104,6 +104,8 @@ static void calculate_curbe_offsets( struct brw_context *brw ) brw->state.dirty.brw |= BRW_NEW_CURBE_OFFSETS; } + + return 0; } @@ -157,7 +159,7 @@ static GLfloat fixed_plane[6][4] = { * cache mechanism, but maybe would benefit from a comparison against * the current uploaded set of constants. */ -static void prepare_constant_buffer(struct brw_context *brw) +static int prepare_constant_buffer(struct brw_context *brw) { const GLuint sz = brw->curbe.total_size; const GLuint bufsz = sz * 16 * sizeof(GLfloat); @@ -170,7 +172,7 @@ static void prepare_constant_buffer(struct brw_context *brw) brw->curbe.last_buf = NULL; brw->curbe.last_bufsz = 0; } - return; + return 0; } buf = (GLfloat *) CALLOC(bufsz, 1); @@ -305,9 +307,11 @@ static void prepare_constant_buffer(struct brw_context *brw) * flushes as necessary when doublebuffering of CURBEs isn't * possible. */ + + return 0; } -static void emit_constant_buffer(struct brw_context *brw) +static int emit_constant_buffer(struct brw_context *brw) { GLuint sz = brw->curbe.total_size; @@ -322,6 +326,7 @@ static void emit_constant_buffer(struct brw_context *brw) (sz - 1) + brw->curbe.curbe_offset); } ADVANCE_BATCH(); + return 0; } const struct brw_tracked_state brw_constant_buffer = { diff --git a/src/gallium/drivers/i965/brw_draw.c b/src/gallium/drivers/i965/brw_draw.c index 7af490bc5a..b5fe7c9601 100644 --- a/src/gallium/drivers/i965/brw_draw.c +++ b/src/gallium/drivers/i965/brw_draw.c @@ -26,15 +26,18 @@ **************************************************************************/ +#include "util/u_prim.h" +#include "util/u_upload_mgr.h" + #include "brw_draw.h" #include "brw_defines.h" #include "brw_context.h" #include "brw_state.h" #include "brw_debug.h" +#include "brw_screen.h" #include "brw_batchbuffer.h" -#define FILE_DEBUG_FLAG DEBUG_BATCH static uint32_t prim_to_hw_prim[PIPE_PRIM_POLYGON+1] = { _3DPRIM_POINTLIST, @@ -56,18 +59,21 @@ static uint32_t prim_to_hw_prim[PIPE_PRIM_POLYGON+1] = { * programs be immune to the active primitive (ie. cope with all * possibilities). That may not be realistic however. */ -static GLuint brw_set_prim(struct brw_context *brw, unsigned prim) +static int brw_set_prim(struct brw_context *brw, unsigned prim ) { if (BRW_DEBUG & DEBUG_PRIMS) debug_printf("PRIM: %s\n", u_prim_name(prim)); if (prim != brw->primitive) { + unsigned reduced_prim; + brw->primitive = prim; brw->state.dirty.brw |= BRW_NEW_PRIMITIVE; - if (reduced_prim[prim] != brw->reduced_primitive) { - brw->reduced_primitive = reduced_prim[prim]; + reduced_prim = u_reduced_prim(prim); + if (reduced_prim != brw->reduced_primitive) { + brw->reduced_primitive = reduced_prim; brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE; } } @@ -77,17 +83,14 @@ static GLuint brw_set_prim(struct brw_context *brw, unsigned prim) -static enum pipe_error brw_emit_prim(struct brw_context *brw, - unsigned prim, - unsigned start, - unsigned count, - boolean indexed, - uint32_t hw_prim) +static int brw_emit_prim(struct brw_context *brw, + unsigned start, + unsigned count, + boolean indexed, + uint32_t hw_prim) { struct brw_3d_primitive prim_packet; - - if (INTEL_DEBUG & DEBUG_PRIMS) - debug_printf("PRIM: %s %d %d\n", u_prim_name(prim), start, count); + int ret; prim_packet.header.opcode = CMD_3D_PRIM; prim_packet.header.length = sizeof(prim_packet)/4 - 2; @@ -101,7 +104,7 @@ static enum pipe_error brw_emit_prim(struct brw_context *brw, prim_packet.start_vert_location += brw->ib.start_vertex_offset; prim_packet.instance_count = 1; prim_packet.start_instance_location = 0; - prim_packet.base_vert_location = prim->basevertex; + prim_packet.base_vert_location = 0; // prim->basevertex; XXX: add this to gallium /* If we're set to always flush, do it before and after the primitive emit. @@ -109,20 +112,20 @@ static enum pipe_error brw_emit_prim(struct brw_context *brw, * and missed flushes of the render cache as it heads to other parts of * the besides the draw code. */ - if (intel->always_flush_cache) { - BEGIN_BATCH(1, IGNORE_CLIPRECTS) - OUT_BATCH(intel->vtbl.flush_cmd()); + if (0) { + BEGIN_BATCH(1, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_MI_FLUSH << 16) | BRW_FLUSH_STATE_CACHE); ADVANCE_BATCH(); } if (prim_packet.verts_per_instance) { - ret = brw_batchbuffer_data( brw->intel.batch, &prim_packet, + ret = brw_batchbuffer_data( brw->batch, &prim_packet, sizeof(prim_packet), LOOP_CLIPRECTS); if (ret) return ret; } - if (intel->always_flush_cache) { + if (0) { BEGIN_BATCH(1, IGNORE_CLIPRECTS); - OUT_BATCH(intel->vtbl.flush_cmd()); + OUT_BATCH((CMD_MI_FLUSH << 16) | BRW_FLUSH_STATE_CACHE); ADVANCE_BATCH(); } @@ -133,44 +136,24 @@ static enum pipe_error brw_emit_prim(struct brw_context *brw, /* May fail if out of video memory for texture or vbo upload, or on * fallback conditions. */ -static GLboolean brw_try_draw_prims( struct brw_context *brw, - const struct gl_client_array *arrays[], - const struct _mesa_prim *prim, - GLuint nr_prims, - const struct _mesa_index_buffer *ib, - GLuint min_index, - GLuint max_index ) +static int +try_draw_range_elements(struct brw_context *brw, + struct pipe_buffer *index_buffer, + unsigned hw_prim, + unsigned start, unsigned count) { - struct brw_context *brw = brw_context(ctx); - GLboolean retval = GL_FALSE; - GLboolean warn = GL_FALSE; - GLboolean first_time = GL_TRUE; - uint32_t hw_prim; - GLuint i; - - if (ctx->NewState) - _mesa_update_state( ctx ); - - /* Bind all inputs, derive varying and size information: - */ - brw_merge_inputs( brw, arrays ); - - brw->ib.ib = ib; - brw->state.dirty.brw |= BRW_NEW_INDICES; - - brw->vb.min_index = min_index; - brw->vb.max_index = max_index; - brw->state.dirty.brw |= BRW_NEW_VERTICES; - - hw_prim = brw_set_prim(brw, prim[i].mode); + int ret; - brw_validate_state(brw); + ret = brw_validate_state(brw); + if (ret) + return ret; /* Check that we can fit our state in with our existing batchbuffer, or * flush otherwise. */ - ret = dri_bufmgr_check_aperture_space(brw->state.validated_bos, - brw->state.validated_bo_count); + ret = brw->sws->check_aperture_space(brw->sws, + brw->state.validated_bos, + brw->state.validated_bo_count); if (ret) return ret; @@ -178,12 +161,12 @@ static GLboolean brw_try_draw_prims( struct brw_context *brw, if (ret) return ret; - ret = brw_emit_prim(brw, &prim[i], hw_prim); + ret = brw_emit_prim(brw, start, count, index_buffer != NULL, hw_prim); if (ret) return ret; - if (intel->always_flush_batch) - brw_batchbuffer_flush(intel->batch); + if (brw->flags.always_flush_batch) + brw_batchbuffer_flush(brw->batch); return 0; } @@ -197,22 +180,45 @@ brw_draw_range_elements(struct pipe_context *pipe, unsigned max_index, unsigned mode, unsigned start, unsigned count) { - enum pipe_error ret; + struct brw_context *brw = brw_context(pipe); + int ret; + uint32_t hw_prim; + + hw_prim = brw_set_prim(brw, mode); - if (!vbo_all_varyings_in_vbos(arrays)) { - if (!index_bounds_valid) - vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index); + if (BRW_DEBUG & DEBUG_PRIMS) + debug_printf("PRIM: %s %d %d\n", u_prim_name(mode), start, count); + + /* Potentially trigger upload of new index buffer. + * + * XXX: do we need to go through state validation to achieve this? + * Could just call upload code directly. + */ + if (brw->curr.index_buffer != index_buffer) { + pipe_buffer_reference( &brw->curr.index_buffer, index_buffer ); + brw->state.dirty.mesa |= PIPE_NEW_INDEX_BUFFER; + } + + /* XXX: do we really care? + */ + if (brw->curr.min_index != min_index || + brw->curr.max_index != max_index) + { + brw->curr.min_index = min_index; + brw->curr.max_index = max_index; + brw->state.dirty.mesa |= PIPE_NEW_INDEX_RANGE; } + /* Make a first attempt at drawing: */ - ret = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); + ret = try_draw_range_elements(brw, index_buffer, hw_prim, start, count ); /* Otherwise, flush and retry: */ if (ret != 0) { - brw_batchbuffer_flush(intel->batch); - ret = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); + brw_batchbuffer_flush(brw->batch); + ret = try_draw_range_elements(brw, index_buffer, hw_prim, start, count ); assert(ret == 0); } @@ -242,28 +248,37 @@ brw_draw_arrays(struct pipe_context *pipe, unsigned mode, -void brw_draw_init( struct brw_context *brw ) +boolean brw_draw_init( struct brw_context *brw ) { /* Register our drawing function: */ brw->base.draw_arrays = brw_draw_arrays; brw->base.draw_elements = brw_draw_elements; brw->base.draw_range_elements = brw_draw_range_elements; -} -void brw_draw_destroy( struct brw_context *brw ) -{ - int i; + /* Create helpers for uploading data in user buffers: + */ + brw->vb.upload_vertex = u_upload_create( &brw->brw_screen->base, + 128 * 1024, + 64, + PIPE_BUFFER_USAGE_VERTEX ); + if (brw->vb.upload_vertex == NULL) + return FALSE; + + brw->vb.upload_index = u_upload_create( &brw->brw_screen->base, + 128 * 1024, + 64, + PIPE_BUFFER_USAGE_INDEX ); + if (brw->vb.upload_index == NULL) + return FALSE; - if (brw->vb.upload.bo != NULL) { - brw->sws->bo_unreference(brw->vb.upload.bo); - brw->vb.upload.bo = NULL; - } + return TRUE; +} - for (i = 0; i < VERT_ATTRIB_MAX; i++) { - brw->sws->bo_unreference(brw->vb.inputs[i].bo); - brw->vb.inputs[i].bo = NULL; - } +void brw_draw_cleanup( struct brw_context *brw ) +{ + u_upload_destroy( brw->vb.upload_vertex ); + u_upload_destroy( brw->vb.upload_index ); brw->sws->bo_unreference(brw->ib.bo); brw->ib.bo = NULL; diff --git a/src/gallium/drivers/i965/brw_draw.h b/src/gallium/drivers/i965/brw_draw.h index 13f0443a81..8dc5dbce62 100644 --- a/src/gallium/drivers/i965/brw_draw.h +++ b/src/gallium/drivers/i965/brw_draw.h @@ -32,8 +32,7 @@ struct brw_context; - -void brw_draw_init( struct brw_context *brw ); +boolean brw_draw_init( struct brw_context *brw ); void brw_draw_cleanup( struct brw_context *brw ); diff --git a/src/gallium/drivers/i965/brw_draw_upload.c b/src/gallium/drivers/i965/brw_draw_upload.c index 7b0860d04c..040d8ca93a 100644 --- a/src/gallium/drivers/i965/brw_draw_upload.c +++ b/src/gallium/drivers/i965/brw_draw_upload.c @@ -26,21 +26,23 @@ **************************************************************************/ #include "pipe/p_context.h" +#include "pipe/p_error.h" #include "util/u_upload_mgr.h" +#include "util/u_math.h" #include "brw_draw.h" #include "brw_defines.h" #include "brw_context.h" #include "brw_state.h" -#include "brw_fallback.h" - +#include "brw_screen.h" #include "brw_batchbuffer.h" +#include "brw_debug.h" -unsigned brw_translate_surface_format( unsigned id ) +static unsigned brw_translate_surface_format( unsigned id ) { switch (id) { case PIPE_FORMAT_R64_FLOAT: @@ -186,70 +188,136 @@ static unsigned get_index_type(int type) } - -static boolean brw_prepare_vertices(struct brw_context *brw) +static int brw_prepare_vertices(struct brw_context *brw) { - GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; + unsigned int min_index = brw->curr.min_index; + unsigned int max_index = brw->curr.max_index; GLuint i; - const unsigned char *ptr = NULL; - GLuint interleave = 0; - unsigned int min_index = brw->vb.min_index; - unsigned int max_index = brw->vb.max_index; - - struct brw_vertex_element *upload[VERT_ATTRIB_MAX]; - GLuint nr_uploads = 0; - - /* First build an array of pointers to ve's in vb.inputs_read - */ - if (0) - _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); - + int ret; + if (BRW_DEBUG & DEBUG_VERTS) + debug_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); - for (i = 0; i < brw->vb.num_vertex_buffer; i++) { - struct brw_vertex_buffer *vb = brw->vb.vertex_buffer[i]; - unsigned size = (vb->stride == 0 ? - vb->size : - vb->stride * (max_index + 1 - min_index)); - - if (brw_is_user_buffer(vb)) { - u_upload_buffer( brw->upload_vertex, - min_index * vb->stride, - size, - &offset, - &buffer ); + for (i = 0; i < brw->curr.num_vertex_buffers; i++) { + struct pipe_vertex_buffer *vb = &brw->curr.vertex_buffer[i]; + struct brw_winsys_buffer *bo; + struct pipe_buffer *upload_buf; + unsigned offset; + + if (BRW_DEBUG & DEBUG_VERTS) + debug_printf("%s vb[%d] user:%d offset:0x%x sz:0x%x stride:0x%x\n", + __FUNCTION__, i, + brw_buffer_is_user_buffer(vb->buffer), + vb->buffer_offset, + vb->buffer->size, + vb->stride); + + if (brw_buffer_is_user_buffer(vb->buffer)) { + + /* XXX: simplify this. Stop the state trackers from generating + * zero-stride buffers & have them use additional constants (or + * add support for >1 constant buffer) instead. + */ + unsigned size = (vb->stride == 0 ? + vb->buffer->size - vb->buffer_offset : + MAX2(vb->buffer->size - vb->buffer_offset, + vb->stride * (max_index + 1 - min_index))); + + ret = u_upload_buffer( brw->vb.upload_vertex, + vb->buffer_offset + min_index * vb->stride, + size, + vb->buffer, + &offset, + &upload_buf ); + if (ret) + return ret; + + bo = brw_buffer(upload_buf)->bo; + + assert(offset + size <= bo->size); } else { - offset = 0; - buffer = vb->buffer; + offset = vb->buffer_offset; + bo = brw_buffer(vb->buffer)->bo; } + + assert(offset < bo->size); /* Set up post-upload info about this vertex buffer: */ - input->offset = (unsigned long)offset; - input->stride = vb->stride; - input->count = count; - brw->sws->bo_unreference(input->bo); - input->bo = intel_bufferobj_buffer(intel, intel_buffer, - INTEL_READ); - brw->sws->bo_reference(input->bo); - - assert(input->offset < input->bo->size); - assert(input->offset + size <= input->bo->size); + brw->vb.vb[i].offset = offset; + brw->vb.vb[i].stride = vb->stride; + brw->vb.vb[i].vertex_count = (vb->stride == 0 ? + 1 : + (bo->size - offset) / vb->stride); + brw->sws->bo_unreference(brw->vb.vb[i].bo); + brw->vb.vb[i].bo = bo; + brw->sws->bo_reference(brw->vb.vb[i].bo); + + /* Don't need to retain this reference. We have a reference on + * the underlying winsys buffer: + */ + pipe_buffer_reference( &upload_buf, NULL ); } + brw->vb.nr_vb = i; brw_prepare_query_begin(brw); - for (i = 0; i < brw->vb.nr_enabled; i++) { - struct brw_vertex_element *input = brw->vb.enabled[i]; + for (i = 0; i < brw->vb.nr_vb; i++) { + brw_add_validated_bo(brw, brw->vb.vb[i].bo); + } + + return 0; +} + +static int brw_emit_vertex_buffers( struct brw_context *brw ) +{ + int i; + + /* If the VS doesn't read any inputs (calculating vertex position from + * a state variable for some reason, for example), just bail. + * + * The stale VB state stays in place, but they don't do anything unless + * a VE loads from them. + */ + if (brw->vb.nr_vb == 0) { + if (BRW_DEBUG & DEBUG_VERTS) + debug_printf("%s: no active vertex buffers\n", __FUNCTION__); - brw_add_validated_bo(brw, input->bo); + return 0; + } + + /* Emit VB state packets. + */ + BEGIN_BATCH(1 + brw->vb.nr_vb * 4, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_BUFFER << 16) | + ((1 + brw->vb.nr_vb * 4) - 2)); + + for (i = 0; i < brw->vb.nr_vb; i++) { + OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) | + BRW_VB0_ACCESS_VERTEXDATA | + (brw->vb.vb[i].stride << BRW_VB0_PITCH_SHIFT)); + OUT_RELOC(brw->vb.vb[i].bo, + I915_GEM_DOMAIN_VERTEX, 0, + brw->vb.vb[i].offset); + if (BRW_IS_IGDNG(brw)) { + OUT_RELOC(brw->vb.vb[i].bo, + I915_GEM_DOMAIN_VERTEX, 0, + brw->vb.vb[i].bo->size - 1); + } else + OUT_BATCH(brw->vb.vb[i].stride ? brw->vb.vb[i].vertex_count : 0); + OUT_BATCH(0); /* Instance data step rate */ } + ADVANCE_BATCH(); + return 0; } -static void brw_emit_vertices(struct brw_context *brw) + + + +static int brw_emit_vertex_elements(struct brw_context *brw) { GLuint i; @@ -262,7 +330,7 @@ static void brw_emit_vertices(struct brw_context *brw) * The stale VB state stays in place, but they don't do anything unless * a VE loads from them. */ - if (brw->vb.nr_enabled == 0) { + if (brw->vb.nr_ve == 0) { BEGIN_BATCH(3, IGNORE_CLIPRECTS); OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1); OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | @@ -274,59 +342,23 @@ static void brw_emit_vertices(struct brw_context *brw) (BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) | (BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT)); ADVANCE_BATCH(); - return; + return 0; } - /* Now emit VB and VEP state packets. + /* Now emit vertex element (VEP) state packets. * - * This still defines a hardware VB for each input, even if they - * are interleaved or from the same VBO. TBD if this makes a - * performance difference. */ - BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS); - OUT_BATCH((CMD_VERTEX_BUFFER << 16) | - ((1 + brw->vb.nr_enabled * 4) - 2)); - - for (i = 0; i < brw->vb.nr_enabled; i++) { - struct brw_vertex_element *input = brw->vb.enabled[i]; - - OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) | - BRW_VB0_ACCESS_VERTEXDATA | - (input->stride << BRW_VB0_PITCH_SHIFT)); - OUT_RELOC(input->bo, - I915_GEM_DOMAIN_VERTEX, 0, - input->offset); - if (BRW_IS_IGDNG(brw)) { - if (input->stride) { - OUT_RELOC(input->bo, - I915_GEM_DOMAIN_VERTEX, 0, - input->offset + input->stride * input->count - 1); - } else { - assert(input->count == 1); - OUT_RELOC(input->bo, - I915_GEM_DOMAIN_VERTEX, 0, - input->offset + input->element_size - 1); - } - } else - OUT_BATCH(input->stride ? input->count : 0); - OUT_BATCH(0); /* Instance data step rate */ - } - ADVANCE_BATCH(); - - BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS); - OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_enabled * 2) - 2)); - for (i = 0; i < brw->vb.nr_enabled; i++) { - struct brw_vertex_element *input = brw->vb.enabled[i]; - uint32_t format = get_surface_type(input->glarray->Type, - input->glarray->Size, - input->glarray->Format, - input->glarray->Normalized); + BEGIN_BATCH(1 + brw->curr.num_vertex_elements * 2, IGNORE_CLIPRECTS); + OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + brw->vb.nr_ve * 2) - 2)); + for (i = 0; i < brw->vb.nr_ve; i++) { + const struct pipe_vertex_element *input = &brw->curr.vertex_element[i]; + uint32_t format = brw_translate_surface_format( input->src_format ); uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC; uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC; uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC; uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC; - switch (input->glarray->Size) { + switch (input->nr_components) { case 0: comp0 = BRW_VE1_COMPONENT_STORE_0; case 1: comp1 = BRW_VE1_COMPONENT_STORE_0; case 2: comp2 = BRW_VE1_COMPONENT_STORE_0; @@ -352,11 +384,29 @@ static void brw_emit_vertices(struct brw_context *brw) ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT)); } ADVANCE_BATCH(); + return 0; +} + + +static int brw_emit_vertices( struct brw_context *brw ) +{ + int ret; + + ret = brw_emit_vertex_buffers( brw ); + if (ret) + return ret; + + ret = brw_emit_vertex_elements( brw ); + if (ret) + return ret; + + return 0; } + const struct brw_tracked_state brw_vertices = { .dirty = { - .mesa = 0, + .mesa = PIPE_NEW_INDEX_RANGE, .brw = BRW_NEW_BATCH | BRW_NEW_VERTICES, .cache = 0, }, @@ -364,104 +414,106 @@ const struct brw_tracked_state brw_vertices = { .emit = brw_emit_vertices, }; -static void brw_prepare_indices(struct brw_context *brw) + +static int brw_prepare_indices(struct brw_context *brw) { - const struct _mesa_index_buffer *index_buffer = brw->ib.ib; - GLuint ib_size; + struct pipe_buffer *index_buffer = brw->curr.index_buffer; struct brw_winsys_buffer *bo = NULL; - struct gl_buffer_object *bufferobj; GLuint offset; - GLuint ib_type_size; + GLuint index_size; + GLuint ib_size; + int ret; if (index_buffer == NULL) - return; + return 0; - ib_type_size = get_size(index_buffer->type); - ib_size = ib_type_size * index_buffer->count; - bufferobj = index_buffer->obj;; + if (DEBUG & DEBUG_VERTS) + debug_printf("%s: index_size:%d index_buffer->size:%d\n", + __FUNCTION__, + brw->curr.index_size, + brw->curr.index_buffer->size); - /* Turn into a proper VBO: - */ - if (!_mesa_is_bufferobj(bufferobj)) { - brw->ib.start_vertex_offset = 0; + ib_size = index_buffer->size; + index_size = brw->curr.index_size; - /* Get new bufferobj, offset: - */ - get_space(brw, ib_size, &bo, &offset); - - /* Straight upload + /* Turn userbuffer into a proper hardware buffer? + */ + if (brw_buffer_is_user_buffer(index_buffer)) { + struct pipe_buffer *upload_buf; + + ret = u_upload_buffer( brw->vb.upload_index, + 0, + ib_size, + index_buffer, + &offset, + &upload_buf ); + if (ret) + return ret; + + bo = brw_buffer(upload_buf)->bo; + brw->sws->bo_reference(bo); + pipe_buffer_reference( &upload_buf, NULL ); + + /* XXX: annotate the userbuffer with the upload information so + * that successive calls don't get re-uploaded. */ - brw_bo_subdata(bo, offset, ib_size, index_buffer->ptr); - - } else { - offset = (GLuint) (unsigned long) index_buffer->ptr; - brw->ib.start_vertex_offset = 0; + } + else { + bo = brw_buffer(index_buffer)->bo; + brw->sws->bo_reference(bo); + + ib_size = bo->size; + offset = 0; + } - /* If the index buffer isn't aligned to its element size, we have to - * rebase it into a temporary. - */ - if ((get_size(index_buffer->type) - 1) & offset) { - GLubyte *map = ctx->Driver.MapBuffer(ctx, - GL_ELEMENT_ARRAY_BUFFER_ARB, - GL_DYNAMIC_DRAW_ARB, - bufferobj); - map += offset; - - get_space(brw, ib_size, &bo, &offset); - - dri_bo_subdata(bo, offset, ib_size, map); - - ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj); - } else { - bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj), - INTEL_READ); - brw->sws->bo_reference(bo); - - /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading - * the index buffer state when we're just moving the start index - * of our drawing. - */ - brw->ib.start_vertex_offset = offset / ib_type_size; - offset = 0; - ib_size = bo->size; - } + /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading the + * index buffer state when we're just moving the start index of our + * drawing. + * + * In gallium this will happen in the case where successive draw + * calls are made with (distinct?) userbuffers, but the upload_mgr + * places the data into a single winsys buffer. + * + * This statechange doesn't raise any state flags and is always + * just merged into the final draw packet: + */ + if (1) { + assert((offset & (index_size - 1)) == 0); + brw->ib.start_vertex_offset = offset / index_size; } + /* These statechanges trigger a new CMD_INDEX_BUFFER packet: + */ if (brw->ib.bo != bo || - brw->ib.offset != offset || brw->ib.size != ib_size) { - drm_intel_bo_unreference(brw->ib.bo); + brw->sws->bo_unreference(brw->ib.bo); brw->ib.bo = bo; - brw->ib.offset = offset; brw->ib.size = ib_size; - brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; - } else { - drm_intel_bo_unreference(bo); + } + else { + brw->sws->bo_unreference(bo); } brw_add_validated_bo(brw, brw->ib.bo); + return 0; } const struct brw_tracked_state brw_indices = { .dirty = { - .mesa = 0, - .brw = BRW_NEW_INDICES, + .mesa = PIPE_NEW_INDEX_BUFFER, + .brw = 0, .cache = 0, }, .prepare = brw_prepare_indices, }; -static void brw_emit_index_buffer(struct brw_context *brw) +static int brw_emit_index_buffer(struct brw_context *brw) { - const struct _mesa_index_buffer *index_buffer = brw->ib.ib; - - if (index_buffer == NULL) - return; - /* Emit the indexbuffer packet: */ + if (brw->ib.bo) { struct brw_indexbuffer ib; @@ -469,7 +521,7 @@ static void brw_emit_index_buffer(struct brw_context *brw) ib.header.bits.opcode = CMD_INDEX_BUFFER; ib.header.bits.length = sizeof(ib)/4 - 2; - ib.header.bits.index_format = get_index_type(index_buffer->type); + ib.header.bits.index_format = get_index_type(brw->ib.size); ib.header.bits.cut_index_enable = 0; BEGIN_BATCH(4, IGNORE_CLIPRECTS); @@ -483,6 +535,8 @@ static void brw_emit_index_buffer(struct brw_context *brw) OUT_BATCH( 0 ); ADVANCE_BATCH(); } + + return 0; } const struct brw_tracked_state brw_index_buffer = { diff --git a/src/gallium/drivers/i965/brw_eu.c b/src/gallium/drivers/i965/brw_eu.c index 1df561386e..df49d4b72f 100644 --- a/src/gallium/drivers/i965/brw_eu.c +++ b/src/gallium/drivers/i965/brw_eu.c @@ -29,6 +29,7 @@ * Keith Whitwell */ +#include "util/u_memory.h" #include "brw_context.h" #include "brw_defines.h" @@ -237,7 +238,7 @@ brw_resolve_cals(struct brw_compile *c) struct brw_glsl_call *call, *next; for (call = c->first_call; call; call = next) { next = call->next; - _mesa_free(call); + FREE(call); } c->first_call = NULL; } @@ -247,7 +248,7 @@ brw_resolve_cals(struct brw_compile *c) struct brw_glsl_label *label, *next; for (label = c->first_label; label; label = next) { next = label->next; - _mesa_free(label); + FREE(label); } c->first_label = NULL; } diff --git a/src/gallium/drivers/i965/brw_eu_debug.c b/src/gallium/drivers/i965/brw_eu_debug.c index ad7ec36e86..5989f5a04e 100644 --- a/src/gallium/drivers/i965/brw_eu_debug.c +++ b/src/gallium/drivers/i965/brw_eu_debug.c @@ -28,7 +28,8 @@ * Authors: * Keith Whitwell */ - + +#include "util/u_debug.h" #include "brw_eu.h" @@ -52,7 +53,7 @@ void brw_print_reg( struct brw_reg hwreg ) "f" }; - _mesa_printf("%s%s", + debug_printf("%s%s", hwreg.abs ? "abs/" : "", hwreg.negate ? "-" : ""); @@ -64,7 +65,7 @@ void brw_print_reg( struct brw_reg hwreg ) hwreg.hstride == BRW_HORIZONTAL_STRIDE_1 && hwreg.type == BRW_REGISTER_TYPE_F) { /* vector register */ - _mesa_printf("vec%d", hwreg.nr); + debug_printf("vec%d", hwreg.nr); } else if (hwreg.file == BRW_GENERAL_REGISTER_FILE && hwreg.vstride == BRW_VERTICAL_STRIDE_0 && @@ -72,13 +73,13 @@ void brw_print_reg( struct brw_reg hwreg ) hwreg.hstride == BRW_HORIZONTAL_STRIDE_0 && hwreg.type == BRW_REGISTER_TYPE_F) { /* "scalar" register */ - _mesa_printf("scl%d.%d", hwreg.nr, hwreg.subnr / 4); + debug_printf("scl%d.%d", hwreg.nr, hwreg.subnr / 4); } else if (hwreg.file == BRW_IMMEDIATE_VALUE) { - _mesa_printf("imm %f", hwreg.dw1.f); + debug_printf("imm %f", hwreg.dw1.f); } else { - _mesa_printf("%s%d.%d<%d;%d,%d>:%s", + debug_printf("%s%d.