diff options
Diffstat (limited to 'src/mesa/drivers/dri/i965')
47 files changed, 578 insertions, 463 deletions
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 7a55333e89..7758a792fd 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -24,7 +24,6 @@ DRIVER_SOURCES = \ intel_pixel_draw.c \ intel_pixel_read.c \ intel_state.c \ - intel_swapbuffers.c \ intel_syncobj.c \ intel_tex.c \ intel_tex_copy.c \ @@ -96,7 +95,8 @@ ASM_SOURCES = DRIVER_DEFINES = -I../intel -I../intel/server -DRI_LIB_DEPS += -ldrm_intel +INCLUDES += $(INTEL_CFLAGS) +DRI_LIB_DEPS += $(INTEL_LIBS) include ../Makefile.template diff --git a/src/mesa/drivers/dri/i965/brw_cc.c b/src/mesa/drivers/dri/i965/brw_cc.c index 9a38631dfe..fa2d394b22 100644 --- a/src/mesa/drivers/dri/i965/brw_cc.c +++ b/src/mesa/drivers/dri/i965/brw_cc.c @@ -293,8 +293,7 @@ cc_unit_create_from_key(struct brw_context *brw, struct brw_cc_unit_key *key) bo = brw_upload_cache(&brw->cache, BRW_CC_UNIT, key, sizeof(*key), &brw->cc.vp_bo, 1, - &cc, sizeof(cc), - NULL, NULL); + &cc, sizeof(cc)); /* Emit CC viewport relocation */ dri_bo_emit_reloc(bo, diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c index dbd10a5297..d3275c7a89 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.c +++ b/src/mesa/drivers/dri/i965/brw_clip.c @@ -50,6 +50,7 @@ static void compile_clip_prog( struct brw_context *brw, struct brw_clip_prog_key *key ) { + struct intel_context *intel = &brw->intel; struct brw_clip_compile c; const GLuint *program; GLuint program_size; @@ -65,14 +66,13 @@ static void compile_clip_prog( struct brw_context *brw, c.func.single_program_flow = 1; c.key = *key; - c.need_ff_sync = BRW_IS_IGDNG(brw); /* Need to locate the two positions present in vertex + header. * These are currently hardcoded: */ c.header_position_offset = ATTR_SIZE; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) delta = 3 * REG_SIZE; else delta = REG_SIZE; @@ -85,7 +85,7 @@ static void compile_clip_prog( struct brw_context *brw, c.nr_attrs = brw_count_bits(c.key.attrs); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) 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? */ @@ -130,20 +130,22 @@ static void compile_clip_prog( struct brw_context *brw, /* Upload */ dri_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 ); + brw->clip.prog_bo = brw_upload_cache_with_auxdata(&brw->cache, + BRW_CLIP_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + sizeof(c.prog_data), + &brw->clip.prog_data); } /* Calculate interpolants for triangle and line rasterization. */ static void upload_clip_prog(struct brw_context *brw) { - GLcontext *ctx = &brw->intel.ctx; + struct intel_context *intel = &brw->intel; + GLcontext *ctx = &intel->ctx; struct brw_clip_prog_key key; memset(&key, 0, sizeof(key)); @@ -160,7 +162,7 @@ static void upload_clip_prog(struct brw_context *brw) /* _NEW_TRANSFORM */ key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) key.clip_mode = BRW_CLIPMODE_KERNEL_CLIP; else key.clip_mode = BRW_CLIPMODE_NORMAL; diff --git a/src/mesa/drivers/dri/i965/brw_clip.h b/src/mesa/drivers/dri/i965/brw_clip.h index 1c6825510a..d71bac7f61 100644 --- a/src/mesa/drivers/dri/i965/brw_clip.h +++ b/src/mesa/drivers/dri/i965/brw_clip.h @@ -118,7 +118,6 @@ struct brw_clip_compile { GLuint header_position_offset; GLuint offset[VERT_ATTRIB_MAX]; - GLboolean need_ff_sync; }; #define ATTR_SIZE (4*4) diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c index 67861ac670..ceb62a3116 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_line.c +++ b/src/mesa/drivers/dri/i965/brw_clip_line.c @@ -45,6 +45,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i = 0,j; /* Register usage is static, precompute here: @@ -84,7 +85,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) i++; } - if (c->need_ff_sync) { + if (intel->needs_ff_sync) { c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; } @@ -125,6 +126,7 @@ static void brw_clip_line_alloc_regs( struct brw_clip_compile *c ) static void clip_and_emit_line( struct brw_clip_compile *c ) { struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; struct brw_indirect vtx0 = brw_indirect(0, 0); struct brw_indirect vtx1 = brw_indirect(1, 0); struct brw_indirect newvtx0 = brw_indirect(2, 0); @@ -151,7 +153,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_clip_init_clipmask(c); /* -ve rhw workaround */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<20)); @@ -188,7 +190,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) * Both can be negative on GM965/G965 due to RHW workaround * if so, this object should be rejected. */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0)); is_neg2 = brw_IF(p, BRW_EXECUTE_1); { @@ -213,7 +215,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) /* If both are positive, do nothing */ /* Only on GM965/G965 */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0)); is_neg2 = brw_IF(p, BRW_EXECUTE_1); } @@ -228,7 +230,7 @@ static void clip_and_emit_line( struct brw_clip_compile *c ) brw_set_predicate_control(p, BRW_PREDICATE_NONE); } - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_ENDIF(p, is_neg2); } } diff --git a/src/mesa/drivers/dri/i965/brw_clip_state.c b/src/mesa/drivers/dri/i965/brw_clip_state.c index 79ae5c7bef..424c9a1f19 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_state.c +++ b/src/mesa/drivers/dri/i965/brw_clip_state.c @@ -73,6 +73,7 @@ static dri_bo * clip_unit_create_from_key(struct brw_context *brw, struct brw_clip_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_clip_unit_state clip; dri_bo *bo; @@ -104,7 +105,7 @@ clip_unit_create_from_key(struct brw_context *brw, /* Although up to 16 concurrent Clip threads are allowed on IGDNG, * only 2 threads can output VUEs at a time. */ - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) clip.thread4.max_threads = 16 - 1; else clip.thread4.max_threads = 2 - 1; @@ -129,7 +130,7 @@ clip_unit_create_from_key(struct brw_context *brw, clip.clip5.api_mode = BRW_CLIP_API_OGL; clip.clip5.clip_mode = key->clip_mode; - if (BRW_IS_G4X(brw)) + if (intel->is_g4x) clip.clip5.negative_w_clip_test = 1; clip.clip6.clipper_viewport_state_ptr = 0; @@ -141,8 +142,7 @@ clip_unit_create_from_key(struct brw_context *brw, bo = brw_upload_cache(&brw->cache, BRW_CLIP_UNIT, key, sizeof(*key), &brw->clip.prog_bo, 1, - &clip, sizeof(clip), - NULL, NULL); + &clip, sizeof(clip)); /* Emit clip program relocation */ assert(brw->clip.prog_bo); diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c index 8d5ca41896..815211acc2 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_tri.c +++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c @@ -50,6 +50,7 @@ static void release_tmps( struct brw_clip_compile *c ) void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, GLuint nr_verts ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i = 0,j; /* Register usage is static, precompute here: @@ -77,7 +78,7 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, for (j = 0; j < 3; j++) { GLuint delta = c->nr_attrs*16 + 32; - if (BRW_IS_IGDNG(c->func.brw)) + if (intel->is_ironlake) delta = c->nr_attrs * 16 + 32 * 3; brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0)); @@ -118,7 +119,7 @@ void brw_clip_tri_alloc_regs( struct brw_clip_compile *c, i++; } - if (c->need_ff_sync) { + if (intel->needs_ff_sync) { c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD); i++; } @@ -570,6 +571,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) { struct brw_instruction *neg_rhw; struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6); brw_clip_tri_init_vertices(c); brw_clip_init_clipmask(c); @@ -577,7 +579,7 @@ void brw_emit_tri_clip( struct brw_clip_compile *c ) /* if -ve rhw workaround bit is set, do cliptest */ - if (BRW_IS_965(p->brw)) { + if (brw->has_negative_rhw_bug) { brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<20)); diff --git a/src/mesa/drivers/dri/i965/brw_clip_util.c b/src/mesa/drivers/dri/i965/brw_clip_util.c index b7f8c25b34..14bc889b0f 100644 --- a/src/mesa/drivers/dri/i965/brw_clip_util.c +++ b/src/mesa/drivers/dri/i965/brw_clip_util.c @@ -134,6 +134,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, GLboolean force_edgeflag) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg tmp = get_tmp(c); GLuint i; @@ -141,7 +142,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, */ /* * After CLIP stage, only first 256 bits of the VUE are read - * back on IGDNG, so needn't change it + * back on Ironlake, so needn't change it */ brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1); @@ -150,7 +151,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, for (i = 0; i < c->nr_attrs; i++) { GLuint delta = i*16 + 32; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) delta = i * 16 + 32 * 3; if (delta == c->offset[VERT_RESULT_EDGE]) { @@ -184,7 +185,7 @@ void brw_clip_interp_vertex( struct brw_clip_compile *c, if (i & 1) { GLuint delta = i*16 + 32; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) delta = i * 16 + 32 * 3; brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0)); @@ -358,7 +359,9 @@ void brw_clip_init_clipmask( struct brw_clip_compile *c ) void brw_clip_ff_sync(struct brw_clip_compile *c) { - if (c->need_ff_sync) { + struct intel_context *intel = &c->func.brw->intel; + + if (intel->needs_ff_sync) { struct brw_compile *p = &c->func; struct brw_instruction *need_ff_sync; @@ -387,7 +390,9 @@ void brw_clip_ff_sync(struct brw_clip_compile *c) void brw_clip_init_ff_sync(struct brw_clip_compile *c) { - if (c->need_ff_sync) { + struct intel_context *intel = &c->func.brw->intel; + + if (intel->needs_ff_sync) { struct brw_compile *p = &c->func; brw_MOV(p, c->reg.ff_sync, brw_imm_ud(0)); diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c index 6b5c3becc2..568db97ec7 100644 --- a/src/mesa/drivers/dri/i965/brw_context.c +++ b/src/mesa/drivers/dri/i965/brw_context.c @@ -68,7 +68,7 @@ static void brwInitDriverFunctions( struct dd_function_table *functions ) } GLboolean brwCreateContext( const __GLcontextModes *mesaVis, - __DRIcontextPrivate *driContextPriv, + __DRIcontext *driContextPriv, void *sharedContextPrivate) { struct dd_function_table functions; @@ -102,6 +102,9 