X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fbrw_tcs.c;h=823933ef77f89c18837138549341b0fd510cbfd9;hb=2567ad28bbbec5750567af0cd9c8a2dd4732f91d;hp=58d080d28767ec3f31a1a8ceb1e4423ee5b23dec;hpb=46a4e4257e42dadc425592719863048d9012f9a7;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/brw_tcs.c b/src/mesa/drivers/dri/i965/brw_tcs.c index 58d080d2876..823933ef77f 100644 --- a/src/mesa/drivers/dri/i965/brw_tcs.c +++ b/src/mesa/drivers/dri/i965/brw_tcs.c @@ -28,20 +28,20 @@ */ #include "brw_context.h" -#include "brw_nir.h" +#include "compiler/brw_nir.h" #include "brw_program.h" -#include "brw_shader.h" #include "brw_state.h" #include "program/prog_parameter.h" #include "nir_builder.h" static nir_shader * -create_passthrough_tcs(const struct brw_compiler *compiler, +create_passthrough_tcs(void *mem_ctx, const struct brw_compiler *compiler, const nir_shader_compiler_options *options, const struct brw_tcs_prog_key *key) { nir_builder b; - nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_TESS_CTRL, options); + nir_builder_init_simple_shader(&b, mem_ctx, MESA_SHADER_TESS_CTRL, + options); nir_shader *nir = b.shader; nir_variable *var; nir_intrinsic_instr *load; @@ -50,10 +50,11 @@ create_passthrough_tcs(const struct brw_compiler *compiler, nir_ssa_def *invoc_id = nir_load_system_value(&b, nir_intrinsic_load_invocation_id, 0); - nir->info->inputs_read = key->outputs_written; - nir->info->outputs_written = key->outputs_written; - nir->info->tcs.vertices_out = key->input_vertices; - nir->info->name = ralloc_strdup(nir, "passthrough"); + nir->info.inputs_read = key->outputs_written & + ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER); + nir->info.outputs_written = key->outputs_written; + nir->info.tess.tcs_vertices_out = key->input_vertices; + nir->info.name = ralloc_strdup(nir, "passthrough"); nir->num_uniforms = 8 * sizeof(uint32_t); var = nir_variable_create(nir, nir_var_uniform, glsl_vec4_type(), "hdr_0"); @@ -80,7 +81,7 @@ create_passthrough_tcs(const struct brw_compiler *compiler, } /* Copy inputs to outputs. */ - uint64_t varyings = key->outputs_written; + uint64_t varyings = nir->info.inputs_read; while (varyings != 0) { const int varying = ffsll(varyings) - 1; @@ -115,31 +116,18 @@ create_passthrough_tcs(const struct brw_compiler *compiler, } static void -brw_tcs_debug_recompile(struct brw_context *brw, - struct gl_shader_program *shader_prog, +brw_tcs_debug_recompile(struct brw_context *brw, struct gl_program *prog, const struct brw_tcs_prog_key *key) { - struct brw_cache_item *c = NULL; - const struct brw_tcs_prog_key *old_key = NULL; - bool found = false; - perf_debug("Recompiling tessellation control shader for program %d\n", - shader_prog->Name); + prog->Id); - for (unsigned int i = 0; i < brw->cache.size; i++) { - for (c = brw->cache.items[i]; c; c = c->next) { - if (c->cache_id == BRW_CACHE_TCS_PROG) { - old_key = c->key; - - if (old_key->program_string_id == key->program_string_id) - break; - } - } - if (c) - break; - } + bool found = false; + const struct brw_tcs_prog_key *old_key = + brw_find_previous_compile(&brw->cache, BRW_CACHE_TCS_PROG, + key->program_string_id); - if (!c) { + if (!old_key) { perf_debug(" Didn't find previous compile in the shader cache for " "debug\n"); return; @@ -163,10 +151,8 @@ brw_tcs_debug_recompile(struct brw_context *brw, } static bool -brw_codegen_tcs_prog(struct brw_context *brw, - struct gl_shader_program *shader_prog, - struct brw_program *tcp, - struct brw_tcs_prog_key *key) +brw_codegen_tcs_prog(struct brw_context *brw, struct brw_program *tcp, + struct brw_program *tep, struct brw_tcs_prog_key *key) { struct gl_context *ctx = &brw->ctx; const struct brw_compiler *compiler = brw->screen->compiler; @@ -177,101 +163,73 @@ brw_codegen_tcs_prog(struct brw_context *brw, bool start_busy = false; double start_time = 0; + void *mem_ctx = ralloc_context(NULL); if (tcp) { nir = tcp->program.nir; } else { - /* Create a dummy nir_shader. We won't actually use NIR code to - * generate assembly (it's easier to generate assembly directly), - * but the whole compiler assumes one of these exists. - */ const nir_shader_compiler_options *options = ctx->Const.ShaderCompilerOptions[MESA_SHADER_TESS_CTRL].NirOptions; - nir = create_passthrough_tcs(compiler, options, key); + nir = create_passthrough_tcs(mem_ctx, compiler, options, key); } memset(&prog_data, 0, sizeof(prog_data)); - /* Allocate the references to the uniforms that will end up in the - * prog_data associated with the compiled program, and which will be freed - * by the state cache. - * - * Note: param_count needs to be num_uniform_components * 4, since we add - * padding around uniform values below vec4 size, so the worst case