%d<%d;%d,%d>:%s", file[hwreg.file], hwreg.nr, hwreg.subnr / type_sz(hwreg.type), diff --git a/src/gallium/drivers/i965/brw_misc_state.c b/src/gallium/drivers/i965/brw_misc_state.c index 0f2612c181..98fec85c1d 100644 --- a/src/gallium/drivers/i965/brw_misc_state.c +++ b/src/gallium/drivers/i965/brw_misc_state.c @@ -315,24 +315,20 @@ const struct brw_tracked_state brw_polygon_stipple = { static void upload_polygon_stipple_offset(struct brw_context *brw) { - __DRIdrawablePrivate *dPriv = brw->intel.driDrawable; 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; - /* If we're drawing to a system window (ctx->DrawBuffer->Name == 0), - * we have to invert the Y axis in order to match the OpenGL - * pixel coordinate system, and our offset must be matched - * to the window position. If we're drawing to a FBO - * (ctx->DrawBuffer->Name != 0), then our native pixel coordinate - * system works just fine, and there's no window system to - * worry about. + /* Never need to offset stipple coordinates. + * + * XXX: is it ever necessary to invert Y values? */ - if (brw->intel.ctx.DrawBuffer->Name == 0) { - bpso.bits0.x_offset = (32 - (dPriv->x & 31)) & 31; - bpso.bits0.y_offset = (32 - ((dPriv->y + dPriv->h) & 31)) & 31; + if (0) { + int x = 0, y = 0, h = 0; + bpso.bits0.x_offset = (32 - (x & 31)) & 31; + bpso.bits0.y_offset = (32 - ((y + h) & 31)) & 31; } else { bpso.bits0.y_offset = 0; diff --git a/src/gallium/drivers/i965/brw_pipe_flush.c b/src/gallium/drivers/i965/brw_pipe_flush.c index e85a1a9c1b..65e7151517 100644 --- a/src/gallium/drivers/i965/brw_pipe_flush.c +++ b/src/gallium/drivers/i965/brw_pipe_flush.c @@ -53,6 +53,9 @@ static void brw_note_fence( struct brw_context *brw, GLuint fence ) static GLuint brw_flush_cmd( void ) { struct brw_mi_flush flush; + + return ; + flush.opcode = CMD_MI_FLUSH; flush.pad = 0; flush.flags = BRW_FLUSH_STATE_CACHE; diff --git a/src/gallium/drivers/i965/brw_pipe_shader.c b/src/gallium/drivers/i965/brw_pipe_shader.c index fbb772d18c..8b61da763c 100644 --- a/src/gallium/drivers/i965/brw_pipe_shader.c +++ b/src/gallium/drivers/i965/brw_pipe_shader.c @@ -33,6 +33,25 @@ #include "brw_util.h" #include "brw_wm.h" + +/** + * Determine if the given fragment program uses GLSL features such + * as flow conditionals, loops, subroutines. + * Some GLSL shaders may use these features, others might not. + */ +GLboolean brw_wm_is_glsl(const struct brw_fragment_shader *fp) +{ + return (fp->info.insn_count[TGSI_OPCODE_ARL] > 0 || + fp->info.insn_count[TGSI_OPCODE_IF] > 0 || + fp->info.insn_count[TGSI_OPCODE_ENDIF] > 0 || /* redundant - IF */ + fp->info.insn_count[TGSI_OPCODE_CAL] > 0 || + fp->info.insn_count[TGSI_OPCODE_BRK] > 0 || /* redundant - BGNLOOP */ + fp->info.insn_count[TGSI_OPCODE_RET] > 0 || /* redundant - CAL */ + fp->info.insn_count[TGSI_OPCODE_BGNLOOP] > 0); +} + + + static void brwBindProgram( struct brw_context *brw, GLenum target, struct gl_program *prog ) diff --git a/src/gallium/drivers/i965/brw_pipe_vertex.c b/src/gallium/drivers/i965/brw_pipe_vertex.c index b0928adbe4..d1d0d7cd43 100644 --- a/src/gallium/drivers/i965/brw_pipe_vertex.c +++ b/src/gallium/drivers/i965/brw_pipe_vertex.c @@ -1,26 +1,11 @@ -static void brw_merge_inputs( struct brw_context *brw, - const struct gl_client_array *arrays[]) -{ - struct brw_vertex_info old = brw->vb.info; - GLuint i; - - for (i = 0; i < VERT_ATTRIB_MAX; i++) - brw->sws->bo_unreference(brw->vb.inputs[i].bo); - memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs)); - memset(&brw->vb.info, 0, sizeof(brw->vb.info)); +void +brw_pipe_vertex_cleanup( struct brw_context *brw ) +{ for (i = 0; i < VERT_ATTRIB_MAX; i++) { - brw->vb.inputs[i].glarray = arrays[i]; - brw->vb.inputs[i].attrib = (gl_vert_attrib) i; - - if (arrays[i]->StrideB != 0) - brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) << - ((i%16) * 2); + brw->sws->bo_unreference(brw->vb.inputs[i].bo); + brw->vb.inputs[i].bo = NULL; } - - /* Raise statechanges if input sizes have changed. */ - if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0) - brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS; } diff --git a/src/gallium/drivers/i965/brw_screen.h b/src/gallium/drivers/i965/brw_screen.h index 79d595d0ad..b0be0e1f8a 100644 --- a/src/gallium/drivers/i965/brw_screen.h +++ b/src/gallium/drivers/i965/brw_screen.h @@ -56,6 +56,14 @@ struct brw_transfer unsigned offset; }; +struct brw_buffer +{ + struct pipe_buffer base; + struct brw_winsys_buffer *bo; + void *ptr; + boolean is_user_buffer; +}; + /* * Cast wrappers @@ -72,5 +80,19 @@ brw_transfer(struct pipe_transfer *transfer) return (struct brw_transfer *)transfer; } +static INLINE struct brw_buffer * +brw_buffer(struct pipe_buffer *buffer) +{ + return (struct brw_buffer *)buffer; +} + + +/* Pipe buffer helpers + */ +static INLINE boolean +brw_buffer_is_user_buffer( const struct pipe_buffer *buf ) +{ + return ((const struct brw_buffer *)buf)->is_user_buffer; +} #endif /* BRW_SCREEN_H */ diff --git a/src/gallium/drivers/i965/brw_sf.c b/src/gallium/drivers/i965/brw_sf.c index 53e8f09e37..e2db2e76e6 100644 --- a/src/gallium/drivers/i965/brw_sf.c +++ b/src/gallium/drivers/i965/brw_sf.c @@ -134,7 +134,7 @@ static void upload_sf_prog(struct brw_context *brw) key.attrs = brw->vs.prog_data->outputs_written; /* BRW_NEW_REDUCED_PRIMITIVE */ - switch (brw->intel.reduced_primitive) { + switch (brw->reduced_primitive) { case GL_TRIANGLES: /* NOTE: We just use the edgeflag attribute as an indicator that * unfilled triangles are active. We don't actually do the diff --git a/src/gallium/drivers/i965/brw_sf_state.c b/src/gallium/drivers/i965/brw_sf_state.c index 0e406f12e1..648a16a038 100644 --- a/src/gallium/drivers/i965/brw_sf_state.c +++ b/src/gallium/drivers/i965/brw_sf_state.c @@ -40,19 +40,12 @@ static void upload_sf_vp(struct brw_context *brw) const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF; struct brw_sf_viewport sfv; GLfloat y_scale, y_bias; - const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0); const GLfloat *v = ctx->Viewport._WindowMap.m; memset(&sfv, 0, sizeof(sfv)); - if (render_to_fbo) { - y_scale = 1.0; - y_bias = 0; - } - else { - y_scale = -1.0; - y_bias = ctx->DrawBuffer->Height; - } + y_scale = 1.0; + y_bias = 0; /* _NEW_VIEWPORT */ @@ -73,20 +66,11 @@ static void upload_sf_vp(struct brw_context *brw) * Note that the hardware's coordinates are inclusive, while Mesa's min is * inclusive but max is exclusive. */ - if (render_to_fbo) { - /* texmemory: Y=0=bottom */ - sfv.scissor.xmin = ctx->DrawBuffer->_Xmin; - sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1; - sfv.scissor.ymin = ctx->DrawBuffer->_Ymin; - sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1; - } - else { - /* memory: Y=0=top */ - sfv.scissor.xmin = ctx->DrawBuffer->_Xmin; - sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1; - sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax; - sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1; - } + /* Y=0=bottom */ + sfv.scissor.xmin = ctx->DrawBuffer->_Xmin; + sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1; + sfv.scissor.ymin = ctx->DrawBuffer->_Ymin; + sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1; brw->sws->bo_unreference(brw->sf.vp_bo); brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 ); @@ -151,7 +135,7 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) /* _NEW_LIGHT */ key->provoking_vertex = ctx->Light.ProvokingVertex; - key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; + key->render_to_fbo = 1; } static struct brw_winsys_buffer * @@ -211,11 +195,6 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, else sf.sf5.front_winding = BRW_FRONTWINDING_CW; - /* The viewport is inverted for rendering to a FBO, and that inverts - * polygon front/back orientation. - */ - sf.sf5.front_winding ^= key->render_to_fbo; - switch (key->cull_face) { case GL_FRONT: sf.sf6.cull_mode = BRW_CULLMODE_FRONT; @@ -245,7 +224,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.sf6.line_width = 0; /* _NEW_BUFFERS */ - key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; + key->render_to_fbo = 1; if (!key->render_to_fbo) { /* Rendering to an OpenGL window */ sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT; diff --git a/src/gallium/drivers/i965/brw_state.h b/src/gallium/drivers/i965/brw_state.h index 02657eaba7..9bf34c3fe4 100644 --- a/src/gallium/drivers/i965/brw_state.h +++ b/src/gallium/drivers/i965/brw_state.h @@ -109,8 +109,8 @@ struct brw_surface_key { /*********************************************************************** * brw_state.c */ -void brw_validate_state(struct brw_context *brw); -void brw_upload_state(struct brw_context *brw); +int brw_validate_state(struct brw_context *brw); +int brw_upload_state(struct brw_context *brw); void brw_init_state(struct brw_context *brw); void brw_destroy_state(struct brw_context *brw); @@ -157,7 +157,7 @@ void brw_state_cache_bo_delete(struct brw_cache *cache, struct brw_winsys_buffer /*********************************************************************** * brw_state_batch.c */ -#define BRW_BATCH_STRUCT(brw, s) brw_batchbuffer_data( brw->intel.batch, (s), sizeof(*(s)), IGNORE_CLIPRECTS) +#define BRW_BATCH_STRUCT(brw, s) brw_batchbuffer_data( brw->batch, (s), sizeof(*(s)), IGNORE_CLIPRECTS) #define BRW_CACHED_BATCH_STRUCT(brw, s) brw_cached_batch_struct( brw, (s), sizeof(*(s)) ) GLboolean brw_cached_batch_struct( struct brw_context *brw, diff --git a/src/gallium/drivers/i965/brw_state_batch.c b/src/gallium/drivers/i965/brw_state_batch.c index b285837070..324fce5163 100644 --- a/src/gallium/drivers/i965/brw_state_batch.c +++ b/src/gallium/drivers/i965/brw_state_batch.c @@ -47,7 +47,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, struct header *newheader = (struct header *)data; if (brw->emit_state_always) { - brw_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + brw_batchbuffer_data(brw->batch, data, sz, IGNORE_CLIPRECTS); return GL_TRUE; } @@ -74,7 +74,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, emit: memcpy(item->header, newheader, sz); - brw_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + brw_batchbuffer_data(brw->batch, data, sz, IGNORE_CLIPRECTS); return GL_TRUE; } diff --git a/src/gallium/drivers/i965/brw_swtnl.c b/src/gallium/drivers/i965/brw_swtnl.c index 83f138f67a..d2df8af9f4 100644 --- a/src/gallium/drivers/i965/brw_swtnl.c +++ b/src/gallium/drivers/i965/brw_swtnl.c @@ -12,13 +12,13 @@ static GLboolean check_fallbacks( struct brw_context *brw, * use fallbacks. If we're forcing fallbacks, always * use fallfacks. */ - if (brw->intel.conformance_mode == 0) + if (brw->flags.no_swtnl) return GL_FALSE; - if (brw->intel.conformance_mode == 2) + if (brw->flags.force_swtnl) return GL_TRUE; - if (ctx->Polygon.SmoothFlag) { + if (brw->curr.rast->tmpl.smooth_polys) { for (i = 0; i < nr_prims; i++) if (reduced_prim[prim[i].mode] == GL_TRIANGLES) return GL_TRUE; diff --git a/src/gallium/drivers/i965/brw_winsys.h b/src/gallium/drivers/i965/brw_winsys.h index 82cd8007ac..51e23b9640 100644 --- a/src/gallium/drivers/i965/brw_winsys.h +++ b/src/gallium/drivers/i965/brw_winsys.h @@ -161,6 +161,13 @@ struct brw_winsys_screen { size_t size, const void *data); + /* XXX: couldn't this be handled by returning true/false on + * bo_emit_reloc? + */ + boolean (*check_aperture_space)( struct brw_winsys_screen *iws, + struct brw_winsys_buffer **buffers, + unsigned count ); + /** * Map a buffer. */ diff --git a/src/gallium/drivers/i965/brw_wm.c b/src/gallium/drivers/i965/brw_wm.c index 284cf42f8b..4948ea0dff 100644 --- a/src/gallium/drivers/i965/brw_wm.c +++ b/src/gallium/drivers/i965/brw_wm.c @@ -158,7 +158,7 @@ static void do_wm_prog( struct brw_context *brw, memcpy(&c->key, key, sizeof(*key)); c->fp = fp; - c->env_param = brw->intel.ctx.FragmentProgram.Parameters; + c->env_param = NULL; /*brw->intel.ctx.FragmentProgram.Parameters;*/ brw_init_compile(brw, &c->func); diff --git a/src/gallium/drivers/i965/brw_wm.h b/src/gallium/drivers/i965/brw_wm.h index 18775830f9..e06de95a8a 100644 --- a/src/gallium/drivers/i965/brw_wm.h +++ b/src/gallium/drivers/i965/brw_wm.h @@ -131,9 +131,9 @@ struct brw_wm_ref { GLuint insn:24; }; -struct brw_wm_constref { +struct brw_wm_imm_ref { const struct brw_wm_ref *ref; - GLfloat constval; + GLfloat imm1f; }; @@ -232,8 +232,8 @@ struct brw_wm_compile { struct brw_wm_instruction instruction[BRW_WM_MAX_INSN]; GLuint nr_insns; - struct brw_wm_constref constref[BRW_WM_MAX_CONST]; - GLuint nr_constrefs; + struct brw_wm_imm_ref imm_ref[BRW_WM_MAX_CONST]; + GLuint nr_imm_refs; struct brw_wm_grf pass2_grf[BRW_WM_MAX_GRF/2]; diff --git a/src/gallium/drivers/i965/brw_wm_glsl.c b/src/gallium/drivers/i965/brw_wm_glsl.c index c4f0711793..a8de5fdd0b 100644 --- a/src/gallium/drivers/i965/brw_wm_glsl.c +++ b/src/gallium/drivers/i965/brw_wm_glsl.c @@ -7,34 +7,6 @@ static struct brw_reg get_dst_reg(struct brw_wm_compile *c, const struct prog_instruction *inst, GLuint component); -/** - * Determine if the given fragment program uses GLSL features such - * as flow conditionals, loops, subroutines. - * Some GLSL shaders may use these features, others might not. - */ -GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) -{ - int i; - - for (i = 0; i < fp->Base.NumInstructions; i++) { - const struct prog_instruction *inst = &fp->Base.Instructions[i]; - switch (inst->Opcode) { - case OPCODE_ARL: - case OPCODE_IF: - case OPCODE_ENDIF: - case OPCODE_CAL: - case OPCODE_BRK: - case OPCODE_RET: - case OPCODE_BGNLOOP: - return GL_TRUE; - default: - break; - } - } - return GL_FALSE; -} - - static void reclaim_temps(struct brw_wm_compile *c); diff --git a/src/gallium/drivers/i965/brw_wm_pass0.c b/src/gallium/drivers/i965/brw_wm_pass0.c index de5f5fe821..31b0270e84 100644 --- a/src/gallium/drivers/i965/brw_wm_pass0.c +++ b/src/gallium/drivers/i965/brw_wm_pass0.c @@ -124,33 +124,33 @@ static const struct brw_wm_ref *get_param_ref( struct brw_wm_compile *c, } -/** Return a ref to a constant/literal value */ -static const struct brw_wm_ref *get_const_ref( struct brw_wm_compile *c, - const GLfloat *constval ) +/** Return a ref to an immediate value */ +static const struct brw_wm_ref *get_imm_ref( struct brw_wm_compile *c, + const GLfloat *imm1f ) { GLuint i; /* Search for an existing const value matching the request: */ - for (i = 0; i < c->nr_constrefs; i++) { - if (c->constref[i].constval == *constval) - return c->constref[i].ref; + for (i = 0; i < c->nr_imm_refs; i++) { + if (c->imm_ref[i].imm_val == *imm1f) + return c->imm_ref[i].ref; } /* Else try to add a new one: */ - if (c->nr_constrefs < BRW_WM_MAX_CONST) { - GLuint i = c->nr_constrefs++; + if (c->nr_imm_refs < BRW_WM_MAX_IMM) { + GLuint i = c->nr_imm_refs++; - /* A constant is a special type of parameter: + /* An immediate is a special type of parameter: */ - c->constref[i].constval = *constval; - c->constref[i].ref = get_param_ref(c, constval); + c->imm_ref[i].imm_val = *imm_val; + c->imm_ref[i].ref = get_param_ref(c, imm_val); - return c->constref[i].ref; + return c->imm_ref[i].ref; } else { - _mesa_printf("%s: out of constrefs\n", __FUNCTION__); + _mesa_printf("%s: out of imm_refs\n", __FUNCTION__); c->prog_data.error = 1; return NULL; } @@ -200,7 +200,7 @@ static const struct brw_wm_ref *pass0_get_reg( struct brw_wm_compile *c, case PROGRAM_CONSTANT: /* These are invarient: */ - ref = get_const_ref(c, &plist->ParameterValues[idx][component]); + ref = get_imm_ref(c, &plist->ParameterValues[idx][component]); break; case PROGRAM_STATE_VAR: @@ -266,9 +266,9 @@ static const struct brw_wm_ref *get_fp_src_reg_ref( struct brw_wm_compile *c, static const GLfloat const_one = 1.0; if (component == SWIZZLE_ZERO) - src_ref = get_const_ref(c, &const_zero); + src_ref = get_imm_ref(c, &const_zero); else if (component == SWIZZLE_ONE) - src_ref = get_const_ref(c, &const_one); + src_ref = get_imm_ref(c, &const_one); else src_ref = pass0_get_reg(c, src.File, src.Index, component); diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c index c76bff9181..ec9c859fcb 100644 --- a/src/mesa/state_tracker/st_draw.c +++ b/src/mesa/state_tracker/st_draw.c @@ -554,7 +554,8 @@ st_draw_vbo(GLcontext *ctx, /* Gallium probably doesn't want this in some cases. */ if (!index_bounds_valid) - vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index); + if (!vbo_all_varyings_in_vbos(arrays)) + vbo_get_minmax_index(ctx, prims, ib, &min_index, &max_index); /* sanity check for pointer arithmetic below */ assert(sizeof(arrays[0]->Ptr[0]) == 1); -- cgit v1.2.3 From 7ba2fe40fa092551f1c493d754c80ca93564d32b Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Tue, 27 Oct 2009 00:29:21 +0000 Subject: i965g: still working on compilation --- src/gallium/drivers/i965/brw_context.h | 1 + src/gallium/drivers/i965/brw_eu.c | 18 ++--- src/gallium/drivers/i965/brw_eu.h | 4 +- src/gallium/drivers/i965/brw_vs.h | 6 ++ src/gallium/drivers/i965/brw_vs_emit.c | 131 ++++++++++++++++----------------- src/gallium/drivers/i965/brw_wm.h | 9 +-- src/gallium/drivers/i965/brw_wm_glsl.c | 2 +- 7 files changed, 83 insertions(+), 88 deletions(-) (limited to 'src/gallium/drivers/i965/brw_eu.c') diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h index 8aaf895d20..7b85363e9f 100644 --- a/src/gallium/drivers/i965/brw_context.h +++ b/src/gallium/drivers/i965/brw_context.h @@ -289,6 +289,7 @@ struct brw_vs_prog_data { GLuint nr_params; /**< number of TGSI_FILE_CONSTANT's */ GLboolean copy_edgeflag; + GLboolean writes_psiz; /* Used for calculating urb partitions: */ diff --git a/src/gallium/drivers/i965/brw_eu.c b/src/gallium/drivers/i965/brw_eu.c index df49d4b72f..1189a35b6f 100644 --- a/src/gallium/drivers/i965/brw_eu.c +++ b/src/gallium/drivers/i965/brw_eu.c @@ -152,7 +152,7 @@ const GLuint *brw_get_program( struct brw_compile *p, */ struct brw_glsl_label { - const char *name; /**< the label string */ + GLuint label; /**< the label number */ GLuint position; /**< the position of the brw instruction for this label */ struct brw_glsl_label *next; /**< next in linked list */ }; @@ -164,7 +164,7 @@ struct brw_glsl_label struct brw_glsl_call { GLuint call_inst_pos; /**< location of the CAL instruction */ - const char *sub_name; /**< name of subroutine to call */ + GLuint label; struct brw_glsl_call *next; /**< next in linked list */ }; @@ -173,10 +173,10 @@ struct brw_glsl_call * Called for each OPCODE_BGNSUB. */ void -brw_save_label(struct brw_compile *c, const char *name, GLuint position) +brw_save_label(struct brw_compile *c, unsigned l, GLuint position) { struct brw_glsl_label *label = CALLOC_STRUCT(brw_glsl_label); - label->name = name; + label->label = l; label->position = position; label->next = c->first_label; c->first_label = label; @@ -187,11 +187,11 @@ brw_save_label(struct brw_compile *c, const char *name, GLuint position) * Called for each OPCODE_CAL. */ void -brw_save_call(struct brw_compile *c, const char *name, GLuint call_pos) +brw_save_call(struct brw_compile *c, GLuint label, GLuint call_pos) { struct brw_glsl_call *call = CALLOC_STRUCT(brw_glsl_call); call->call_inst_pos = call_pos; - call->sub_name = name; + call->label = label; call->next = c->first_call; c->first_call = call; } @@ -201,11 +201,11 @@ brw_save_call(struct brw_compile *c, const char *name, GLuint call_pos) * Lookup a label, return label's position/offset. */ static GLuint -brw_lookup_label(struct brw_compile *c, const char *name) +brw_lookup_label(struct brw_compile *c, unsigned l) { const struct brw_glsl_label *label; for (label = c->first_label; label; label = label->next) { - if (strcmp(name, label->name) == 0) { + if (l == label->label) { return label->position; } } @@ -224,7 +224,7 @@ brw_resolve_cals(struct brw_compile *c) const struct brw_glsl_call *call; for (call = c->first_call; call; call = call->next) { - const GLuint sub_loc = brw_lookup_label(c, call->sub_name); + const GLuint sub_loc = brw_lookup_label(c, call->label); struct brw_instruction *brw_call_inst = &c->store[call->call_inst_pos]; struct brw_instruction *brw_sub_inst = &c->store[sub_loc]; GLint offset = brw_sub_inst - brw_call_inst; diff --git a/src/gallium/drivers/i965/brw_eu.h b/src/gallium/drivers/i965/brw_eu.h index ac5a623cac..3379522104 100644 --- a/src/gallium/drivers/i965/brw_eu.h +++ b/src/gallium/drivers/i965/brw_eu.h @@ -136,10 +136,10 @@ struct brw_compile { void -brw_save_label(struct brw_compile *c, const char *name, GLuint position); +brw_save_label(struct brw_compile *c, unsigned label, GLuint position); void -brw_save_call(struct brw_compile *c, const char *name, GLuint call_pos); +brw_save_call(struct brw_compile *c, unsigned label, GLuint call_pos); void brw_resolve_cals(struct brw_compile *c); diff --git a/src/gallium/drivers/i965/brw_vs.h b/src/gallium/drivers/i965/brw_vs.h index 58119567dc..2a2dbb3457 100644 --- a/src/gallium/drivers/i965/brw_vs.h +++ b/src/gallium/drivers/i965/brw_vs.h @@ -54,6 +54,7 @@ struct brw_vs_compile { struct brw_compile func; struct brw_vs_prog_key key; struct brw_vs_prog_data prog_data; + struct brw_chipset chipset; struct brw_vertex_shader *vp; @@ -88,7 +89,12 @@ struct brw_vs_compile { struct brw_instruction *if_inst[MAX_IF_DEPTH]; struct brw_instruction *loop_inst[MAX_LOOP_DEPTH]; + GLuint insn; + GLuint if_depth; + GLuint loop_depth; + GLuint end_offset; + struct brw_indirect stack_index; }; diff --git a/src/gallium/drivers/i965/brw_vs_emit.c b/src/gallium/drivers/i965/brw_vs_emit.c index 4daa98b29e..5366ab8514 100644 --- a/src/gallium/drivers/i965/brw_vs_emit.c +++ b/src/gallium/drivers/i965/brw_vs_emit.c @@ -35,19 +35,15 @@ #include "util/u_math.h" #include "tgsi/tgsi_ureg.h" +#include "tgsi/tgsi_ureg_parse.h" +#include "tgsi/tgsi_dump.h" +#include "tgsi/tgsi_info.h" #include "brw_context.h" #include "brw_vs.h" #include "brw_debug.h" -struct ureg_instruction { - unsigned opcode:8; - unsigned tex_target:3; - struct ureg_dst dst; - struct ureg_src src[3]; -}; - static struct brw_reg get_tmp( struct brw_vs_compile *c ) { @@ -149,7 +145,7 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) c->first_output = reg; c->first_overflow_output = 0; - if (BRW_IS_IGDNG(c->func.brw)) + if (c->chipset.is_igdng) mrf = 8; else mrf = 4; @@ -251,7 +247,7 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) */ attributes_in_vue = MAX2(c->nr_outputs, c->nr_inputs); - if (BRW_IS_IGDNG(c->func.brw)) + if (c->chipset.is_igdng) c->prog_data.urb_entry_size = (attributes_in_vue + 6 + 3) / 4; else c->prog_data.urb_entry_size = (attributes_in_vue + 2 + 3) / 4; @@ -1058,7 +1054,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) */ if (c->prog_data.writes_psiz || c->key.nr_userclip || - BRW_IS_965(p->brw)) + c->chipset.is_965) { struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); GLuint i; @@ -1089,7 +1085,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) * Later, clipping will detect ucp[6] and ensure the primitive is * clipped against all fixed planes. */ - if (BRW_IS_965(p->brw)) { + if (c->chipset.is_965) { brw_CMP(p, vec8(brw_null_reg()), BRW_CONDITIONAL_L, @@ -1117,7 +1113,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) brw_set_access_mode(p, BRW_ALIGN_1); brw_MOV(p, offset(m0, 2), ndc); - if (BRW_IS_IGDNG(p->brw)) { + if (c->chipset.is_igdng) { /* There are 20 DWs (D0-D19) in VUE vertex header on IGDNG */ brw_MOV(p, offset(m0, 3), pos); /* a portion of vertex header */ /* m4, m5 contain the distances from vertex to the user clip planeXXX. @@ -1205,6 +1201,9 @@ post_vs_emit( struct brw_vs_compile *c, static uint32_t get_predicate(const struct ureg_instruction *inst) { + /* XXX: disabling for now + */ +#if 0 if (inst->dst.CondMask == COND_TR) return BRW_PREDICATE_NONE; @@ -1237,11 +1236,15 @@ get_predicate(const struct ureg_instruction *inst) inst->dst.CondMask); return BRW_PREDICATE_NORMAL; } +#else + return BRW_PREDICATE_NORMAL; +#endif } static void emit_insn(struct brw_vs_compile *c, - const struct tgsi_full_instruction *insn) + const struct ureg_instruction *inst) { + struct brw_compile *p = &c->func; struct brw_reg args[3], dst; GLuint i; @@ -1253,9 +1256,6 @@ static void emit_insn(struct brw_vs_compile *c, /* Get argument regs. */ for (i = 0; i < 3; i++) { - const struct ureg_src src = inst->src[i]; - index = src.Index; - file = src.File; args[i] = get_arg(c, inst, i); } @@ -1263,16 +1263,13 @@ static void emit_insn(struct brw_vs_compile *c, * dst and arg, given the static allocation of registers. So * care needs to be taken emitting multi-operation instructions. */ - index = inst->dst.Index; - file = inst->dst.File; dst = get_dst(c, inst->dst); - if (inst->SaturateMode != SATURATE_OFF) { - debug_printf("Unsupported saturate %d in vertex shader", - inst->SaturateMode); + if (inst->dst.Saturate) { + debug_printf("Unsupported saturate in vertex shader"); } - switch (inst->Opcode) { + switch (inst->opcode) { case TGSI_OPCODE_ABS: brw_MOV(p, dst, brw_abs(args[0])); break; @@ -1291,7 +1288,7 @@ static void emit_insn(struct brw_vs_compile *c, case TGSI_OPCODE_DPH: brw_DPH(p, dst, args[0], args[1]); break; - case TGSI_OPCODE_NRM3: + case TGSI_OPCODE_NRM: emit_nrm(c, dst, args[0], 3); break; case TGSI_OPCODE_NRM4: @@ -1384,21 +1381,21 @@ static void emit_insn(struct brw_vs_compile *c, emit_xpd(p, dst, args[0], args[1]); break; case TGSI_OPCODE_IF: - assert(if_depth < MAX_IF_DEPTH); - if_inst[if_depth] = brw_IF(p, BRW_EXECUTE_8); + assert(c->if_depth < MAX_IF_DEPTH); + c->if_inst[c->if_depth] = brw_IF(p, BRW_EXECUTE_8); /* Note that brw_IF smashes the predicate_control field. */ - if_inst[if_depth]->header.predicate_control = get_predicate(inst); - if_depth++; + c->if_inst[c->if_depth]->header.predicate_control = get_predicate(inst); + c->if_depth++; break; case TGSI_OPCODE_ELSE: - if_inst[if_depth-1] = brw_ELSE(p, if_inst[if_depth-1]); + c->if_inst[c->if_depth-1] = brw_ELSE(p, c->if_inst[c->if_depth-1]); break; case TGSI_OPCODE_ENDIF: - assert(if_depth > 0); - brw_ENDIF(p, if_inst[--if_depth]); + assert(c->if_depth > 0); + brw_ENDIF(p, c->if_inst[--c->if_depth]); break; case TGSI_OPCODE_BGNLOOP: - loop_inst[loop_depth++] = brw_DO(p, BRW_EXECUTE_8); + c->loop_inst[c->loop_depth++] = brw_DO(p, BRW_EXECUTE_8); break; case TGSI_OPCODE_BRK: brw_set_predicate_control(p, get_predicate(inst)); @@ -1415,14 +1412,14 @@ static void emit_insn(struct brw_vs_compile *c, struct brw_instruction *inst0, *inst1; GLuint br = 1; - loop_depth--; + c->loop_depth--; - if (BRW_IS_IGDNG(brw)) + if (c->chipset.is_igdng) br = 2; - inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); + inst0 = inst1 = brw_WHILE(p, c->loop_inst[c->loop_depth]); /* patch all the BREAK/CONT instructions from last BEGINLOOP */ - while (inst0 > loop_inst[loop_depth]) { + while (inst0 > c->loop_inst[c->loop_depth]) { inst0--; if (inst0->header.opcode == TGSI_OPCODE_BRK) { inst0->bits3.if_else.jump_count = br * (inst1 - inst0 + 1); @@ -1442,41 +1439,37 @@ static void emit_insn(struct brw_vs_compile *c, break; case TGSI_OPCODE_CAL: brw_set_access_mode(p, BRW_ALIGN_1); - brw_ADD(p, deref_1d(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16)); + brw_ADD(p, deref_1d(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(stack_index), - get_addr_reg(stack_index), brw_imm_d(4)); - brw_save_call(p, inst->Comment, p->nr_insn); + brw_ADD(p, get_addr_reg(c->stack_index), + get_addr_reg(c->stack_index), brw_imm_d(4)); + brw_save_call(p, inst->label, p->nr_insn); brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); break; case TGSI_OPCODE_RET: - brw_ADD(p, get_addr_reg(stack_index), - get_addr_reg(stack_index), brw_imm_d(-4)); + 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_1d(stack_index, 0)); + brw_MOV(p, brw_ip_reg(), deref_1d(c->stack_index, 0)); brw_set_access_mode(p, BRW_ALIGN_16); break; case TGSI_OPCODE_END: - end_offset = p->nr_insn; + c->end_offset = p->nr_insn; /* this instruction will get patched later to jump past subroutine * code, etc. */ brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); break; - case TGSI_OPCODE_PRINT: - /* no-op */ - break; case TGSI_OPCODE_BGNSUB: - brw_save_label(p, inst->Comment, p->nr_insn); + brw_save_label(p, p->nr_insn, p->nr_insn); break; case TGSI_OPCODE_ENDSUB: /* no-op */ break; default: debug_printf("Unsupported opcode %i (%s) in vertex shader", - inst->Opcode, inst->Opcode < MAX_OPCODE ? - _mesa_opcode_string(inst->Opcode) : - "unknown"); + inst->opcode, + tgsi_get_opcode_name(inst->opcode)); } /* Set the predication update on the last instruction of the native @@ -1485,12 +1478,16 @@ static void emit_insn(struct brw_vs_compile *c, * This would be problematic if it was set on a math instruction, * but that shouldn't be the case with the current GLSL compiler. */ +#if 0 + /* XXX: disabled + */ if (inst->CondUpdate) { struct brw_instruction *hw_insn = &p->store[p->nr_insn - 1]; assert(hw_insn->header.destreg__conditionalmod == 0); hw_insn->header.destreg__conditionalmod = BRW_CONDITIONAL_NZ; } +#endif release_tmps(c); } @@ -1498,24 +1495,19 @@ static void emit_insn(struct brw_vs_compile *c, /* Emit the vertex program instructions here. */ -void brw_vs_emit(struct brw_vs_compile *c ) +void brw_vs_emit(struct brw_vs_compile *c) { struct brw_compile *p = &c->func; - struct brw_context *brw = p->brw; - GLuint insn, if_depth = 0, loop_depth = 0; - GLuint end_offset = 0; struct brw_instruction *end_inst, *last_inst; - const struct brw_indirect