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureCoordUnits, ctx->Const.MaxTextureImageUnits); ctx->Const.MaxVertexTextureImageUnits = 0; /* no vertex shader textures */ + ctx->Const.MaxCombinedTextureImageUnits = + ctx->Const.MaxVertexTextureImageUnits + + ctx->Const.MaxTextureImageUnits; /* Mesa limits textures to 4kx4k; it would be nice to fix that someday */ @@ -146,6 +149,38 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, MIN2(ctx->Const.FragmentProgram.MaxNativeParameters, ctx->Const.FragmentProgram.MaxEnvParams); + if (intel->is_ironlake || intel->is_g4x) { + brw->CMD_VF_STATISTICS = CMD_VF_STATISTICS_GM45; + brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_GM45; + brw->has_surface_tile_offset = GL_TRUE; + brw->has_compr4 = GL_TRUE; + brw->has_aa_line_parameters = GL_TRUE; + } else { + brw->CMD_VF_STATISTICS = CMD_VF_STATISTICS_965; + brw->CMD_PIPELINE_SELECT = CMD_PIPELINE_SELECT_965; + } + + /* WM maximum threads is number of EUs times number of threads per EU. */ + if (intel->is_ironlake) { + brw->urb.size = 1024; + brw->vs_max_threads = 72; + brw->wm_max_threads = 12 * 6; + } else if (intel->is_g4x) { + brw->urb.size = 384; + brw->vs_max_threads = 32; + brw->wm_max_threads = 10 * 5; + } else { + brw->urb.size = 256; + brw->vs_max_threads = 16; + brw->wm_max_threads = 8 * 4; + brw->has_negative_rhw_bug = GL_TRUE; + } + + if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) { + brw->vs_max_threads = 1; + brw->wm_max_threads = 1; + } + brw_init_state( brw ); brw->state.dirty.mesa = ~0; diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index fded47aa2f..79818b92b7 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -131,7 +131,6 @@ struct brw_context; #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100 #define BRW_NEW_PSP 0x800 #define BRW_NEW_WM_SURFACES 0x1000 -#define BRW_NEW_FENCE 0x2000 #define BRW_NEW_INDICES 0x4000 #define BRW_NEW_VERTICES 0x8000 /** @@ -172,8 +171,8 @@ struct brw_fragment_program { GLuint id; /**< serial no. to identify frag progs, never re-used */ GLboolean isGLSL; /**< really, any IF/LOOP/CONT/BREAK instructions */ - dri_bo *const_buffer; /** Program constant buffer/surface */ GLboolean use_const_buffer; + dri_bo *const_buffer; /** Program constant buffer/surface */ /** for debugging, which texture units are referenced */ GLbitfield tex_units_used; @@ -332,7 +331,6 @@ struct brw_cache { struct brw_cache_item **items; GLuint size, n_items; - GLuint aux_size[BRW_MAX_CACHE]; char *name[BRW_MAX_CACHE]; /* Record of the last BOs chosen for each cache_id. Used to set @@ -438,8 +436,11 @@ struct brw_context GLuint primitive; GLboolean emit_state_always; - GLboolean no_batch_wrap; - + GLboolean has_surface_tile_offset; + GLboolean has_compr4; + GLboolean has_negative_rhw_bug; + GLboolean has_aa_line_parameters; +; struct { struct brw_state_flags dirty; @@ -515,6 +516,12 @@ struct brw_context */ GLuint next_free_page; + /* hw-dependent 3DSTATE_VF_STATISTICS opcode */ + uint32_t CMD_VF_STATISTICS; + /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */ + uint32_t CMD_PIPELINE_SELECT; + int vs_max_threads; + int wm_max_threads; /* BRW_NEW_URB_ALLOCATIONS: */ @@ -542,6 +549,7 @@ struct brw_context GLuint clip_start; GLuint sf_start; GLuint cs_start; + GLuint size; /* Hardware URB size, in KB. */ } urb; @@ -573,6 +581,7 @@ struct brw_context struct { struct brw_vs_prog_data *prog_data; + int8_t *constant_map; /* variable array following prog_data */ dri_bo *prog_bo; dri_bo *state_bo; @@ -669,7 +678,7 @@ void brwInitVtbl( struct brw_context *brw ); * brw_context.c */ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, - __DRIcontextPrivate *driContextPriv, + __DRIcontext *driContextPriv, void *sharedContextPrivate); /*====================================================================== diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c index aadcfbe2da..22e3e732f4 100644 --- a/src/mesa/drivers/dri/i965/brw_curbe.c +++ b/src/mesa/drivers/dri/i965/brw_curbe.c @@ -256,13 +256,24 @@ static void prepare_constant_buffer(struct brw_context *brw) */ _mesa_load_state_parameters(ctx, vp->program.Base.Parameters); - /* XXX just use a memcpy here */ - for (i = 0; i < nr; i++) { - const GLfloat *value = vp->program.Base.Parameters->ParameterValues[i]; - buf[offset + i * 4 + 0] = value[0]; - buf[offset + i * 4 + 1] = value[1]; - buf[offset + i * 4 + 2] = value[2]; - buf[offset + i * 4 + 3] = value[3]; + if (vp->use_const_buffer) { + /* Load the subset of push constants that will get used when + * we also have a pull constant buffer. + */ + for (i = 0; i < vp->program.Base.Parameters->NumParameters; i++) { + if (brw->vs.constant_map[i] != -1) { + assert(brw->vs.constant_map[i] <= nr); + memcpy(buf + offset + brw->vs.constant_map[i] * 4, + vp->program.Base.Parameters->ParameterValues[i], + 4 * sizeof(float)); + } + } + } else { + for (i = 0; i < nr; i++) { + memcpy(buf + offset + i * 4, + vp->program.Base.Parameters->ParameterValues[i], + 4 * sizeof(float)); + } } } @@ -340,7 +351,7 @@ static void emit_constant_buffer(struct brw_context *brw) struct intel_context *intel = &brw->intel; GLuint sz = brw->curbe.total_size; - BEGIN_BATCH(2, IGNORE_CLIPRECTS); + BEGIN_BATCH(2); if (sz == 0) { OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2)); OUT_BATCH(0); diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h index c19510bbd4..ea0d7e05d4 100644 --- a/src/mesa/drivers/dri/i965/brw_defines.h +++ b/src/mesa/drivers/dri/i965/brw_defines.h @@ -832,12 +832,4 @@ #include "intel_chipset.h" -#define BRW_IS_G4X(brw) (IS_G4X((brw)->intel.intelScreen->deviceID)) -#define BRW_IS_IGDNG(brw) (IS_IGDNG((brw)->intel.intelScreen->deviceID)) -#define BRW_IS_965(brw) (!(BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw))) -#define CMD_PIPELINE_SELECT(brw) ((BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? CMD_PIPELINE_SELECT_GM45 : CMD_PIPELINE_SELECT_965) -#define CMD_VF_STATISTICS(brw) ((BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? CMD_VF_STATISTICS_GM45 : CMD_VF_STATISTICS_965) -#define URB_SIZES(brw) (BRW_IS_IGDNG(brw) ? 1024 : \ - (BRW_IS_G4X(brw) ? 384 : 256)) /* 512 bit units */ - #endif diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c index 130bd0f3e5..a8f6b993ac 100644 --- a/src/mesa/drivers/dri/i965/brw_disasm.c +++ b/src/mesa/drivers/dri/i965/brw_disasm.c @@ -239,7 +239,7 @@ char *imm_encoding[8] = { [2] = "UW", [3] = "W", [5] = "VF", - [5] = "V", + [6] = "V", [7] = "F" }; diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c index e17680a525..d510d767f9 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.c +++ b/src/mesa/drivers/dri/i965/brw_draw.c @@ -143,7 +143,7 @@ static void brw_emit_prim(struct brw_context *brw, prim_packet.base_vert_location = prim->basevertex; /* Can't wrap here, since we rely on the validated state. */ - brw->no_batch_wrap = GL_TRUE; + intel->no_batch_wrap = GL_TRUE; /* If we're set to always flush, do it before and after the primitive emit. * We want to catch both missed flushes that hurt instruction/state cache @@ -155,13 +155,13 @@ static void brw_emit_prim(struct brw_context *brw, } if (prim_packet.verts_per_instance) { intel_batchbuffer_data( brw->intel.batch, &prim_packet, - sizeof(prim_packet), LOOP_CLIPRECTS); + sizeof(prim_packet)); } if (intel->always_flush_cache) { intel_batchbuffer_emit_mi_flush(intel->batch); } - brw->no_batch_wrap = GL_FALSE; + intel->no_batch_wrap = GL_FALSE; } static void brw_merge_inputs( struct brw_context *brw, @@ -337,13 +337,6 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, * so can't access it earlier. */ - LOCK_HARDWARE(intel); - - if (!intel->constant_cliprect && intel->driDrawable->numClipRects == 0) { - UNLOCK_HARDWARE(intel); - return GL_TRUE; - } - for (i = 0; i < nr_prims; i++) { uint32_t hw_prim; @@ -354,8 +347,7 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, * an upper bound of how much we might emit in a single * brw_try_draw_prims(). */ - intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4, - LOOP_CLIPRECTS); + intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4); hw_prim = brw_set_prim(brw, prim[i].mode); @@ -402,7 +394,6 @@ static GLboolean brw_try_draw_prims( GLcontext *ctx, if (intel->always_flush_batch) intel_batchbuffer_flush(intel->batch); out: - UNLOCK_HARDWARE(intel); brw_state_cache_check_size(brw); diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index 16b2529fad..c46b9ba89c 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -55,6 +55,14 @@ static GLuint float_types[5] = { BRW_SURFACEFORMAT_R32G32B32A32_FLOAT }; +static GLuint half_float_types[5] = { + 0, + BRW_SURFACEFORMAT_R16_FLOAT, + BRW_SURFACEFORMAT_R16G16_FLOAT, + 0, /* can't seem to render this one */ + BRW_SURFACEFORMAT_R16G16B16A16_FLOAT +}; + static GLuint uint_types_norm[5] = { 0, BRW_SURFACEFORMAT_R32_UNORM, @@ -168,6 +176,7 @@ static GLuint get_surface_type( GLenum type, GLuint size, switch (type) { case GL_DOUBLE: return double_types[size]; case GL_FLOAT: return float_types[size]; + case GL_HALF_FLOAT: return half_float_types[size]; case GL_INT: return int_types_norm[size]; case GL_SHORT: return short_types_norm[size]; case GL_BYTE: return byte_types_norm[size]; @@ -190,6 +199,7 @@ static GLuint get_surface_type( GLenum type, GLuint size, switch (type) { case GL_DOUBLE: return double_types[size]; case GL_FLOAT: return float_types[size]; + case GL_HALF_FLOAT: return half_float_types[size]; case GL_INT: return int_types_scale[size]; case GL_SHORT: return short_types_scale[size]; case GL_BYTE: return byte_types_scale[size]; @@ -207,6 +217,7 @@ static GLuint get_size( GLenum type ) switch (type) { case GL_DOUBLE: return sizeof(GLdouble); case GL_FLOAT: return sizeof(GLfloat); + case GL_HALF_FLOAT: return sizeof(GLhalfARB); case GL_INT: return sizeof(GLint); case GL_SHORT: return sizeof(GLshort); case GL_BYTE: return sizeof(GLbyte); @@ -239,16 +250,6 @@ static void wrap_buffers( struct brw_context *brw, dri_bo_unreference(brw->vb.upload.bo); brw->vb.upload.bo = dri_bo_alloc(brw->intel.bufmgr, "temporary VBO", size, 1); - - /* Set the internal VBO\ to no-backing-store. We only use them as a - * temporary within a brw_try_draw_prims while the lock is held. - */ - /* DON'T