stack_index = brw_indirect(0, 0); - struct tgsi_parse_context parse; - struct tgsi_full_declaration *decl; - GLuint index; - GLuint file; + struct ureg_parse_context parse; + struct ureg_declaration *decl; + struct ureg_declaration *imm; + struct ureg_declaration *insn; - if (BRW_DEBUG & DEBUG_VS) { - debug_printf("vs-mesa:\n"); - _mesa_print_program(&c->vp->program.Base); - debug_printf("\n"); - } + if (BRW_DEBUG & DEBUG_VS) + tgsi_dump(c->vp->tokens, 0); + + c->stack_index = brw_indirect(0, 0); brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_set_access_mode(p, BRW_ALIGN_16); @@ -1523,12 +1515,15 @@ void brw_vs_emit(struct brw_vs_compile *c ) /* Static register allocation */ brw_vs_alloc_regs(c); - brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); + brw_MOV(p, get_addr_reg(c->stack_index), brw_address(c->stack)); - for (insn = 0; insn < nr_insns; insn++) { + while (ureg_next_decl(&parse, &decl)) { + } - const struct ureg_instruction *inst = &c->vp->program.Base.Instructions[insn]; - + while (ureg_next_immediate(&parse, &imm)) { + } + + while (ureg_next_instruction(&parse, &insn)) { } end_inst = &p->store[end_offset]; diff --git a/src/gallium/drivers/i965/brw_wm.h b/src/gallium/drivers/i965/brw_wm.h index 5bc2a49c1f..084430cf28 100644 --- a/src/gallium/drivers/i965/brw_wm.h +++ b/src/gallium/drivers/i965/brw_wm.h @@ -34,6 +34,7 @@ #define BRW_WM_H #include "tgsi/tgsi_ureg.h" +#include "tgsi/tgsi_ureg_parse.h" #include "brw_context.h" #include "brw_eu.h" @@ -163,14 +164,6 @@ struct brw_wm_instruction { #define BRW_WM_MAX_SUBROUTINE 16 -struct ureg_instruction { - unsigned opcode:8; - unsigned tex_target:3; - struct ureg_dst dst; - struct ureg_src src[3]; -}; - - /* New opcodes to track internal operations required for WM unit. * These are added early so that the registers used can be tracked, * freed and reused like those of other instructions. diff --git a/src/gallium/drivers/i965/brw_wm_glsl.c b/src/gallium/drivers/i965/brw_wm_glsl.c index 23f7ba16fd..59bc4ef701 100644 --- a/src/gallium/drivers/i965/brw_wm_glsl.c +++ b/src/gallium/drivers/i965/brw_wm_glsl.c @@ -1867,7 +1867,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) 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)); - brw_save_call(&c->func, inst->Comment, p->nr_insn); + brw_save_call(&c->func, inst->label, p->nr_insn); brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16)); brw_pop_insn_state(p); break; -- cgit v1.2.3 From 99cc0fd67597cbcd6106afcf437a0d5e2431c9df Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 29 Oct 2009 20:18:01 +0000 Subject: i965g: work in progress on fragment shaders --- src/gallium/drivers/i965/brw_context.h | 10 +- src/gallium/drivers/i965/brw_eu.c | 20 +- src/gallium/drivers/i965/brw_eu.h | 8 +- src/gallium/drivers/i965/brw_pipe_depth.c | 42 +- src/gallium/drivers/i965/brw_pipe_rast.c | 18 + src/gallium/drivers/i965/brw_pipe_rast.h | 1 + src/gallium/drivers/i965/brw_pipe_shader.c | 4 +- src/gallium/drivers/i965/brw_screen.h | 7 + src/gallium/drivers/i965/brw_vs_emit.c | 2 - src/gallium/drivers/i965/brw_wm.c | 167 ++--- src/gallium/drivers/i965/brw_wm.h | 41 +- src/gallium/drivers/i965/brw_wm_debug.c | 17 +- src/gallium/drivers/i965/brw_wm_emit.c | 195 +++--- src/gallium/drivers/i965/brw_wm_fp.c | 1031 ++++++++++------------------ src/gallium/drivers/i965/brw_wm_glsl.c | 12 +- src/gallium/drivers/i965/brw_wm_pass0.c | 73 +- src/gallium/drivers/i965/brw_wm_pass1.c | 26 +- src/gallium/drivers/i965/brw_wm_state.c | 8 +- 18 files changed, 682 insertions(+), 1000 deletions(-) (limited to 'src/gallium/drivers/i965/brw_eu.c') diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h index 7b85363e9f..e6c3161066 100644 --- a/src/gallium/drivers/i965/brw_context.h +++ b/src/gallium/drivers/i965/brw_context.h @@ -132,6 +132,8 @@ struct brw_depth_stencil_state { struct brw_cc2 cc2; struct brw_cc3 cc3; struct brw_cc7 cc7; + + unsigned iz_lookup; }; @@ -164,7 +166,10 @@ struct brw_fragment_shader { const struct tgsi_token *tokens; struct tgsi_shader_info info; - GLboolean isGLSL; + unsigned iz_lookup; + + boolean uses_depth:1; + boolean has_flow_control:1; unsigned id; struct brw_winsys_buffer *const_buffer; /** Program constant buffer/surface */ @@ -194,6 +199,7 @@ struct brw_fragment_shader { #define PIPE_NEW_COLOR_BUFFERS 0x40000 #define PIPE_NEW_QUERY 0x80000 #define PIPE_NEW_SCISSOR 0x100000 +#define PIPE_NEW_BOUND_TEXTURES 0x200000 @@ -487,7 +493,7 @@ struct brw_context const struct brw_rasterizer_state *rast; const struct brw_depth_stencil_state *zstencil; - const struct pipe_texture *texture[PIPE_MAX_SAMPLERS]; + const struct brw_texture *texture[PIPE_MAX_SAMPLERS]; const struct pipe_sampler *sampler[PIPE_MAX_SAMPLERS]; unsigned num_textures; unsigned num_samplers; diff --git a/src/gallium/drivers/i965/brw_eu.c b/src/gallium/drivers/i965/brw_eu.c index 1189a35b6f..de43b14512 100644 --- a/src/gallium/drivers/i965/brw_eu.c +++ b/src/gallium/drivers/i965/brw_eu.c @@ -150,22 +150,22 @@ const GLuint *brw_get_program( struct brw_compile *p, /** * For each OPCODE_BGNSUB we create one of these. */ -struct brw_glsl_label +struct brw_eu_label { GLuint label; /**< the label number */ GLuint position; /**< the position of the brw instruction for this label */ - struct brw_glsl_label *next; /**< next in linked list */ + struct brw_eu_label *next; /**< next in linked list */ }; /** * For each OPCODE_CAL we create one of these. */ -struct brw_glsl_call +struct brw_eu_call { GLuint call_inst_pos; /**< location of the CAL instruction */ GLuint label; - struct brw_glsl_call *next; /**< next in linked list */ + struct brw_eu_call *next; /**< next in linked list */ }; @@ -175,7 +175,7 @@ struct brw_glsl_call void brw_save_label(struct brw_compile *c, unsigned l, GLuint position) { - struct brw_glsl_label *label = CALLOC_STRUCT(brw_glsl_label); + struct brw_eu_label *label = CALLOC_STRUCT(brw_eu_label); label->label = l; label->position = position; label->next = c->first_label; @@ -189,7 +189,7 @@ brw_save_label(struct brw_compile *c, unsigned l, GLuint position) void brw_save_call(struct brw_compile *c, GLuint label, GLuint call_pos) { - struct brw_glsl_call *call = CALLOC_STRUCT(brw_glsl_call); + struct brw_eu_call *call = CALLOC_STRUCT(brw_eu_call); call->call_inst_pos = call_pos; call->label = label; call->next = c->first_call; @@ -203,7 +203,7 @@ brw_save_call(struct brw_compile *c, GLuint label, GLuint call_pos) static GLuint brw_lookup_label(struct brw_compile *c, unsigned l) { - const struct brw_glsl_label *label; + const struct brw_eu_label *label; for (label = c->first_label; label; label = label->next) { if (l == label->label) { return label->position; @@ -221,7 +221,7 @@ brw_lookup_label(struct brw_compile *c, unsigned l) void brw_resolve_cals(struct brw_compile *c) { - const struct brw_glsl_call *call; + const struct brw_eu_call *call; for (call = c->first_call; call; call = call->next) { const GLuint sub_loc = brw_lookup_label(c, call->label); @@ -235,7 +235,7 @@ brw_resolve_cals(struct brw_compile *c) /* free linked list of calls */ { - struct brw_glsl_call *call, *next; + struct brw_eu_call *call, *next; for (call = c->first_call; call; call = next) { next = call->next; FREE(call); @@ -245,7 +245,7 @@ brw_resolve_cals(struct brw_compile *c) /* free linked list of labels */ { - struct brw_glsl_label *label, *next; + struct brw_eu_label *label, *next; for (label = c->first_label; label; label = next) { next = label->next; FREE(label); diff --git a/src/gallium/drivers/i965/brw_eu.h b/src/gallium/drivers/i965/brw_eu.h index 3379522104..7bddc3859c 100644 --- a/src/gallium/drivers/i965/brw_eu.h +++ b/src/gallium/drivers/i965/brw_eu.h @@ -109,8 +109,8 @@ struct brw_indirect { }; -struct brw_glsl_label; -struct brw_glsl_call; +struct brw_eu_label; +struct brw_eu_call; @@ -130,8 +130,8 @@ struct brw_compile { GLboolean single_program_flow; struct brw_context *brw; - struct brw_glsl_label *first_label; /**< linked list of labels */ - struct brw_glsl_call *first_call; /**< linked list of CALs */ + struct brw_eu_label *first_label; /**< linked list of labels */ + struct brw_eu_call *first_call; /**< linked list of CALs */ }; diff --git a/src/gallium/drivers/i965/brw_pipe_depth.c b/src/gallium/drivers/i965/brw_pipe_depth.c index 33fe517e0b..e010d76e0d 100644 --- a/src/gallium/drivers/i965/brw_pipe_depth.c +++ b/src/gallium/drivers/i965/brw_pipe_depth.c @@ -5,6 +5,10 @@ #include "brw_context.h" #include "brw_defines.h" +/* XXX: Fixme - include this to get IZ_ defines + */ +#include "brw_wm.h" + static unsigned brw_translate_compare_func(unsigned func) { switch (func) { @@ -55,13 +59,9 @@ static unsigned translate_stencil_op(unsigned op) } } - -static void * -brw_create_depth_stencil_state( struct pipe_context *pipe, - const struct pipe_depth_stencil_alpha_state *templ ) +static void create_bcc_state( struct brw_depth_stencil_state *zstencil, + const struct pipe_depth_stencil_alpha_state *templ ) { - struct brw_depth_stencil_state *zstencil = CALLOC_STRUCT(brw_depth_stencil_state); - if (templ->stencil[0].enabled) { zstencil->cc0.stencil_enable = 1; zstencil->cc0.stencil_func = @@ -108,6 +108,36 @@ brw_create_depth_stencil_state( struct pipe_context *pipe, zstencil->cc2.depth_test_function = brw_translate_compare_func(templ->depth.func); zstencil->cc2.depth_write_enable = templ->depth.writemask; } +} + +static void create_wm_iz_state( struct brw_depth_stencil_state *zstencil ) +{ + if (zstencil->cc3.alpha_test) + zstencil->iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT; + + if (zstencil->cc2.depth_test) + zstencil->iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT; + + if (zstencil->cc2.depth_write_enable) + zstencil->iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT; + + if (zstencil->cc0.stencil_enable) + zstencil->iz_lookup |= IZ_STENCIL_TEST_ENABLE_BIT; + + if (zstencil->cc0.stencil_write_enable) + zstencil->iz_lookup |= IZ_STENCIL_WRITE_ENABLE_BIT; + +} + + +static void * +brw_create_depth_stencil_state( struct pipe_context *pipe, + const struct pipe_depth_stencil_alpha_state *templ ) +{ + struct brw_depth_stencil_state *zstencil = CALLOC_STRUCT(brw_depth_stencil_state); + + create_bcc_state( zstencil, templ ); + create_wm_iz_state( zstencil ); return (void *)zstencil; } diff --git a/src/gallium/drivers/i965/brw_pipe_rast.c b/src/gallium/drivers/i965/brw_pipe_rast.c index 86822d478a..51159bf147 100644 --- a/src/gallium/drivers/i965/brw_pipe_rast.c +++ b/src/gallium/drivers/i965/brw_pipe_rast.c @@ -64,3 +64,21 @@ calculate_line_stipple_rast() bls.bits1.inverse_repeat_count = tmpi; } + + + +static void +calculate_wm_lookup() +{ + if (rast->fill_cw == PIPE_POLYGON_MODE_LINE && + rast->fill_ccw == PIPE_POLYGON_MODE_LINE) { + line_aa = AA_ALWAYS; + } + else if (rast->fill_cw == PIPE_POLYGON_MODE_LINE || + rast->fill_ccw == PIPE_POLYGON_MODE_LINE) { + line_aa = AA_SOMETIMES; + } + else { + line_aa = AA_NEVER; + } +} diff --git a/src/gallium/drivers/i965/brw_pipe_rast.h b/src/gallium/drivers/i965/brw_pipe_rast.h index 800a9208a7..9354f01e18 100644 --- a/src/gallium/drivers/i965/brw_pipe_rast.h +++ b/src/gallium/drivers/i965/brw_pipe_rast.h @@ -10,6 +10,7 @@ struct brw_rasterizer_state { */ struct brw_clip_prog_key clip_key; struct brw_line_stipple bls; + unsigned unfilled_aa_line; }; #endif diff --git a/src/gallium/drivers/i965/brw_pipe_shader.c b/src/gallium/drivers/i965/brw_pipe_shader.c index 8b61da763c..6e37eac634 100644 --- a/src/gallium/drivers/i965/brw_pipe_shader.c +++ b/src/gallium/drivers/i965/brw_pipe_shader.c @@ -39,7 +39,7 @@ * as flow conditionals, loops, subroutines. * Some GLSL shaders may use these features, others might not. */ -GLboolean brw_wm_is_glsl(const struct brw_fragment_shader *fp) +GLboolean brw_wm_has_flow_control(const struct brw_fragment_shader *fp) { return (fp->info.insn_count[TGSI_OPCODE_ARL] > 0 || fp->info.insn_count[TGSI_OPCODE_IF] > 0 || @@ -144,7 +144,7 @@ static void brwProgramStringNotify( struct brw_context *brw, if (newFP == curFP) brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM; newFP->id = brw->program_id++; - newFP->isGLSL = brw_wm_is_glsl(fprog); + newFP->has_flow_control = brw_wm_has_flow_control(fprog); } else if (target == GL_VERTEX_PROGRAM_ARB) { struct gl_vertex_program *vprog = (struct gl_vertex_program *) prog; diff --git a/src/gallium/drivers/i965/brw_screen.h b/src/gallium/drivers/i965/brw_screen.h index eafd8ddf77..efa27db1e0 100644 --- a/src/gallium/drivers/i965/brw_screen.h +++ b/src/gallium/drivers/i965/brw_screen.h @@ -64,6 +64,13 @@ struct brw_buffer boolean is_user_buffer; }; +struct brw_texture +{ + struct pipe_texture base; + + ubyte shader_swizzle; +}; + /* * Cast wrappers diff --git a/src/gallium/drivers/i965/brw_vs_emit.c b/src/gallium/drivers/i965/brw_vs_emit.c index 6809bccdec..bcc5c5f713 100644 --- a/src/gallium/drivers/i965/brw_vs_emit.c +++ b/src/gallium/drivers/i965/brw_vs_emit.c @@ -1013,8 +1013,6 @@ static struct brw_reg get_arg( struct brw_vs_compile *c, src->SrcRegister.SwizzleZ, src->SrcRegister.SwizzleW); - /* Note this is ok for non-swizzle instructions: - */ reg.negate = src->SrcRegister.Negate ? 1 : 0; /* XXX: abs, absneg diff --git a/src/gallium/drivers/i965/brw_wm.c b/src/gallium/drivers/i965/brw_wm.c index f0dabfcfd0..33602b59c1 100644 --- a/src/gallium/drivers/i965/brw_wm.c +++ b/src/gallium/drivers/i965/brw_wm.c @@ -28,14 +28,17 @@ * Authors: * Keith Whitwell */ +#include "pipe/p_error.h" #include "tgsi/tgsi_info.h" #include "brw_context.h" +#include "brw_screen.h" #include "brw_util.h" #include "brw_wm.h" #include "brw_state.h" #include "brw_debug.h" +#include "brw_pipe_rast.h" /** Return number of src args for given instruction */ @@ -85,12 +88,12 @@ GLuint brw_wm_is_scalar_result( GLuint opcode ) /** - * Do GPU code generation for non-GLSL shader. non-GLSL shaders have - * no flow control instructions so we can more readily do SSA-style - * optimizations. + * Do GPU code generation for shaders without flow control. Shaders + * without flow control instructions can more readily be analysed for + * SSA-style optimizations. */ static void -brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c) +brw_wm_linear_shader_emit(struct brw_context *brw, struct brw_wm_compile *c) { /* Augment fragment program. Add instructions for pre- and * post-fragment-program tasks such as interpolation and fogging. @@ -136,7 +139,7 @@ brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c) * Depending on the instructions used (i.e. flow control instructions) * we'll use one of two code generators. */ -static void do_wm_prog( struct brw_context *brw, +static int do_wm_prog( struct brw_context *brw, struct brw_fragment_shader *fp, struct brw_wm_prog_key *key) { @@ -153,7 +156,7 @@ static void do_wm_prog( struct brw_context *brw, * without triggering a segfault, no way to signal, * so just return. */ - return; + return PIPE_ERROR_OUT_OF_MEMORY; } } else { memset(c, 0, sizeof(*brw->wm.compile_data)); @@ -166,19 +169,19 @@ static void do_wm_prog( struct brw_context *brw, brw_init_compile(brw, &c->func); /* temporary sanity check assertion */ - assert(fp->isGLSL == brw_wm_is_glsl(&c->fp->program)); + assert(fp->has_flow_control == brw_wm_has_flow_control(c->fp)); /* * Shader which use GLSL features such as flow control are handled * differently from "simple" shaders. */ - if (fp->isGLSL) { + if (fp->has_flow_control) { c->dispatch_width = 8; - brw_wm_glsl_emit(brw, c); + brw_wm_branching_shader_emit(brw, c); } else { c->dispatch_width = 16; - brw_wm_non_glsl_emit(brw, c); + brw_wm_linear_shader_emit(brw, c); } if (BRW_DEBUG & DEBUG_WM) @@ -195,6 +198,8 @@ static void do_wm_prog( struct brw_context *brw, program, program_size, &c->prog_data, &brw->wm.prog_data ); + + return 0; } @@ -202,71 +207,36 @@ static void do_wm_prog( struct brw_context *brw, static void brw_wm_populate_key( struct brw_context *brw, struct brw_wm_prog_key *key ) { - /* BRW_NEW_FRAGMENT_PROGRAM */ - const struct brw_fragment_program *fp = brw->curr.fragment_shader; - GLboolean uses_depth = (fp->program.Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0; - GLuint lookup = 0; - GLuint line_aa; - GLuint i; + unsigned lookup, line_aa; + unsigned i; memset(key, 0, sizeof(*key)); - /* Build the index for table lookup + /* PIPE_NEW_FRAGMENT_SHADER + * PIPE_NEW_DEPTH_STENCIL_ALPHA */ - /* _NEW_COLOR */ - if (fp->program.UsesKill || - ctx->Color.AlphaEnabled) - lookup |= IZ_PS_KILL_ALPHATEST_BIT; - - if (fp->program.Base.OutputsWritten & (1<Depth.Test) - lookup |= IZ_DEPTH_TEST_ENABLE_BIT; - - if (ctx->Depth.Test && - ctx->Depth.Mask) /* ?? */ - lookup |= IZ_DEPTH_WRITE_ENABLE_BIT; + lookup = (brw->curr.zstencil->iz_lookup | + brw->curr.fragment_shader->iz_lookup); - /* _NEW_STENCIL */ - if (ctx->Stencil._Enabled) { - lookup |= IZ_STENCIL_TEST_ENABLE_BIT; - if (ctx->Stencil.WriteMask[0] || - ctx->Stencil.WriteMask[ctx->Stencil._BackFace]) - lookup |= IZ_STENCIL_WRITE_ENABLE_BIT; - } - - line_aa = AA_NEVER; - - /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */ - if (ctx->Line.SmoothFlag) { - if (brw->intel.reduced_primitive == GL_LINES) { - line_aa = AA_ALWAYS; - } - else if (brw->intel.reduced_primitive == GL_TRIANGLES) { - if (ctx->Polygon.FrontMode == GL_LINE) { - line_aa = AA_SOMETIMES; - - if (ctx->Polygon.BackMode == GL_LINE || - (ctx->Polygon.CullFlag && - ctx->Polygon.CullFaceMode == GL_BACK)) - line_aa = AA_ALWAYS; - } - else if (ctx->Polygon.BackMode == GL_LINE) { - line_aa = AA_SOMETIMES; - - if ((ctx->Polygon.CullFlag && - ctx->Polygon.CullFaceMode == GL_FRONT)) - line_aa = AA_ALWAYS; - } - } + /* PIPE_NEW_RAST + * BRW_NEW_REDUCED_PRIMITIVE + */ + switch (brw->reduced_primitive) { + case PIPE_PRIM_POINTS: + line_aa = AA_NEVER; + break; + case PIPE_PRIM_LINES: + line_aa = AA_ALWAYS; + break; + default: + line_aa = brw->curr.rast->unfilled_aa_line; + break; } brw_wm_lookup_iz(line_aa, lookup, - uses_depth, + brw->curr.fragment_shader->uses_depth, key); /* Revisit this, figure out if it's really useful, and either push @@ -276,54 +246,39 @@ static void brw_wm_populate_key( struct brw_context *brw, key->proj_attrib_mask = ~0; /*brw->wm.input_size_masks[4-1];*/ /* PIPE_NEW_RAST */ - key->flat_shade = brw->rast.flat_shade; + key->flat_shade = brw->curr.rast->templ.flatshade; /* This can be determined by looking at the INTERP mode each input decl. */ - key->linear_color = 0; - - /* _NEW_TEXTURE */ - for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { - if (i < brw->nr_textures) { - const struct gl_texture_unit *unit = &ctx->Texture.Unit[i]; - const struct gl_texture_object *t = unit->_Current; - const struct gl_texture_image *img = t->Image[0][t->BaseLevel]; - - if (img->InternalFormat == GL_YCBCR_MESA) { - key->yuvtex_mask |= 1 << i; - if (img->TexFormat->MesaFormat == MESA_FORMAT_YCBCR) - key->yuvtex_swap_mask |= 1 << i; - } + key->linear_attrib_mask = 0; - key->tex_swizzles[i] = t->_Swizzle; + /* PIPE_NEW_BOUND_TEXTURES */ + for (i = 0; i < brw->curr.num_textures; i++) { + const struct brw_texture *tex = brw->curr.texture[i]; - if (0) - key->shadowtex_mask |= 1<tex_swizzles[i] = SWIZZLE_NOOP; - } - } + if (tex->base.format == PIPE_FORMAT_YCBCR) + key->yuvtex_mask |= 1 << i; + if (tex->base.format == PIPE_FORMAT_YCBCR_REV) + key->yuvtex_swap_mask |= 1 << i; - /* _NEW_FRAMEBUFFER */ - if (brw->intel.driDrawable != NULL) { - key->drawable_height = brw->fb.cbufs[0].height; + /* XXX: shadow texture + */ + /* key->shadowtex_mask |= 1<vp_nr_outputs_written = brw->vs.prog_data->nr_outputs_written; + key->vp_nr_outputs = brw->vs.prog_data->nr_outputs; /* The unique fragment program ID */ - key->program_string_id = fp->id; + key->program_string_id = brw->curr.fragment_shader->id; } -static void brw_prepare_wm_prog(struct brw_context *brw) +static int brw_prepare_wm_prog(struct brw_context *brw) { struct brw_wm_prog_key key; - struct brw_fragment_program *fp = (struct brw_fragment_program *) - brw->fragment_program; + struct brw_fragment_shader *fs = brw->curr.fragment_shader; brw_wm_populate_key(brw, &key); @@ -335,23 +290,19 @@ static void brw_prepare_wm_prog(struct brw_context *brw) NULL, 0, &brw->wm.prog_data); if (brw->wm.prog_bo == NULL) - do_wm_prog(brw, fp, &key); + return do_wm_prog(brw, fs, &key); + + return 0; } const struct brw_tracked_state brw_wm_prog = { .dirty = { - .mesa = (_NEW_COLOR | - _NEW_DEPTH | - _NEW_HINT | - _NEW_STENCIL | - _NEW_POLYGON | - _NEW_LINE | - _NEW_LIGHT | - _NEW_BUFFERS | - _NEW_TEXTURE), - .brw = (BRW_NEW_FRAGMENT_PROGRAM | - BRW_NEW_WM_INPUT_DIMENSIONS | + .mesa = (PIPE_NEW_FRAGMENT_SHADER | + PIPE_NEW_DEPTH_STENCIL_ALPHA | + PIPE_NEW_RAST | + PIPE_NEW_BOUND_TEXTURES), + .brw = (BRW_NEW_WM_INPUT_DIMENSIONS | BRW_NEW_REDUCED_PRIMITIVE), .cache = CACHE_NEW_VS_PROG, }, diff --git a/src/gallium/drivers/i965/brw_wm.h b/src/gallium/drivers/i965/brw_wm.h index 084430cf28..2cd5bb7081 100644 --- a/src/gallium/drivers/i965/brw_wm.h +++ b/src/gallium/drivers/i965/brw_wm.h @@ -33,9 +33,6 @@ #ifndef BRW_WM_H #define BRW_WM_H -#include "tgsi/tgsi_ureg.h" -#include "tgsi/tgsi_ureg_parse.h" - #include "brw_context.h" #include "brw_eu.h" @@ -59,8 +56,8 @@ #define AA_ALWAYS 2 struct brw_wm_prog_key { - unsigned proj_attrib_mask; /**< one bit per fragment program attribute */ - unsigned linear_attrib_mask:1; /**< linear interpolation vs perspective interp */ + unsigned proj_attrib_mask; /**< one bit per fragment program attribute */ + unsigned linear_attrib_mask; /**< linear interpolation vs perspective interp */ GLuint source_depth_reg:3; GLuint aa_dest_stencil_reg:3; @@ -75,11 +72,10 @@ struct brw_wm_prog_key { GLuint yuvtex_mask:16; GLuint yuvtex_swap_mask:16; /* UV swaped */ - GLuint tex_swizzles[BRW_MAX_TEX_UNIT]; - - GLuint program_string_id:32; + GLuint vp_nr_outputs:6; + GLuint nr_cbufs:3; - GLuint vp_nr_outputs_written; + GLuint program_string_id; }; @@ -146,9 +142,8 @@ struct brw_wm_instruction { GLuint opcode:8; GLuint saturate:1; GLuint writemask:4; - GLuint tex_unit:4; /* texture unit for TEX, TXD, TXP instructions */ - GLuint tex_idx:3; /* TEXTURE_1D,2D,3D,CUBE,RECT_INDEX source target */ - GLuint tex_shadow:1; /* do shadow comparison? */ + GLuint tex_unit:4; /* texture/sampler unit for texture instructions */ + GLuint tex_target:4; /* TGSI_TEXTURE_x for texture instructions*/ GLuint eot:1; /* End of thread indicator for FB_WRITE*/ GLuint target:10; /* target binding table index for FB_WRITE*/ }; @@ -180,15 +175,17 @@ struct brw_wm_instruction { #define WM_FRONTFACING (MAX_OPCODE + 8) #define MAX_WM_OPCODE (MAX_OPCODE + 9) -#define PROGRAM_PAYLOAD (TGSI_FILE_COUNT) -#define PAYLOAD_DEPTH (FRAG_ATTRIB_MAX) +#define BRW_FILE_PAYLOAD (TGSI_FILE_COUNT) +#define PAYLOAD_DEPTH (FRAG_ATTRIB_MAX) /* ?? */ + +struct brw_passfp_program; struct brw_wm_compile { struct brw_compile func; struct brw_wm_prog_key key; struct brw_wm_prog_data prog_data; - struct brw_fragment_program *fp; + struct brw_fragment_shader *fp; GLfloat (*env_param)[4]; @@ -201,15 +198,7 @@ struct brw_wm_compile { * simplifying and adding instructions for interpolation and * framebuffer writes. */ - struct ureg_instruction prog_instructions[BRW_WM_MAX_INSN]; - GLuint nr_fp_insns; - GLuint fp_temp; - GLuint fp_interp_emitted; - GLuint fp_fragcolor_emitted; - - struct ureg_src pixel_xy; - struct ureg_src delta_xy; - struct ureg_src pixel_w; + struct brw_passfp_program *pass_fp; struct brw_wm_value vreg[BRW_WM_MAX_VREG]; @@ -298,8 +287,8 @@ void brw_wm_lookup_iz( GLuint line_aa, GLboolean ps_uses_depth, struct brw_wm_prog_key *key ); -//GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp); -void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c); +GLboolean brw_wm_has_flow_control(const struct brw_fragment_shader *fp); +void brw_wm_branching_shader_emit(struct brw_context *brw, struct brw_wm_compile *c); void emit_ddxy(struct brw_compile *p, const struct brw_reg *dst, diff --git a/src/gallium/drivers/i965/brw_wm_debug.c b/src/gallium/drivers/i965/brw_wm_debug.c index 04dec5ba39..65d7626eea 100644 --- a/src/gallium/drivers/i965/brw_wm_debug.c +++ b/src/gallium/drivers/i965/brw_wm_debug.c @@ -28,7 +28,8 @@ * Authors: * Keith Whitwell */ - + +#include "tgsi/tgsi_info.h" #include "brw_context.h" #include "brw_wm.h" @@ -49,10 +50,10 @@ void brw_wm_print_value( struct brw_wm_compile *c, value - c->creg < BRW_WM_MAX_PARAM) debug_printf("c%d", value - c->creg); else if (value - c->payload.input_interp >= 0 && - value - c->payload.input_interp < FRAG_ATTRIB_MAX) + value - c->payload.input_interp < PIPE_MAX_SHADER_INPUTS) debug_printf("i%d", value - c->payload.input_interp); else if (value - c->payload.depth >= 0 && - value - c->payload.depth < FRAG_ATTRIB_MAX) + value - c->payload.depth < PIPE_MAX_SHADER_INPUTS) debug_printf("d%d", value - c->payload.depth); else debug_printf("?"); @@ -100,10 +101,10 @@ void brw_wm_print_insn( struct brw_wm_compile *c, if (inst->writemask != BRW_WRITEMASK_XYZW) debug_printf(".%s%s%s%s", - GET_BIT(inst->writemask, 0) ? "x" : "", - GET_BIT(inst->writemask, 1) ? "y" : "", - GET_BIT(inst->writemask, 2) ? "z" : "", - GET_BIT(inst->writemask, 3) ? "w" : ""); + (inst->writemask & BRW_WRITEMASK_X) ? "x" : "", + (inst->writemask & BRW_WRITEMASK_Y) ? "y" : "", + (inst->writemask & BRW_WRITEMASK_Z) ? "z" : "", + (inst->writemask & BRW_WRITEMASK_W) ? "w" : ""); switch (inst->opcode) { case WM_PIXELXY: @@ -134,7 +135,7 @@ void brw_wm_print_insn( struct brw_wm_compile *c, debug_printf(" = FRONTFACING"); break; default: - debug_printf(" = %s", _mesa_opcode_string(inst->opcode)); + debug_printf(" = %s", tgsi_get_opcode_info(inst->opcode)->mnemonic); break; } diff --git a/src/gallium/drivers/i965/brw_wm_emit.c b/src/gallium/drivers/i965/brw_wm_emit.c index 5f7ae6592c..a705d8b344 100644 --- a/src/gallium/drivers/i965/brw_wm_emit.c +++ b/src/gallium/drivers/i965/brw_wm_emit.c @@ -28,10 +28,13 @@ * Authors: * Keith Whitwell */ - + +#include "util/u_math.h" +#include "tgsi/tgsi_info.h" #include "brw_context.h" #include "brw_wm.h" +#include "brw_debug.h" /* Not quite sure how correct this is - need to understand horiz * vs. vertical strides a little better. @@ -45,15 +48,15 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) /* Payload R0: * - * R0.0 -- pixel mask, one bit for each of 4 pixels in 4 tiles, + * R0.0 -- pixel mask, one bit for each of 4 pixels in 4 quads, * corresponding to each of the 16 execution channels. * R0.1..8 -- ? * R1.0 -- triangle vertex 0.X * R1.1 -- triangle vertex 0.Y - * R1.2 -- tile 0 x,y coords (2 packed uwords) - * R1.3 -- tile 1 x,y coords (2 packed uwords) - * R1.4 -- tile 2 x,y coords (2 packed uwords) - * R1.5 -- tile 3 x,y coords (2 packed uwords) + * R1.2 -- quad 0 x,y coords (2 packed uwords) + * R1.3 -- quad 1 x,y coords (2 packed uwords) + * R1.4 -- quad 2 x,y coords (2 packed uwords) + * R1.5 -- quad 3 x,y coords (2 packed uwords) * R1.6 -- ? * R1.7 -- ? * R1.8 -- ? @@ -134,11 +137,17 @@ static void emit_wpos_xy(struct brw_wm_compile *c, /* XXX: is this needed any more, or is this a NOOP? */ if (mask & BRW_WRITEMASK_Y) { +#if 0 /* Y' = height - 1 - Y */ brw_ADD(p, dst[1], negate(retype(arg0[1], BRW_REGISTER_TYPE_W)), brw_imm_d(c->key.drawable_height - 1)); +#else + brw_MOV(p, + dst[0], + retype(arg0[0], BRW_REGISTER_TYPE_W)); +#endif } } @@ -279,28 +288,28 @@ static void emit_frontfacing( struct brw_compile *p, /* For OPCODE_DDX and OPCODE_DDY, per channel of output we've got input * looking like: * - * arg0: ss0.tl ss0.tr ss0.bl ss0.br ss1.tl ss1.tr ss1.bl ss1.br + * arg0: q0.tl q0.tr q0.bl q0.br q1.tl q1.tr q1.bl q1.br * * and we're trying to produce: * * DDX DDY - * dst: (ss0.tr - ss0.tl) (ss0.tl - ss0.bl) - * (ss0.tr - ss0.tl) (ss0.tr - ss0.br) - * (ss0.br - ss0.bl) (ss0.tl - ss0.bl) - * (ss0.br - ss0.bl) (ss0.tr - ss0.br) - * (ss1.tr - ss1.tl) (ss1.tl - ss1.bl) - * (ss1.tr - ss1.tl) (ss1.tr - ss1.br) - * (ss1.br - ss1.bl) (ss1.tl - ss1.bl) - * (ss1.br - ss1.bl) (ss1.tr - ss1.br) + * dst: (q0.tr - q0.tl) (q0.tl - q0.bl) + * (q0.tr - q0.tl) (q0.tr - q0.br) + * (q0.br - q0.bl) (q0.tl - q0.bl) + * (q0.br - q0.bl) (q0.tr - q0.br) + * (q1.tr - q1.tl) (q1.tl - q1.bl) + * (q1.tr - q1.tl) (q1.tr - q1.br) + * (q1.br - q1.bl) (q1.tl - q1.bl) + * (q1.br - q1.bl) (q1.tr - q1.br) * - * and add another set of two more subspans if in 16-pixel dispatch mode. + * and add two more quads if in 16-pixel dispatch mode. * * For DDX, it ends up being easy: width = 2, horiz=0 gets us the same result * for each pair, and vertstride = 2 jumps us 2 elements after processing a * pair. But for DDY, it's harder, as we want to produce the pairs swizzled * between each other. We could probably do it like ddx and swizzle the right * order later, but bail for now and just produce - * ((ss0.tl - ss0.bl)x4 (ss1.tl - ss1.bl)x4) + * ((q0.tl - q0.bl)x4 (q1.tl - q1.bl)x4) */ void emit_ddxy(struct brw_compile *p, const struct brw_reg *dst, @@ -611,12 +620,12 @@ static void emit_dp3( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { - int dst_chan = _mesa_ffs(mask & BRW_WRITEMASK_XYZW) - 1; + int dst_chan = ffs(mask & BRW_WRITEMASK_XYZW) - 1; if (!