DO THIS AS IF WE HAVE TO RE-ORG MEMORY WE NEED SOMEWHERE WITH - FAKE TO PUSH THIS STUFF */ - /* - if (!brw->intel.ttm) - dri_bo_fake_disable_backing_store(brw->vb.upload.bo, NULL, NULL); - */ } static void get_space( struct brw_context *brw, @@ -500,7 +501,7 @@ static void brw_emit_vertices(struct brw_context *brw) * a VE loads from them. */ if (brw->vb.nr_enabled == 0) { - BEGIN_BATCH(3, IGNORE_CLIPRECTS); + BEGIN_BATCH(3); OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1); OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) | BRW_VE0_VALID | @@ -520,7 +521,7 @@ static void brw_emit_vertices(struct brw_context *brw) * are interleaved or from the same VBO. TBD if this makes a * performance difference. */ - BEGIN_BATCH(1 + brw->vb.nr_enabled * 4, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 4); OUT_BATCH((CMD_VERTEX_BUFFER << 16) | ((1 + brw->vb.nr_enabled * 4) - 2)); @@ -533,7 +534,7 @@ static void brw_emit_vertices(struct brw_context *brw) OUT_RELOC(input->bo, I915_GEM_DOMAIN_VERTEX, 0, input->offset); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { OUT_RELOC(input->bo, I915_GEM_DOMAIN_VERTEX, 0, input->bo->size - 1); @@ -543,7 +544,7 @@ static void brw_emit_vertices(struct brw_context *brw) } ADVANCE_BATCH(); - BEGIN_BATCH(1 + brw->vb.nr_enabled * 2, IGNORE_CLIPRECTS); + BEGIN_BATCH(1 + brw->vb.nr_enabled * 2); 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]; @@ -569,7 +570,7 @@ static void brw_emit_vertices(struct brw_context *brw) (format << BRW_VE0_FORMAT_SHIFT) | (0 << BRW_VE0_SRC_OFFSET_SHIFT)); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) | (comp1 << BRW_VE1_COMPONENT_1_SHIFT) | (comp2 << BRW_VE1_COMPONENT_2_SHIFT) | @@ -710,7 +711,7 @@ static void brw_emit_index_buffer(struct brw_context *brw) ib.header.bits.index_format = get_index_type(index_buffer->type); ib.header.bits.cut_index_enable = 0; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH( ib.header.dword ); OUT_RELOC(brw->ib.bo, I915_GEM_DOMAIN_VERTEX, 0, diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index c8fb0a288f..8d6ac00839 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -252,9 +252,10 @@ static void brw_set_math_message( struct brw_context *brw, GLboolean saturate, GLuint dataType ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.math_igdng.function = function; insn->bits3.math_igdng.int_type = integer_type; insn->bits3.math_igdng.precision = low_precision; @@ -319,9 +320,10 @@ static void brw_set_urb_message( struct brw_context *brw, GLuint offset, GLuint swizzle_control ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.urb_igdng.opcode = 0; /* ? */ insn->bits3.urb_igdng.offset = offset; insn->bits3.urb_igdng.swizzle_control = swizzle_control; @@ -358,9 +360,10 @@ static void brw_set_dp_write_message( struct brw_context *brw, GLuint response_length, GLuint end_of_thread ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.dp_write_igdng.binding_table_index = binding_table_index; insn->bits3.dp_write_igdng.msg_control = msg_control; insn->bits3.dp_write_igdng.pixel_scoreboard_clear = pixel_scoreboard_clear; @@ -395,9 +398,10 @@ static void brw_set_dp_read_message( struct brw_context *brw, GLuint response_length, GLuint end_of_thread ) { + struct intel_context *intel = &brw->intel; brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.dp_read_igdng.binding_table_index = binding_table_index; insn->bits3.dp_read_igdng.msg_control = msg_control; insn->bits3.dp_read_igdng.msg_type = msg_type; @@ -433,10 +437,11 @@ static void brw_set_sampler_message(struct brw_context *brw, GLuint header_present, GLuint simd_mode) { + struct intel_context *intel = &brw->intel; assert(eot == 0); brw_set_src1(insn, brw_imm_d(0)); - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { insn->bits3.sampler_igdng.binding_table_index = binding_table_index; insn->bits3.sampler_igdng.sampler = sampler; insn->bits3.sampler_igdng.msg_type = msg_type; @@ -447,7 +452,7 @@ static void brw_set_sampler_message(struct brw_context *brw, insn->bits3.sampler_igdng.end_of_thread = eot; insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_SAMPLER; insn->bits2.send_igdng.end_of_thread = eot; - } else if (BRW_IS_G4X(brw)) { + } else if (intel->is_g4x) { insn->bits3.sampler_g4x.binding_table_index = binding_table_index; insn->bits3.sampler_g4x.sampler = sampler; insn->bits3.sampler_g4x.msg_type = msg_type; @@ -648,10 +653,11 @@ struct brw_instruction *brw_IF(struct brw_compile *p, GLuint execute_size) struct brw_instruction *brw_ELSE(struct brw_compile *p, struct brw_instruction *if_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *insn; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) { @@ -690,9 +696,10 @@ struct brw_instruction *brw_ELSE(struct brw_compile *p, void brw_ENDIF(struct brw_compile *p, struct brw_instruction *patch_insn) { + struct intel_context *intel = &p->brw->intel; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) { @@ -803,10 +810,11 @@ struct brw_instruction *brw_DO(struct brw_compile *p, GLuint execute_size) struct brw_instruction *brw_WHILE(struct brw_compile *p, struct brw_instruction *do_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *insn; GLuint br = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) br = 2; if (p->single_program_flow) @@ -846,10 +854,11 @@ struct brw_instruction *brw_WHILE(struct brw_compile *p, void brw_land_fwd_jump(struct brw_compile *p, struct brw_instruction *jmp_insn) { + struct intel_context *intel = &p->brw->intel; struct brw_instruction *landing = &p->store[p->nr_insn]; GLuint jmpi = 1; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; assert(jmp_insn->header.opcode == BRW_OPCODE_JMPI); diff --git a/src/mesa/drivers/dri/i965/brw_fallback.c b/src/mesa/drivers/dri/i965/brw_fallback.c index 1914febf26..79691c3a0f 100644 --- a/src/mesa/drivers/dri/i965/brw_fallback.c +++ b/src/mesa/drivers/dri/i965/brw_fallback.c @@ -40,13 +40,10 @@ #include "intel_fbo.h" #include "intel_regions.h" -#include "glapi/glapi.h" - #define FILE_DEBUG_FLAG DEBUG_FALLBACKS static GLboolean do_check_fallback(struct brw_context *brw) { - struct intel_context *intel = &brw->intel; GLcontext *ctx = &brw->intel.ctx; GLuint i; @@ -85,8 +82,7 @@ static GLboolean do_check_fallback(struct brw_context *brw) } /* _NEW_BUFFERS */ - if (IS_965(intel->intelScreen->deviceID) && - !IS_G4X(intel->intelScreen->deviceID)) { + if (!brw->has_surface_tile_offset) { for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i]; struct intel_renderbuffer *irb = intel_renderbuffer(rb); diff --git a/src/mesa/drivers/dri/i965/brw_gs.c b/src/mesa/drivers/dri/i965/brw_gs.c index 610b6c35e2..7261b316c1 100644 --- a/src/mesa/drivers/dri/i965/brw_gs.c +++ b/src/mesa/drivers/dri/i965/brw_gs.c @@ -47,6 +47,7 @@ static void compile_gs_prog( struct brw_context *brw, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &brw->intel; struct brw_gs_compile c; const GLuint *program; GLuint program_size; @@ -54,13 +55,12 @@ static void compile_gs_prog( struct brw_context *brw, memset(&c, 0, sizeof(c)); c.key = *key; - c.need_ff_sync = BRW_IS_IGDNG(brw); /* Need to locate the two positions present in vertex + header. * These are currently hardcoded: */ c.nr_attrs = brw_count_bits(c.key.attrs); - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) 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? */ @@ -125,12 +125,13 @@ static void compile_gs_prog( struct brw_context *brw, /* Upload */ dri_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 ); + brw->gs.prog_bo = brw_upload_cache_with_auxdata(&brw->cache, BRW_GS_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + sizeof(c.prog_data), + &brw->gs.prog_data); } static const GLenum gs_prim[GL_POLYGON+1] = { diff --git a/src/mesa/drivers/dri/i965/brw_gs.h b/src/mesa/drivers/dri/i965/brw_gs.h index 010c1c2352..813b8d447a 100644 --- a/src/mesa/drivers/dri/i965/brw_gs.h +++ b/src/mesa/drivers/dri/i965/brw_gs.h @@ -63,7 +63,6 @@ struct brw_gs_compile { GLuint nr_attrs; GLuint nr_regs; GLuint nr_bytes; - GLboolean need_ff_sync; }; #define ATTR_SIZE (4*4) diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c index e9b71ed0c3..dd7b057d62 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c @@ -121,12 +121,14 @@ static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim) void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 4); /* Use polygons for correct edgeflag behaviour. Note that vertex 3 * is the PV for quads, but vertex 0 for polygons: */ - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); if (key->pv_first) { brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); @@ -144,9 +146,11 @@ void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 4); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); if (key->pv_first) { brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START)); @@ -164,9 +168,11 @@ void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key ) void brw_gs_tris( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 3); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_TRILIST << 2) | R02_PRIM_START)); brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_TRILIST << 2)); @@ -175,9 +181,11 @@ void brw_gs_tris( struct brw_gs_compile *c ) void brw_gs_lines( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 2); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_START)); brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_LINESTRIP << 2) | R02_PRIM_END)); @@ -185,9 +193,11 @@ void brw_gs_lines( struct brw_gs_compile *c ) void brw_gs_points( struct brw_gs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; + brw_gs_alloc_regs(c, 1); - if (c->need_ff_sync) + if (intel->needs_ff_sync) brw_gs_ff_sync(c, 1); brw_gs_emit_vue(c, c->reg.vertex[0], 1, ((_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END)); } diff --git a/src/mesa/drivers/dri/i965/brw_gs_state.c b/src/mesa/drivers/dri/i965/brw_gs_state.c index d291f5b595..d8ad5cecf3 100644 --- a/src/mesa/drivers/dri/i965/brw_gs_state.c +++ b/src/mesa/drivers/dri/i965/brw_gs_state.c @@ -71,6 +71,7 @@ gs_unit_populate_key(struct brw_context *brw, struct brw_gs_unit_key *key) static dri_bo * gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_gs_unit_state