(mask & BRW_WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert(is_power_of_two(mask & BRW_WRITEMASK_XYZW)); + assert(util_is_power_of_two(mask & BRW_WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); @@ -633,12 +642,12 @@ static void emit_dp4( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { - int dst_chan = _mesa_ffs(mask & BRW_WRITEMASK_XYZW) - 1; + int dst_chan = ffs(mask & BRW_WRITEMASK_XYZW) - 1; if (!(mask & BRW_WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert(is_power_of_two(mask & BRW_WRITEMASK_XYZW)); + assert(util_is_power_of_two(mask & BRW_WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); @@ -656,12 +665,12 @@ static void emit_dph( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1 ) { - const int dst_chan = _mesa_ffs(mask & BRW_WRITEMASK_XYZW) - 1; + const int dst_chan = ffs(mask & BRW_WRITEMASK_XYZW) - 1; if (!(mask & BRW_WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert(is_power_of_two(mask & BRW_WRITEMASK_XYZW)); + assert(util_is_power_of_two(mask & BRW_WRITEMASK_XYZW)); brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]); brw_MAC(p, brw_null_reg(), arg0[1], arg1[1]); @@ -704,12 +713,12 @@ static void emit_math1( struct brw_compile *p, GLuint mask, const struct brw_reg *arg0 ) { - int dst_chan = _mesa_ffs(mask & BRW_WRITEMASK_XYZW) - 1; + int dst_chan = ffs(mask & BRW_WRITEMASK_XYZW) - 1; if (!(mask & BRW_WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert(is_power_of_two(mask & BRW_WRITEMASK_XYZW)); + assert(util_is_power_of_two(mask & BRW_WRITEMASK_XYZW)); brw_MOV(p, brw_message_reg(2), arg0[0]); @@ -732,12 +741,12 @@ static void emit_math2( struct brw_compile *p, const struct brw_reg *arg0, const struct brw_reg *arg1) { - int dst_chan = _mesa_ffs(mask & BRW_WRITEMASK_XYZW) - 1; + int dst_chan = ffs(mask & BRW_WRITEMASK_XYZW) - 1; if (!(mask & BRW_WRITEMASK_XYZW)) return; /* Do not emit dead code */ - assert(is_power_of_two(mask & BRW_WRITEMASK_XYZW)); + assert(util_is_power_of_two(mask & BRW_WRITEMASK_XYZW)); brw_push_insn_state(p); @@ -790,21 +799,32 @@ static void emit_tex( struct brw_wm_compile *c, GLuint i, nr; GLuint emit; GLuint msg_type; + GLboolean shadow = FALSE; /* How many input regs are there? */ - switch (inst->tex_idx) { - case TEXTURE_1D_INDEX: + switch (inst->tex_target) { + case TGSI_TEXTURE_1D: emit = BRW_WRITEMASK_X; nr = 1; break; - case TEXTURE_2D_INDEX: - case TEXTURE_RECT_INDEX: + case TGSI_TEXTURE_SHADOW1D: + emit = BRW_WRITEMASK_XW; + nr = 4; + shadow = TRUE; + break; + case TGSI_TEXTURE_2D: emit = BRW_WRITEMASK_XY; nr = 2; break; - case TEXTURE_3D_INDEX: - case TEXTURE_CUBE_INDEX: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: + emit = BRW_WRITEMASK_XYW; + nr = 4; + shadow = TRUE; + break; + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: emit = BRW_WRITEMASK_XYZ; nr = 3; break; @@ -813,11 +833,6 @@ static void emit_tex( struct brw_wm_compile *c, abort(); } - if (inst->tex_shadow) { - nr = 4; - emit |= BRW_WRITEMASK_W; - } - msgLength = 1; for (i = 0; i < nr; i++) { @@ -832,12 +847,12 @@ static void emit_tex( struct brw_wm_compile *c, responseLength = 8; /* always */ if (BRW_IS_IGDNG(p->brw)) { - if (inst->tex_shadow) + if (shadow) msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE_IGDNG; else msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_IGDNG; } else { - if (inst->tex_shadow) + if (shadow) msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE; else msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE; @@ -870,20 +885,23 @@ static void emit_txb( struct brw_wm_compile *c, GLuint msg_type; /* Shadow ignored for txb. */ - switch (inst->tex_idx) { - case TEXTURE_1D_INDEX: + switch (inst->tex_target) { + case TGSI_TEXTURE_1D: + case TGSI_TEXTURE_SHADOW1D: brw_MOV(p, brw_message_reg(2), arg[0]); brw_MOV(p, brw_message_reg(4), brw_imm_f(0)); brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); break; - case TEXTURE_2D_INDEX: - case TEXTURE_RECT_INDEX: + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: brw_MOV(p, brw_message_reg(2), arg[0]); brw_MOV(p, brw_message_reg(4), arg[1]); brw_MOV(p, brw_message_reg(6), brw_imm_f(0)); break; - case TEXTURE_3D_INDEX: - case TEXTURE_CUBE_INDEX: + case TGSI_TEXTURE_3D: + case TGSI_TEXTURE_CUBE: brw_MOV(p, brw_message_reg(2), arg[0]); brw_MOV(p, brw_message_reg(4), arg[1]); brw_MOV(p, brw_message_reg(6), arg[2]); @@ -976,10 +994,10 @@ static void emit_kil( struct brw_wm_compile *c, } } -/* KIL_NV kills the pixels that are currently executing, not based on a test +/* KILLP kills the pixels that are currently executing, not based on a test * of the arguments. */ -static void emit_kil_nv( struct brw_wm_compile *c ) +static void emit_killp( struct brw_wm_compile *c ) { struct brw_compile *p = &c->func; struct brw_reg r0uw = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW); @@ -1259,7 +1277,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) */ spill_values(c, c->payload.depth, 4); spill_values(c, c->creg, c->nr_creg); - spill_values(c, c->payload.input_interp, FRAG_ATTRIB_MAX); + spill_values(c, c->payload.input_interp, PIPE_MAX_SHADER_INPUTS); for (insn = 0; insn < c->nr_insns; insn++) { @@ -1328,89 +1346,89 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Straightforward arithmetic: */ - case OPCODE_ADD: + case TGSI_OPCODE_ADD: emit_alu2(p, brw_ADD, dst, dst_flags, args[0], args[1]); break; - case OPCODE_FRC: + case TGSI_OPCODE_FRC: emit_alu1(p, brw_FRC, dst, dst_flags, args[0]); break; - case OPCODE_FLR: + case TGSI_OPCODE_FLR: emit_alu1(p, brw_RNDD, dst, dst_flags, args[0]); break; - case OPCODE_DDX: + case TGSI_OPCODE_DDX: emit_ddxy(p, dst, dst_flags, GL_TRUE, args[0]); break; - case OPCODE_DDY: + case TGSI_OPCODE_DDY: emit_ddxy(p, dst, dst_flags, GL_FALSE, args[0]); break; - case OPCODE_DP3: + case TGSI_OPCODE_DP3: emit_dp3(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_DP4: + case TGSI_OPCODE_DP4: emit_dp4(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_DPH: + case TGSI_OPCODE_DPH: emit_dph(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_TRUNC: + case TGSI_OPCODE_TRUNC: emit_trunc(p, dst, dst_flags, args[0]); break; - case OPCODE_LRP: + case TGSI_OPCODE_LRP: emit_lrp(p, dst, dst_flags, args[0], args[1], args[2]); break; - case OPCODE_MAD: + case TGSI_OPCODE_MAD: emit_mad(p, dst, dst_flags, args[0], args[1], args[2]); break; - case OPCODE_MOV: + case TGSI_OPCODE_MOV: emit_alu1(p, brw_MOV, dst, dst_flags, args[0]); break; - case OPCODE_MUL: + case TGSI_OPCODE_MUL: emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]); break; - case OPCODE_XPD: + case TGSI_OPCODE_XPD: emit_xpd(p, dst, dst_flags, args[0], args[1]); break; /* Higher math functions: */ - case OPCODE_RCP: + case TGSI_OPCODE_RCP: emit_math1(p, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]); break; - case OPCODE_RSQ: + case TGSI_OPCODE_RSQ: emit_math1(p, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]); break; - case OPCODE_SIN: + case TGSI_OPCODE_SIN: emit_math1(p, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]); break; - case OPCODE_COS: + case TGSI_OPCODE_COS: emit_math1(p, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]); break; - case OPCODE_EX2: + case TGSI_OPCODE_EX2: emit_math1(p, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]); break; - case OPCODE_LG2: + case TGSI_OPCODE_LG2: emit_math1(p, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]); break; - case OPCODE_SCS: + case TGSI_OPCODE_SCS: /* There is an scs math function, but it would need some * fixup for 16-element execution. */ @@ -1420,71 +1438,70 @@ void brw_wm_emit( struct brw_wm_compile *c ) emit_math1(p, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|BRW_WRITEMASK_X, args[0]); break; - case OPCODE_POW: + case TGSI_OPCODE_POW: emit_math2(p, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]); break; /* Comparisons: */ - case OPCODE_CMP: + case TGSI_OPCODE_CMP: emit_cmp(p, dst, dst_flags, args[0], args[1], args[2]); break; - case OPCODE_MAX: + case TGSI_OPCODE_MAX: emit_max(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_MIN: + case TGSI_OPCODE_MIN: emit_min(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SLT: + case TGSI_OPCODE_SLT: emit_slt(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SLE: + case TGSI_OPCODE_SLE: emit_sle(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SGT: + case TGSI_OPCODE_SGT: emit_sgt(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SGE: + case TGSI_OPCODE_SGE: emit_sge(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SEQ: + case TGSI_OPCODE_SEQ: emit_seq(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_SNE: + case TGSI_OPCODE_SNE: emit_sne(p, dst, dst_flags, args[0], args[1]); break; - case OPCODE_LIT: + case TGSI_OPCODE_LIT: emit_lit(p, dst, dst_flags, args[0]); break; /* Texturing operations: */ - case OPCODE_TEX: + case TGSI_OPCODE_TEX: emit_tex(c, inst, dst, dst_flags, args[0]); break; - case OPCODE_TXB: + case TGSI_OPCODE_TXB: emit_txb(c, inst, dst, dst_flags, args[0]); break; - case OPCODE_KIL: + case TGSI_OPCODE_KIL: emit_kil(c, args[0]); break; - case OPCODE_KIL_NV: - emit_kil_nv(c); + case TGSI_OPCODE_KILP: + emit_killp(c); break; default: debug_printf("Unsupported opcode %i (%s) in fragment shader\n", - inst->opcode, inst->opcode < MAX_OPCODE ? - _mesa_opcode_string(inst->opcode) : - "unknown"); + inst->opcode, + tgsi_get_opcode_info(inst->opcode)->mnemonic); } for (i = 0; i < 4; i++) diff --git a/src/gallium/drivers/i965/brw_wm_fp.c b/src/gallium/drivers/i965/brw_wm_fp.c index d594730730..8ba037cdae 100644 --- a/src/gallium/drivers/i965/brw_wm_fp.c +++ b/src/gallium/drivers/i965/brw_wm_fp.c @@ -30,9 +30,8 @@ */ -#include "pipe/p_shader_constants.h" +#include "pipe/p_shader_tokens.h" -#include "brw_context.h" #include "brw_wm.h" #include "brw_util.h" @@ -43,7 +42,7 @@ #define W 3 -static const char *wm_opcode_strings[] = { +static const char *wm_opcode_strings[] = { "PIXELXY", "DELTAXY", "PIXELW", @@ -57,143 +56,6 @@ static const char *wm_opcode_strings[] = { -/*********************************************************************** - * Source regs - */ - -static struct prog_src_register src_reg(GLuint file, GLuint idx) -{ - struct prog_src_register reg; - reg.File = file; - reg.Index = idx; - reg.Swizzle = SWIZZLE_NOOP; - reg.RelAddr = 0; - reg.Negate = NEGATE_NONE; - reg.Abs = 0; - return reg; -} - -static struct prog_src_register src_reg_from_dst(struct prog_dst_register dst) -{ - return src_reg(dst.File, dst.Index); -} - -static struct prog_src_register src_undef( void ) -{ - return src_reg(PROGRAM_UNDEFINED, 0); -} - -static GLboolean src_is_undef(struct prog_src_register src) -{ - return src.File == PROGRAM_UNDEFINED; -} - -static struct prog_src_register src_swizzle( struct prog_src_register reg, int x, int y, int z, int w ) -{ - reg.Swizzle = MAKE_SWIZZLE4(x,y,z,w); - return reg; -} - -static struct prog_src_register src_swizzle1( struct prog_src_register reg, int x ) -{ - return src_swizzle(reg, x, x, x, x); -} - -static struct prog_src_register src_swizzle4( struct prog_src_register reg, uint swizzle ) -{ - reg.Swizzle = swizzle; - return reg; -} - - -/*********************************************************************** - * Dest regs - */ - -static struct prog_dst_register dst_reg(GLuint file, GLuint idx) -{ - struct prog_dst_register reg; - reg.File = file; - reg.Index = idx; - reg.WriteMask = BRW_WRITEMASK_XYZW; - reg.RelAddr = 0; - reg.CondMask = COND_TR; - reg.CondSwizzle = 0; - reg.CondSrc = 0; - reg.pad = 0; - return reg; -} - -static struct prog_dst_register dst_mask( struct prog_dst_register reg, int mask ) -{ - reg.WriteMask &= mask; - return reg; -} - -static struct prog_dst_register dst_undef( void ) -{ - return dst_reg(PROGRAM_UNDEFINED, 0); -} - - - -static struct prog_dst_register get_temp( struct brw_wm_compile *c ) -{ - int bit = _mesa_ffs( ~c->fp_temp ); - - if (!bit) { - debug_printf("%s: out of temporaries\n", __FILE__); - exit(1); - } - - c->fp_temp |= 1<<(bit-1); - return dst_reg(PROGRAM_TEMPORARY, c->first_internal_temp+(bit-1)); -} - - -static void release_temp( struct brw_wm_compile *c, struct prog_dst_register temp ) -{ - c->fp_temp &= ~(1 << (temp.Index - c->first_internal_temp)); -} - - -/*********************************************************************** - * Instructions - */ - -static struct prog_instruction *get_fp_inst(struct brw_wm_compile *c) -{ - return &c->prog_instructions[c->nr_fp_insns++]; -} - -static struct prog_instruction *emit_insn(struct brw_wm_compile *c, - const struct prog_instruction *inst0) -{ - struct prog_instruction *inst = get_fp_inst(c); - *inst = *inst0; - return inst; -} - -static struct prog_instruction * emit_op(struct brw_wm_compile *c, - GLuint op, - struct prog_dst_register dest, - GLuint saturate, - struct prog_src_register src0, - struct prog_src_register src1, - struct prog_src_register src2 ) -{ - struct prog_instruction *inst = get_fp_inst(c); - - memset(inst, 0, sizeof(*inst)); - - inst->Opcode = op; - inst->DstReg = dest; - inst->SaturateMode = saturate; - inst->SrcReg[0] = src0; - inst->SrcReg[1] = src1; - inst->SrcReg[2] = src2; - return inst; -} /* Many opcodes produce the same value across all the result channels. @@ -202,32 +64,28 @@ static struct prog_instruction * emit_op(struct brw_wm_compile *c, * anyway. We can easily get both by emitting the opcode to one channel, and * then MOVing it to the others, which brw_wm_pass*.c already understands. */ -static struct prog_instruction *emit_scalar_insn(struct brw_wm_compile *c, - const struct prog_instruction *inst0) -{ - struct prog_instruction *inst; - unsigned int dst_chan; - unsigned int other_channel_mask; - - if (inst0->DstReg.WriteMask == 0) - return NULL; - - dst_chan = _mesa_ffs(inst0->DstReg.WriteMask) - 1; - inst = get_fp_inst(c); - *inst = *inst0; - inst->DstReg.WriteMask = 1 << dst_chan; - - other_channel_mask = inst0->DstReg.WriteMask & ~(1 << dst_chan); - if (other_channel_mask != 0) { - inst = emit_op(c, - TGSI_OPCODE_MOV, - dst_mask(inst0->DstReg, other_channel_mask), - 0, - src_swizzle1(src_reg_from_dst(inst0->DstReg), dst_chan), - src_undef(), - src_undef()); +static void emit_scalar_insn(struct brw_wm_compile *c, + unsigned opcode, + struct brw_dst dst, + struct brw_src src0, + struct brw_src src1, + struct brw_src src2 ) +{ + unsigned first_chan = ffs(dst.writemask) - 1; + unsigned first_mask = 1 << first_chan; + + if (dst.writemask == 0) + return; + + emit_op( c, opcode, + brw_writemask(dst, first_mask), + src0, src1, src2 ); + + if (dst.writemask != first_mask) { + emit_op1(c, TGSI_OPCODE_MOV, + brw_writemask(dst, ~first_mask), + src_swizzle1(brw_src(dst), first_chan)); } - return inst; } @@ -235,11 +93,11 @@ static struct prog_instruction *emit_scalar_insn(struct brw_wm_compile *c, * Special instructions for interpolation and other tasks */ -static struct prog_src_register get_pixel_xy( struct brw_wm_compile *c ) +static struct ureg_src get_pixel_xy( struct brw_wm_compile *c ) { if (src_is_undef(c->pixel_xy)) { - struct prog_dst_register pixel_xy = get_temp(c); - struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH); + struct ureg_dst pixel_xy = get_temp(c); + struct ureg_src payload_r0_depth = src_reg(TGSI_FILE_PAYLOAD, PAYLOAD_DEPTH); /* Emit the out calculations, and hold onto the results. Use @@ -250,7 +108,6 @@ static struct prog_src_register get_pixel_xy( struct brw_wm_compile *c ) emit_op(c, WM_PIXELXY, dst_mask(pixel_xy, BRW_WRITEMASK_XY), - 0, payload_r0_depth, src_undef(), src_undef()); @@ -261,19 +118,18 @@ static struct prog_src_register get_pixel_xy( struct brw_wm_compile *c ) return c->pixel_xy; } -static struct prog_src_register get_delta_xy( struct brw_wm_compile *c ) +static struct ureg_src get_delta_xy( struct brw_wm_compile *c ) { if (src_is_undef(c->delta_xy)) { - struct prog_dst_register delta_xy = get_temp(c); - struct prog_src_register pixel_xy = get_pixel_xy(c); - struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH); + struct ureg_dst delta_xy = get_temp(c); + struct ureg_src pixel_xy = get_pixel_xy(c); + struct ureg_src payload_r0_depth = src_reg(TGSI_FILE_PAYLOAD, PAYLOAD_DEPTH); /* deltas.xy = DELTAXY pixel_xy, payload[0] */ emit_op(c, WM_DELTAXY, dst_mask(delta_xy, BRW_WRITEMASK_XY), - 0, pixel_xy, payload_r0_depth, src_undef()); @@ -284,19 +140,18 @@ static struct prog_src_register get_delta_xy( struct brw_wm_compile *c ) return c->delta_xy; } -static struct prog_src_register get_pixel_w( struct brw_wm_compile *c ) +static struct ureg_src get_pixel_w( struct brw_wm_compile *c ) { if (src_is_undef(c->pixel_w)) { - struct prog_dst_register pixel_w = get_temp(c); - struct prog_src_register deltas = get_delta_xy(c); - struct prog_src_register interp_wpos = src_reg(PROGRAM_PAYLOAD, FRAG_ATTRIB_WPOS); + struct ureg_dst pixel_w = get_temp(c); + struct ureg_src deltas = get_delta_xy(c); + struct ureg_src interp_wpos = src_reg(TGSI_FILE_PAYLOAD, FRAG_ATTRIB_WPOS); /* deltas.xyw = DELTAS2 deltas.xy, payload.interp_wpos.x */ emit_op(c, WM_PIXELW, dst_mask(pixel_w, BRW_WRITEMASK_W), - 0, interp_wpos, deltas, src_undef()); @@ -313,9 +168,9 @@ static void emit_interp( struct brw_wm_compile *c, GLuint semantic_index, GLuint interp_mode ) { - struct prog_dst_register dst = dst_reg(PROGRAM_INPUT, idx); - struct prog_src_register interp = src_reg(PROGRAM_PAYLOAD, idx); - struct prog_src_register deltas = get_delta_xy(c); + struct ureg_dst dst = dst_reg(TGSI_FILE_INPUT, idx); + struct ureg_src interp = src_reg(TGSI_FILE_PAYLOAD, idx); + struct ureg_src deltas = get_delta_xy(c); /* Need to use PINTERP on attributes which have been * multiplied by 1/W in the SF program, and LINTERP on those @@ -325,271 +180,197 @@ static void emit_interp( struct brw_wm_compile *c, case FRAG_ATTRIB_WPOS: /* Have to treat wpos.xy specially: */ - emit_op(c, + emit_op1(c, WM_WPOSXY, dst_mask(dst, BRW_WRITEMASK_XY), - 0, - get_pixel_xy(c), - src_undef(), - src_undef()); + get_pixel_xy(c)); - dst = dst_mask(dst, BRW_WRITEMASK_ZW); - - /* PROGRAM_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw + /* TGSI_FILE_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw */ - emit_op(c, - WM_LINTERP, - dst, - 0, - interp, - deltas, - src_undef()); + emit_op2(c, + WM_LINTERP, + dst_mask(dst, BRW_WRITEMASK_ZW), + interp, + deltas); break; case TGSI_SEMANTIC_COLOR: if (c->key.flat_shade) { - emit_op(c, + emit_op1(c, WM_CINTERP, dst, - 0, - interp, - src_undef(), - src_undef()); + interp); + } + else if (interp_mode == TGSI_INTERPOLATE_LINEAR) { + emit_op2(c, + WM_LINTERP, + dst, + interp, + deltas); } else { - emit_op(c, - translate_interp_mode(interp_mode), - dst, - 0, - interp, - deltas, - src_undef()); + emit_op3(c, + WM_PINTERP, + dst, + interp, + deltas, + get_pixel_w(c)); } + break; case FRAG_ATTRIB_FOGC: /* Interpolate the fog coordinate */ - emit_op(c, + emit_op3(c, WM_PINTERP, dst_mask(dst, BRW_WRITEMASK_X), - 0, interp, deltas, get_pixel_w(c)); - emit_op(c, + emit_op1(c, TGSI_OPCODE_MOV, - dst_mask(dst, BRW_WRITEMASK_YZW), - 0, - src_swizzle(interp, - SWIZZLE_ZERO, - SWIZZLE_ZERO, - SWIZZLE_ZERO, - SWIZZLE_ONE), - src_undef(), - src_undef()); + dst_mask(dst, BRW_WRITEMASK_YZ), + brw_imm1f(0.0)); + + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_W), + brw_imm1f(1.0)); break; case FRAG_ATTRIB_FACE: /* XXX review/test this case */ - emit_op(c, - WM_FRONTFACING, - dst_mask(dst, BRW_WRITEMASK_X), - 0, - src_undef(), - src_undef(), - src_undef()); + emit_op0(c, + WM_FRONTFACING, + dst_mask(dst, BRW_WRITEMASK_X)); + + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_YZ), + brw_imm1f(0.0)); + + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_W), + brw_imm1f(1.0)); break; case FRAG_ATTRIB_PNTC: /* XXX review/test this case */ - emit_op(c, - WM_PINTERP, - dst_mask(dst, BRW_WRITEMASK_XY), - 0, - interp, - deltas, - get_pixel_w(c)); - - emit_op(c, + emit_op3(c, + WM_PINTERP, + dst_mask(dst, BRW_WRITEMASK_XY), + interp, + deltas, + get_pixel_w(c)); + + emit_op1(c, TGSI_OPCODE_MOV, - dst_mask(dst, BRW_WRITEMASK_ZW), - 0, - src_swizzle(interp, - SWIZZLE_ZERO, - SWIZZLE_ZERO, - SWIZZLE_ZERO, - SWIZZLE_ONE), - src_undef(), - src_undef()); - break; + dst_mask(dst, BRW_WRITEMASK_Z), + brw_imm1f(c->pass_fp, 0.0f)); - default: - emit_op(c, - translate_interp_mode(interp_mode), - dst, - 0, - interp, - deltas, - get_pixel_w(c)); + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_W), + brw_imm1f(c->pass_fp, 1.0f)); break; - } -} - -/*********************************************************************** - * Hacks to extend the program parameter and constant lists. - */ - -/* Add the fog parameters to the parameter list of the original - * program, rather than creating a new list. Doesn't really do any - * harm and it's not as if the parameter handling isn't a big hack - * anyway. - */ -static struct prog_src_register search_or_add_param5(struct brw_wm_compile *c, - GLint s0, - GLint s1, - GLint s2, - GLint s3, - GLint s4) -{ - struct gl_program_parameter_list *paramList = c->fp->program.Base.Parameters; - gl_state_index tokens[STATE_LENGTH]; - GLuint idx; - tokens[0] = s0; - tokens[1] = s1; - tokens[2] = s2; - tokens[3] = s3; - tokens[4] = s4; - - for (idx = 0; idx < paramList->NumParameters; idx++) { - if (paramList->Parameters[idx].Type == PROGRAM_STATE_VAR && - memcmp(paramList->Parameters[idx].StateIndexes, tokens, sizeof(tokens)) == 0) - return src_reg(PROGRAM_STATE_VAR, idx); - } - - idx = _mesa_add_state_reference( paramList, tokens ); - - return src_reg(PROGRAM_STATE_VAR, idx); -} + default: + switch (interp_mode) { + case TGSI_INTERPOLATE_CONSTANT: + emit_op1(c, + WM_CINTERP, + dst, + interp); + break; -static struct prog_src_register search_or_add_const4f( struct brw_wm_compile *c, - GLfloat s0, - GLfloat s1, - GLfloat s2, - GLfloat s3) -{ - struct gl_program_parameter_list *paramList = c->fp->program.Base.Parameters; - GLfloat values[4]; - GLuint idx; - GLuint swizzle; - - values[0] = s0; - values[1] = s1; - values[2] = s2; - values[3] = s3; - - /* Have to search, otherwise multiple compilations will each grow - * the parameter list. - */ - for (idx = 0; idx < paramList->NumParameters; idx++) { - if (paramList->Parameters[idx].Type == PROGRAM_CONSTANT && - memcmp(paramList->ParameterValues[idx], values, sizeof(values)) == 0) + case TGSI_INTERPOLATE_LINEAR: + emit_op2(c, + WM_LINTERP, + dst, + interp, + deltas); + break; - /* XXX: this mimics the mesa bug which puts all constants and - * parameters into the "PROGRAM_STATE_VAR" category: - */ - return src_reg(PROGRAM_STATE_VAR, idx); + case TGSI_INTERPOLATE_PERSPECTIVE: + emit_op3(c, + WM_PINTERP, + dst, + interp, + deltas, + get_pixel_w(c)); + break; + } + break; } - - idx = _mesa_add_unnamed_constant( paramList, values, 4, &swizzle ); - assert(swizzle == SWIZZLE_NOOP); /* Need to handle swizzle in reg setup */ - return src_reg(PROGRAM_STATE_VAR, idx); } - /*********************************************************************** * Expand various instructions here to simpler forms. */ static void precalc_dst( struct brw_wm_compile *c, - const struct prog_instruction *inst ) + struct brw_dst dst, + struct brw_src src0, + struct brw_src src1 ) { - struct prog_src_register src0 = inst->SrcReg[0]; - struct prog_src_register src1 = inst->SrcReg[1]; - struct prog_dst_register dst = inst->DstReg; - if (dst.WriteMask & BRW_WRITEMASK_Y) { /* dst.y = mul src0.y, src1.y */ - emit_op(c, - TGSI_OPCODE_MUL, - dst_mask(dst, BRW_WRITEMASK_Y), - inst->SaturateMode, - src0, - src1, - src_undef()); + emit_op2(c, + TGSI_OPCODE_MUL, + dst_mask(dst, BRW_WRITEMASK_Y), + src0, + src1); } if (dst.WriteMask & BRW_WRITEMASK_XZ) { struct prog_instruction *swz; GLuint z = GET_SWZ(src0.Swizzle, Z); - /* dst.xz = swz src0.1zzz + /* dst.z = mov src0.zzzz + */ + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_Z), + src_swizzle1(src0, Z)); + + /* dst.x = immf(1.0) */ - swz = emit_op(c, - TGSI_OPCODE_MOV, - dst_mask(dst, BRW_WRITEMASK_XZ), - inst->SaturateMode, - src_swizzle(src0, SWIZZLE_ONE, z, z, z), - src_undef(), - src_undef()); - /* Avoid letting negation flag of src0 affect our 1 constant. */ - swz->SrcReg[0].Negate &= ~NEGATE_X; + emit_op1(c, + TGSI_OPCODE_MOV, + brw_saturate(dst_mask(dst, BRW_WRITEMASK_X), 0), + src_immf(c, 1.0)); } if (dst.WriteMask & BRW_WRITEMASK_W) { /* dst.w = mov src1.w */ - emit_op(c, - TGSI_OPCODE_MOV, - dst_mask(dst, BRW_WRITEMASK_W), - inst->SaturateMode, - src1, - src_undef(), - src_undef()); + emit_op1(c, + TGSI_OPCODE_MOV, + dst_mask(dst, BRW_WRITEMASK_W), + src1); } } static void precalc_lit( struct brw_wm_compile *c, - const struct prog_instruction *inst ) + struct ureg_dst dst, + struct ureg_src src0 ) { - struct prog_src_register src0 = inst->SrcReg[0]; - struct prog_dst_register dst = inst->DstReg; - if (dst.WriteMask & BRW_WRITEMASK_XW) { - struct prog_instruction *swz; - - /* dst.xw = swz src0.1111 + /* dst.xw = imm(1.0f) */ - swz = emit_op(c, - TGSI_OPCODE_MOV, - dst_mask(dst, BRW_WRITEMASK_XW), - 0, - src_swizzle1(src0, SWIZZLE_ONE), - src_undef(), - src_undef()); - /* Avoid letting the negation flag of src0 affect our 1 constant. */ - swz->SrcReg[0].Negate = NEGATE_NONE; + emit_op1(c, + TGSI_OPCODE_MOV, + brw_saturate(brw_writemask(dst, BRW_WRITEMASK_XW), 0), + brw_imm1f(1.0f)); } if (dst.WriteMask & BRW_WRITEMASK_YZ) { - emit_op(c, - TGSI_OPCODE_LIT, - dst_mask(dst, BRW_WRITEMASK_YZ), - inst->SaturateMode, - src0, - src_undef(), - src_undef()); + emit_op1(c, + TGSI_OPCODE_LIT, + brw_writemask(dst, BRW_WRITEMASK_YZ), + src0); } } @@ -601,99 +382,62 @@ static void precalc_lit( struct brw_wm_compile *c, * instruction itself. */ static void precalc_tex( struct brw_wm_compile *c, - const struct prog_instruction *inst ) + struct brw_dst dst, + unsigned unit, + struct brw_src src0 ) { - struct prog_src_register coord; - struct prog_dst_register tmpcoord; - const GLuint unit = c->fp->program.Base.SamplerUnits[inst->TexSrcUnit]; + struct ureg_src coord = src_undef(); + struct ureg_dst tmp = dst_undef(); assert(unit < BRW_MAX_TEX_UNIT); + /* Cubemap: find longest component of