gs; dri_bo *bo; @@ -97,7 +98,7 @@ gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) else gs.thread4.max_threads = 0; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) gs.thread4.rendering_enable = 1; if (INTEL_DEBUG & DEBUG_STATS) @@ -106,8 +107,7 @@ gs_unit_create_from_key(struct brw_context *brw, struct brw_gs_unit_key *key) bo = brw_upload_cache(&brw->cache, BRW_GS_UNIT, key, sizeof(*key), &brw->gs.prog_bo, 1, - &gs, sizeof(gs), - NULL, NULL); + &gs, sizeof(gs)); if (key->prog_active) { /* Emit GS program relocation */ diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c index 4b0d598336..7b70f787b7 100644 --- a/src/mesa/drivers/dri/i965/brw_misc_state.c +++ b/src/mesa/drivers/dri/i965/brw_misc_state.c @@ -78,10 +78,7 @@ static void upload_drawing_rect(struct brw_context *brw) struct intel_context *intel = &brw->intel; GLcontext *ctx = &intel->ctx; - if (!intel->constant_cliprect) - return; - - BEGIN_BATCH(4, NO_LOOP_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_DRAWRECT_INFO_I965); OUT_BATCH(0); /* xmin, ymin */ OUT_BATCH(((ctx->DrawBuffer->Width - 1) & 0xffff) | @@ -116,7 +113,7 @@ static void upload_binding_table_pointers(struct brw_context *brw) { struct intel_context *intel = &brw->intel; - BEGIN_BATCH(6, IGNORE_CLIPRECTS); + BEGIN_BATCH(6); OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2)); if (brw->vs.bind_bo != NULL) OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */ @@ -150,7 +147,7 @@ static void upload_pipelined_state_pointers(struct brw_context *brw ) { struct intel_context *intel = &brw->intel; - BEGIN_BATCH(7, IGNORE_CLIPRECTS); + BEGIN_BATCH(7); OUT_BATCH(CMD_PIPELINED_STATE_POINTERS << 16 | (7 - 2)); OUT_RELOC(brw->vs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0); if (brw->gs.prog_active) @@ -212,10 +209,10 @@ static void emit_depthbuffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; struct intel_region *region = brw->state.depth_region; - unsigned int len = (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) ? 6 : 5; + unsigned int len = (intel->is_g4x || intel->is_ironlake) ? 6 : 5; if (region == NULL) { - BEGIN_BATCH(len, IGNORE_CLIPRECTS); + BEGIN_BATCH(len); OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2)); OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) | (BRW_SURFACE_NULL << 29)); @@ -223,7 +220,7 @@ static void emit_depthbuffer(struct brw_context *brw) OUT_BATCH(0); OUT_BATCH(0); - if (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) + if (intel->is_g4x || intel->is_ironlake) OUT_BATCH(0); ADVANCE_BATCH(); @@ -247,7 +244,7 @@ static void emit_depthbuffer(struct brw_context *brw) assert(region->tiling != I915_TILING_X); - BEGIN_BATCH(len, IGNORE_CLIPRECTS); + BEGIN_BATCH(len); OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2)); OUT_BATCH(((region->pitch * region->cpp) - 1) | (format << 18) | @@ -262,7 +259,7 @@ static void emit_depthbuffer(struct brw_context *brw) ((region->height - 1) << 19)); OUT_BATCH(0); - if (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw)) + if (intel->is_g4x || intel->is_ironlake) OUT_BATCH(0); ADVANCE_BATCH(); @@ -330,7 +327,7 @@ const struct brw_tracked_state brw_polygon_stipple = { static void upload_polygon_stipple_offset(struct brw_context *brw) { - __DRIdrawablePrivate *dPriv = brw->intel.driDrawable; + __DRIdrawable *dPriv = brw->intel.driDrawable; struct brw_polygon_stipple_offset bpso; memset(&bpso, 0, sizeof(bpso)); @@ -374,8 +371,8 @@ const struct brw_tracked_state brw_polygon_stipple_offset = { static void upload_aa_line_parameters(struct brw_context *brw) { struct brw_aa_line_parameters balp; - - if (BRW_IS_965(brw)) + + if (!brw->has_aa_line_parameters) return; /* use legacy aa line coverage computation */ @@ -444,7 +441,7 @@ static void upload_invarient_state( struct brw_context *brw ) struct brw_pipeline_select ps; memset(&ps, 0, sizeof(ps)); - ps.header.opcode = CMD_PIPELINE_SELECT(brw); + ps.header.opcode = brw->CMD_PIPELINE_SELECT; ps.header.pipeline_select = 0; BRW_BATCH_STRUCT(brw, &ps); } @@ -480,7 +477,7 @@ static void upload_invarient_state( struct brw_context *brw ) struct brw_vf_statistics vfs; memset(&vfs, 0, sizeof(vfs)); - vfs.opcode = CMD_VF_STATISTICS(brw); + vfs.opcode = brw->CMD_VF_STATISTICS; if (INTEL_DEBUG & DEBUG_STATS) vfs.statistics_enable = 1; @@ -512,8 +509,8 @@ static void upload_state_base_address( struct brw_context *brw ) /* Output the structure (brw_state_base_address) directly to the * batchbuffer, so we can emit relocations inline. */ - if (BRW_IS_IGDNG(brw)) { - BEGIN_BATCH(8, IGNORE_CLIPRECTS); + if (intel->is_ironlake) { + BEGIN_BATCH(8); OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (8 - 2)); OUT_BATCH(1); /* General state base address */ OUT_BATCH(1); /* Surface state base address */ @@ -524,7 +521,7 @@ static void upload_state_base_address( struct brw_context *brw ) OUT_BATCH(1); /* Instruction access upper bound */ ADVANCE_BATCH(); } else { - BEGIN_BATCH(6, IGNORE_CLIPRECTS); + BEGIN_BATCH(6); OUT_BATCH(CMD_STATE_BASE_ADDRESS << 16 | (6 - 2)); OUT_BATCH(1); /* General state base address */ OUT_BATCH(1); /* Surface state base address */ diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c index a195bc32b0..5399a74244 100644 --- a/src/mesa/drivers/dri/i965/brw_queryobj.c +++ b/src/mesa/drivers/dri/i965/brw_queryobj.c @@ -188,7 +188,7 @@ brw_emit_query_begin(struct brw_context *brw) if (brw->query.active || is_empty_list(&brw->query.active_head)) return; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_PIPE_CONTROL | PIPE_CONTROL_DEPTH_STALL | PIPE_CONTROL_WRITE_DEPTH_COUNT); @@ -227,7 +227,7 @@ brw_emit_query_end(struct brw_context *brw) if (!brw->query.active) return; - BEGIN_BATCH(4, IGNORE_CLIPRECTS); + BEGIN_BATCH(4); OUT_BATCH(_3DSTATE_PIPE_CONTROL | PIPE_CONTROL_DEPTH_STALL | PIPE_CONTROL_WRITE_DEPTH_COUNT); diff --git a/src/mesa/drivers/dri/i965/brw_sf.c b/src/mesa/drivers/dri/i965/brw_sf.c index 968890f7fb..8e6839b812 100644 --- a/src/mesa/drivers/dri/i965/brw_sf.c +++ b/src/mesa/drivers/dri/i965/brw_sf.c @@ -117,12 +117,13 @@ static void compile_sf_prog( struct brw_context *brw, /* Upload */ dri_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 ); + brw->sf.prog_bo = brw_upload_cache_with_auxdata(&brw->cache, BRW_SF_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + sizeof(c.prog_data), + &brw->sf.prog_data); } /* Calculate interpolants for triangle and line rasterization. diff --git a/src/mesa/drivers/dri/i965/brw_sf_emit.c b/src/mesa/drivers/dri/i965/brw_sf_emit.c index 3eae41ee74..bb08055e3b 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_emit.c +++ b/src/mesa/drivers/dri/i965/brw_sf_emit.c @@ -149,6 +149,7 @@ static void copy_colors( struct brw_sf_compile *c, static void do_flatshade_triangle( struct brw_sf_compile *c ) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg ip = brw_ip_reg(); GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS); GLuint jmpi = 1; @@ -161,7 +162,7 @@ static void do_flatshade_triangle( struct brw_sf_compile *c ) if (c->key.primitive == SF_UNFILLED_TRIS) return; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; brw_push_insn_state(p); @@ -187,6 +188,7 @@ static void do_flatshade_triangle( struct brw_sf_compile *c ) static void do_flatshade_line( struct brw_sf_compile *c ) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg ip = brw_ip_reg(); GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS); GLuint jmpi = 1; @@ -199,7 +201,7 @@ static void do_flatshade_line( struct brw_sf_compile *c ) if (c->key.primitive == SF_UNFILLED_TRIS) return; - if (BRW_IS_IGDNG(p->brw)) + if (intel->is_ironlake) jmpi = 2; brw_push_insn_state(p); diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index b76ed98c4b..8ebbd09e43 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -164,6 +164,7 @@ static dri_bo * sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, dri_bo **reloc_bufs) { + struct intel_context *intel = &brw->intel; struct brw_sf_unit_state sf; dri_bo *bo; int chipset_max_threads; @@ -176,7 +177,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.thread3.dispatch_grf_start_reg = 3; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) sf.thread3.urb_entry_read_offset = 3; else sf.thread3.urb_entry_read_offset = 1; @@ -186,10 +187,10 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, sf.thread4.nr_urb_entries = key->nr_urb_entries; sf.thread4.urb_entry_allocation_size = key->sfsize - 1; - /* Each SF thread produces 1 PUE, and there can be up to 24(Pre-IGDNG) or - * 48(IGDNG) threads + /* Each SF thread produces 1 PUE, and there can be up to 24 (Pre-Ironlake) or + * 48 (Ironlake) threads. */ - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) chipset_max_threads = 48; else chipset_max_threads = 24; @@ -307,8 +308,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key, bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT, key, sizeof(*key), reloc_bufs, 2, - &sf, sizeof(sf), - NULL, NULL); + &sf, sizeof(sf)); /* STATE_PREFETCH command description describes this state as being * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain. diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index 14d5319796..536fe8b249 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -124,16 +124,26 @@ dri_bo *brw_cache_data(struct brw_cache *cache, dri_bo **reloc_bufs, GLuint nr_reloc_bufs); -dri_bo *brw_upload_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_sz, - dri_bo **reloc_bufs, - GLuint nr_reloc_bufs, - const void *data, - GLuint data_sz, - const void *aux, - void *aux_return ); +drm_intel_bo *brw_upload_cache(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_sz, + dri_bo **reloc_bufs, + GLuint nr_reloc_bufs, + const void *data, + GLuint data_sz); + +drm_intel_bo *brw_upload_cache_with_auxdata(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_sz, + dri_bo **reloc_bufs, + GLuint nr_reloc_bufs, + const void *data, + GLuint data_sz, + const void *aux, + GLuint aux_sz, + void *aux_return); dri_bo *brw_search_cache( struct brw_cache *cache, enum brw_cache_id cache_id, @@ -151,7 +161,7 @@ void brw_state_cache_bo_delete(struct brw_cache *cache, dri_bo *bo); /*********************************************************************** * brw_state_batch.c */ -#define BRW_BATCH_STRUCT(brw, s) intel_batchbuffer_data( brw->intel.batch, (s), sizeof(*(s)), IGNORE_CLIPRECTS) +#define BRW_BATCH_STRUCT(brw, s) intel_batchbuffer_data( brw->intel.batch, (s), sizeof(*(s))) #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/mesa/drivers/dri/i965/brw_state_batch.c b/src/mesa/drivers/dri/i965/brw_state_batch.c index 7821898cf9..ed8120d617 100644 --- a/src/mesa/drivers/dri/i965/brw_state_batch.c +++ b/src/mesa/drivers/dri/i965/brw_state_batch.c @@ -48,7 +48,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, struct header *newheader = (struct header *)data; if (brw->emit_state_always) { - intel_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + intel_batchbuffer_data(brw->intel.batch, data, sz); return GL_TRUE; } @@ -75,7 +75,7 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, emit: memcpy(item->header, newheader, sz); - intel_batchbuffer_data(brw->intel.batch, data, sz, IGNORE_CLIPRECTS); + intel_batchbuffer_data(brw->intel.batch, data, sz); return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c index e4c9ba7d87..5fc47b0420 100644 --- a/src/mesa/drivers/dri/i965/brw_state_cache.c +++ b/src/mesa/drivers/dri/i965/brw_state_cache.c @@ -71,25 +71,23 @@ static GLuint -hash_key(const void *key, GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs) +hash_key(struct brw_cache_item *item) { - GLuint *ikey = (GLuint *)key; - GLuint hash = 0, i; + GLuint *ikey = (GLuint *)item->key; + GLuint hash = item->cache_id, i; - assert(key_size % 4 == 0); + assert(item->key_size % 4 == 0); /* I'm sure this can be improved on: */ - for (i = 0; i < key_size/4; i++) { + for (i = 0; i < item->key_size/4; i++) { hash ^= ikey[i]; hash = (hash << 5) | (hash >> 27); } /* Include the BO pointers as key data as well */ - ikey = (GLuint *)reloc_bufs; - key_size = nr_reloc_bufs * sizeof(dri_bo *); - for (i = 0; i < key_size/4; i++) { + ikey = (GLuint *)item->reloc_bufs; + for (i = 0; i < item->nr_reloc_bufs * sizeof(drm_intel_bo *) / 4; i++) { hash ^= ikey[i]; hash = (hash << 5) | (hash >> 27); } @@ -114,11 +112,22 @@ update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id, cache->brw->state.dirty.cache |= 1 << cache_id; } +static int +brw_cache_item_equals(const struct brw_cache_item *a, + const struct brw_cache_item *b) +{ + return a->cache_id == b->cache_id && + a->hash == b->hash && + a->key_size == b->key_size && + (memcmp(a->key, b->key, a->key_size) == 0) && + a->nr_reloc_bufs == b->nr_reloc_bufs && + (memcmp(a->reloc_bufs, b->reloc_bufs, + a->nr_reloc_bufs * sizeof(dri_bo *)) == 0); +} static struct brw_cache_item * -search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, - GLuint hash, const void *key, GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs) +search_cache(struct brw_cache *cache, GLuint hash, + struct brw_cache_item *lookup) { struct brw_cache_item *c; @@ -133,13 +142,7 @@ search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, #endif for (c = cache->items[hash % cache->size]; c; c = c->next) { - if (c->cache_id == cache_id && - c->hash == hash && - c->key_size == key_size && - memcmp(c->key, key, key_size) == 0 && - c->nr_reloc_bufs == nr_reloc_bufs && - memcmp(c->reloc_bufs, reloc_bufs, - nr_reloc_bufs * sizeof(dri_bo *)) == 0) + if (brw_cache_item_equals(lookup, c)) return c; } @@ -182,10 +185,18 @@ brw_search_cache(struct brw_cache *cache, void *aux_return) { struct brw_cache_item *item; - GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); + struct brw_cache_item lookup; + GLuint hash; - item = search_cache(cache, cache_id, hash, key, key_size, - reloc_bufs, nr_reloc_bufs); + lookup.cache_id = cache_id; + lookup.key = key; + lookup.key_size = key_size; + lookup.reloc_bufs = reloc_bufs; + lookup.nr_reloc_bufs = nr_reloc_bufs; + hash = hash_key(&lookup); + lookup.hash = hash; + + item = search_cache(cache, hash, &lookup); if (item == NULL) return NULL; @@ -200,26 +211,34 @@ brw_search_cache(struct brw_cache *cache, } -dri_bo * -brw_upload_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_size, - dri_bo **reloc_bufs, - GLuint nr_reloc_bufs, - const void *data, - GLuint data_size, - const void *aux, - void *aux_return ) +drm_intel_bo * +brw_upload_cache_with_auxdata(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_size, + dri_bo **reloc_bufs, + GLuint nr_reloc_bufs, + const void *data, + GLuint data_size, + const void *aux, + GLuint aux_size, + void *aux_return) { struct brw_cache_item *item = CALLOC_STRUCT(brw_cache_item); - GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); + GLuint hash; GLuint relocs_size = nr_reloc_bufs * sizeof(dri_bo *); - GLuint aux_size = cache->aux_size[cache_id]; void *tmp; dri_bo *bo; int i; + item->cache_id = cache_id; + item->key = key; + item->key_size = key_size; + item->reloc_bufs = reloc_bufs; + item->nr_reloc_bufs = nr_reloc_bufs; + hash = hash_key(item); + item->hash = hash; + /* Create the buffer object to contain the data */ bo = dri_bo_alloc(cache->brw->intel.bufmgr, cache->name[cache_id], data_size, 1 << 6); @@ -229,19 +248,15 @@ brw_upload_cache( struct brw_cache *cache, tmp = _mesa_malloc(key_size + aux_size + relocs_size); memcpy(tmp, key, key_size); - memcpy(tmp + key_size, aux, cache->aux_size[cache_id]); + memcpy(tmp + key_size, aux, aux_size); memcpy(tmp + key_size + aux_size, reloc_bufs, relocs_size); for (i = 0; i < nr_reloc_bufs; i++) { if (reloc_bufs[i] != NULL) dri_bo_reference(reloc_bufs[i]); } - item->cache_id = cache_id; item->key = tmp; - item->hash = hash; - item->key_size = key_size; item->reloc_bufs = tmp + key_size + aux_size; - item->nr_reloc_bufs = nr_reloc_bufs; item->bo = bo; dri_bo_reference(bo); @@ -255,7 +270,6 @@ brw_upload_cache( struct brw_cache *cache, cache->n_items++; if (aux_return) { - assert(cache->aux_size[cache_id]); *(void **)aux_return = (void *)((char *)item->key + item->key_size); } @@ -272,6 +286,23 @@ brw_upload_cache( struct brw_cache *cache, return bo; } +drm_intel_bo * +brw_upload_cache(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_size, + dri_bo **reloc_bufs, + GLuint nr_reloc_bufs, + const void *data, + GLuint data_size) +{ + return brw_upload_cache_with_auxdata(cache, cache_id, + key, key_size, + reloc_bufs, nr_reloc_bufs, + data, data_size, + NULL, 0, + NULL); +} /** * Wrapper around brw_cache_data_sz using the cache_id's canonical key size. @@ -292,11 +323,18 @@ brw_cache_data(struct brw_cache *cache, GLuint nr_reloc_bufs) { dri_bo *bo; - struct brw_cache_item *item; - GLuint hash = hash_key(data, data_size, reloc_bufs, nr_reloc_bufs); - - item = search_cache(cache, cache_id, hash, data, data_size, - reloc_bufs, nr_reloc_bufs); + struct brw_cache_item *item, lookup; + GLuint hash; + + lookup.cache_id = cache_id; + lookup.key = data; + lookup.key_size = data_size; + lookup.reloc_bufs = reloc_bufs; + lookup.nr_reloc_bufs = nr_reloc_bufs; + hash = hash_key(&lookup); + lookup.hash = hash; + + item = search_cache(cache, hash, &lookup); if (item) { update_cache_last(cache, cache_id, item->bo); dri_bo_reference(item->bo); @@ -306,8 +344,7 @@ brw_cache_data(struct brw_cache *cache, bo = brw_upload_cache(cache, cache_id, data, data_size, reloc_bufs, nr_reloc_bufs, - data, data_size, - NULL, NULL); + data, data_size); return bo; } @@ -321,11 +358,9 @@ enum pool_type { static void brw_init_cache_id(struct brw_cache *cache, const char *name, - enum brw_cache_id id, - GLuint aux_size) + enum brw_cache_id id) { cache->name[id] = strdup(name); - cache->aux_size[id] = aux_size; } @@ -341,80 +376,28 @@ brw_init_non_surface_cache(struct brw_context *brw) cache->items = (struct brw_cache_item **) _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); - brw_init_cache_id(cache, - "CC_VP", - BRW_CC_VP, - 0); - - brw_init_cache_id(cache, - "CC_UNIT", - BRW_CC_UNIT, - 0); - - brw_init_cache_id(cache, - "WM_PROG", - BRW_WM_PROG, - sizeof(struct brw_wm_prog_data)); - - brw_init_cache_id(cache, - "SAMPLER_DEFAULT_COLOR", - BRW_SAMPLER_DEFAULT_COLOR, - 0); - - brw_init_cache_id(cache, - "SAMPLER", - BRW_SAMPLER, - 0); - - brw_init_cache_id(cache, - "WM_UNIT", - BRW_WM_UNIT, - 0); - - brw_init_cache_id(cache, - "SF_PROG", - BRW_SF_PROG, - sizeof(struct brw_sf_prog_data)); - - brw_init_cache_id(cache, - "SF_VP", - BRW_SF_VP, - 0); - - brw_init_cache_id(cache, - "SF_UNIT", - BRW_SF_UNIT, - 0); - - brw_init_cache_id(cache, - "VS_UNIT", - BRW_VS_UNIT, - 0); - - brw_init_cache_id(cache, - "VS_PROG", - BRW_VS_PROG, - sizeof(struct brw_vs_prog_data)); - - brw_init_cache_id(cache, - "CLIP_UNIT", - BRW_CLIP_UNIT, - 0); - - brw_init_cache_id(cache, - "CLIP_PROG", - BRW_CLIP_PROG, - sizeof(struct brw_clip_prog_data)); - - brw_init_cache_id(cache, - "GS_UNIT", - BRW_GS_UNIT, - 0); - - brw_init_cache_id(cache, - "GS_PROG", - BRW_GS_PROG, - sizeof(struct brw_gs_prog_data)); + brw_init_cache_id(cache, "CC_VP", BRW_CC_VP); + brw_init_cache_id(cache, "CC_UNIT", BRW_CC_UNIT); + brw_init_cache_id(cache, "WM_PROG", BRW_WM_PROG); + brw_init_cache_id(cache, "SAMPLER_DEFAULT_COLOR", BRW_SAMPLER_DEFAULT_COLOR); + brw_init_cache_id(cache, "SAMPLER", BRW_SAMPLER); + brw_init_cache_id(cache, "WM_UNIT", BRW_WM_UNIT); + brw_init_cache_id(cache, "SF_PROG", BRW_SF_PROG); + brw_init_cache_id(cache, "SF_VP", BRW_SF_VP); + + brw_init_cache_id(cache, "SF_UNIT", BRW_SF_UNIT); + + brw_init_cache_id(cache, "VS_UNIT", BRW_VS_UNIT); + + brw_init_cache_id(cache, "VS_PROG", BRW_VS_PROG); + + brw_init_cache_id(cache, "CLIP_UNIT", BRW_CLIP_UNIT); + + brw_init_cache_id(cache, "CLIP_PROG", BRW_CLIP_PROG); + + brw_init_cache_id(cache, "GS_UNIT", BRW_GS_UNIT); + + brw_init_cache_id(cache, "GS_PROG", BRW_GS_PROG); } @@ -430,15 +413,8 @@ brw_init_surface_cache(struct brw_context *brw) cache->items = (struct brw_cache_item **) _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); - brw_init_cache_id(cache, - "SS_SURFACE", - BRW_SS_SURFACE, - 0); - - brw_init_cache_id(cache, - "SS_SURF_BIND", - BRW_SS_SURF_BIND, - 0); + brw_init_cache_id(cache, "SS_SURFACE", BRW_SS_SURFACE); + brw_init_cache_id(cache, "SS_SURF_BIND", BRW_SS_SURF_BIND); } diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index af8dfb4c15..0ecbef1ef9 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -36,13 +36,7 @@ #include "intel_batchbuffer.h" #include "intel_buffers.h" -/* This is used to initialize brw->state.atoms[]. We could use this - * list directly except for a single atom, brw_constant_buffer, which - * has a .dirty value which changes according to the parameters of the - * current fragment and vertex programs, and so cannot be a static - * value. - */ -const struct brw_tracked_state *atoms[] = +static const struct brw_tracked_state *atoms[] = { &brw_check_fallback, @@ -208,7 +202,6 @@ static struct dirty_bit_map brw_bits[] = { DEFINE_BIT(BRW_NEW_CONTEXT), DEFINE_BIT(BRW_NEW_WM_INPUT_DIMENSIONS), DEFINE_BIT(BRW_NEW_PSP), - DEFINE_BIT(BRW_NEW_FENCE), DEFINE_BIT(BRW_NEW_INDICES), DEFINE_BIT(BRW_NEW_INDEX_BUFFER), DEFINE_BIT(BRW_NEW_VERTICES), diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index e59e52ed86..64a9535282 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -49,7 +49,7 @@ GLboolean brw_miptree_layout(struct intel_context *intel, switch (mt->target) { case GL_TEXTURE_CUBE_MAP: - if (IS_IGDNG(intel->intelScreen->deviceID)) { + if (intel->is_ironlake) { GLuint align_h = 2, align_w = 4; GLuint level; GLuint x = 0; diff --git a/src/mesa/drivers/dri/i965/brw_urb.c b/src/mesa/drivers/dri/i965/brw_urb.c index 8c6f4355a6..f2cdb203b8 100644 --- a/src/mesa/drivers/dri/i965/brw_urb.c +++ b/src/mesa/drivers/dri/i965/brw_urb.c @@ -105,7 +105,8 @@ static GLboolean check_urb_layout( struct brw_context *brw ) brw->urb.sf_start = brw->urb.clip_start + brw->urb.nr_clip_entries * brw->urb.vsize; brw->urb.cs_start = brw->urb.sf_start + brw->urb.nr_sf_entries * brw->urb.sfsize; - return brw->urb.cs_start + brw->urb.nr_cs_entries * brw->urb.csize <= URB_SIZES(brw); + return brw->urb.cs_start + brw->urb.nr_cs_entries * + brw->urb.csize <= brw->urb.size; } /* Most minimal update, forces re-emit of URB fence packet after GS @@ -113,6 +114,7 @@ static GLboolean check_urb_layout( struct brw_context *brw ) */ static void recalculate_urb_fence( struct brw_context *brw ) { + struct intel_context *intel = &brw->intel; GLuint csize = brw->curbe.total_size; GLuint vsize = brw->vs.prog_data->urb_entry_size; GLuint sfsize = brw->sf.prog_data->urb_entry_size; @@ -146,7 +148,7 @@ static void recalculate_urb_fence( struct brw_context *brw ) brw->urb.constrained = 0; - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { brw->urb.nr_vs_entries = 128; brw->urb.nr_sf_entries = 48; if (check_urb_layout(brw)) { @@ -156,7 +158,7 @@ static void recalculate_urb_fence( struct brw_context *brw ) brw->urb.nr_vs_entries = limits[VS].preferred_nr_entries; brw->urb.nr_sf_entries = limits[SF].preferred_nr_entries; } - } else if (BRW_IS_G4X(brw)) { + } else if (intel->is_g4x) { brw->urb.nr_vs_entries = 64; if (check_urb_layout(brw)) { goto done; @@ -200,7 +202,7 @@ done: brw->urb.clip_start, brw->urb.sf_start, brw->urb.cs_start, - URB_SIZES(brw)); + brw->urb.size); brw->state.dirty.brw |= BRW_NEW_URB_FENCE; } @@ -244,7 +246,7 @@ void brw_upload_urb_fence(struct brw_context *brw) uf.bits0.gs_fence = brw->urb.clip_start; uf.bits0.clp_fence = brw->urb.sf_start; uf.bits1.sf_fence = brw->urb.cs_start; - uf.bits1.cs_fence = URB_SIZES(brw); + uf.bits1.cs_fence = brw->urb.size; BRW_BATCH_STRUCT(brw, &uf); } diff --git a/src/mesa/drivers/dri/i965/brw_vs.c b/src/mesa/drivers/dri/i965/brw_vs.c index fd055e225e..44b085e214 100644 --- a/src/mesa/drivers/dri/i965/brw_vs.c +++ b/src/mesa/drivers/dri/i965/brw_vs.c @@ -35,6 +35,7 @@ #include "brw_util.h" #include "brw_state.h" #include "shader/prog_print.h" +#include "shader/prog_parameter.h" @@ -42,9 +43,11 @@ static void do_vs_prog( struct brw_context *brw, struct brw_vertex_program *vp, struct brw_vs_prog_key *key ) { + GLcontext *ctx = &brw->intel.ctx; GLuint program_size; const GLuint *program; struct brw_vs_compile c; + int aux_size; memset(&c, 0, sizeof(c)); memcpy(&c.key, key, sizeof(*key)); @@ -73,13 +76,27 @@ static void do_vs_prog( struct brw_context *brw, */ program = brw_get_program(&c.func, &program_size); + /* We upload from &c.prog_data including the constant_map assuming + * they're packed together. It would be nice to have a + * compile-time assert macro here. + */ + assert(c.constant_map == (int8_t *)&c.prog_data + + sizeof(c.prog_data)); + assert(ctx->Const.VertexProgram.MaxNativeParameters == + ARRAY_SIZE(c.constant_map)); + + aux_size = sizeof(c.prog_data); + if (c.vp->use_const_buffer) + aux_size += c.vp->program.Base.Parameters->NumParameters; + dri_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 ); + brw->vs.prog_bo = brw_upload_cache_with_auxdata(&brw->cache, BRW_VS_PROG, + &c.key, sizeof(c.key), + NULL, 0, + program, program_size, + &c.prog_data, + aux_size, + &brw->vs.prog_data); } @@ -109,6 +126,8 @@ static void brw_upload_vs_prog(struct brw_context *brw) &brw->vs.prog_data); if (brw->vs.prog_bo == NULL) do_vs_prog(brw, vp, &key); + brw->vs.constant_map = ((int8_t *)brw->vs.prog_data + + sizeof(*brw->vs.prog_data)); } diff --git a/src/mesa/drivers/dri/i965/brw_vs.h b/src/mesa/drivers/dri/i965/brw_vs.h index 4a591365c9..95e0501b1e 100644 --- a/src/mesa/drivers/dri/i965/brw_vs.h +++ b/src/mesa/drivers/dri/i965/brw_vs.h @@ -51,6 +51,7 @@ struct brw_vs_compile { struct brw_compile func; struct brw_vs_prog_key key; struct brw_vs_prog_data prog_data; + int8_t constant_map[1024]; struct brw_vertex_program *vp; @@ -81,6 +82,8 @@ struct brw_vs_compile { GLint index; struct brw_reg reg; } current_const[3]; + + GLboolean needs_stack; }; void brw_vs_emit( struct brw_vs_compile *c ); diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index 27aac8b54a..52cc04fee8 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -67,6 +67,7 @@ static void release_tmps( struct brw_vs_compile *c ) */ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { + struct intel_context *intel = &c->func.brw->intel; GLuint i, reg = 0, mrf; int attributes_in_vue; @@ -103,9 +104,47 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) /* Vertex program parameters from curbe: */ if (c->vp->use_const_buffer) { - /* get constants from a real constant buffer */ - c->prog_data.curb_read_length = 0; - c->prog_data.nr_params = 4; /* XXX 0 causes a bug elsewhere... */ + int max_constant = BRW_MAX_GRF - 20 - c->vp->program.Base.NumTemporaries; + int constant = 0; + + /* We've got more constants than we can load with the push + * mechanism. This is often correlated with reladdr loads where + * we should probably be using a pull mechanism anyway to avoid + * excessive reading. However, the pull mechanism is slow in + * general. So, we try to allocate as many non-reladdr-loaded + * constants through the push buffer as we can before giving up. + */ + memset(c->constant_map, -1, c->vp->program.Base.Parameters->NumParameters); + for (i = 0; + i < c->vp->program.Base.NumInstructions && constant < max_constant; + i++) { + struct prog_instruction *inst = &c->vp->program.Base.Instructions[i]; + int arg; + + for (arg = 0; arg < 3 && constant < max_constant; arg++) { + if ((inst->SrcReg[arg].File != PROGRAM_STATE_VAR && + inst->SrcReg[arg].File != PROGRAM_CONSTANT && + inst->SrcReg[arg].File != PROGRAM_UNIFORM && + inst->SrcReg[arg].File != PROGRAM_ENV_PARAM && + inst->SrcReg[arg].File != PROGRAM_LOCAL_PARAM) || + inst->SrcReg[arg].RelAddr) + continue; + + if (c->constant_map[inst->SrcReg[arg].Index] == -1) { + c->constant_map[inst->SrcReg[arg].Index] = constant++; + } + } + } + + for (i = 0; i < constant; i++) { + c->regs[PROGRAM_STATE_VAR][i] = stride( brw_vec4_grf(reg+i/2, + (i%2) * 4), + 0, 4, 1); + } + reg += (constant + 1) / 2; + c->prog_data.curb_read_length = reg - 1; + /* XXX 0 causes a bug elsewhere... */ + c->prog_data.nr_params = MAX2(constant * 4, 4); } else { /* use a section of the GRF for constants */ @@ -141,7 +180,7 @@ 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 (intel->is_ironlake) mrf = 8; else mrf = 4; @@ -213,8 +252,10 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) } } - c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg, 0); - reg += 2; + if (c->needs_stack) { + c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg, 0); + reg += 2; + } /* Some opcodes need an internal temporary: */ @@ -238,7 +279,7 @@ 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 (intel->is_ironlake) 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; @@ -761,15 +802,14 @@ get_constant(struct brw_vs_compile *c, { const struct prog_src_register *src = &inst->SrcReg[argIndex]; struct brw_compile *p = &c->func; - struct brw_reg const_reg; - struct brw_reg const2_reg; - const GLboolean relAddr = src->RelAddr; + struct brw_reg const_reg = c->current_const[argIndex].reg; assert(argIndex < 3); - if (c->current_const[argIndex].index != src->Index || relAddr) { + if (c->current_const[argIndex].index != src->Index) { struct brw_reg addrReg = c->regs[PROGRAM_ADDRESS][0]; + /* Keep track of the last constant loaded in this slot, for reuse. */ c->current_const[argIndex].index = src->Index; #if 0 @@ -778,48 +818,74 @@ get_constant(struct brw_vs_compile *c, #endif /* need to fetch the constant now */ brw_dp_READ_4_vs(p, - c->current_const[argIndex].reg,/* writeback dest */ + const_reg, /* writeback dest */ 0, /* oword */ - relAddr, /* relative indexing? */ + 0, /* relative indexing? */ addrReg, /* address register */ 16 * src->Index, /* byte offset */ SURF_INDEX_VERT_CONST_BUFFER /* binding table index */ ); - - if (relAddr) { - /* second read */ - const2_reg = get_tmp(c); - - /* use upper half of address reg for second read */ - addrReg = stride(addrReg, 0, 4, 0); - addrReg.subnr = 16; - - brw_dp_READ_4_vs(p, - const2_reg, /* writeback dest */ - 1, /* oword */ - relAddr, /* relative indexing? */ - addrReg, /* address register */ - 16 * src->Index, /* byte offset */ - SURF_INDEX_VERT_CONST_BUFFER - ); - } } - const_reg = c->current_const[argIndex].reg; + /* replicate lower four floats into upper half (to get XYZWXYZW) */ + const_reg = stride(const_reg, 0, 4, 0); + const_reg.subnr = 0; - if (relAddr) { - /* merge the two Owords into the constant register */ - /* const_reg[7..4] = const2_reg[7..4] */ - brw_MOV(p, - suboffset(stride(const_reg, 0, 4, 1), 