coord vector and normalize + * it. + */ if (inst->TexSrcTarget == TEXTURE_CUBE_INDEX) { - struct prog_instruction *out; - struct prog_dst_register tmp0 = get_temp(c); - struct prog_src_register tmp0src = src_reg_from_dst(tmp0); - struct prog_dst_register tmp1 = get_temp(c); - struct prog_src_register tmp1src = src_reg_from_dst(tmp1); - struct prog_src_register src0 = inst->SrcReg[0]; - - /* find longest component of coord vector and normalize it */ - tmpcoord = get_temp(c); - coord = src_reg_from_dst(tmpcoord); - - /* tmpcoord = src0 (i.e.: coord = src0) */ - out = emit_op(c, TGSI_OPCODE_MOV, - tmpcoord, - 0, - src0, - src_undef(), - src_undef()); - out->SrcReg[0].Negate = NEGATE_NONE; - out->SrcReg[0].Abs = 1; - - /* tmp0 = MAX(coord.X, coord.Y) */ - emit_op(c, TGSI_OPCODE_MAX, - tmp0, - 0, - src_swizzle1(coord, X), - src_swizzle1(coord, Y), - src_undef()); - - /* tmp1 = MAX(tmp0, coord.Z) */ - emit_op(c, TGSI_OPCODE_MAX, - tmp1, - 0, - tmp0src, - src_swizzle1(coord, Z), - src_undef()); - - /* tmp0 = 1 / tmp1 */ - emit_op(c, TGSI_OPCODE_RCP, - dst_mask(tmp0, BRW_WRITEMASK_X), - 0, - tmp1src, - src_undef(), - src_undef()); - - /* tmpCoord = src0 * tmp0 */ - emit_op(c, TGSI_OPCODE_MUL, - tmpcoord, - 0, - src0, - src_swizzle1(tmp0src, SWIZZLE_X), - src_undef()); - - release_temp(c, tmp0); - release_temp(c, tmp1); + struct ureg_src tmpsrc; + + tmp = get_temp(c); + tmpsrc = brw_src(tmpcoord) + + /* tmp = abs(src0) */ + emit_op1(c, + TGSI_OPCODE_MOV, + tmp, + brw_abs(src0)); + + /* tmp.X = MAX(tmp.X, tmp.Y) */ + emit_op2(c, TGSI_OPCODE_MAX, + brw_writemask(tmp, BRW_WRITEMASK_X), + src_swizzle1(tmpsrc, X), + src_swizzle1(tmpsrc, Y)); + + /* tmp.X = MAX(tmp.X, tmp.Z) */ + emit_op2(c, TGSI_OPCODE_MAX, + brw_writemask(tmp, BRW_WRITEMASK_X), + tmpsrc, + src_swizzle1(tmpsrc, Z)); + + /* tmp.X = 1 / tmp.X */ + emit_op1(c, TGSI_OPCODE_RCP, + dst_mask(tmp, BRW_WRITEMASK_X), + tmpsrc); + + /* tmp = src0 * tmp.xxxx */ + emit_op2(c, TGSI_OPCODE_MUL, + tmp, + src0, + src_swizzle1(tmpsrc, SWIZZLE_X)); + + coord = tmpsrc; } else if (inst->TexSrcTarget == TEXTURE_RECT_INDEX) { - struct prog_src_register scale = - search_or_add_param5( c, - STATE_INTERNAL, - STATE_TEXRECT_SCALE, - unit, - 0,0 ); - - tmpcoord = get_temp(c); - - /* coord.xy = MUL inst->SrcReg[0], { 1/width, 1/height } + /* XXX: need a mechanism for internally generated constants. */ - emit_op(c, - TGSI_OPCODE_MUL, - tmpcoord, - 0, - inst->SrcReg[0], - src_swizzle(scale, - SWIZZLE_X, - SWIZZLE_Y, - SWIZZLE_ONE, - SWIZZLE_ONE), - src_undef()); - - coord = src_reg_from_dst(tmpcoord); + coord = src0; } else { - coord = inst->SrcReg[0]; + coord = src0; } /* Need to emit YUV texture conversions by hand. Probably need to @@ -704,58 +448,36 @@ static void precalc_tex( struct brw_wm_compile *c, if (c->key.yuvtex_mask & (1 << unit)) { /* convert ycbcr to RGBA */ GLboolean swap_uv = c->key.yuvtex_swap_mask & (1<DstReg; - struct prog_dst_register tmp = get_temp(c); - struct prog_src_register tmpsrc = src_reg_from_dst(tmp); - struct prog_src_register C0 = search_or_add_const4f( c, -.5, -.0625, -.5, 1.164 ); - struct prog_src_register C1 = search_or_add_const4f( c, 1.596, -0.813, 2.018, -.391 ); + struct ureg_dst dst = inst->DstReg; + struct ureg_dst tmp = get_temp(c); + struct ureg_src tmpsrc = src_reg_from_dst(tmp); + struct ureg_src C0 = ureg_imm4f( c->ureg, -.5, -.0625, -.5, 1.164 ); + struct ureg_src C1 = ureg_imm4f( c->ureg, 1.596, -0.813, 2.018, -.391 ); /* tmp = TEX ... */ emit_tex_op(c, TGSI_OPCODE_TEX, - tmp, - inst->SaturateMode, + brw_saturate(tmp, dst.Saturate), unit, inst->TexSrcTarget, - inst->TexShadow, coord, src_undef(), src_undef()); /* tmp.xyz = ADD TMP, C0 */ - emit_op(c, - TGSI_OPCODE_ADD, - dst_mask(tmp, BRW_WRITEMASK_XYZ), - 0, - tmpsrc, - C0, - src_undef()); + emit_op2(c, TGSI_OPCODE_ADD, + dst_mask(tmp, BRW_WRITEMASK_XYZ), + tmpsrc, + C0); /* YUV.y = MUL YUV.y, C0.w */ - - emit_op(c, - TGSI_OPCODE_MUL, - dst_mask(tmp, BRW_WRITEMASK_Y), - 0, - tmpsrc, - src_swizzle1(C0, W), - src_undef()); + emit_op2(c, TGSI_OPCODE_MUL, + dst_mask(tmp, BRW_WRITEMASK_Y), + tmpsrc, + src_swizzle1(C0, W)); /* * if (UV swaped) @@ -764,23 +486,22 @@ static void precalc_tex( struct brw_wm_compile *c, * RGB.xyz = MAD YUV.xxz, C1, YUV.y */ - emit_op(c, - TGSI_OPCODE_MAD, - dst_mask(dst, BRW_WRITEMASK_XYZ), - 0, - swap_uv?src_swizzle(tmpsrc, Z,Z,X,X):src_swizzle(tmpsrc, X,X,Z,Z), - C1, - src_swizzle1(tmpsrc, Y)); + emit_op3(c, TGSI_OPCODE_MAD, + dst_mask(dst, BRW_WRITEMASK_XYZ), + ( swap_uv ? + src_swizzle(tmpsrc, Z,Z,X,X) : + src_swizzle(tmpsrc, X,X,Z,Z)), + C1, + src_swizzle1(tmpsrc, Y)); /* RGB.y = MAD YUV.z, C1.w, RGB.y */ - emit_op(c, - TGSI_OPCODE_MAD, - dst_mask(dst, BRW_WRITEMASK_Y), - 0, - src_swizzle1(tmpsrc, Z), - src_swizzle1(C1, W), - src_swizzle1(src_reg_from_dst(dst), Y)); + emit_op3(c, + TGSI_OPCODE_MAD, + dst_mask(dst, BRW_WRITEMASK_Y), + src_swizzle1(tmpsrc, Z), + src_swizzle1(C1, W), + src_swizzle1(src_reg_from_dst(dst), Y)); release_temp(c, tmp); } @@ -789,29 +510,20 @@ static void precalc_tex( struct brw_wm_compile *c, emit_tex_op(c, TGSI_OPCODE_TEX, inst->DstReg, - inst->SaturateMode, unit, inst->TexSrcTarget, - inst->TexShadow, coord, src_undef(), src_undef()); } - /* For GL_EXT_texture_swizzle: */ - if (c->key.tex_swizzles[unit] != SWIZZLE_NOOP) { - /* swizzle the result of the TEX instruction */ - struct prog_src_register tmpsrc = src_reg_from_dst(inst->DstReg); - emit_op(c, TGSI_OPCODE_MOV, - inst->DstReg, - SATURATE_OFF, /* saturate already done above */ - src_swizzle4(tmpsrc, c->key.tex_swizzles[unit]), - src_undef(), - src_undef()); - } + /* XXX: add GL_EXT_texture_swizzle support to gallium -- by + * generating shader varients in mesa state tracker. + */ - if ((inst->TexSrcTarget == TEXTURE_RECT_INDEX) || - (inst->TexSrcTarget == TEXTURE_CUBE_INDEX)) + /* Release this temp if we ended up allocating it: + */ + if (!brw_dst_is_undef(tmpcoord)) release_temp(c, tmpcoord); } @@ -822,7 +534,7 @@ static void precalc_tex( struct brw_wm_compile *c, static GLboolean projtex( struct brw_wm_compile *c, const struct prog_instruction *inst ) { - const struct prog_src_register src = inst->SrcReg[0]; + const struct ureg_src src = inst->SrcReg[0]; GLboolean retVal; assert(inst->Opcode == TGSI_OPCODE_TXP); @@ -836,7 +548,7 @@ static GLboolean projtex( struct brw_wm_compile *c, */ if (inst->TexSrcTarget == TEXTURE_CUBE_INDEX) retVal = GL_FALSE; /* ut2004 gun rendering !?! */ - else if (src.File == PROGRAM_INPUT && + else if (src.File == TGSI_FILE_INPUT && GET_SWZ(src.Swizzle, W) == W && (c->key.proj_attrib_mask & (1 << src.Index)) == 0) retVal = GL_FALSE; @@ -853,10 +565,10 @@ static GLboolean projtex( struct brw_wm_compile *c, static void precalc_txp( struct brw_wm_compile *c, const struct prog_instruction *inst ) { - struct prog_src_register src0 = inst->SrcReg[0]; + struct ureg_src src0 = inst->SrcReg[0]; if (projtex(c, inst)) { - struct prog_dst_register tmp = get_temp(c); + struct ureg_dst tmp = get_temp(c); struct prog_instruction tmp_inst; /* tmp0.w = RCP inst.arg[0][3] @@ -864,7 +576,6 @@ static void precalc_txp( struct brw_wm_compile *c, emit_op(c, TGSI_OPCODE_RCP, dst_mask(tmp, BRW_WRITEMASK_W), - 0, src_swizzle1(src0, GET_SWZ(src0.Swizzle, W)), src_undef(), src_undef()); @@ -874,7 +585,6 @@ static void precalc_txp( struct brw_wm_compile *c, emit_op(c, TGSI_OPCODE_MUL, dst_mask(tmp, BRW_WRITEMASK_XYZ), - 0, src0, src_swizzle1(src_reg_from_dst(tmp), W), src_undef()); @@ -899,43 +609,30 @@ static void precalc_txp( struct brw_wm_compile *c, static void emit_fb_write( struct brw_wm_compile *c ) { - struct prog_src_register payload_r0_depth = src_reg(PROGRAM_PAYLOAD, PAYLOAD_DEPTH); - struct prog_src_register outdepth = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DEPTH); - struct prog_src_register outcolor; + struct ureg_src payload_r0_depth = src_reg(TGSI_FILE_PAYLOAD, PAYLOAD_DEPTH); + struct ureg_src outdepth = src_reg(TGSI_FILE_OUTPUT, FRAG_RESULT_DEPTH); + struct ureg_src outcolor; + struct prog_instruction *inst; GLuint i; - struct prog_instruction *inst, *last_inst; - struct brw_context *brw = c->func.brw; /* The inst->Aux field is used for FB write target and the EOT marker */ - if (brw->state.nr_color_regions > 1) { - for (i = 0 ; i < brw->state.nr_color_regions; i++) { - outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i); - last_inst = inst = emit_op(c, - WM_FB_WRITE, dst_mask(dst_undef(),0), 0, - outcolor, payload_r0_depth, outdepth); - inst->Aux = (i<<1); - if (c->fp_fragcolor_emitted) { - outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR); - last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0), - 0, outcolor, payload_r0_depth, outdepth); - inst->Aux = (i<<1); - } - } - last_inst->Aux |= 1; //eot - } - else { - /* if gl_FragData[0] is written, use it, else use gl_FragColor */ - if (c->fp->program.Base.OutputsWritten & (1 << FRAG_RESULT_DATA0)) - outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0); - else - outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR); - - inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0), - 0, outcolor, payload_r0_depth, outdepth); - inst->Aux = 1|(0<<1); + for (i = 0 ; i < c->key.nr_cbufs; i++) { + outcolor = find_output_by_semantic(c, TGSI_SEMANTIC_COLOR, i); + + inst = emit_op(c, WM_FB_WRITE, + dst_mask(dst_undef(), 0), + outcolor, + payload_r0_depth, + outdepth); + + inst->Aux = (i<<1); } + + /* Set EOT flag on last inst: + */ + inst->Aux |= 1; //eot } @@ -952,7 +649,7 @@ static void validate_src_regs( struct brw_wm_compile *c, GLuint i; for (i = 0; i < nr_args; i++) { - if (inst->SrcReg[i].File == PROGRAM_INPUT) { + if (inst->SrcReg[i].File == TGSI_FILE_INPUT) { GLuint idx = inst->SrcReg[i].Index; if (!(c->fp_interp_emitted & (1<DstReg.File == PROGRAM_OUTPUT) { + if (inst->DstReg.File == TGSI_FILE_OUTPUT) { GLuint idx = inst->DstReg.Index; if (idx == FRAG_RESULT_COLOR) - c->fp_fragcolor_emitted = 1; + c->fp_fragcolor_emitted |= inst->DstReg.WriteMask; } } -static void print_insns( const struct prog_instruction *insn, - GLuint nr ) + + +static void emit_insn( struct brw_wm_compile *c, + const struct tgsi_full_instruction *inst ) { - GLuint i; - for (i = 0; i < nr; i++, insn++) { - debug_printf("%3d: ", i); - if (insn->Opcode < MAX_OPCODE) - _mesa_print_instruction(insn); - else if (insn->Opcode < MAX_WM_OPCODE) { - GLuint idx = insn->Opcode - MAX_OPCODE; - - _mesa_print_alu_instruction(insn, - wm_opcode_strings[idx], - 3); - } - else - debug_printf("965 Opcode %d\n", insn->Opcode); + + switch (inst->Opcode) { + case TGSI_OPCODE_ABS: + emit_op1(c, TGSI_OPCODE_MOV, + dst, + brw_abs(src[0])); + break; + + case TGSI_OPCODE_SUB: + emit_op2(c, TGSI_OPCODE_ADD, + dst, + src[0], + brw_negate(src[1])); + break; + + case TGSI_OPCODE_SCS: + emit_op1(c, TGSI_OPCODE_SCS, + brw_writemask(dst, BRW_WRITEMASK_XY), + src[0]); + break; + + case TGSI_OPCODE_DST: + precalc_dst(c, inst); + break; + + case TGSI_OPCODE_LIT: + precalc_lit(c, inst); + break; + + case TGSI_OPCODE_TEX: + precalc_tex(c, inst); + break; + + case TGSI_OPCODE_TXP: + precalc_txp(c, inst); + break; + + case TGSI_OPCODE_TXB: + out = emit_insn(c, inst); + out->TexSrcUnit = fp->program.Base.SamplerUnits[inst->TexSrcUnit]; + assert(out->TexSrcUnit < BRW_MAX_TEX_UNIT); + break; + + case TGSI_OPCODE_XPD: + emit_op2(c, TGSI_OPCODE_XPD, + brw_writemask(dst, BRW_WRITEMASK_XYZ), + src[0], + src[1]); + break; + + case TGSI_OPCODE_KIL: + emit_op1(c, TGSI_OPCODE_KIL, + brw_writemask(dst_undef(), 0), + src[0]); + break; + + case TGSI_OPCODE_END: + emit_fb_write(c); + break; + default: + if (brw_wm_is_scalar_result(inst->Opcode)) + emit_scalar_insn(c, opcode, dst, src[0], src[1], src[2]); + else + emit_op(c, opcode, dst, src[0], src[1], src[2]); + break; } } - /** * Initial pass for fragment program code generation. * This function is used by both the GLSL and non-GLSL paths. @@ -1004,108 +753,62 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) if (BRW_DEBUG & DEBUG_WM) { debug_printf("pre-fp:\n"); - _mesa_print_program(&fp->program.Base); - debug_printf("\n"); + tgsi_dump(fp->tokens, 0); } - c->pixel_xy = src_undef(); - c->delta_xy = src_undef(); - c->pixel_w = src_undef(); + c->pixel_xy = brw_src_undef(); + c->delta_xy = brw_src_undef(); + c->pixel_w = brw_src_undef(); c->nr_fp_insns = 0; c->fp->tex_units_used = 0x0; - /* Emit preamble instructions. This is where special instructions such as - * WM_CINTERP, WM_LINTERP, WM_PINTERP and WM_WPOSXY are emitted to - * compute shader inputs from varying vars. - */ - for (insn = 0; insn < fp->program.Base.NumInstructions; insn++) { - const struct prog_instruction *inst = &fp->program.Base.Instructions[insn]; - validate_src_regs(c, inst); - validate_dst_regs(c, inst); - } /* Loop over all instructions doing assorted simplifications and * transformations. */ - for (insn = 0; insn < fp->program.Base.NumInstructions; insn++) { - const struct prog_instruction *inst = &fp->program.Base.Instructions[insn]; - struct prog_instruction *out; - - /* Check for INPUT values, emit INTERP instructions where - * necessary: - */ - - switch (inst->Opcode) { - case TGSI_OPCODE_ABS: - out = emit_insn(c, inst); - out->Opcode = TGSI_OPCODE_MOV; - out->SrcReg[0].Negate = NEGATE_NONE; - out->SrcReg[0].Abs = 1; - break; - - case TGSI_OPCODE_SUB: - out = emit_insn(c, inst); - out->Opcode = TGSI_OPCODE_ADD; - out->SrcReg[1].Negate ^= NEGATE_XYZW; - break; - - case TGSI_OPCODE_SCS: - out = emit_insn(c, inst); - /* This should probably be done in the parser. + tgsi_parse_init( &parse, tokens ); + while( !tgsi_parse_end_of_tokens( &parse ) ) { + tgsi_parse_token( &parse ); + + switch( parse.FullToken.Token.Type ) { + case TGSI_TOKEN_TYPE_DECLARATION: + /* If branching shader, emit preamble instructions at decl time, as + * instruction order in the shader does not correspond to the order + * instructions are executed in the wild. + * + * This is where special instructions such as WM_CINTERP, + * WM_LINTERP, WM_PINTERP and WM_WPOSXY are emitted to compute + * shader inputs from varying vars. + * + * XXX: For non-branching shaders, consider deferring variable + * initialization as late as possible to minimize register + * usage. This is how the original BRW driver worked. */ - out->DstReg.WriteMask &= BRW_WRITEMASK_XY; - break; - - case TGSI_OPCODE_DST: - precalc_dst(c, inst); - break; - - case TGSI_OPCODE_LIT: - precalc_lit(c, inst); - break; - - case TGSI_OPCODE_TEX: - precalc_tex(c, inst); - break; - - case TGSI_OPCODE_TXP: - precalc_txp(c, inst); - break; - - case TGSI_OPCODE_TXB: - out = emit_insn(c, inst); - out->TexSrcUnit = fp->program.Base.SamplerUnits[inst->TexSrcUnit]; - assert(out->TexSrcUnit < BRW_MAX_TEX_UNIT); - break; - - case TGSI_OPCODE_XPD: - out = emit_insn(c, inst); - /* This should probably be done in the parser. + validate_src_regs(c, inst); + validate_dst_regs(c, inst); + break; + + case TGSI_TOKEN_TYPE_IMMEDIATE: + /* Unlike VS programs we can probably manage fine encoding + * immediate values directly into the emitted EU + * instructions, as we probably only need to reference one + * float value per instruction. Just save the data for now + * and use directly later. */ - out->DstReg.WriteMask &= BRW_WRITEMASK_XYZ; break; - case TGSI_OPCODE_KIL: - out = emit_insn(c, inst); - /* This should probably be done in the parser. - */ - out->DstReg.WriteMask = 0; - break; - case TGSI_OPCODE_END: - emit_fb_write(c); - break; - default: - if (brw_wm_is_scalar_result(inst->Opcode)) - emit_scalar_insn(c, inst); - else - emit_insn(c, inst); + case TGSI_TOKEN_TYPE_INSTRUCTION: + inst = &parse.FullToken.FullInstruction; + emit_insn( c, inst ); break; } } + c->brw_program = brw_finalize( c->builder ); + if (BRW_DEBUG & DEBUG_WM) { debug_printf("pass_fp:\n"); - print_insns( c->prog_instructions, c->nr_fp_insns ); + brw_print_program( c->brw_program ); debug_printf("\n"); } } diff --git a/src/gallium/drivers/i965/brw_wm_glsl.c b/src/gallium/drivers/i965/brw_wm_glsl.c index 59bc4ef701..cdc10484a6 100644 --- a/src/gallium/drivers/i965/brw_wm_glsl.c +++ b/src/gallium/drivers/i965/brw_wm_glsl.c @@ -332,7 +332,7 @@ static void prealloc_reg(struct brw_wm_compile *c) for (j = 0; j < 4; j++) set_reg(c, PROGRAM_PAYLOAD, fp_input, j, reg); } - if (c->key.vp_outputs_written & (1 << i)) { + if (c->key.nr_vp_outputs > i) { reg_index += 2; } } @@ -1670,7 +1670,7 @@ get_argument_regs(struct brw_wm_compile *c, } } -static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) +static void brw_wm_emit_branching_shader(struct brw_context *brw, struct brw_wm_compile *c) { #define MAX_IF_DEPTH 32 #define MAX_LOOP_DEPTH 32 @@ -1943,20 +1943,20 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) * Do GPU code generation for shaders that use GLSL features such as * flow control. Other shaders will be compiled with the */ -void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c) +void brw_wm_branching_shader_emit(struct brw_context *brw, struct brw_wm_compile *c) { if (BRW_DEBUG & DEBUG_WM) { - debug_printf("brw_wm_glsl_emit:\n"); + debug_printf("%s:\n", __FUNCTION__); } /* initial instruction translation/simplification */ brw_wm_pass_fp(c); /* actual code generation */ - brw_wm_emit_glsl(brw, c); + brw_wm_emit_branching_shader(brw, c); if (BRW_DEBUG & DEBUG_WM) { - brw_wm_print_program(c, "brw_wm_glsl_emit done"); + brw_wm_print_program(c, "brw_wm_branching_shader_emit done"); } c->prog_data.total_grf = num_grf_used(c); diff --git a/src/gallium/drivers/i965/brw_wm_pass0.c b/src/gallium/drivers/i965/brw_wm_pass0.c index 71e4c56835..d8b9028927 100644 --- a/src/gallium/drivers/i965/brw_wm_pass0.c +++ b/src/gallium/drivers/i965/brw_wm_pass0.c @@ -168,54 +168,20 @@ static const struct brw_wm_ref *pass0_get_reg( struct brw_wm_compile *c, if (!ref) { switch (file) { - case PROGRAM_INPUT: - case PROGRAM_PAYLOAD: - case PROGRAM_TEMPORARY: - case PROGRAM_OUTPUT: - case PROGRAM_VARYING: + case TGSI_FILE_INPUT: + case TGSI_FILE_TEMPORARY: + case TGSI_FILE_OUTPUT: + case BRW_FILE_PAYLOAD: + /* should already be done?? */ break; - case PROGRAM_LOCAL_PARAM: - ref = get_param_ref(c, &c->fp->program.Base.LocalParams[idx][component]); - break; - - case PROGRAM_ENV_PARAM: + case TGSI_FILE_CONSTANT: ref = get_param_ref(c, &c->env_param[idx][component]); break; - case PROGRAM_STATE_VAR: - case PROGRAM_UNIFORM: - case PROGRAM_CONSTANT: - case PROGRAM_NAMED_PARAM: { - struct gl_program_parameter_list *plist = c->fp->program.Base.Parameters; - - /* There's something really hokey about parameters parsed in - * arb programs - they all end up in here, whether they be - * state values, parameters or constants. This duplicates the - * structure above & also seems to subvert the limits set for - * each type of constant/param. - */ - switch (plist->Parameters[idx].Type) { - case PROGRAM_NAMED_PARAM: - case PROGRAM_CONSTANT: - /* These are invarient: - */ - ref = get_imm_ref(c, &plist->ParameterValues[idx][component]); - break; - - case PROGRAM_STATE_VAR: - case PROGRAM_UNIFORM: - /* These may change from run to run: - */ - ref = get_param_ref(c, &plist->ParameterValues[idx][component] ); - break; - - default: - assert(0); - break; - } + case TGSI_FILE_IMMEDIATE: + ref = get_imm_ref(c, &plist->ParameterValues[idx][component]); break; - } default: assert(0); @@ -310,17 +276,16 @@ translate_insn(struct brw_wm_compile *c, const struct prog_instruction *inst) { struct brw_wm_instruction *out = get_instruction(c); - GLuint writemask = inst->DstReg.WriteMask; + GLuint writemask = inst->dst.WriteMask; GLuint nr_args = brw_wm_nr_args(inst->Opcode); GLuint i, j; /* Copy some data out of the instruction */ out->opcode = inst->Opcode; - out->saturate = (inst->SaturateMode != SATURATE_OFF); + out->saturate = inst->dst.Saturate; out->tex_unit = inst->TexSrcUnit; - out->tex_idx = inst->TexSrcTarget; - out->tex_shadow = inst->TexShadow; + out->tex_target = inst->TexSrcTarget; out->eot = inst->Aux & 1; out->target = inst->Aux >> 1; @@ -328,7 +293,7 @@ translate_insn(struct brw_wm_compile *c, */ for (i = 0; i < nr_args; i++) { for (j = 0; j < 4; j++) { - out->src[i][j] = get_new_ref(c, inst->SrcReg[i], j, out); + out->src[i][j] = get_new_ref(c, inst->src[i], j, out); } } @@ -380,15 +345,6 @@ static void pass0_init_payload( struct brw_wm_compile *c ) &c->payload.depth[j] ); } -#if 0 - /* This seems to be an alternative to the INTERP_WPOS stuff I do - * elsewhere: - */ - if (c->key.source_depth_reg) - pass0_set_fpreg_value(c, PROGRAM_INPUT, FRAG_ATTRIB_WPOS, 2, - &c->payload.depth[c->key.source_depth_reg/2]); -#endif - for (i = 0; i < FRAG_ATTRIB_MAX; i++) pass0_set_fpreg_value( c, PROGRAM_PAYLOAD, i, 0, &c->payload.input_interp[i] ); @@ -403,6 +359,9 @@ static void pass0_init_payload( struct brw_wm_compile *c ) * the same number. * * Translate away swizzling and eliminate non-saturating moves. + * + * Translate instructions from Mesa's prog_instruction structs to our + * internal brw_wm_instruction representation. */ void brw_wm_pass0( struct brw_wm_compile *c ) { @@ -421,7 +380,7 @@ void brw_wm_pass0( struct brw_wm_compile *c ) */ switch (inst->Opcode) { case OPCODE_MOV: - if (!inst->SaturateMode) { + if (!inst->dst.Saturate) { pass0_precalc_mov(c, inst); } else { diff --git a/src/gallium/drivers/i965/brw_wm_pass1.c b/src/gallium/drivers/i965/brw_wm_pass1.c index 85a3a55ca4..b0356b1bd5 100644 --- a/src/gallium/drivers/i965/brw_wm_pass1.c +++ b/src/gallium/drivers/i965/brw_wm_pass1.c @@ -90,17 +90,24 @@ static void track_arg(struct brw_wm_compile *c, static GLuint get_texcoord_mask( GLuint tex_idx ) { switch (tex_idx) { - case TEXTURE_1D_INDEX: + case TGSI_TEXTURE_1D: return BRW_WRITEMASK_X; - case TEXTURE_2D_INDEX: + case TGSI_TEXTURE_2D: + case TGSI_TEXTURE_RECT: return BRW_WRITEMASK_XY; - case TEXTURE_3D_INDEX: + case TGSI_TEXTURE_3D: return BRW_WRITEMASK_XYZ; - case TEXTURE_CUBE_INDEX: + case TGSI_TEXTURE_CUBE: return BRW_WRITEMASK_XYZ; - case TEXTURE_RECT_INDEX: - return BRW_WRITEMASK_XY; - default: return 0; + + case TGSI_TEXTURE_SHADOW1D: + return BRW_WRITEMASK_XZ; + case TGSI_TEXTURE_SHADOW2D: + case TGSI_TEXTURE_SHADOWRECT: + return BRW_WRITEMASK_XYZ; + default: + assert(0); + return 0; } } @@ -217,14 +224,9 @@ void brw_wm_pass1( struct brw_wm_compile *c ) case TGSI_OPCODE_TEX: case TGSI_OPCODE_TXP: read0 = get_texcoord_mask(inst->tex_idx); - - if (inst->tex_shadow) - read0 |= BRW_WRITEMASK_Z; break; case TGSI_OPCODE_TXB: - /* Shadow ignored for txb. - */ read0 = get_texcoord_mask(inst->tex_idx) | BRW_WRITEMASK_W; break; diff --git a/src/gallium/drivers/i965/brw_wm_state.c b/src/gallium/drivers/i965/brw_wm_state.c index edabf6ceb6..1898f38cef 100644 --- a/src/gallium/drivers/i965/brw_wm_state.c +++ b/src/gallium/drivers/i965/brw_wm_state.c @@ -52,7 +52,7 @@ struct brw_wm_unit_key { unsigned int max_threads; unsigned int nr_surfaces, sampler_count; - GLboolean uses_depth, computes_depth, uses_kill, is_glsl; + GLboolean uses_depth, computes_depth, uses_kill, has_flow_control; GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable; GLfloat offset_units, offset_factor; }; @@ -114,10 +114,10 @@ wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) /* _NEW_COLOR */ key->uses_kill = fp->UsesKill || ctx->Color.AlphaEnabled; - key->is_glsl = bfp->isGLSL; + key->has_flow_control = bfp->has_flow_control; /* temporary sanity check assertion */ - ASSERT(bfp->isGLSL == brw_wm_is_glsl(fp)); + ASSERT(bfp->has_flow_control == brw_wm_has_flow_control(fp)); /* _NEW_QUERY */ key->stats_wm = (brw->query.stats_wm != 0); @@ -184,7 +184,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, wm.wm5.program_computes_depth = key->computes_depth; wm.wm5.program_uses_killpixel = key->uses_kill; - if (key->is_glsl) + if (key->has_flow_control) wm.wm5.enable_8_pix = 1; else wm.wm5.enable_16_pix = 1; -- cgit v1.2.3 From c796aed5ddad011d66e631c4cafdbf779e73f213 Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Thu, 5 Nov 2009 13:57:05 +0000 Subject: i965g: add lots of error checks and early returns Any allocation that may fail should be checked, and propogate the error upwards. At the highest level we will flush batch and retry. This is an alternate strategy to what the original DRI driver did of attempting to flush batch from the lowest levels (eg inside BEGIN_BATCH). The trouble with that strategy was that flushes could occur at unexpected times, and additionally there was a need for a wierd notification mechanism to propogate the 'lost context' state back up to higher levels. Propogating the errors directly gives us a lot of flexibility how to deal with these states, at the expense of a lot more checking in the code. Will add some sanity checks later to make sure that out-of-memory conditions are properly escalated and not lost halfway up the stack. --- src/gallium/drivers/i965/brw_batchbuffer.c | 19 +- src/gallium/drivers/i965/brw_batchbuffer.h | 3 +- src/gallium/drivers/i965/brw_cc.c | 73 ++++--- src/gallium/drivers/i965/brw_clip.c | 60 ++++-- src/gallium/drivers/i965/brw_clip_state.c | 60 +++--- src/gallium/drivers/i965/brw_context.c | 46 +++-- src/gallium/drivers/i965/brw_context.h | 2 +- src/gallium/drivers/i965/brw_curbe.c | 18 +- src/gallium/drivers/i965/brw_draw.c | 3 +- src/gallium/drivers/i965/brw_draw_upload.c | 18 +- src/gallium/drivers/i965/brw_eu.c | 13 +- src/gallium/drivers/i965/brw_eu.h | 8 +- src/gallium/drivers/i965/brw_gs.c | 69 ++++--- src/gallium/drivers/i965/brw_gs_state.c | 48 +++-- src/gallium/drivers/i965/brw_pipe_query.c | 31 +-- src/gallium/drivers/i965/brw_pipe_shader.c | 3 +- src/gallium/drivers/i965/brw_pipe_vertex.c | 2 +- src/gallium/drivers/i965/brw_screen_buffers.c | 16 +- src/gallium/drivers/i965/brw_screen_surface.c | 7 +- src/gallium/drivers/i965/brw_screen_texture.c | 17 +- src/gallium/drivers/i965/brw_sf.c | 52 +++-- src/gallium/drivers/i965/brw_sf_state.c | 86 ++++---- src/gallium/drivers/i965/brw_state.h | 71 +++---- src/gallium/drivers/i965/brw_state_cache.c | 115 +++++------ src/gallium/drivers/i965/brw_state_upload.c | 3 +- src/gallium/drivers/i965/brw_vs.c | 56 +++--- src/gallium/drivers/i965/brw_vs_state.c | 58 +++--- src/gallium/drivers/i965/brw_vs_surface_state.c | 97 ++++++---- src/gallium/drivers/i965/brw_winsys.h | 56 ++++-- src/gallium/drivers/i965/brw_wm.c | 78 ++++---- src/gallium/drivers/i965/brw_wm_constant_buffer.c | 87 +++++---- src/gallium/drivers/i965/brw_wm_sampler_state.c | 98 ++++++---- src/gallium/drivers/i965/brw_wm_state.c | 103 ++++++---- src/gallium/drivers/i965/brw_wm_surface_state.c | 226 ++++++++++++---------- src/gallium/winsys/drm/i965/xlib/xlib_i965.c | 46 ++--- 35 files changed, 1003 insertions(+), 745 deletions(-) (limited to 'src/gallium/drivers/i965/brw_eu.c') diff --git a/src/gallium/drivers/i965/brw_batchbuffer.c b/src/gallium/drivers/i965/brw_batchbuffer.c index ca612e5ed0..e5f73bd6a3 100644 --- a/src/gallium/drivers/i965/brw_batchbuffer.c +++ b/src/gallium/drivers/i965/brw_batchbuffer.c @@ -38,17 +38,17 @@ #define USE_MALLOC_BUFFER 1 #define ALWAYS_EMIT_MI_FLUSH 1 -void +enum pipe_error brw_batchbuffer_reset(struct brw_batchbuffer *batch) { - if (batch->buf) { - batch->sws->bo_unreference(batch->buf); - batch->buf = NULL; - } + enum pipe_error ret; - batch->buf = batch->sws->bo_alloc(batch->sws, - BRW_BUFFER_TYPE_BATCH, - BRW_BATCH_SIZE, 4096); + ret = batch->sws->bo_alloc( batch->sws, + BRW_BUFFER_TYPE_BATCH, + BRW_BATCH_SIZE, 4096, + &batch->buf ); + if (ret) + return ret; if (batch->malloc_buffer) batch->map = batch->malloc_buffer; @@ -59,6 +59,7 @@ brw_batchbuffer_reset(struct brw_batchbuffer *batch) batch->size = BRW_BATCH_SIZE; batch->ptr = batch->map; + return PIPE_OK; } struct brw_batchbuffer * @@ -91,7 +92,7 @@ brw_batchbuffer_free(struct brw_batchbuffer *batch) batch->map = NULL; } - batch->sws->bo_unreference(batch->buf); + bo_reference(&batch->buf, NULL); FREE(batch); } diff --git a/src/gallium/drivers/i965/brw_batchbuffer.h b/src/gallium/drivers/i965/brw_batchbuffer.h index 1f04826aea..288a9d2755 100644 --- a/src/gallium/drivers/i965/brw_batchbuffer.h +++ b/src/gallium/drivers/i965/brw_batchbuffer.h @@ -65,7 +65,8 @@ void _brw_batchbuffer_flush(struct brw_batchbuffer *batch, const char *file, int line); -void brw_batchbuffer_reset(struct brw_batchbuffer *batch); +enum pipe_error +brw_batchbuffer_reset(struct brw_batchbuffer *batch); /* Unlike bmBufferData, this currently requires the buffer be mapped. diff --git a/src/gallium/drivers/i965/brw_cc.c b/src/gallium/drivers/i965/brw_cc.c index 20967f0191..8e25fe8585 100644 --- a/src/gallium/drivers/i965/brw_cc.c +++ b/src/gallium/drivers/i965/brw_cc.c @@ -57,10 +57,11 @@ static void calc_sane_viewport( const struct pipe_viewport_state *vp, svp->far = 1; } -static int prepare_cc_vp( struct brw_context *brw ) +static enum pipe_error prepare_cc_vp( struct brw_context *brw ) { struct brw_cc_viewport ccv; struct sane_viewport svp; + enum pipe_error ret; memset(&ccv, 0, sizeof(ccv)); @@ -70,10 +71,12 @@ static int prepare_cc_vp( struct brw_context *brw ) ccv.min_depth = svp.near; ccv.max_depth = svp.far; - brw->sws->bo_unreference(brw->cc.vp_bo); - brw->cc.vp_bo = brw_cache_data( &brw->cache, BRW_CC_VP, &ccv, NULL, 0 ); - - return 0; + ret = brw_cache_data( &brw->cache, BRW_CC_VP, &ccv, NULL, 0, + &brw->cc.vp_bo ); + if (ret) + return ret; + + return PIPE_OK; } const struct brw_tracked_state brw_cc_vp = { @@ -123,11 +126,13 @@ cc_unit_populate_key(const struct brw_context *brw, /** * Creates the state cache entry for the given CC unit key. */ -static struct brw_winsys_buffer * -cc_unit_create_from_key(struct brw_context *brw, struct brw_cc_unit_key *key) +static enum pipe_error +cc_unit_create_from_key(struct brw_context *brw, + struct brw_cc_unit_key *key, + struct brw_winsys_buffer **bo_out) { struct brw_cc_unit_state cc; - struct brw_winsys_buffer *bo; + enum pipe_error ret; memset(&cc, 0, sizeof(cc)); @@ -143,38 +148,48 @@ cc_unit_create_from_key(struct brw_context *brw, struct brw_cc_unit_key *key) cc.cc6 = key->cc6; cc.cc7 = key->cc7; - bo = brw_upload_cache(&brw->cache, BRW_CC_UNIT, - key, sizeof(*key), - &brw->cc.vp_bo, 1, - &cc, sizeof(cc), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_CC_UNIT, + key, sizeof(*key), + &brw->cc.vp_bo, 1, + &cc, sizeof(cc), + NULL, NULL, + bo_out); + if (ret) + return ret; - /* Emit CC viewport relocation */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - 0, - offsetof(struct brw_cc_unit_state, cc4), - brw->cc.vp_bo); - return bo; + /* Emit CC viewport relocation */ + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + 0, + offsetof(struct brw_cc_unit_state, cc4), + brw->cc.vp_bo); + if (ret) + return ret; + + return PIPE_OK; } static int prepare_cc_unit( struct brw_context *brw ) { struct brw_cc_unit_key key; + enum pipe_error ret; cc_unit_populate_key(brw, &key); - brw->sws->bo_unreference(brw->cc.state_bo); - brw->cc.state_bo = brw_search_cache(&brw->cache, BRW_CC_UNIT, - &key, sizeof(key), - &brw->cc.vp_bo, 1, - NULL); - - if (brw->cc.state_bo == NULL) - brw->cc.state_bo = cc_unit_create_from_key(brw, &key); + if (brw_search_cache(&brw->cache, BRW_CC_UNIT, + &key, sizeof(key), + &brw->cc.vp_bo, 1, + NULL, + &brw->cc.state_bo)) + return PIPE_OK; + + ret = cc_unit_create_from_key(brw, &key, + &brw->cc.state_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_cc_unit = { diff --git a/src/gallium/drivers/i965/brw_clip.c b/src/gallium/drivers/i965/brw_clip.c index 1a52fa771b..35e1d2fdbd 100644 --- a/src/gallium/drivers/i965/brw_clip.c +++ b/src/gallium/drivers/i965/brw_clip.c @@ -48,9 +48,12 @@ #define BACK_UNFILLED_BIT 0x2 -static void compile_clip_prog( struct brw_context *brw, - struct brw_clip_prog_key *key ) +static enum pipe_error +compile_clip_prog( struct brw_context *brw, + struct brw_clip_prog_key *key, + struct brw_winsys_buffer **bo_out ) { + enum pipe_error ret; struct brw_clip_compile c; const GLuint *program; GLuint program_size; @@ -123,31 +126,39 @@ static void compile_clip_prog( struct brw_context *brw, break; default: assert(0); - return; + return PIPE_ERROR_BAD_INPUT; } /* get the program */ - program = brw_get_program(&c.func, &program_size); + ret = brw_get_program(&c.func, &program, &program_size); + if (ret) + return ret; /* Upload */ - brw->sws->bo_unreference(brw->clip.prog_bo); - brw->clip.prog_bo = brw_upload_cache( &brw->cache, - BRW_CLIP_PROG, - &c.key, sizeof(c.key), - NULL, 0, - program, program_size, - &c.prog_data, - &brw->clip.prog_data ); + ret = brw_upload_cache( &brw->cache, + BRW_CLIP_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + &brw->clip.prog_data, + bo_out ); + if (ret) + return ret; + + return PIPE_OK; } /* Calculate interpolants for triangle and line rasterization. */ -static int upload_clip_prog(struct brw_context *brw) +static enum pipe_error +upload_clip_prog(struct brw_context *brw) { + enum pipe_error ret; struct brw_clip_prog_key key; /* Populate the key, starting from the almost-complete version from @@ -166,15 +177,22 @@ static int upload_clip_prog(struct brw_context *brw) /* PIPE_NEW_CLIP */ key.nr_userclip = brw->curr.ucp.nr; - brw->sws->bo_unreference(brw->clip.prog_bo); - brw->clip.prog_bo = brw_search_cache(&brw->cache, BRW_CLIP_PROG, - &key, sizeof(key), - NULL, 0, - &brw->clip.prog_data); - if (brw->clip.prog_bo == NULL) - compile_clip_prog( brw, &key ); + /* Already cached? + */ + if (brw_search_cache(&brw->cache, BRW_CLIP_PROG, + &key, sizeof(key), + NULL, 0, + &brw->clip.prog_data, + &brw->clip.prog_bo)) + return PIPE_OK; + + /* Compile new program: + */ + ret = compile_clip_prog( brw, &key, &brw->clip.prog_bo ); + if (ret) + return ret; - return 0; + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_clip_state.c b/src/gallium/drivers/i965/brw_clip_state.c index 6f8309fea9..d4e3c43c61 100644 --- a/src/gallium/drivers/i965/brw_clip_state.c +++ b/src/gallium/drivers/i965/brw_clip_state.c @@ -72,12 +72,13 @@ clip_unit_populate_key(struct brw_context *brw, struct brw_clip_unit_key *key) key->depth_clamp = 0; // XXX: add this to gallium: ctx->Transform.DepthClamp; } -static struct brw_winsys_buffer * +static enum pipe_error clip_unit_create_from_key(struct brw_context *brw, - struct brw_clip_unit_key *key) + struct brw_clip_unit_key *key, + struct brw_winsys_buffer **bo_out) { struct brw_clip_unit_state clip; - struct brw_winsys_buffer *bo; + enum pipe_error ret; memset(&clip, 0, sizeof(clip)); @@ -141,39 +142,50 @@ clip_unit_create_from_key(struct brw_context *brw, clip.viewport_ymin = -1; clip.viewport_ymax = 1; - bo = brw_upload_cache(&brw->cache, BRW_CLIP_UNIT, - key, sizeof(*key), - &brw->clip.prog_bo, 1, - &clip, sizeof(clip), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_CLIP_UNIT, + key, sizeof(*key), + &brw->clip.prog_bo, 1, + &clip, sizeof(clip), + NULL, NULL, + bo_out); + if (ret) + return ret; /* Emit clip program relocation */ assert(brw->clip.prog_bo); - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - clip.thread0.grf_reg_count << 1, - offsetof(struct brw_clip_unit_state, thread0), - brw->clip.prog_bo); - - return bo; + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + clip.thread0.grf_reg_count << 1, + offsetof(struct brw_clip_unit_state, thread0), + brw->clip.prog_bo); + if (ret) + return ret; + + return PIPE_OK; } static int upload_clip_unit( struct brw_context *brw ) { struct brw_clip_unit_key key; + enum pipe_error ret; clip_unit_populate_key(brw, &key); - brw->sws->bo_unreference(brw->clip.state_bo); - brw->clip.state_bo = brw_search_cache(&brw->cache, BRW_CLIP_UNIT, - &key, sizeof(key), - &brw->clip.prog_bo, 1, - NULL); - if (brw->clip.state_bo == NULL) { - brw->clip.state_bo = clip_unit_create_from_key(brw, &key); - } + if (brw_search_cache(&brw->cache, BRW_CLIP_UNIT, + &key, sizeof(key), + &brw->clip.prog_bo, 1, + NULL, + &brw->clip.state_bo)) + return PIPE_OK; + + /* Create new: + */ + ret = clip_unit_create_from_key(brw, &key, + &brw->clip.state_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_clip_unit = { diff --git a/src/gallium/drivers/i965/brw_context.c b/src/gallium/drivers/i965/brw_context.c index aaf7d1834e..2cee7a7a3c 100644 --- a/src/gallium/drivers/i965/brw_context.c +++ b/src/gallium/drivers/i965/brw_context.c @@ -72,29 +72,33 @@ static void brw_destroy_context( struct pipe_context *pipe ) brw->curr.fb.nr_cbufs = 0; pipe_surface_reference(&brw->curr.fb.zsbuf, NULL); - brw->sws->bo_unreference(brw->curbe.curbe_bo); - brw->sws->bo_unreference(brw->vs.prog_bo); - brw->sws->bo_unreference(brw->vs.state_bo); - brw->sws->bo_unreference(brw->vs.bind_bo); - brw->sws->bo_unreference(brw->gs.prog_bo); - brw->sws->bo_unreference(brw->gs.state_bo); - brw->sws->bo_unreference(brw->clip.prog_bo); - brw->sws->bo_unreference(brw->clip.state_bo); - brw->sws->bo_unreference(brw->clip.vp_bo); - brw->sws->bo_unreference(brw->sf.prog_bo); - brw->sws->bo_unreference(brw->sf.state_bo); - brw->sws->bo_unreference(brw->sf.vp_bo); + bo_reference(&brw->curbe.curbe_bo, NULL); + bo_reference(&brw->vs.prog_bo, NULL); + bo_reference(&brw->vs.state_bo, NULL); + bo_reference(&brw->vs.bind_bo, NULL); + bo_reference(&brw->gs.prog_bo, NULL); + bo_reference(&brw->gs.state_bo, NULL); + bo_reference(&brw->clip.prog_bo, NULL); + bo_reference(&brw->clip.state_bo, NULL); + bo_reference(&brw->clip.vp_bo, NULL); + bo_reference(&brw->sf.prog_bo, NULL); + bo_reference(&brw->sf.state_bo, NULL); + bo_reference(&brw->sf.vp_bo, NULL); + for (i = 0; i < BRW_MAX_TEX_UNIT; i++) - brw->sws->bo_unreference(brw->wm.sdc_bo[i]); - brw->sws->bo_unreference(brw->wm.bind_bo); + bo_reference(&brw->wm.sdc_bo[i], NULL); + + bo_reference(&brw->wm.bind_bo, NULL); + for (i = 0; i < BRW_WM_MAX_SURF; i++) - brw->sws->bo_unreference(brw->wm.surf_bo[i]); - brw->sws->bo_unreference(brw->wm.sampler_bo); - brw->sws->bo_unreference(brw->wm.prog_bo); - brw->sws->bo_unreference(brw->wm.state_bo); - brw->sws->bo_unreference(brw->cc.prog_bo); - brw->sws->bo_unreference(brw->cc.state_bo); - brw->sws->bo_unreference(brw->cc.vp_bo); + bo_reference(&brw->wm.surf_bo[i], NULL); + + bo_reference(&brw->wm.sampler_bo, NULL); + bo_reference(&brw->wm.prog_bo, NULL); + bo_reference(&brw->wm.state_bo, NULL); + bo_reference(&brw->cc.prog_bo, NULL); + bo_reference(&brw->cc.state_bo, NULL); + bo_reference(&brw->cc.vp_bo, NULL); } diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h index 09d34615c7..580251d2f1 100644 --- a/src/gallium/drivers/i965/brw_context.h +++ b/src/gallium/drivers/i965/brw_context.h @@ -744,7 +744,7 @@ struct brw_context * brw_queryobj.c */ void brw_init_query(struct brw_context *brw); -void brw_prepare_query_begin(struct brw_context *brw); +enum pipe_error brw_prepare_query_begin(struct brw_context *brw); void brw_emit_query_begin(struct brw_context *brw); void brw_emit_query_end(struct brw_context *brw); diff --git a/src/gallium/drivers/i965/brw_curbe.c b/src/gallium/drivers/i965/brw_curbe.c index 1e2e232204..ca7774a7cc 100644 --- a/src/gallium/drivers/i965/brw_curbe.c +++ b/src/gallium/drivers/i965/brw_curbe.c @@ -160,10 +160,11 @@ static GLfloat fixed_plane[6][4] = { * cache mechanism, but maybe would benefit from a comparison against * the current uploaded set of constants. */ -static int prepare_curbe_buffer(struct brw_context *brw) +static enum pipe_error prepare_curbe_buffer(struct brw_context *brw) { const GLuint sz = brw->curbe.total_size; const GLuint bufsz = sz * 16 * sizeof(GLfloat); + enum pipe_error ret; GLfloat *buf; GLuint i; @@ -267,17 +268,20 @@ static int prepare_curbe_buffer(struct brw_context *brw) (brw->curbe.need_new_bo || brw->curbe.curbe_next_offset + bufsz > brw->curbe.curbe_bo->size)) { - brw->sws->bo_unreference(brw->curbe.curbe_bo); - brw->curbe.curbe_bo = NULL; + bo_reference(&brw->curbe.curbe_bo, NULL); } if (brw->curbe.curbe_bo == NULL) { /* Allocate a single page for CURBE entries for this batchbuffer. * They're generally around 64b. */ - brw->curbe.curbe_bo = brw->sws->bo_alloc(brw->sws, - BRW_BUFFER_TYPE_CURBE, - 4096, 1 << 6); + ret = brw->sws->bo_alloc(brw->sws, + BRW_BUFFER_TYPE_CURBE, + 4096, 1 << 6, + &brw->curbe.curbe_bo); + if (ret) + return ret; + brw->curbe.curbe_next_offset = 0; } @@ -313,7 +317,7 @@ static int prepare_curbe_buffer(struct brw_context *brw) return 0; } -static int emit_curbe_buffer(struct brw_context *brw) +static enum pipe_error emit_curbe_buffer(struct brw_context *brw) { GLuint sz = brw->curbe.total_size; diff --git a/src/gallium/drivers/i965/brw_draw.c b/src/gallium/drivers/i965/brw_draw.c index 6d6b1c7c5c..88cb31ad54 100644 --- a/src/gallium/drivers/i965/brw_draw.c +++ b/src/gallium/drivers/i965/brw_draw.c @@ -280,6 +280,5 @@ void brw_draw_cleanup( struct brw_context *brw ) u_upload_destroy( brw->vb.upload_vertex ); u_upload_destroy( brw->vb.upload_index ); - brw->sws->bo_unreference(brw->ib.bo); - brw->ib.bo = NULL; + bo_reference(&brw->ib.bo, NULL); } diff --git a/src/gallium/drivers/i965/brw_draw_upload.c b/src/gallium/drivers/i965/brw_draw_upload.c index 4fa7d549eb..188605a0c1 100644 --- a/src/gallium/drivers/i965/brw_draw_upload.c +++ b/src/gallium/drivers/i965/brw_draw_upload.c @@ -251,9 +251,8 @@ static int brw_prepare_vertices(struct brw_context *brw) brw->vb.vb[i].vertex_count = (vb->stride == 0 ? 1 : (bo->size - offset) / vb->stride); - brw->sws->bo_unreference(brw->vb.vb[i].bo); - brw->vb.vb[i].bo = bo; - brw->sws->bo_reference(brw->vb.vb[i].bo); + + bo_reference( &brw->vb.vb[i].bo, bo ); /* Don't need to retain this reference. We have a reference on * the underlying winsys buffer: @@ -417,6 +416,7 @@ const struct brw_tracked_state brw_vertices = { static int brw_prepare_indices(struct brw_context *brw) { struct pipe_buffer *index_buffer = brw->curr.index_buffer; + struct pipe_buffer *upload_buf = NULL; struct brw_winsys_buffer *bo = NULL; GLuint offset; GLuint index_size; @@ -438,7 +438,6 @@ static int brw_prepare_indices(struct brw_context *brw) /* Turn userbuffer into a proper hardware buffer? */ if (brw_buffer_is_user_buffer(index_buffer)) { - struct pipe_buffer *upload_buf; ret = u_upload_buffer( brw->vb.upload_index, 0, @@ -450,8 +449,6 @@ static int brw_prepare_indices(struct brw_context *brw) return ret; bo = brw_buffer(upload_buf)->bo; - brw->sws->bo_reference(bo); - pipe_buffer_reference( &upload_buf, NULL ); /* XXX: annotate the userbuffer with the upload information so * that successive calls don't get re-uploaded. @@ -459,8 +456,6 @@ static int brw_prepare_indices(struct brw_context *brw) } else { bo = brw_buffer(index_buffer)->bo; - brw->sws->bo_reference(bo); - ib_size = bo->size; offset = 0; } @@ -486,15 +481,12 @@ static int brw_prepare_indices(struct brw_context *brw) if (brw->ib.bo != bo || brw->ib.size != ib_size) { - brw->sws->bo_unreference(brw->ib.bo); - brw->ib.bo = bo; + bo_reference(&brw->ib.bo, bo); brw->ib.size = ib_size; brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER; } - else { - brw->sws->bo_unreference(bo); - } + pipe_buffer_reference( &upload_buf, NULL ); brw_add_validated_bo(brw, brw->ib.bo); return 0; } diff --git a/src/gallium/drivers/i965/brw_eu.c b/src/gallium/drivers/i965/brw_eu.c index de43b14512..a8fcb5f97e 100644 --- a/src/gallium/drivers/i965/brw_eu.c +++ b/src/gallium/drivers/i965/brw_eu.c @@ -118,16 +118,23 @@ void brw_init_compile( struct brw_context *brw, struct brw_compile *p ) } -const GLuint *brw_get_program( struct brw_compile *p, - GLuint *sz ) +enum pipe_error brw_get_program( struct brw_compile *p, + const GLuint **data, + GLuint *sz ) { GLuint i; for (i = 0; i < 8; i++) brw_NOP(p); + /* Is the generated program malformed for some reason? + */ + if (p->error) + return PIPE_ERROR_BAD_INPUT; + *sz = p->nr_insn * sizeof(struct brw_instruction); - return (const GLuint *)p->store; + *data = (const GLuint *)p->store; + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_eu.h b/src/gallium/drivers/i965/brw_eu.h index 7bddc3859c..565f4ef1c5 100644 --- a/src/gallium/drivers/i965/brw_eu.h +++ b/src/gallium/drivers/i965/brw_eu.h @@ -34,6 +34,7 @@ #define BRW_EU_H #include "util/u_debug.h" +#include "pipe/p_error.h" #include "brw_structs.h" #include "brw_defines.h" @@ -132,6 +133,8 @@ struct brw_compile { struct brw_eu_label *first_label; /**< linked list of labels */ struct brw_eu_call *first_call; /**< linked list of CALs */ + + boolean error; }; @@ -772,7 +775,10 @@ void brw_set_predicate_control( struct brw_compile *p, GLuint pc ); void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional ); void brw_init_compile( struct brw_context *, struct brw_compile *p ); -const GLuint *brw_get_program( struct brw_compile *p, GLuint *sz ); + +enum pipe_error brw_get_program( struct brw_compile *p, + const GLuint **program, + GLuint *sz ); /* Helpers for regular instructions: diff --git a/src/gallium/drivers/i965/brw_gs.c b/src/gallium/drivers/i965/brw_gs.c index 693d8bfdf8..ce77be24f6 100644 --- a/src/gallium/drivers/i965/brw_gs.c +++ b/src/gallium/drivers/i965/brw_gs.c @@ -40,10 +40,12 @@ -static void compile_gs_prog( struct brw_context *brw, - struct brw_gs_prog_key *key ) +static enum pipe_error compile_gs_prog( struct brw_context *brw, + struct brw_gs_prog_key *key, + struct brw_winsys_buffer **bo_out ) { struct brw_gs_compile c; + enum pipe_error ret; const GLuint *program; GLuint program_size; @@ -57,9 +59,9 @@ static void compile_gs_prog( struct brw_context *brw, c.nr_attrs = c.key.nr_attrs; if (BRW_IS_IGDNG(brw)) - c.nr_regs = (c.nr_attrs + 1) / 2 + 3; /* are vertices packed, or reg-aligned? */ + c.nr_regs = (c.nr_attrs + 1) / 2 + 3; /* are vertices packed, or reg-aligned? */ else - c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ + c.nr_regs = (c.nr_attrs + 1) / 2 + 1; /* are vertices packed, or reg-aligned? */ c.nr_bytes = c.nr_regs * REG_SIZE; @@ -93,40 +95,47 @@ static void compile_gs_prog( struct brw_context *brw, if (key->hint_gs_always) brw_gs_lines( &c ); else { - return; + return PIPE_OK; } break; case PIPE_PRIM_TRIANGLES: if (key->hint_gs_always) brw_gs_tris( &c ); else { - return; + return PIPE_OK; } break; case PIPE_PRIM_POINTS: if (key->hint_gs_always) brw_gs_points( &c ); else { - return; + return PIPE_OK; } - break; + break; default: - return; + assert(0); + return PIPE_ERROR_BAD_INPUT; } /* get the program */ - program = brw_get_program(&c.func, &program_size); + ret = brw_get_program(&c.func, &program, &program_size); + if (ret) + return ret; /* Upload */ - brw->sws->bo_unreference(brw->gs.prog_bo); - brw->gs.prog_bo = brw_upload_cache( &brw->cache, BRW_GS_PROG, - &c.key, sizeof(c.key), - NULL, 0, - program, program_size, - &c.prog_data, - &brw->gs.prog_data ); + ret = brw_upload_cache( &brw->cache, BRW_GS_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + &brw->gs.prog_data, + bo_out ); + if (ret) + return ret; + + return PIPE_OK; } static const unsigned gs_prim[PIPE_PRIM_MAX] = { @@ -166,6 +175,8 @@ static void populate_key( struct brw_context *brw, static int prepare_gs_prog(struct brw_context *brw) { struct brw_gs_prog_key key; + enum pipe_error ret; + /* Populate the key: */ populate_key(brw, &key); @@ -175,17 +186,21 @@ static int prepare_gs_prog(struct brw_context *brw) brw->gs.prog_active = key.need_gs_prog; } - if (brw->gs.prog_active) { - brw->sws->bo_unreference(brw->gs.prog_bo); - brw->gs.prog_bo = brw_search_cache(&brw->cache, BRW_GS_PROG, - &key, sizeof(key), - NULL, 0, - &brw->gs.prog_data); - if (brw->gs.prog_bo == NULL) - compile_gs_prog( brw, &key ); - } + if (!brw->gs.prog_active) + return PIPE_OK; + + if (brw_search_cache(&brw->cache, BRW_GS_PROG, + &key, sizeof(key), + NULL, 0, + &brw->gs.prog_data, + &brw->gs.prog_bo)) + return PIPE_OK; + + ret = compile_gs_prog( brw, &key, &brw->gs.prog_bo ); + if (ret) + return ret; - return 0; + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_gs_state.c b/src/gallium/drivers/i965/brw_gs_state.c index f27f886a65..18a66da538 100644 --- a/src/gallium/drivers/i965/brw_gs_state.c +++ b/src/gallium/drivers/i965/brw_gs_state.c @@ -69,11 +69,13 @@ gs_unit_populate_key(struct brw_context *brw, struct brw_gs_unit_key *key) key->urb_size = brw->urb.vsize; } -static struct brw_winsys_buffer * -gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) +static enum pipe_error +gs_unit_create_from_key(struct brw_context *brw, + struct brw_gs_unit_key *key, + struct brw_winsys_buffer **bo_out) { struct brw_gs_unit_state gs; - struct brw_winsys_buffer *bo; + enum pipe_error ret; memset(&gs, 0, sizeof(gs)); @@ -104,40 +106,46 @@ gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) if (BRW_DEBUG & DEBUG_STATS) gs.thread4.stats_enable = 1; - bo = brw_upload_cache(&brw->cache, BRW_GS_UNIT, - key, sizeof(*key), - &brw->gs.prog_bo, 1, - &gs, sizeof(gs), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_GS_UNIT, + key, sizeof(*key), + &brw->gs.prog_bo, 1, + &gs, sizeof(gs), + NULL, NULL, + bo_out); + if (ret) + return ret; if (key->prog_active) { /* Emit GS program relocation */ - brw->sws->bo_emit_reloc(bo, + brw->sws->bo_emit_reloc(*bo_out, BRW_USAGE_STATE, gs.thread0.grf_reg_count << 1, offsetof(struct brw_gs_unit_state, thread0), brw->gs.prog_bo); } - return bo; + return PIPE_OK; } -static int prepare_gs_unit(struct brw_context *brw) +static enum pipe_error prepare_gs_unit(struct brw_context *brw) { struct brw_gs_unit_key key; + enum pipe_error ret; gs_unit_populate_key(brw, &key); - brw->sws->bo_unreference(brw->gs.state_bo); - brw->gs.state_bo = brw_search_cache(&brw->cache, BRW_GS_UNIT, - &key, sizeof(key), - &brw->gs.prog_bo, 1, - NULL); - if (brw->gs.state_bo == NULL) { - brw->gs.state_bo = gs_unit_create_from_key(brw, &key); - } + if (brw_search_cache(&brw->cache, BRW_GS_UNIT, + &key, sizeof(key), + &brw->gs.prog_bo, 1, + NULL, + &brw->gs.state_bo)) + return PIPE_OK; + + ret = gs_unit_create_from_key(brw, &key, &brw->gs.state_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_gs_unit = { diff --git a/src/gallium/drivers/i965/brw_pipe_query.c b/src/gallium/drivers/i965/brw_pipe_query.c index 3370ebd262..6a01173787 100644 --- a/src/gallium/drivers/i965/brw_pipe_query.c +++ b/src/gallium/drivers/i965/brw_pipe_query.c @@ -72,8 +72,7 @@ brw_query_get_result(struct pipe_context *pipe, } brw->sws->bo_unmap(query->bo); - brw->sws->bo_unreference(query->bo); - query->bo = NULL; + bo_reference(&query->bo, NULL); } *result = query->result; @@ -100,10 +99,9 @@ brw_query_create(struct pipe_context *pipe, unsigned type ) static void brw_query_destroy(struct pipe_context *pipe, struct pipe_query *q) { - struct brw_context *brw = brw_context(pipe); struct brw_query_object *query = (struct brw_query_object *)q; - brw->sws->bo_unreference(query->bo); + bo_reference(&query->bo, NULL); FREE(query); } @@ -114,9 +112,8 @@ brw_query_begin(struct pipe_context *pipe, struct pipe_query *q) struct brw_query_object *query = (struct brw_query_object *)q; /* Reset our driver's tracking of query state. */ - brw->sws->bo_unreference(query->bo); + bo_reference(&query->bo, NULL); query->result = 0; - query->bo = NULL; query->first_index = -1; query->last_index = -1; @@ -139,8 +136,7 @@ brw_query_end(struct pipe_context *pipe, struct pipe_query *q) brw_emit_query_end(brw); brw_context_flush( brw ); - brw->sws->bo_unreference(brw->query.bo); - brw->query.bo = NULL; + bo_reference(&brw->query.bo, NULL); } remove_from_list(query); @@ -153,24 +149,30 @@ brw_query_end(struct pipe_context *pipe, struct pipe_query *q) */ /** Called to set up the query BO and account for its aperture space */ -void +enum pipe_error brw_prepare_query_begin(struct brw_context *brw) { + enum pipe_error ret; + /* Skip if we're not doing any queries. */ if (is_empty_list(&brw->query.active_head)) - return; + return PIPE_OK; /* Get a new query BO if we're going to need it. */ if (brw->query.bo == NULL || brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) { - brw->sws->bo_unreference(brw->query.bo); - brw->query.bo = NULL; - brw->query.bo = brw->sws->bo_alloc(brw->sws, BRW_BUFFER_TYPE_QUERY, 4096, 1); + ret = brw->sws->bo_alloc(brw->sws, BRW_BUFFER_TYPE_QUERY, 4096, 1, + &brw->query.bo); + if (ret) + return ret; + brw->query.index = 0; } brw_add_validated_bo(brw, brw->query.bo); + + return PIPE_OK; } /** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */ @@ -213,8 +215,7 @@ brw_emit_query_begin(struct brw_context *brw) FALSE, &tmp ); - brw->sws->bo_reference(brw->query.bo); - query->bo = brw->query.bo; + bo_reference( &query->bo, brw->query.bo ); query->first_index = brw->query.index; } query->last_index = brw->query.index; diff --git a/src/gallium/drivers/i965/brw_pipe_shader.c b/src/gallium/drivers/i965/brw_pipe_shader.c index 2833f2bce0..662c43c3e5 100644 --- a/src/gallium/drivers/i965/brw_pipe_shader.c +++ b/src/gallium/drivers/i965/brw_pipe_shader.c @@ -146,10 +146,9 @@ fail: static void brw_delete_fs_state( struct pipe_context *pipe, void *prog ) { - struct brw_context *brw = brw_context(pipe); struct brw_fragment_shader *fs = (struct brw_fragment_shader *)prog; - brw->sws->bo_unreference(fs->const_buffer); + bo_reference(&fs->const_buffer, NULL); FREE( (void *)fs->tokens ); FREE( fs ); } diff --git a/src/gallium/drivers/i965/brw_pipe_vertex.c b/src/gallium/drivers/i965/brw_pipe_vertex.c index 97e9a23688..73bba5b088 100644 --- a/src/gallium/drivers/i965/brw_pipe_vertex.c +++ b/src/gallium/drivers/i965/brw_pipe_vertex.c @@ -56,7 +56,7 @@ brw_pipe_vertex_cleanup( struct brw_context *brw ) */ #if 0 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { - brw->sws->bo_unreference(brw->vb.inputs[i].bo); + bo_reference(&brw->vb.inputs[i].bo, NULL); brw->vb.inputs[i].bo = NULL; } #endif diff --git a/src/gallium/drivers/i965/brw_screen_buffers.c b/src/gallium/drivers/i965/brw_screen_buffers.c index ba54740225..7ae386ffb3 100644 --- a/src/gallium/drivers/i965/brw_screen_buffers.c +++ b/src/gallium/drivers/i965/brw_screen_buffers.c @@ -43,15 +43,11 @@ brw_buffer_unmap( struct pipe_screen *screen, static void brw_buffer_destroy( struct pipe_buffer *buffer ) { - struct brw_screen *bscreen = brw_screen( buffer->screen ); - struct brw_winsys_screen *sws = bscreen->sws; struct brw_buffer *buf = brw_buffer( buffer ); assert(!p_atomic_read(&buffer->reference.count)); - if (buf->bo) - sws->bo_unreference(buf->bo); - + bo_reference(&buf->bo, NULL); FREE(buf); } @@ -66,6 +62,7 @@ brw_buffer_create(struct pipe_screen *screen, struct brw_winsys_screen *sws = bscreen->sws; struct brw_buffer *buf; unsigned buffer_type; + enum pipe_error ret; buf = CALLOC_STRUCT(brw_buffer); if (!buf) @@ -101,10 +98,11 @@ brw_buffer_create(struct pipe_screen *screen, break; } - buf->bo = sws->bo_alloc( sws, - buffer_type, - size, - alignment ); + ret = sws->bo_alloc( sws, buffer_type, + size, alignment, + &buf->bo ); + if (ret != PIPE_OK) + return NULL; return &buf->base; } diff --git a/src/gallium/drivers/i965/brw_screen_surface.c b/src/gallium/drivers/i965/brw_screen_surface.c index 1c408e9f2e..21a7382873 100644 --- a/src/gallium/drivers/i965/brw_screen_surface.c +++ b/src/gallium/drivers/i965/brw_screen_surface.c @@ -150,9 +150,7 @@ static struct brw_surface *create_in_place_view( struct brw_screen *brw_screen, surface->pitch = tex->pitch; surface->tiling = tex->tiling; - surface->bo = tex->bo; - brw_screen->sws->bo_reference(surface->bo); - + bo_reference( &surface->bo, tex->bo ); pipe_texture_reference( &surface->base.texture, &tex->base ); surface->ss.ss0.surface_format = tex->ss.ss0.surface_format; @@ -244,11 +242,10 @@ static struct pipe_surface *brw_get_tex_surface(struct pipe_screen *screen, static void brw_tex_surface_destroy( struct pipe_surface *surf ) { struct brw_surface *surface = brw_surface(surf); - struct brw_screen *screen = brw_screen(surf->texture->screen); /* Unreference texture, shared buffer: */ - screen->sws->bo_unreference(surface->bo); + bo_reference(&surface->bo, NULL); pipe_texture_reference( &surface->base.texture, NULL ); diff --git a/src/gallium/drivers/i965/brw_screen_texture.c b/src/gallium/drivers/i965/brw_screen_texture.c index ba6dc7dfde..355abf0b89 100644 --- a/src/gallium/drivers/i965/brw_screen_texture.c +++ b/src/gallium/drivers/i965/brw_screen_texture.c @@ -187,6 +187,7 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen, struct brw_screen *bscreen = brw_screen(screen); struct brw_texture *tex; enum brw_buffer_type buffer_type; + enum pipe_error ret; tex = CALLOC_STRUCT(brw_texture); if (tex == NULL) @@ -235,10 +236,13 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen, buffer_type = BRW_BUFFER_TYPE_TEXTURE; } - tex->bo = bscreen->sws->bo_alloc( bscreen->sws, - buffer_type, - tex->pitch * tex->total_height * tex->cpp, - 64 ); + ret = bscreen->sws->bo_alloc( bscreen->sws, + buffer_type, + tex->pitch * tex->total_height * tex->cpp, + 64, + &tex->bo ); + if (ret) + goto fail; tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW; tex->ss.ss0.surface_type = translate_tex_target(tex->base.target); @@ -289,7 +293,7 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen, return &tex->base; fail: - bscreen->sws->bo_unreference(tex->bo); + bo_reference(&tex->bo, NULL); FREE(tex); return NULL; } @@ -306,7 +310,8 @@ static struct pipe_texture *brw_texture_blanket(struct pipe_screen *screen, static void brw_texture_destroy(struct pipe_texture *pt) { - //bscreen->sws->bo_unreference(tex->bo); + struct brw_texture *tex = brw_texture(pt); + bo_reference(&tex->bo, NULL); FREE(pt); } diff --git a/src/gallium/drivers/i965/brw_sf.c b/src/gallium/drivers/i965/brw_sf.c index 013d839e37..24d1015bbd 100644 --- a/src/gallium/drivers/i965/brw_sf.c +++ b/src/gallium/drivers/i965/brw_sf.c @@ -40,9 +40,11 @@ #include "brw_sf.h" #include "brw_state.h" -static void compile_sf_prog( struct brw_context *brw, - struct brw_sf_prog_key *key ) +static enum pipe_error compile_sf_prog( struct brw_context *brw, + struct brw_sf_prog_key *key, + struct brw_winsys_buffer **bo_out ) { + enum pipe_error ret; struct brw_sf_compile c; const GLuint *program; GLuint program_size; @@ -87,28 +89,35 @@ static void compile_sf_prog( struct brw_context *brw, break; default: assert(0); - return; + return PIPE_ERROR_BAD_INPUT; } /* get the program */ - program = brw_get_program(&c.func, &program_size); + ret = brw_get_program(&c.func, &program, &program_size); + if (ret) + return ret; /* Upload */ - brw->sws->bo_unreference(brw->sf.prog_bo); - brw->sf.prog_bo = brw_upload_cache( &brw->cache, BRW_SF_PROG, - &c.key, sizeof(c.key), - NULL, 0, - program, program_size, - &c.prog_data, - &brw->sf.prog_data ); + ret = brw_upload_cache( &brw->cache, BRW_SF_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + &brw->sf.prog_data, + bo_out); + if (ret) + return ret; + + return PIPE_OK; } /* Calculate interpolants for triangle and line rasterization. */ -static int upload_sf_prog(struct brw_context *brw) +static enum pipe_error upload_sf_prog(struct brw_context *brw) { + enum pipe_error ret; struct brw_sf_prog_key key; memset(&key, 0, sizeof(key)); @@ -161,15 +170,18 @@ static int upload_sf_prog(struct brw_context *brw) PIPE_WINDING_CCW); } - brw->sws->bo_unreference(brw->sf.prog_bo); - brw->sf.prog_bo = brw_search_cache(&brw->cache, BRW_SF_PROG, - &key, sizeof(key), - NULL, 0, - &brw->sf.prog_data); - if (brw->sf.prog_bo == NULL) - compile_sf_prog( brw, &key ); + if (brw_search_cache(&brw->cache, BRW_SF_PROG, + &key, sizeof(key), + NULL, 0, + &brw->sf.prog_data, + &brw->sf.prog_bo)) + return PIPE_OK; - return 0; + ret = compile_sf_prog( brw, &key, &brw->sf.prog_bo ); + if (ret) + return ret; + + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_sf_state.c b/src/gallium/drivers/i965/brw_sf_state.c index 31343ff245..f030f26c19 100644 --- a/src/gallium/drivers/i965/brw_sf_state.c +++ b/src/gallium/drivers/i965/brw_sf_state.c @@ -39,11 +39,12 @@ #include "brw_debug.h" #include "brw_pipe_rast.h" -static int upload_sf_vp(struct brw_context *brw) +static enum pipe_error upload_sf_vp(struct brw_context *brw) { const struct pipe_viewport_state *vp = &brw->curr.vp; const struct pipe_scissor_state *scissor = &brw->curr.scissor; struct brw_sf_viewport sfv; + enum pipe_error ret; memset(&sfv, 0, sizeof(sfv)); @@ -61,10 +62,12 @@ static int upload_sf_vp(struct brw_context *brw) sfv.scissor.ymin = scissor->miny; sfv.scissor.ymax = scissor->maxy; /* -1 ?? */ - brw->sws->bo_unreference(brw->sf.vp_bo); - brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 ); + ret = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0, + &brw->sf.vp_bo ); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_sf_vp = { @@ -128,12 +131,13 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) rast->point_size_max); } -static struct brw_winsys_buffer * +static enum pipe_error sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, - struct brw_winsys_buffer **reloc_bufs) + struct brw_winsys_buffer **reloc_bufs, + struct brw_winsys_buffer **bo_out) { struct brw_sf_unit_state sf; - struct brw_winsys_buffer *bo; + enum pipe_error ret; int chipset_max_threads; memset(&sf, 0, sizeof(sf)); @@ -273,51 +277,65 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.sf6.dest_org_hbias = 0x0; } - bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT, - key, sizeof(*key), - reloc_bufs, 2, - &sf, sizeof(sf), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_SF_UNIT, + key, sizeof(*key), + reloc_bufs, 2, + &sf, sizeof(sf), + NULL, NULL, + bo_out); + if (ret) + return ret; /* STATE_PREFETCH command description describes this state as being * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain. */ /* Emit SF program relocation */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - sf.thread0.grf_reg_count << 1, - offsetof(struct brw_sf_unit_state, thread0), - brw->sf.prog_bo); + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + sf.thread0.grf_reg_count << 1, + offsetof(struct brw_sf_unit_state, thread0), + brw->sf.prog_bo); + if (ret) + return ret; - /* Emit SF viewport relocation */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - sf.sf5.front_winding | (sf.sf5.viewport_transform << 1), - offsetof(struct brw_sf_unit_state, sf5), - brw->sf.vp_bo); - return bo; + /* Emit SF viewport relocation */ + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + sf.sf5.front_winding | (sf.sf5.viewport_transform << 1), + offsetof(struct brw_sf_unit_state, sf5), + brw->sf.vp_bo); + if (ret) + return ret; + + return PIPE_OK; } -static int upload_sf_unit( struct brw_context *brw ) +static enum pipe_error upload_sf_unit( struct brw_context *brw ) { struct brw_sf_unit_key key; struct brw_winsys_buffer *reloc_bufs[2]; + enum pipe_error ret; sf_unit_populate_key(brw, &key); reloc_bufs[0] = brw->sf.prog_bo; reloc_bufs[1] = brw->sf.vp_bo; - brw->sws->bo_unreference(brw->sf.state_bo); - brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT, - &key, sizeof(key), - reloc_bufs, 2, - NULL); - if (brw->sf.state_bo == NULL) { - brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs); - } - return 0; + if (brw_search_cache(&brw->cache, BRW_SF_UNIT, + &key, sizeof(key), + reloc_bufs, 2, + NULL, + &brw->sf.state_bo)) + return PIPE_OK; + + + ret = sf_unit_create_from_key(brw, &key, reloc_bufs, + &brw->sf.state_bo); + if (ret) + return ret; + + return PIPE_OK; } const struct brw_tracked_state brw_sf_unit = { diff --git a/src/gallium/drivers/i965/brw_state.h b/src/gallium/drivers/i965/brw_state.h index 94d2cb6f82..e219a1d870 100644 --- a/src/gallium/drivers/i965/brw_state.h +++ b/src/gallium/drivers/i965/brw_state.h @@ -44,8 +44,8 @@ brw_add_validated_bo(struct brw_context *brw, struct brw_winsys_buffer *bo) assert(brw->state.validated_bo_count < Elements(brw->state.validated_bos)); if (bo != NULL) { - brw->sws->bo_reference(bo); - brw->state.validated_bos[brw->state.validated_bo_count++] = bo; + bo_reference( &brw->state.validated_bos[brw->state.validated_bo_count++], + bo ); } } @@ -106,37 +106,42 @@ void brw_destroy_state(struct brw_context *brw); /*********************************************************************** * brw_state_cache.c */ -struct brw_winsys_buffer *brw_cache_data(struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *data, - struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs); - -struct brw_winsys_buffer *brw_cache_data_sz(struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *data, - GLuint data_size, - struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs); - -struct brw_winsys_buffer *brw_upload_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_sz, - struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs, - const void *data, - GLuint data_sz, - const void *aux, - void *aux_return ); - -struct brw_winsys_buffer *brw_search_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_size, - struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs, - void *aux_return); +enum pipe_error brw_cache_data(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *data, + struct brw_winsys_buffer **reloc_bufs, + GLuint nr_reloc_bufs, + struct brw_winsys_buffer **bo_out ); + +enum pipe_error brw_cache_data_sz(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *data, + GLuint data_size, + struct brw_winsys_buffer **reloc_bufs, + GLuint nr_reloc_bufs, + struct brw_winsys_buffer **bo_out); + +enum pipe_error brw_upload_cache( struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_sz, + struct brw_winsys_buffer **reloc_bufs, + GLuint nr_reloc_bufs, + const void *data, + GLuint data_sz, + const void *aux, + void *aux_return , + struct brw_winsys_buffer **bo_out); + +boolean brw_search_cache( struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_size, + struct brw_winsys_buffer **reloc_bufs, + GLuint nr_reloc_bufs, + void *aux_return, + struct brw_winsys_buffer **bo_out); + void brw_state_cache_check_size( struct brw_context *brw ); void brw_init_caches( struct brw_context *brw ); diff --git a/src/gallium/drivers/i965/brw_state_cache.c b/src/gallium/drivers/i965/brw_state_cache.c index cbd1f02d77..f8369d31ec 100644 --- a/src/gallium/drivers/i965/brw_state_cache.c +++ b/src/gallium/drivers/i965/brw_state_cache.c @@ -109,9 +109,8 @@ update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id, if (bo == cache->last_bo[cache_id]) return; /* no change */ - cache->sws->bo_unreference(cache->last_bo[cache_id]); - cache->last_bo[cache_id] = bo; - cache->sws->bo_reference(cache->last_bo[cache_id]); + bo_reference( &cache->last_bo[cache_id], bo ); + cache->brw->state.dirty.cache |= 1 << cache_id; } @@ -174,14 +173,15 @@ rehash(struct brw_cache *cache) /** * Returns the buffer object matching cache_id and key, or NULL. */ -struct brw_winsys_buffer * +boolean brw_search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, const void *key, GLuint key_size, struct brw_winsys_buffer **reloc_bufs, GLuint nr_reloc_bufs, - void *aux_return) + void *aux_return, + struct brw_winsys_buffer **bo_out) { struct brw_cache_item *item; GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); @@ -189,20 +189,20 @@ brw_search_cache(struct brw_cache *cache, item = search_cache(cache, cache_id, hash, key, key_size, reloc_bufs, nr_reloc_bufs); - if (item == NULL) - return NULL; - - if (aux_return) - *(void **)aux_return = (void *)((char *)item->key + item->key_size); - - update_cache_last(cache, cache_id, item->bo); - - cache->sws->bo_reference(item->bo); - return item->bo; + if (item) { + if (aux_return) + *(void **)aux_return = (void *)((char *)item->key + item->key_size); + + update_cache_last(cache, cache_id, item->bo); + bo_reference(bo_out, item->bo); + return TRUE; + } + + return FALSE; } -struct brw_winsys_buffer * +enum pipe_error brw_upload_cache( struct brw_cache *cache, enum brw_cache_id cache_id, const void *key, @@ -212,14 +212,15 @@ brw_upload_cache( struct brw_cache *cache, const void *data, GLuint data_size, const void *aux, - void *aux_return ) + void *aux_return, + struct brw_winsys_buffer **bo_out) { struct brw_cache_item *item = CALLOC_STRUCT(brw_cache_item); GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); GLuint relocs_size = nr_reloc_bufs * sizeof(struct brw_winsys_buffer *); GLuint aux_size = cache->aux_size[cache_id]; + enum pipe_error ret; void *tmp; - struct brw_winsys_buffer *bo; int i; /* Create the buffer object to contain the data. For now, use a @@ -227,9 +228,12 @@ brw_upload_cache( struct brw_cache *cache, * may want to take advantage of hardware distinctions between * these various entities. */ - bo = cache->sws->bo_alloc(cache->sws, - cache->buffer_type, - data_size, 1 << 6); + ret = cache->sws->bo_alloc(cache->sws, + cache->buffer_type, + data_size, 1 << 6, + bo_out); + if (ret) + return ret; /* Set up the memory containing the key, aux_data, and reloc_bufs */ @@ -240,7 +244,7 @@ brw_upload_cache( struct brw_cache *cache, memcpy((char *)tmp + key_size + aux_size, reloc_bufs, relocs_size); for (i = 0; i < nr_reloc_bufs; i++) { if (reloc_bufs[i] != NULL) - cache->sws->bo_reference(reloc_bufs[i]); + p_atomic_inc(&reloc_bufs[i]->reference.count); } item->cache_id = cache_id; @@ -249,9 +253,7 @@ brw_upload_cache( struct brw_cache *cache, item->key_size = key_size; item->reloc_bufs = (struct brw_winsys_buffer **)((char *)tmp + key_size + aux_size); item->nr_reloc_bufs = nr_reloc_bufs; - - item->bo = bo; - cache->sws->bo_reference(bo); + bo_reference( &item->bo, *bo_out ); item->data_size = data_size; if (cache->n_items > cache->size * 1.5) @@ -273,28 +275,28 @@ brw_upload_cache( struct brw_cache *cache, data_size, cache_id); /* Copy data to the buffer */ - cache->sws->bo_subdata(bo, + cache->sws->bo_subdata(item->bo, cache_id, 0, data_size, data); - update_cache_last(cache, cache_id, bo); + update_cache_last(cache, cache_id, item->bo); - return bo; + return PIPE_OK; } /** * This doesn't really work with aux data. Use search/upload instead */ -struct brw_winsys_buffer * +enum pipe_error brw_cache_data_sz(struct brw_cache *cache, enum brw_cache_id cache_id, const void *data, GLuint data_size, struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs) + GLuint nr_reloc_bufs, + struct brw_winsys_buffer **bo_out) { - struct brw_winsys_buffer *bo; struct brw_cache_item *item; GLuint hash = hash_key(data, data_size, reloc_bufs, nr_reloc_bufs); @@ -302,17 +304,17 @@ brw_cache_data_sz(struct brw_cache *cache, reloc_bufs, nr_reloc_bufs); if (item) { update_cache_last(cache, cache_id, item->bo); - cache->sws->bo_reference(item->bo); - return item->bo; - } - bo = brw_upload_cache(cache, cache_id, - data, data_size, - reloc_bufs, nr_reloc_bufs, - data, data_size, - NULL, NULL); + bo_reference(bo_out, item->bo); + return PIPE_OK; + } - return bo; + return brw_upload_cache(cache, cache_id, + data, data_size, + reloc_bufs, nr_reloc_bufs, + data, data_size, + NULL, NULL, + bo_out); } @@ -323,15 +325,16 @@ brw_cache_data_sz(struct brw_cache *cache, * better to use, as the potentially changing offsets in the data-used-as-key * will result in excessive cache misses. */ -struct brw_winsys_buffer * +enum pipe_error brw_cache_data(struct brw_cache *cache, enum brw_cache_id cache_id, const void *data, struct brw_winsys_buffer **reloc_bufs, - GLuint nr_reloc_bufs) + GLuint nr_reloc_bufs, + struct brw_winsys_buffer **bo_out) { return brw_cache_data_sz(cache, cache_id, data, cache->key_size[cache_id], - reloc_bufs, nr_reloc_bufs); + reloc_bufs, nr_reloc_bufs, bo_out); } @@ -506,11 +509,13 @@ brw_clear_cache(struct brw_context *brw, struct brw_cache *cache) int j; next = c->next; + for (j = 0; j < c->nr_reloc_bufs; j++) - brw->sws->bo_unreference(c->reloc_bufs[j]); - brw->sws->bo_unreference(c->bo); - free((void *)c->key); - free(c); + bo_reference(&c->reloc_bufs[j], NULL); + + bo_reference(&c->bo, NULL); + FREE((void *)c->key); + FREE(c); } cache->items[i] = NULL; } @@ -551,10 +556,12 @@ brw_state_cache_bo_delete(struct brw_cache *cache, struct brw_winsys_buffer *bo) *prev = c->next; for (j = 0; j < c->nr_reloc_bufs; j++) - cache->sws->bo_unreference(c->reloc_bufs[j]); - cache->sws->bo_unreference(c->bo); - free((void *)c->key); - free(c); + bo_reference(&c->reloc_bufs[j], NULL); + + bo_reference(&c->bo, NULL); + + FREE((void *)c->key); + FREE(c); cache->n_items--; } else { prev = &c->next; @@ -590,10 +597,10 @@ brw_destroy_cache(struct brw_context *brw, struct brw_cache *cache) brw_clear_cache(brw, cache); for (i = 0; i < BRW_MAX_CACHE; i++) { - brw->sws->bo_unreference(cache->last_bo[i]); - free(cache->name[i]); + bo_reference(&cache->last_bo[i], NULL); + FREE(cache->name[i]); } - free(cache->items); + FREE(cache->items); cache->items = NULL; cache->size = 0; } diff --git a/src/gallium/drivers/i965/brw_state_upload.c b/src/gallium/drivers/i965/brw_state_upload.c index a71af4d2b9..fdcdd59129 100644 --- a/src/gallium/drivers/i965/brw_state_upload.c +++ b/src/gallium/drivers/i965/brw_state_upload.c @@ -140,8 +140,7 @@ brw_clear_validated_bos(struct brw_context *brw) /* Clear the last round of validated bos */ for (i = 0; i < brw->state.validated_bo_count; i++) { - brw->sws->bo_unreference(brw->state.validated_bos[i]); - brw->state.validated_bos[i] = NULL; + bo_reference(&brw->state.validated_bos[i], NULL); } brw->state.validated_bo_count = 0; } diff --git a/src/gallium/drivers/i965/brw_vs.c b/src/gallium/drivers/i965/brw_vs.c index 26a28114d9..966940ceac 100644 --- a/src/gallium/drivers/i965/brw_vs.c +++ b/src/gallium/drivers/i965/brw_vs.c @@ -39,10 +39,12 @@ -static void do_vs_prog( struct brw_context *brw, - struct brw_vertex_shader *vp, - struct brw_vs_prog_key *key ) +static enum pipe_error do_vs_prog( struct brw_context *brw, + struct brw_vertex_shader *vp, + struct brw_vs_prog_key *key, + struct brw_winsys_buffer **bo_out) { + enum pipe_error ret; GLuint program_size; const GLuint *program; struct brw_vs_compile c; @@ -66,22 +68,29 @@ static void do_vs_prog( struct brw_context *brw, /* get the program */ - program = brw_get_program(&c.func, &program_size); - - brw->sws->bo_unreference(brw->vs.prog_bo); - brw->vs.prog_bo = brw_upload_cache( &brw->cache, BRW_VS_PROG, - &c.key, sizeof(c.key), - NULL, 0, - program, program_size, - &c.prog_data, - &brw->vs.prog_data ); + ret = brw_get_program(&c.func, &program, &program_size); + if (ret) + return ret; + + ret = brw_upload_cache( &brw->cache, BRW_VS_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + &brw->vs.prog_data, + bo_out); + if (ret) + return ret; + + return PIPE_OK; } -static int brw_upload_vs_prog(struct brw_context *brw) +static enum pipe_error brw_upload_vs_prog(struct brw_context *brw) { struct brw_vs_prog_key key; struct brw_vertex_shader *vp = brw->curr.vertex_shader; + enum pipe_error ret; memset(&key, 0, sizeof(key)); @@ -95,15 +104,18 @@ static int brw_upload_vs_prog(struct brw_context *brw) /* Make an early check for the key. */ - brw->sws->bo_unreference(brw->vs.prog_bo); - brw->vs.prog_bo = brw_search_cache(&brw->cache, BRW_VS_PROG, - &key, sizeof(key), - NULL, 0, - &brw->vs.prog_data); - if (brw->vs.prog_bo == NULL) - do_vs_prog(brw, vp, &key); - - return 0; + if (brw_search_cache(&brw->cache, BRW_VS_PROG, + &key, sizeof(key), + NULL, 0, + &brw->vs.prog_data, + &brw->vs.prog_bo)) + return PIPE_OK; + + ret = do_vs_prog(brw, vp, &key, &brw->vs.prog_bo); + if (ret) + return ret; + + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_vs_state.c b/src/gallium/drivers/i965/brw_vs_state.c index 26d5d005fa..22a4d7f01b 100644 --- a/src/gallium/drivers/i965/brw_vs_state.c +++ b/src/gallium/drivers/i965/brw_vs_state.c @@ -78,11 +78,13 @@ vs_unit_populate_key(struct brw_context *brw, struct brw_vs_unit_key *key) } } -static struct brw_winsys_buffer * -vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) +static enum pipe_error +vs_unit_create_from_key(struct brw_context *brw, + struct brw_vs_unit_key *key, + struct brw_winsys_buffer **bo_out) { + enum pipe_error ret; struct brw_vs_unit_state vs; - struct brw_winsys_buffer *bo; int chipset_max_threads; memset(&vs, 0, sizeof(vs)); @@ -141,38 +143,46 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) */ vs.vs6.vs_enable = 1; - bo = brw_upload_cache(&brw->cache, BRW_VS_UNIT, - key, sizeof(*key), - &brw->vs.prog_bo, 1, - &vs, sizeof(vs), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_VS_UNIT, + key, sizeof(*key), + &brw->vs.prog_bo, 1, + &vs, sizeof(vs), + NULL, NULL, + bo_out); + if (ret) + return ret; /* Emit VS program relocation */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - vs.thread0.grf_reg_count << 1, - offsetof(struct brw_vs_unit_state, thread0), - brw->vs.prog_bo); - - return bo; + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + vs.thread0.grf_reg_count << 1, + offsetof(struct brw_vs_unit_state, thread0), + brw->vs.prog_bo); + if (ret) + return ret; + + return PIPE_OK; } static int prepare_vs_unit(struct brw_context *brw) { struct brw_vs_unit_key key; + enum pipe_error ret; vs_unit_populate_key(brw, &key); - brw->sws->bo_unreference(brw->vs.state_bo); - brw->vs.state_bo = brw_search_cache(&brw->cache, BRW_VS_UNIT, - &key, sizeof(key), - &brw->vs.prog_bo, 1, - NULL); - if (brw->vs.state_bo == NULL) { - brw->vs.state_bo = vs_unit_create_from_key(brw, &key); - } + if (brw_search_cache(&brw->cache, BRW_VS_UNIT, + &key, sizeof(key), + &brw->vs.prog_bo, 1, + NULL, + &brw->vs.state_bo)) + return PIPE_OK; + + ret = vs_unit_create_from_key(brw, &key, &brw->vs.state_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_vs_unit = { diff --git a/src/gallium/drivers/i965/brw_vs_surface_state.c b/src/gallium/drivers/i965/brw_vs_surface_state.c index 32fb9b2a8b..b12df0ec03 100644 --- a/src/gallium/drivers/i965/brw_vs_surface_state.c +++ b/src/gallium/drivers/i965/brw_vs_surface_state.c @@ -83,22 +83,23 @@ brw_update_vs_constant_surface( struct brw_context *brw, { struct brw_surface_key key; struct pipe_buffer *cb = brw->curr.vs_constants; + enum pipe_error ret; assert(surf == 0); /* If we're in this state update atom, we need to update VS constants, so * free the old buffer and create a new one for the new contents. */ - brw->sws->bo_unreference(vp->const_buffer); - vp->const_buffer = brw_vs_update_constant_buffer(brw); + ret = brw_vs_update_constant_buffer(brw, &vp->const_buffer); + if (ret) + return ret; /* If there's no constant buffer, then no surface BO is needed to point at * it. */ - if (vp->const_buffer == 0) { - drm_intel_bo_unreference(brw->vs.surf_bo[surf]); - brw->vs.surf_bo[surf] = NULL; - return; + if (vp->const_buffer == NULL) { + bo_reference(brw->vs.surf_bo[surf], NULL); + return PIPE_OK; } memset(&key, 0, sizeof(key)); @@ -118,15 +119,20 @@ brw_update_vs_constant_surface( struct brw_context *brw, key.width, key.height, key.depth, key.cpp, key.pitch); */ - drm_intel_bo_unreference(brw->vs.surf_bo[surf]); - brw->vs.surf_bo[surf] = brw_search_cache(&brw->surface_cache, - BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, key.bo ? 1 : 0, - NULL); - if (brw->vs.surf_bo[surf] == NULL) { - brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key); - } + if (brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL, + &brw->vs.surf_bo[surf])) + return PIPE_OK; + + ret = brw_create_constant_surface(brw, &key + &brw->vs.surf_bo[surf]); + if (ret) + return ret; + + return PIPE_OK; } #endif @@ -134,18 +140,20 @@ brw_update_vs_constant_surface( struct brw_context *brw, /** * Constructs the binding table for the VS surface state. */ -static struct brw_winsys_buffer * -brw_vs_get_binding_table(struct brw_context *brw) +static enum pipe_error +brw_vs_get_binding_table(struct brw_context *brw, + struct brw_winsys_buffer **bo_out) { #if 0 - struct brw_winsys_buffer *bind_bo; - - bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->vs.surf_bo, BRW_VS_MAX_SURF, - NULL); - - if (bind_bo == NULL) { + if (brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + NULL, + bo_out)) + { + return PIPE_OK; + } + else { GLuint data_size = BRW_VS_MAX_SURF * sizeof(GLuint); uint32_t *data = malloc(data_size); int i; @@ -156,11 +164,14 @@ brw_vs_get_binding_table(struct brw_context *brw) else data[i] = 0; - bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->vs.surf_bo, BRW_VS_MAX_SURF, - data, data_size, - NULL, NULL); + ret = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + data, data_size, + NULL, NULL, + bo_out); + if (ret) + return ret; /* Emit binding table relocations to surface state */ for (i = 0; i < BRW_VS_MAX_SURF; i++) { @@ -168,18 +179,19 @@ brw_vs_get_binding_table(struct brw_context *brw) /* The presumed offsets were set in the data values for * brw_upload_cache. */ - drm_intel_bo_emit_reloc(bind_bo, i * 4, - brw->vs.surf_bo[i], 0, - BRW_USAGE_STATE); + ret = sws->bo_emit_reloc(*bo_out, i * 4, + brw->vs.surf_bo[i], 0, + BRW_USAGE_STATE); + if (ret) + return ret; } } - free(data); + FREE(data); + return PIPE_OK; } - - return bind_bo; #else - return NULL; + return PIPE_OK; #endif } @@ -190,8 +202,10 @@ brw_vs_get_binding_table(struct brw_context *brw) * to be updated, and produces BRW_NEW_NR_VS_SURFACES for the VS unit and * CACHE_NEW_SURF_BIND for the binding table upload. */ -static int prepare_vs_surfaces(struct brw_context *brw ) +static enum pipe_error prepare_vs_surfaces(struct brw_context *brw ) { + enum pipe_error ret; + #if 0 int i; int nr_surfaces = 0; @@ -215,11 +229,12 @@ static int prepare_vs_surfaces(struct brw_context *brw ) * just slightly increases our working set size. */ if (brw->vs.nr_surfaces != 0) { - brw->sws->bo_unreference(brw->vs.bind_bo); - brw->vs.bind_bo = brw_vs_get_binding_table(brw); + ret = brw_vs_get_binding_table(brw, &brw->vs.bind_bo); + if (ret) + return ret; } - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_vs_surfaces = { diff --git a/src/gallium/drivers/i965/brw_winsys.h b/src/gallium/drivers/i965/brw_winsys.h index d941fbcebe..f61c541ad1 100644 --- a/src/gallium/drivers/i965/brw_winsys.h +++ b/src/gallium/drivers/i965/brw_winsys.h @@ -28,6 +28,7 @@ #include "pipe/p_compiler.h" #include "pipe/p_error.h" +#include "pipe/p_refcnt.h" struct brw_winsys; struct pipe_fence_handle; @@ -36,10 +37,13 @@ struct pipe_fence_handle; */ #define BRW_BATCH_SIZE (32*1024) +struct brw_winsys_screen; /* Need a tiny bit of information inside the abstract buffer struct: */ struct brw_winsys_buffer { + struct pipe_reference reference; + struct brw_winsys_screen *sws; unsigned *offset; unsigned size; }; @@ -105,6 +109,10 @@ enum brw_buffer_data_type { BRW_DATA_MAX }; + + + + struct brw_winsys_screen { @@ -116,33 +124,33 @@ struct brw_winsys_screen { /** * Create a buffer. */ - struct brw_winsys_buffer *(*bo_alloc)( struct brw_winsys_screen *sws, - enum brw_buffer_type type, - unsigned size, - unsigned alignment ); + enum pipe_error (*bo_alloc)( struct brw_winsys_screen *sws, + enum brw_buffer_type type, + unsigned size, + unsigned alignment, + struct brw_winsys_buffer **bo_out ); - /* Reference and unreference buffers: + /* Destroy a buffer when our refcount goes to zero: */ - void (*bo_reference)( struct brw_winsys_buffer *buffer ); - void (*bo_unreference)( struct brw_winsys_buffer *buffer ); + void (*bo_destroy)( struct brw_winsys_buffer *buffer ); /* delta -- added to b2->offset, and written into buffer * offset -- location above value is written to within buffer */ - int (*bo_emit_reloc)( struct brw_winsys_buffer *buffer, - enum brw_buffer_usage usage, - unsigned delta, - unsigned offset, - struct brw_winsys_buffer *b2); + enum pipe_error (*bo_emit_reloc)( struct brw_winsys_buffer *buffer, + enum brw_buffer_usage usage, + unsigned delta, + unsigned offset, + struct brw_winsys_buffer *b2); - int (*bo_exec)( struct brw_winsys_buffer *buffer, - unsigned bytes_used ); + enum pipe_error (*bo_exec)( struct brw_winsys_buffer *buffer, + unsigned bytes_used ); - int (*bo_subdata)(struct brw_winsys_buffer *buffer, - enum brw_buffer_data_type data_type, - size_t offset, - size_t size, - const void *data); + enum pipe_error (*bo_subdata)(struct brw_winsys_buffer *buffer, + enum brw_buffer_data_type data_type, + size_t offset, + size_t size, + const void *data); boolean (*bo_is_busy)(struct brw_winsys_buffer *buffer); boolean (*bo_references)(struct brw_winsys_buffer *a, @@ -175,6 +183,16 @@ struct brw_winsys_screen { }; +static INLINE void +bo_reference(struct brw_winsys_buffer **ptr, struct brw_winsys_buffer *buf) +{ + struct brw_winsys_buffer *old_buf = *ptr; + + if (pipe_reference((struct pipe_reference **)ptr, &buf->reference)) + old_buf->sws->bo_destroy(old_buf); +} + + /** * Create brw pipe_screen. */ diff --git a/src/gallium/drivers/i965/brw_wm.c b/src/gallium/drivers/i965/brw_wm.c index 815ae8c51a..93f90bf329 100644 --- a/src/gallium/drivers/i965/brw_wm.c +++ b/src/gallium/drivers/i965/brw_wm.c @@ -137,30 +137,26 @@ brw_wm_linear_shader_emit(struct brw_context *brw, struct brw_wm_compile *c) * Depending on the instructions used (i.e. flow control instructions) * we'll use one of two code generators. */ -static int do_wm_prog( struct brw_context *brw, - struct brw_fragment_shader *fp, - struct brw_wm_prog_key *key) +static enum pipe_error do_wm_prog( struct brw_context *brw, + struct brw_fragment_shader *fp, + struct brw_wm_prog_key *key, + struct brw_winsys_buffer **bo_out) { + enum pipe_error ret; struct brw_wm_compile *c; const GLuint *program; GLuint program_size; - c = brw->wm.compile_data; - if (c == NULL) { - brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data)); - c = brw->wm.compile_data; - if (c == NULL) { - /* Ouch - big out of memory problem. Can't continue - * without triggering a segfault, no way to signal, - * so just return. - */ + if (brw->wm.compile_data == NULL) { + brw->wm.compile_data = MALLOC(sizeof(*brw->wm.compile_data)); + if (!brw->wm.compile_data) return PIPE_ERROR_OUT_OF_MEMORY; - } - } else { - memset(c, 0, sizeof(*brw->wm.compile_data)); } - memcpy(&c->key, key, sizeof(*key)); + c = brw->wm.compile_data; + memset(c, 0, sizeof *c); + + c->key = *key; c->fp = fp; c->env_param = NULL; /*brw->intel.ctx.FragmentProgram.Parameters;*/ @@ -190,17 +186,21 @@ static int do_wm_prog( struct brw_context *brw, /* get the program */ - program = brw_get_program(&c->func, &program_size); - - brw->sws->bo_unreference(brw->wm.prog_bo); - brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG, - &c->key, sizeof(c->key), - NULL, 0, - program, program_size, - &c->prog_data, - &brw->wm.prog_data ); - - return 0; + ret = brw_get_program(&c->func, &program, &program_size); + if (ret) + return ret; + + ret = brw_upload_cache( &brw->cache, BRW_WM_PROG, + &c->key, sizeof(c->key), + NULL, 0, + program, program_size, + &c->prog_data, + &brw->wm.prog_data, + bo_out ); + if (ret) + return ret; + + return PIPE_OK; } @@ -267,24 +267,28 @@ static void brw_wm_populate_key( struct brw_context *brw, } -static int brw_prepare_wm_prog(struct brw_context *brw) +static enum pipe_error brw_prepare_wm_prog(struct brw_context *brw) { struct brw_wm_prog_key key; struct brw_fragment_shader *fs = brw->curr.fragment_shader; + enum pipe_error ret; brw_wm_populate_key(brw, &key); /* Make an early check for the key. */ - brw->sws->bo_unreference(brw->wm.prog_bo); - brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG, - &key, sizeof(key), - NULL, 0, - &brw->wm.prog_data); - if (brw->wm.prog_bo == NULL) - return do_wm_prog(brw, fs, &key); - - return 0; + if (brw_search_cache(&brw->cache, BRW_WM_PROG, + &key, sizeof(key), + NULL, 0, + &brw->wm.prog_data, + &brw->wm.prog_bo)) + return PIPE_OK; + + ret = do_wm_prog(brw, fs, &key, &brw->wm.prog_bo); + if (ret) + return ret; + + return PIPE_OK; } diff --git a/src/gallium/drivers/i965/brw_wm_constant_buffer.c b/src/gallium/drivers/i965/brw_wm_constant_buffer.c index 50ecef29a4..14568265dd 100644 --- a/src/gallium/drivers/i965/brw_wm_constant_buffer.c +++ b/src/gallium/drivers/i965/brw_wm_constant_buffer.c @@ -6,12 +6,14 @@ * Create the constant buffer surface. Vertex/fragment shader constants will be * read from this buffer with Data Port Read instructions/messages. */ -struct brw_winsys_buffer * +enum pipe_error brw_create_constant_surface( struct brw_context *brw, - struct brw_surface_key *key ) + struct brw_surface_key *key, + struct brw_winsys_buffer **bo_out ) { const GLint w = key->width - 1; struct brw_winsys_buffer *bo; + enum pipe_error ret; memset(&surf, 0, sizeof(surf)); @@ -28,22 +30,27 @@ brw_create_constant_surface( struct brw_context *brw, surf.ss3.pitch = (key->pitch * key->cpp) - 1; /* ignored?? */ brw_set_surface_tiling(&surf, key->tiling); /* tiling now allowed */ - bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, - key, sizeof(*key), - &key->bo, key->bo ? 1 : 0, - &surf, sizeof(surf), - NULL, NULL); + ret = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, + key, sizeof(*key), + &key->bo, key->bo ? 1 : 0, + &surf, sizeof(surf), + NULL, NULL, + &bo_out); + if (ret) + return ret; if (key->bo) { /* Emit relocation to surface contents */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_SAMPLER, - 0, - offsetof(struct brw_surface_state, ss1), - key->bo); + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_SAMPLER, + 0, + offsetof(struct brw_surface_state, ss1), + key->bo); + if (ret) + return ret; } - return bo; + return PIPE_OK; } @@ -52,7 +59,7 @@ brw_create_constant_surface( struct brw_context *brw, * Update the surface state for a WM constant buffer. * The constant buffer will be (re)allocated here if needed. */ -static void +static enum pipe_error brw_update_wm_constant_surface( struct brw_context *brw, GLuint surf) { @@ -60,20 +67,21 @@ brw_update_wm_constant_surface( struct brw_context *brw, struct brw_fragment_shader *fp = brw->curr.fragment_shader; struct pipe_buffer *cbuf = brw->curr.fragment_constants; int pitch = cbuf->size / (4 * sizeof(float)); + enum pipe_error ret; /* If we're in this state update atom, we need to update WM constants, so * free the old buffer and create a new one for the new contents. */ - brw->sws->bo_unreference(fp->const_buffer); - fp->const_buffer = brw_wm_update_constant_buffer(brw); + ret = brw_wm_update_constant_buffer(brw, &fp->const_buffer); + if (ret) + return ret; /* If there's no constant buffer, then no surface BO is needed to point at * it. */ if (cbuf == NULL) { - drm_intel_bo_unreference(brw->wm.surf_bo[surf]); - brw->wm.surf_bo[surf] = NULL; - return; + bo_reference(&brw->wm.surf_bo[surf], NULL); + return PIPE_OK; } memset(&key, 0, sizeof(key)); @@ -97,16 +105,20 @@ brw_update_wm_constant_surface( struct brw_context *brw, key.width, key.height, key.depth, key.cpp, key.pitch); */ - brw->sws->bo_unreference(brw->wm.surf_bo[surf]); - brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, - BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, 1, - NULL); - if (brw->wm.surf_bo[surf] == NULL) { - brw->wm.surf_bo[surf] = brw_create_constant_surface(brw, &key); - } + if (brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, 1, + NULL, + &brw->wm.surf_bo[surf])) + return PIPE_OK; + + ret = brw_create_constant_surface(brw, &key, &brw->wm.surf_bo[surf]); + if (ret) + return ret; + brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; + return PIPE_OK; } /** @@ -117,28 +129,33 @@ brw_update_wm_constant_surface( struct brw_context *brw, * BRW_NEW_WM_SURFACES to get picked up by brw_prepare_wm_surfaces for * inclusion in the binding table. */ -static void prepare_wm_constant_surface(struct brw_context *brw ) +static enum pipe_error prepare_wm_constant_surface(struct brw_context *brw ) { struct brw_fragment_program *fp = (struct brw_fragment_program *) brw->fragment_program; GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER; - drm_intel_bo_unreference(fp->const_buffer); - fp->const_buffer = brw_wm_update_constant_buffer(brw); + ret = brw_wm_update_constant_buffer(brw, + &fp->const_buffer); + if (ret) + return ret; /* If there's no constant buffer, then no surface BO is needed to point at * it. */ if (fp->const_buffer == 0) { if (brw->wm.surf_bo[surf] != NULL) { - drm_intel_bo_unreference(brw->wm.surf_bo[surf]); - brw->wm.surf_bo[surf] = NULL; + bo_reference(&brw->wm.surf_bo[surf], NULL); brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; } - return; + return PIPE_OK; } - brw_update_wm_constant_surface(ctx, surf); + ret = brw_update_wm_constant_surface(ctx, surf); + if (ret) + return ret; + + return PIPE_OK } const struct brw_tracked_state brw_wm_constant_surface = { diff --git a/src/gallium/drivers/i965/brw_wm_sampler_state.c b/src/gallium/drivers/i965/brw_wm_sampler_state.c index 2fddb4ad89..2861aa979f 100644 --- a/src/gallium/drivers/i965/brw_wm_sampler_state.c +++ b/src/gallium/drivers/i965/brw_wm_sampler_state.c @@ -43,16 +43,22 @@ -static struct brw_winsys_buffer * +static enum pipe_error upload_default_color( struct brw_context *brw, - const GLfloat *color ) + const GLfloat *color, + struct brw_winsys_buffer **bo_out ) { struct brw_sampler_default_color sdc; + enum pipe_error ret; COPY_4V(sdc.color, color); - return brw_cache_data( &brw->cache, BRW_SAMPLER_DEFAULT_COLOR, &sdc, - NULL, 0 ); + ret = brw_cache_data( &brw->cache, BRW_SAMPLER_DEFAULT_COLOR, &sdc, + NULL, 0, bo_out ); + if (ret) + return ret; + + return PIPE_OK; } @@ -111,9 +117,10 @@ brw_wm_sampler_populate_key(struct brw_context *brw, } -static void +static enum pipe_error brw_wm_sampler_update_default_colors(struct brw_context *brw) { + enum pipe_error ret; int nr = MIN2(brw->curr.num_textures, brw->curr.num_samplers); int i; @@ -121,8 +128,7 @@ brw_wm_sampler_update_default_colors(struct brw_context *brw) for (i = 0; i < nr; i++) { const struct brw_texture *tex = brw_texture(brw->curr.texture[i]); const struct brw_sampler *sampler = brw->curr.sampler[i]; - - brw->sws->bo_unreference(brw->wm.sdc_bo[i]); + const float *bc; if (pf_is_depth_or_stencil(tex->base.format)) { float bordercolor[4] = { @@ -131,15 +137,25 @@ brw_wm_sampler_update_default_colors(struct brw_context *brw) sampler->border_color[0], sampler->border_color[0] }; - /* GL specs that border color for depth textures is taken from the - * R channel, while the hardware uses A. Spam R into all the - * channels for safety. - */ - brw->wm.sdc_bo[i] = upload_default_color(brw, bordercolor); - } else { - brw->wm.sdc_bo[i] = upload_default_color(brw, sampler->border_color); + + bc = bordercolor; + } + else { + bc = sampler->border_color; } + + /* GL specs that border color for depth textures is taken from the + * R channel, while the hardware uses A. Spam R into all the + * channels for safety. + */ + ret = upload_default_color(brw, + bc, + &brw->wm.sdc_bo[i]); + if (ret) + return ret; } + + return PIPE_OK; } @@ -149,6 +165,7 @@ brw_wm_sampler_update_default_colors(struct brw_context *brw) static int upload_wm_samplers( struct brw_context *brw ) { struct wm_sampler_key key; + enum pipe_error ret; int i; brw_wm_sampler_update_default_colors(brw); @@ -159,35 +176,40 @@ static int upload_wm_samplers( struct brw_context *brw ) brw->state.dirty.cache |= CACHE_NEW_SAMPLER; } - brw->sws->bo_unreference(brw->wm.sampler_bo); - brw->wm.sampler_bo = NULL; - if (brw->wm.sampler_count == 0) - return 0; + if (brw->wm.sampler_count == 0) { + bo_reference(&brw->wm.sampler_bo, NULL); + return PIPE_OK; + } - brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER, - &key, sizeof(key), - brw->wm.sdc_bo, key.sampler_count, - NULL); + if (brw_search_cache(&brw->cache, BRW_SAMPLER, + &key, sizeof(key), + brw->wm.sdc_bo, key.sampler_count, + NULL, + &brw->wm.sampler_bo)) + return PIPE_OK; /* If we didnt find it in the cache, compute the state and put it in the * cache. */ - if (brw->wm.sampler_bo == NULL) { - brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER, - &key, sizeof(key), - brw->wm.sdc_bo, key.sampler_count, - &key.sampler, sizeof(key.sampler), - NULL, NULL); - - /* Emit SDC relocations */ - for (i = 0; i < key.sampler_count; i++) { - brw->sws->bo_emit_reloc(brw->wm.sampler_bo, - BRW_USAGE_SAMPLER, - 0, - i * sizeof(struct brw_sampler_state) + - offsetof(struct brw_sampler_state, ss2), - brw->wm.sdc_bo[i]); - } + ret = brw_upload_cache(&brw->cache, BRW_SAMPLER, + &key, sizeof(key), + brw->wm.sdc_bo, key.sampler_count, + &key.sampler, sizeof(key.sampler), + NULL, NULL, + &brw->wm.sampler_bo); + if (ret) + return ret; + + /* Emit SDC relocations */ + for (i = 0; i < key.sampler_count; i++) { + ret = brw->sws->bo_emit_reloc(brw->wm.sampler_bo, + BRW_USAGE_SAMPLER, + 0, + i * sizeof(struct brw_sampler_state) + + offsetof(struct brw_sampler_state, ss2), + brw->wm.sdc_bo[i]); + if (ret) + return ret; } return 0; diff --git a/src/gallium/drivers/i965/brw_wm_state.c b/src/gallium/drivers/i965/brw_wm_state.c index ccbb647bcd..86dc10540d 100644 --- a/src/gallium/drivers/i965/brw_wm_state.c +++ b/src/gallium/drivers/i965/brw_wm_state.c @@ -138,12 +138,13 @@ wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) /** * Setup wm hardware state. See page 225 of Volume 2 */ -static struct brw_winsys_buffer * +static enum pipe_error wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, - struct brw_winsys_buffer **reloc_bufs) + struct brw_winsys_buffer **reloc_bufs, + struct brw_winsys_buffer **bo_out) { struct brw_wm_unit_state wm; - struct brw_winsys_buffer *bo; + enum pipe_error ret; memset(&wm, 0, sizeof(wm)); @@ -222,45 +223,56 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, if (BRW_DEBUG & DEBUG_STATS || key->stats_wm) wm.wm4.stats_enable = 1; - bo = brw_upload_cache(&brw->cache, BRW_WM_UNIT, - key, sizeof(*key), - reloc_bufs, 3, - &wm, sizeof(wm), - NULL, NULL); + ret = brw_upload_cache(&brw->cache, BRW_WM_UNIT, + key, sizeof(*key), + reloc_bufs, 3, + &wm, sizeof(wm), + NULL, NULL, + bo_out); + if (ret) + return ret; /* Emit WM program relocation */ - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - wm.thread0.grf_reg_count << 1, - offsetof(struct brw_wm_unit_state, thread0), - brw->wm.prog_bo); + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + wm.thread0.grf_reg_count << 1, + offsetof(struct brw_wm_unit_state, thread0), + brw->wm.prog_bo); + if (ret) + return ret; /* Emit scratch space relocation */ if (key->total_scratch != 0) { - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_SCRATCH, - wm.thread2.per_thread_scratch_space, - offsetof(struct brw_wm_unit_state, thread2), - brw->wm.scratch_bo); + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_SCRATCH, + wm.thread2.per_thread_scratch_space, + offsetof(struct brw_wm_unit_state, thread2), + brw->wm.scratch_bo); + if (ret) + return ret; } /* Emit sampler state relocation */ if (key->sampler_count != 0) { - brw->sws->bo_emit_reloc(bo, - BRW_USAGE_STATE, - wm.wm4.stats_enable | (wm.wm4.sampler_count << 2), - offsetof(struct brw_wm_unit_state, wm4), - brw->wm.sampler_bo); + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + wm.wm4.stats_enable | (wm.wm4.sampler_count << 2), + offsetof(struct brw_wm_unit_state, wm4), + brw->wm.sampler_bo); + if (ret) + return ret; } - return bo; + return PIPE_OK; } -static int upload_wm_unit( struct brw_context *brw ) +static enum pipe_error upload_wm_unit( struct brw_context *brw ) { struct brw_wm_unit_key key; struct brw_winsys_buffer *reloc_bufs[3]; + enum pipe_error ret; + wm_unit_populate_key(brw, &key); /* Allocate the necessary scratch space if we haven't already. Don't @@ -271,15 +283,19 @@ static int upload_wm_unit( struct brw_context *brw ) if (key.total_scratch) { GLuint total = key.total_scratch * key.max_threads; - if (brw->wm.scratch_bo && total > brw->wm.scratch_bo->size) { - brw->sws->bo_unreference(brw->wm.scratch_bo); - brw->wm.scratch_bo = NULL; - } + /* Do we need a new buffer: + */ + if (brw->wm.scratch_bo && total > brw->wm.scratch_bo->size) + bo_reference(&brw->wm.scratch_bo, NULL); + if (brw->wm.scratch_bo == NULL) { - brw->wm.scratch_bo = brw->sws->bo_alloc(brw->sws, - BRW_BUFFER_TYPE_SHADER_SCRATCH, - total, - 4096); + ret = brw->sws->bo_alloc(brw->sws, + BRW_BUFFER_TYPE_SHADER_SCRATCH, + total, + 4096, + &brw->wm.scratch_bo); + if (ret) + return ret; } } @@ -287,16 +303,19 @@ static int upload_wm_unit( struct brw_context *brw ) reloc_bufs[1] = brw->wm.scratch_bo; reloc_bufs[2] = brw->wm.sampler_bo; - brw->sws->bo_unreference(brw->wm.state_bo); - brw->wm.state_bo = brw_search_cache(&brw->cache, BRW_WM_UNIT, - &key, sizeof(key), - reloc_bufs, 3, - NULL); - if (brw->wm.state_bo == NULL) { - brw->wm.state_bo = wm_unit_create_from_key(brw, &key, reloc_bufs); - } + if (brw_search_cache(&brw->cache, BRW_WM_UNIT, + &key, sizeof(key), + reloc_bufs, 3, + NULL, + &brw->wm.state_bo)) + return PIPE_OK; + + ret = wm_unit_create_from_key(brw, &key, reloc_bufs, + &brw->wm.state_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_wm_unit = { diff --git a/src/gallium/drivers/i965/brw_wm_surface_state.c b/src/gallium/drivers/i965/brw_wm_surface_state.c index b055dde20c..e5d0329967 100644 --- a/src/gallium/drivers/i965/brw_wm_surface_state.c +++ b/src/gallium/drivers/i965/brw_wm_surface_state.c @@ -40,31 +40,40 @@ -static void +static enum pipe_error brw_update_texture_surface( struct brw_context *brw, struct brw_texture *tex, - GLuint surf ) + struct brw_winsys_buffer **bo_out) { - brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, - BRW_SS_SURFACE, - &tex->ss, sizeof tex->ss, - &tex->bo, 1, - NULL); - - if (brw->wm.surf_bo[surf] == NULL) { - brw->wm.surf_bo[surf] = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, - &tex->ss, sizeof tex->ss, - &tex->bo, 1, - &tex->ss, sizeof tex->ss, - NULL, NULL); + enum pipe_error ret; + + if (brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &tex->ss, sizeof tex->ss, + &tex->bo, 1, + NULL, + bo_out)) + return PIPE_OK; + + ret = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, + &tex->ss, sizeof tex->ss, + &tex->bo, 1, + &tex->ss, sizeof tex->ss, + NULL, NULL, + bo_out); + if (ret) + return ret; - /* Emit relocation to surface contents */ - brw->sws->bo_emit_reloc(brw->wm.surf_bo[surf], - BRW_USAGE_SAMPLER, - 0, - offsetof(struct brw_surface_state, ss1), - tex->bo); - } + /* Emit relocation to surface contents */ + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_SAMPLER, + 0, + offsetof(struct brw_surface_state, ss1), + tex->bo); + if (ret) + return ret; + + return PIPE_OK; } @@ -79,13 +88,14 @@ brw_update_texture_surface( struct brw_context *brw, * While it is only used for the front/back buffer currently, it should be * usable for further buffers when doing ARB_draw_buffer support. */ -static void -brw_update_renderbuffer_surface(struct brw_context *brw, - struct brw_surface *surface, - unsigned int unit) +static enum pipe_error +brw_update_render_surface(struct brw_context *brw, + struct brw_surface *surface, + struct brw_winsys_buffer **bo_out) { struct brw_surf_ss0 blend_ss0 = brw->curr.blend->ss0; struct brw_surface_state ss; + enum pipe_error ret; /* Surfaces are potentially shared between contexts, so can't * scribble the in-place ss0 value in the surface. @@ -98,30 +108,35 @@ brw_update_renderbuffer_surface(struct brw_context *brw, ss.ss0.writedisable_red = blend_ss0.writedisable_red; ss.ss0.writedisable_alpha = blend_ss0.writedisable_alpha; - brw->sws->bo_unreference(brw->wm.surf_bo[unit]); - brw->wm.surf_bo[unit] = brw_search_cache(&brw->surface_cache, - BRW_SS_SURFACE, - &ss, sizeof(ss), - &surface->bo, 1, - NULL); - - if (brw->wm.surf_bo[unit] == NULL) { - - brw->wm.surf_bo[unit] = brw_upload_cache(&brw->surface_cache, - BRW_SS_SURFACE, - &ss, sizeof ss, - &surface->bo, 1, - &ss, sizeof ss, - NULL, NULL); + if (brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &ss, sizeof(ss), + &surface->bo, 1, + NULL, + bo_out)) + return PIPE_OK; + + ret = brw_upload_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &ss, sizeof ss, + &surface->bo, 1, + &ss, sizeof ss, + NULL, NULL, + bo_out); + if (ret) + return ret; /* XXX: we will only be rendering to this surface: */ - brw->sws->bo_emit_reloc(brw->wm.surf_bo[unit], - BRW_USAGE_RENDER_TARGET, - ss.ss1.base_addr - surface->bo->offset[0], /* XXX */ - offsetof(struct brw_surface_state, ss1), - surface->bo); - } + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_RENDER_TARGET, + ss.ss1.base_addr - surface->bo->offset[0], /* XXX */ + offsetof(struct brw_surface_state, ss1), + surface->bo); + if (ret) + return ret; + + return PIPE_OK; } @@ -129,60 +144,60 @@ brw_update_renderbuffer_surface(struct brw_context *brw, * Constructs the binding table for the WM surface state, which maps unit * numbers to surface state objects. */ -static struct brw_winsys_buffer * -brw_wm_get_binding_table(struct brw_context *brw) +static enum pipe_error +brw_wm_get_binding_table(struct brw_context *brw, + struct brw_winsys_buffer **bo_out ) { - struct brw_winsys_buffer *bind_bo; + enum pipe_error ret; + uint32_t data[BRW_WM_MAX_SURF]; + GLuint data_size = brw->wm.nr_surfaces * sizeof data[0]; + int i; assert(brw->wm.nr_surfaces <= BRW_WM_MAX_SURF); + assert(brw->wm.nr_surfaces > 0); /* Note there is no key for this search beyond the values in the * relocation array: */ - bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->wm.surf_bo, brw->wm.nr_surfaces, - NULL); - - if (bind_bo == NULL) { - uint32_t data[BRW_WM_MAX_SURF]; - GLuint data_size = brw->wm.nr_surfaces * sizeof data[0]; - int i; - - for (i = 0; i < brw->wm.nr_surfaces; i++) - data[i] = brw->wm.surf_bo[i]->offset[0]; - - bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->wm.surf_bo, brw->wm.nr_surfaces, - data, data_size, - NULL, NULL); - - /* Emit binding table relocations to surface state */ - for (i = 0; i < brw->wm.nr_surfaces; i++) { - brw->sws->bo_emit_reloc(bind_bo, - BRW_USAGE_STATE, - 0, - i * sizeof(GLuint), - brw->wm.surf_bo[i]); - } + if (brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->wm.surf_bo, + brw->wm.nr_surfaces, + NULL, + bo_out)) + return PIPE_OK; + + for (i = 0; i < brw->wm.nr_surfaces; i++) + data[i] = brw->wm.surf_bo[i]->offset[0]; + + ret = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->wm.surf_bo, brw->wm.nr_surfaces, + data, data_size, + NULL, NULL, + bo_out); + if (ret) + return ret; + + /* Emit binding table relocations to surface state */ + for (i = 0; i < brw->wm.nr_surfaces; i++) { + ret = brw->sws->bo_emit_reloc(*bo_out, + BRW_USAGE_STATE, + 0, + i * sizeof(GLuint), + brw->wm.surf_bo[i]); + if (ret) + return ret; } - return bind_bo; + return PIPE_OK; } -static int prepare_wm_surfaces(struct brw_context *brw ) +static enum pipe_error prepare_wm_surfaces(struct brw_context *brw ) { - GLuint i; + enum pipe_error ret; int nr_surfaces = 0; - - /* Unreference old buffers - */ - for (i = 0; i < brw->wm.nr_surfaces; i++) { - brw->sws->bo_unreference(brw->wm.surf_bo[i]); - brw->wm.surf_bo[i] = NULL; - } - + GLuint i; /* PIPE_NEW_COLOR_BUFFERS | PIPE_NEW_BLEND * @@ -192,38 +207,51 @@ static int prepare_wm_surfaces(struct brw_context *brw ) * XXX: no color buffer case */ for (i = 0; i < brw->curr.fb.nr_cbufs; i++) { - brw_update_renderbuffer_surface(brw, - brw_surface(brw->curr.fb.cbufs[i]), - nr_surfaces++); + ret = brw_update_render_surface(brw, + brw_surface(brw->curr.fb.cbufs[i]), + &brw->wm.surf_bo[nr_surfaces++]); + if (ret) + return ret; } /* PIPE_NEW_TEXTURE */ for (i = 0; i < brw->curr.num_textures; i++) { - brw_update_texture_surface(brw, - brw_texture(brw->curr.texture[i]), - nr_surfaces++); + ret = brw_update_texture_surface(brw, + brw_texture(brw->curr.texture[i]), + &brw->wm.surf_bo[nr_surfaces++]); + if (ret) + return ret; } /* PIPE_NEW_FRAGMENT_CONSTANTS */ #if 0 if (brw->curr.fragment_constants) { - brw_update_fragment_constant_surface(brw, - brw->curr.fragment_constants, - nr_surfaces++); + ret = brw_update_fragment_constant_surface(brw, + brw->curr.fragment_constants, + &brw->wm.surf_bo[nr_surfaces++]); + if (ret) + return ret; } #endif if (brw->wm.nr_surfaces != nr_surfaces) { + + /* Unreference any left-over old buffers + */ + for (i = nr_surfaces; i < brw->wm.nr_surfaces; i++) + bo_reference(&brw->wm.surf_bo[i], NULL); + brw->wm.nr_surfaces = nr_surfaces; brw->state.dirty.brw |= BRW_NEW_NR_WM_SURFACES; } - brw->sws->bo_unreference(brw->wm.bind_bo); - brw->wm.bind_bo = brw_wm_get_binding_table(brw); + ret = brw_wm_get_binding_table(brw, &brw->wm.bind_bo); + if (ret) + return ret; - return 0; + return PIPE_OK; } const struct brw_tracked_state brw_wm_surfaces = { diff --git a/src/gallium/winsys/drm/i965/xlib/xlib_i965.c b/src/gallium/winsys/drm/i965/xlib/xlib_i965.c index b1edca818a..fc465d7c14 100644 --- a/src/gallium/winsys/drm/i965/xlib/xlib_i965.c +++ b/src/gallium/winsys/drm/i965/xlib/xlib_i965.c @@ -134,11 +134,12 @@ const char *data_types[BRW_DATA_MAX] = }; -static struct brw_winsys_buffer * +static enum pipe_error xlib_brw_bo_alloc( struct brw_winsys_screen *sws, - enum brw_buffer_type type, - unsigned size, - unsigned alignment ) + enum brw_buffer_type type, + unsigned size, + unsigned alignment, + struct brw_winsys_buffer **bo_out ) { struct xlib_brw_winsys *xbw = xlib_brw_winsys(sws); struct xlib_brw_buffer *buf; @@ -148,12 +149,13 @@ xlib_brw_bo_alloc( struct brw_winsys_screen *sws, buf = CALLOC_STRUCT(xlib_brw_buffer); if (!buf) - return NULL; + return PIPE_ERROR_OUT_OF_MEMORY; + + pipe_reference_init(&buf->base.reference, 1); buf->offset = align(xbw->offset, alignment); buf->type = type; buf->virtual = MALLOC(size); - buf->cheesy_refcount = 1; buf->base.offset = &buf->offset; /* hmm, cheesy */ buf->base.size = size; @@ -161,36 +163,25 @@ xlib_brw_bo_alloc( struct brw_winsys_screen *sws, if (xbw->offset > MAX_VRAM) goto err; - return &buf->base; + /* XXX: possibly rentrant call to bo_destroy: + */ + bo_reference(bo_out, &buf->base); + return PIPE_OK; err: assert(0); + FREE(buf->virtual); FREE(buf); - return NULL; -} - -static void -xlib_brw_bo_reference( struct brw_winsys_buffer *buffer ) -{ - struct xlib_brw_buffer *buf = xlib_brw_buffer(buffer); - - buf->cheesy_refcount++; + return PIPE_ERROR_OUT_OF_MEMORY; } static void -xlib_brw_bo_unreference( struct brw_winsys_buffer *buffer ) +xlib_brw_bo_destroy( struct brw_winsys_buffer *buffer ) { struct xlib_brw_buffer *buf = xlib_brw_buffer(buffer); - /* As a special favor in this call only, buffer is allowed to be - * NULL: - */ - if (buffer == NULL) - return; - - if (--buf->cheesy_refcount == 0) { - FREE(buffer); - } + FREE(buf->virtual); + FREE(buf); } static int @@ -378,8 +369,7 @@ xlib_create_brw_winsys_screen( void ) ws->base.destroy = xlib_brw_winsys_destroy; ws->base.bo_alloc = xlib_brw_bo_alloc; - ws->base.bo_reference = xlib_brw_bo_reference; - ws->base.bo_unreference = xlib_brw_bo_unreference; + ws->base.bo_destroy = xlib_brw_bo_destroy; ws->base.bo_emit_reloc = xlib_brw_bo_emit_reloc; ws->base.bo_exec = xlib_brw_bo_exec; ws->base.bo_subdata = xlib_brw_bo_subdata; -- cgit v1.2.3