4), - suboffset(stride(const2_reg, 0, 4, 1), 4)); - release_tmp(c, const2_reg); - } - else { - /* replicate lower four floats into upper half (to get XYZWXYZW) */ - const_reg = stride(const_reg, 0, 4, 0); - const_reg.subnr = 0; - } + return const_reg; +} + +static struct brw_reg +get_reladdr_constant(struct brw_vs_compile *c, + const struct prog_instruction *inst, + GLuint argIndex) +{ + const struct prog_src_register *src = &inst->SrcReg[argIndex]; + struct brw_compile *p = &c->func; + struct brw_reg const_reg = c->current_const[argIndex].reg; + struct brw_reg const2_reg; + struct brw_reg addrReg = c->regs[PROGRAM_ADDRESS][0]; + + assert(argIndex < 3); + + /* Can't reuse a reladdr constant load. */ + c->current_const[argIndex].index = -1; + + #if 0 + printf(" fetch const[a0.x+%d] for arg %d into reg %d\n", + src->Index, argIndex, c->current_const[argIndex].reg.nr); +#endif + + /* fetch the first vec4 */ + brw_dp_READ_4_vs(p, + const_reg, /* writeback dest */ + 0, /* oword */ + 1, /* relative indexing? */ + addrReg, /* address register */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_VERT_CONST_BUFFER /* binding table index */ + ); + /* second vec4 */ + const2_reg = get_tmp(c); + + /* use upper half of address reg for second read */ + addrReg = stride(addrReg, 0, 4, 0); + addrReg.subnr = 16; + + brw_dp_READ_4_vs(p, + const2_reg, /* writeback dest */ + 1, /* oword */ + 1, /* relative indexing? */ + addrReg, /* address register */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_VERT_CONST_BUFFER + ); + + /* merge the two Owords into the constant register */ + /* const_reg[7..4] = const2_reg[7..4] */ + brw_MOV(p, + suboffset(stride(const_reg, 0, 4, 1), 4), + suboffset(stride(const2_reg, 0, 4, 1), 4)); + release_tmp(c, const2_reg); return const_reg; } @@ -927,7 +993,13 @@ get_src_reg( struct brw_vs_compile *c, case PROGRAM_ENV_PARAM: case PROGRAM_LOCAL_PARAM: if (c->vp->use_const_buffer) { - return get_constant(c, inst, argIndex); + if (!relAddr && c->constant_map[index] != -1) { + assert(c->regs[PROGRAM_STATE_VAR][c->constant_map[index]].nr != 0); + return c->regs[PROGRAM_STATE_VAR][c->constant_map[index]]; + } else if (relAddr) + return get_reladdr_constant(c, inst, argIndex); + else + return get_constant(c, inst, argIndex); } else if (relAddr) { return deref(c, c->regs[PROGRAM_STATE_VAR][0], index); @@ -1113,6 +1185,8 @@ static void emit_swz( struct brw_vs_compile *c, static void emit_vertex_write( struct brw_vs_compile *c) { struct brw_compile *p = &c->func; + struct brw_context *brw = p->brw; + struct intel_context *intel = &brw->intel; struct brw_reg m0 = brw_message_reg(0); struct brw_reg pos = c->regs[PROGRAM_OUTPUT][VERT_RESULT_HPOS]; struct brw_reg ndc; @@ -1136,7 +1210,7 @@ static void emit_vertex_write( struct brw_vs_compile *c) * workaround. */ if ((c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_PSIZ)) || - c->key.nr_userclip || BRW_IS_965(p->brw)) + c->key.nr_userclip || brw->has_negative_rhw_bug) { struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD); GLuint i; @@ -1167,7 +1241,7 @@ 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 (brw->has_negative_rhw_bug) { brw_CMP(p, vec8(brw_null_reg()), BRW_CONDITIONAL_L, @@ -1195,8 +1269,8 @@ 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)) { - /* There are 20 DWs (D0-D19) in VUE vertex header on IGDNG */ + if (intel->is_ironlake) { + /* There are 20 DWs (D0-D19) in VUE vertex header on Ironlake */ brw_MOV(p, offset(m0, 3), pos); /* a portion of vertex header */ /* m4, m5 contain the distances from vertex to the user clip planeXXX. * Seems it is useless for us. @@ -1359,6 +1433,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) #define MAX_LOOP_DEPTH 32 struct brw_compile *p = &c->func; struct brw_context *brw = p->brw; + struct intel_context *intel = &brw->intel; const GLuint nr_insns = c->vp->program.Base.NumInstructions; GLuint insn, if_depth = 0, loop_depth = 0; GLuint end_offset = 0; @@ -1376,12 +1451,14 @@ void brw_vs_emit(struct brw_vs_compile *c ) brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_set_access_mode(p, BRW_ALIGN_16); - - /* Message registers can't be read, so copy the output into GRF register - if they are used in source registers */ + for (insn = 0; insn < nr_insns; insn++) { GLuint i; struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn]; + + /* Message registers can't be read, so copy the output into GRF + * register if they are used in source registers + */ for (i = 0; i < 3; i++) { struct prog_src_register *src = &inst->SrcReg[i]; GLuint index = src->Index; @@ -1389,12 +1466,23 @@ void brw_vs_emit(struct brw_vs_compile *c ) if (file == PROGRAM_OUTPUT && index != VERT_RESULT_HPOS) c->output_regs[index].used_in_src = GL_TRUE; } + + switch (inst->Opcode) { + case OPCODE_CAL: + case OPCODE_RET: + c->needs_stack = GL_TRUE; + break; + default: + break; + } } /* Static register allocation */ brw_vs_alloc_regs(c); - brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); + + if (c->needs_stack) + brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); for (insn = 0; insn < nr_insns; insn++) { @@ -1592,7 +1680,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) loop_depth--; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) br = 2; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); diff --git a/src/mesa/drivers/dri/i965/brw_vs_state.c b/src/mesa/drivers/dri/i965/brw_vs_state.c index 7285466645..fd9f2fee42 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_state.c +++ b/src/mesa/drivers/dri/i965/brw_vs_state.c @@ -82,9 +82,9 @@ vs_unit_populate_key(struct brw_context *brw, struct brw_vs_unit_key *key) static dri_bo * vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) { + struct intel_context *intel = &brw->intel; struct brw_vs_unit_state vs; dri_bo *bo; - int chipset_max_threads; memset(&vs, 0, sizeof(vs)); @@ -98,7 +98,7 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) */ vs.thread1.single_program_flow = 0; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) vs.thread1.binding_table_entry_count = 0; /* hardware requirement */ else vs.thread1.binding_table_entry_count = key->nr_surfaces; @@ -109,7 +109,7 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) vs.thread3.urb_entry_read_offset = 0; vs.thread3.const_urb_entry_read_offset = key->curbe_offset * 2; - if (BRW_IS_IGDNG(brw)) { + if (intel->is_ironlake) { switch (key->nr_urb_entries) { case 8: case 12: @@ -135,7 +135,7 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) case 32: break; case 64: - assert(BRW_IS_G4X(brw)); + assert(intel->is_g4x); break; default: assert(0); @@ -145,17 +145,8 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) vs.thread4.urb_entry_allocation_size = key->urb_size - 1; - if (BRW_IS_IGDNG(brw)) - chipset_max_threads = 72; - else if (BRW_IS_G4X(brw)) - chipset_max_threads = 32; - else - chipset_max_threads = 16; vs.thread4.max_threads = CLAMP(key->nr_urb_entries / 2, - 1, chipset_max_threads) - 1; - - if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) - vs.thread4.max_threads = 0; + 1, brw->vs_max_threads) - 1; /* No samplers for ARB_vp programs: */ @@ -173,8 +164,7 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) bo = brw_upload_cache(&brw->cache, BRW_VS_UNIT, key, sizeof(*key), &brw->vs.prog_bo, 1, - &vs, sizeof(vs), - NULL, NULL); + &vs, sizeof(vs)); /* Emit VS program relocation */ dri_bo_emit_reloc(bo, diff --git a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c index 22ff3410ae..ca93ca7506 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c @@ -167,8 +167,7 @@ brw_vs_get_binding_table(struct brw_context *brw) 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); + data, data_size); /* Emit binding table relocations to surface state */ for (i = 0; i < BRW_VS_MAX_SURF; i++) { diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c index daf1964553..e08f32c4e2 100644 --- a/src/mesa/drivers/dri/i965/brw_vtbl.c +++ b/src/mesa/drivers/dri/i965/brw_vtbl.c @@ -149,9 +149,6 @@ static void brw_new_batch( struct intel_context *intel ) { struct brw_context *brw = brw_context(&intel->ctx); - /* Check that we didn't just wrap our batchbuffer at a bad time. */ - assert(!brw->no_batch_wrap); - brw->curbe.need_new_bo = GL_TRUE; /* Mark all context state as needing to be re-emitted. @@ -174,12 +171,6 @@ static void brw_new_batch( struct intel_context *intel ) } } - -static void brw_note_fence( struct intel_context *intel, GLuint fence ) -{ - brw_context(&intel->ctx)->state.dirty.brw |= BRW_NEW_FENCE; -} - static void brw_invalidate_state( struct intel_context *intel, GLuint new_state ) { /* nothing */ @@ -195,7 +186,6 @@ void brwInitVtbl( struct brw_context *brw ) brw->intel.vtbl.update_texture_state = 0; brw->intel.vtbl.invalidate_state = brw_invalidate_state; - brw->intel.vtbl.note_fence = brw_note_fence; brw->intel.vtbl.new_batch = brw_new_batch; brw->intel.vtbl.finish_batch = brw_finish_batch; brw->intel.vtbl.destroy = brw_destroy_context; diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 802236e8e3..8afd1f4aef 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -198,12 +198,13 @@ static void do_wm_prog( struct brw_context *brw, program = brw_get_program(&c->func, &program_size); dri_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 ); + brw->wm.prog_bo = brw_upload_cache_with_auxdata(&brw->cache, BRW_WM_PROG, + &c->key, sizeof(c->key), + NULL, 0, + program, program_size, + &c->prog_data, + sizeof(c->prog_data), + &brw->wm.prog_data); } diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index 9dcb6e14bb..b9b987ea70 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -76,10 +76,10 @@ struct brw_wm_prog_key { GLushort tex_swizzles[BRW_MAX_TEX_UNIT]; - GLuint program_string_id:32; GLushort origin_x, origin_y; GLushort drawable_height; GLbitfield64 vp_outputs_written; + GLuint program_string_id:32; }; diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index c6d10f5570..f316e0cda4 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -830,6 +830,7 @@ void emit_tex(struct brw_wm_compile *c, GLboolean shadow) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; struct brw_reg dst_retyped; GLuint cur_mrf = 2, response_length; GLuint i, nr_texcoords; @@ -873,7 +874,7 @@ void emit_tex(struct brw_wm_compile *c, } /* Pre-Ironlake, the 8-wide sampler always took u,v,r. */ - if (!BRW_IS_IGDNG(p->brw) && c->dispatch_width == 8) + if (!intel->is_ironlake && c->dispatch_width == 8) nr_texcoords = 3; /* For shadow comparisons, we have to supply u,v,r. */ @@ -891,7 +892,7 @@ void emit_tex(struct brw_wm_compile *c, /* Fill in the shadow comparison reference value. */ if (shadow) { - if (BRW_IS_IGDNG(p->brw)) { + if (intel->is_ironlake) { /* Fill in the cube map array index value. */ brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0)); cur_mrf += mrf_per_channel; @@ -904,7 +905,7 @@ void emit_tex(struct brw_wm_compile *c, cur_mrf += mrf_per_channel; } - if (BRW_IS_IGDNG(p->brw)) { + if (intel->is_ironlake) { if (shadow) msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_IGDNG; else @@ -944,6 +945,7 @@ void emit_txb(struct brw_wm_compile *c, GLuint sampler) { struct brw_compile *p = &c->func; + struct intel_context *intel = &p->brw->intel; GLuint msgLength; GLuint msg_type; GLuint mrf_per_channel; @@ -955,8 +957,8 @@ void emit_txb(struct brw_wm_compile *c, * undefined, and trust the execution mask to keep the undefined pixels * from mattering. */ - if (c->dispatch_width == 16 || !BRW_IS_IGDNG(p->brw)) { - if (BRW_IS_IGDNG(p->brw)) + if (c->dispatch_width == 16 || !intel->is_ironlake) { + if (intel->is_ironlake) msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_IGDNG; else msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS; @@ -1174,7 +1176,7 @@ void emit_fb_write(struct brw_wm_compile *c, brw_push_insn_state(p); for (channel = 0; channel < 4; channel++) { - if (c->dispatch_width == 16 && (BRW_IS_G4X(brw) || BRW_IS_IGDNG(brw))) { + if (c->dispatch_width == 16 && brw->has_compr4) { /* By setting the high bit of the MRF register number, we indicate * that we want COMPR4 mode - instead of doing the usual destination * + 1 for the second half we get destination + 4. diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index 7d03179588..3737faf26f 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -138,7 +138,6 @@ static struct prog_dst_register dst_reg(GLuint file, GLuint idx) reg.CondMask = COND_TR; reg.CondSwizzle = 0; reg.CondSrc = 0; - reg.pad = 0; return reg; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 7e5533e258..fde83eea62 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -1826,6 +1826,7 @@ get_argument_regs(struct brw_wm_compile *c, static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) { + struct intel_context *intel = &brw->intel; #define MAX_IF_DEPTH 32 #define MAX_LOOP_DEPTH 32 struct brw_instruction *if_inst[MAX_IF_DEPTH], *loop_inst[MAX_LOOP_DEPTH]; @@ -1876,10 +1877,6 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) else brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); - dst_flags = inst->DstReg.WriteMask; - if (inst->SaturateMode == SATURATE_ZERO_ONE) - dst_flags |= SATURATE; - switch (inst->Opcode) { case WM_PIXELXY: emit_pixel_xy(c, dst, dst_flags); @@ -2043,6 +2040,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) if_inst[if_depth++] = brw_IF(p, BRW_EXECUTE_8); break; case OPCODE_ELSE: + assert(if_depth > 0); if_inst[if_depth-1] = brw_ELSE(p, if_inst[if_depth-1]); break; case OPCODE_ENDIF: @@ -2096,9 +2094,10 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) struct brw_instruction *inst0, *inst1; GLuint br = 1; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) br = 2; - + + assert(loop_depth > 0); loop_depth--; inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]); /* patch all the BREAK/CONT instructions from last BGNLOOP */ diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index aa2e519588..87387b1e2d 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -262,10 +262,10 @@ brw_wm_sampler_populate_key(struct brw_context *brw, dri_bo_unreference(brw->wm.sdc_bo[unit]); if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) { float bordercolor[4] = { - texObj->BorderColor[0], - texObj->BorderColor[0], - texObj->BorderColor[0], - texObj->BorderColor[0] + texObj->BorderColor.f[0], + texObj->BorderColor.f[0], + texObj->BorderColor.f[0], + texObj->BorderColor.f[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 @@ -274,7 +274,7 @@ brw_wm_sampler_populate_key(struct brw_context *brw, brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor); } else { brw->wm.sdc_bo[unit] = upload_default_color(brw, - texObj->BorderColor); + texObj->BorderColor.f); } key->sampler_count = unit + 1; } @@ -326,8 +326,7 @@ static void upload_wm_samplers( struct brw_context *brw ) brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER, &key, sizeof(key), brw->wm.sdc_bo, key.sampler_count, - &sampler, sizeof(sampler), - NULL, NULL); + &sampler, sizeof(sampler)); /* Emit SDC relocations */ for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c index f89ed9bce7..a7f80db554 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_state.c @@ -49,8 +49,6 @@ struct brw_wm_unit_key { unsigned int curbe_offset; unsigned int urb_size; - unsigned int max_threads; - unsigned int nr_surfaces, sampler_count; GLboolean uses_depth, computes_depth, uses_kill, is_glsl; GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable; @@ -67,18 +65,6 @@ wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) memset(key, 0, sizeof(*key)); - if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) - key->max_threads = 1; - else { - /* WM maximum threads is number of EUs times number of threads per EU. */ - if (BRW_IS_IGDNG(brw)) - key->max_threads = 12 * 6; - else if (BRW_IS_G4X(brw)) - key->max_threads = 10 * 5; - else - key->max_threads = 8 * 4; - } - /* CACHE_NEW_WM_PROG */ key->total_grf = brw->wm.prog_data->total_grf; key->urb_entry_read_length = brw->wm.prog_data->urb_read_length; @@ -140,6 +126,7 @@ static dri_bo * wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, dri_bo **reloc_bufs) { + struct intel_context *intel = &brw->intel; struct brw_wm_unit_state wm; dri_bo *bo; @@ -150,7 +137,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, wm.thread1.depth_coef_urb_read_offset = 1; wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) wm.thread1.binding_table_entry_count = 0; /* hardware requirement */ else wm.thread1.binding_table_entry_count = key->nr_surfaces; @@ -170,7 +157,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, wm.thread3.const_urb_entry_read_length = key->curb_entry_read_length; wm.thread3.const_urb_entry_read_offset = key->curbe_offset * 2; - if (BRW_IS_IGDNG(brw)) + if (intel->is_ironlake) wm.wm4.sampler_count = 0; /* hardware requirement */ else wm.wm4.sampler_count = (key->sampler_count + 1) / 4; @@ -191,7 +178,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, else wm.wm5.enable_16_pix = 1; - wm.wm5.max_threads = key->max_threads - 1; + wm.wm5.max_threads = brw->wm_max_threads - 1; wm.wm5.thread_dispatch_enable = 1; /* AKA: color_write */ wm.wm5.legacy_line_rast = 0; wm.wm5.legacy_global_depth_bias = 0; @@ -223,8 +210,7 @@ wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, bo = brw_upload_cache(&brw->cache, BRW_WM_UNIT, key, sizeof(*key), reloc_bufs, 3, - &wm, sizeof(wm), - NULL, NULL); + &wm, sizeof(wm)); /* Emit WM program relocation */ dri_bo_emit_reloc(bo, @@ -268,7 +254,7 @@ static void upload_wm_unit( struct brw_context *brw ) */ assert(key.total_scratch <= 12 * 1024); if (key.total_scratch) { - GLuint total = key.total_scratch * key.max_threads; + GLuint total = key.total_scratch * brw->wm_max_threads; if (brw->wm.scratch_bo && total > brw->wm.scratch_bo->size) { dri_bo_unreference(brw->wm.scratch_bo); diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index afb36b7bd7..357c8c90de 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -256,8 +256,7 @@ brw_create_texture_surface( struct brw_context *brw, bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, - &surf, sizeof(surf), - NULL, NULL); + &surf, sizeof(surf)); if (key->bo) { /* Emit relocation to surface contents */ @@ -351,8 +350,7 @@ brw_create_constant_surface( struct brw_context *brw, bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, - &surf, sizeof(surf), - NULL, NULL); + &surf, sizeof(surf)); if (key->bo) { /* Emit relocation to surface contents. Section 5.1.1 of the gen4 @@ -511,7 +509,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw, struct gl_renderbuffer *rb, unsigned int unit) { - GLcontext *ctx = &brw->intel.ctx; + struct intel_context *intel = &brw->intel;; + GLcontext *ctx = &intel->ctx; dri_bo *region_bo = NULL; struct intel_renderbuffer *irb = intel_renderbuffer(rb); struct intel_region *region = irb ? irb->region : NULL; @@ -578,7 +577,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw, key.draw_y = 0; } /* _NEW_COLOR */ - memcpy(key.color_mask, ctx->Color.ColorMask, + memcpy(key.color_mask, ctx->Color.ColorMask[0], sizeof(key.color_mask)); /* As mentioned above, disable writes to the alpha component when the @@ -622,7 +621,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw, tile_base = ((key.draw_y / 32) * (32 * pitch)); tile_base += (key.draw_x - tile_x) / (128 / key.cpp) * 4096; } - assert(BRW_IS_G4X(brw) || (tile_x == 0 && tile_y == 0)); + assert(intel->is_g4x || (tile_x == 0 && tile_y == 0)); assert(tile_x % 4 == 0); assert(tile_y % 2 == 0); /* Note that the low bits of these fields are missing, so @@ -652,8 +651,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw, BRW_SS_SURFACE, &key, sizeof(key), ®ion_bo, 1, - &surf, sizeof(surf), - NULL, NULL); + &surf, sizeof(surf)); if (region_bo != NULL) { /* We might sample from it, and we might render to it, so flag * them both. We might be able to figure out from other state @@ -700,8 +698,7 @@ brw_wm_get_binding_table(struct brw_context *brw) 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); + data, data_size); /* Emit binding table relocations to surface state */ for (i = 0; i < BRW_WM_MAX_SURF; i++) { diff --git a/src/mesa/drivers/dri/i965/intel_swapbuffers.c b/src/mesa/drivers/dri/i965/intel_swapbuffers.c deleted file mode 120000 index 148d5215aa..0000000000 --- a/src/mesa/drivers/dri/i965/intel_swapbuffers.c +++ /dev/null @@ -1 +0,0 @@ -../intel/intel_swapbuffers.c
\ No newline at end of file |