is that - * every uniform is a float which gets padded to the size of a vec4. - */ - struct gl_linked_shader *tcs = shader_prog ? - shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL] : NULL; - int param_count = nir->num_uniforms / 4; - - prog_data.base.base.param = - rzalloc_array(NULL, const gl_constant_value *, param_count); - prog_data.base.base.pull_param = - rzalloc_array(NULL, const gl_constant_value *, param_count); - prog_data.base.base.nr_params = param_count; - - if (tcs) { - brw_assign_common_binding_table_offsets(MESA_SHADER_TESS_CTRL, devinfo, - shader_prog, &tcp->program, + if (tcp) { + brw_assign_common_binding_table_offsets(devinfo, &tcp->program, &prog_data.base.base, 0); - prog_data.base.base.image_param = - rzalloc_array(NULL, struct brw_image_param, tcs->NumImages); - prog_data.base.base.nr_image_params = tcs->NumImages; - - brw_nir_setup_glsl_uniforms(nir, shader_prog, &tcp->program, + brw_nir_setup_glsl_uniforms(mem_ctx, nir, &tcp->program, &prog_data.base.base, compiler->scalar_stage[MESA_SHADER_TESS_CTRL]); + brw_nir_analyze_ubo_ranges(compiler, tcp->program.nir, NULL, + prog_data.base.base.ubo_ranges); } else { /* Upload the Patch URB Header as the first two uniforms. * Do the annoying scrambling so the shader doesn't have to. */ - const float **param = (const float **) prog_data.base.base.param; - static float zero = 0.0f; + assert(nir->num_uniforms == 32); + prog_data.base.base.param = rzalloc_array(mem_ctx, uint32_t, 8); + prog_data.base.base.nr_params = 8; + + uint32_t *param = prog_data.base.base.param; for (int i = 0; i < 8; i++) - param[i] = &zero; + param[i] = BRW_PARAM_BUILTIN_ZERO; if (key->tes_primitive_mode == GL_QUADS) { for (int i = 0; i < 4; i++) - param[7 - i] = &ctx->TessCtrlProgram.patch_default_outer_level[i]; + param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i; - param[3] = &ctx->TessCtrlProgram.patch_default_inner_level[0]; - param[2] = &ctx->TessCtrlProgram.patch_default_inner_level[1]; + param[3] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X; + param[2] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y; } else if (key->tes_primitive_mode == GL_TRIANGLES) { for (int i = 0; i < 3; i++) - param[7 - i] = &ctx->TessCtrlProgram.patch_default_outer_level[i]; + param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i; - param[4] = &ctx->TessCtrlProgram.patch_default_inner_level[0]; + param[4] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X; } else { assert(key->tes_primitive_mode == GL_ISOLINES); - param[7] = &ctx->TessCtrlProgram.patch_default_outer_level[1]; - param[6] = &ctx->TessCtrlProgram.patch_default_outer_level[0]; + param[7] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_Y; + param[6] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X; } } - if (unlikely(INTEL_DEBUG & DEBUG_TCS) && tcs) - brw_dump_ir("tessellation control", shader_prog, tcs, NULL); - int st_index = -1; - if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME)) - st_index = brw_get_shader_time_index(brw, shader_prog, NULL, ST_TCS); + if (unlikely((INTEL_DEBUG & DEBUG_SHADER_TIME) && tep)) + st_index = brw_get_shader_time_index(brw, &tep->program, ST_TCS, true); if (unlikely(brw->perf_debug)) { - start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo); + start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo); start_time = get_time(); } - void *mem_ctx = ralloc_context(NULL); - unsigned program_size; char *error_str; const unsigned *program = brw_compile_tcs(compiler, brw, mem_ctx, key, &prog_data, nir, st_index, - &program_size, &error_str); + &error_str); if (program == NULL) { - if (shader_prog) { - shader_prog->LinkStatus = false; - ralloc_strcat(&shader_prog->InfoLog, error_str); - } else { - ralloc_free(nir); + if (tep) { + tep->program.sh.data->LinkStatus = LINKING_FAILURE; + ralloc_strcat(&tep->program.sh.data->InfoLog, error_str); } _mesa_problem(NULL, "Failed to compile tessellation control shader: " @@ -282,14 +240,14 @@ brw_codegen_tcs_prog(struct brw_context *brw, } if (unlikely(brw->perf_debug)) { - struct brw_shader *btcs = (struct brw_shader *) tcs; - if (btcs) { - if (btcs->compiled_once) { - brw_tcs_debug_recompile(brw, shader_prog, key); + if (tcp) { + if (tcp->compiled_once) { + brw_tcs_debug_recompile(brw, &tcp->program, key); } - btcs->compiled_once = true; + tcp->compiled_once = true; } - if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) { + + if (start_busy && !brw_bo_busy(brw->batch.last_bo)) { perf_debug("TCS compile took %.03f ms and stalled the GPU\n", (get_time() - start_time) * 1000); } @@ -297,17 +255,17 @@ brw_codegen_tcs_prog(struct brw_context *brw, /* Scratch space is used for register spilling */ brw_alloc_stage_scratch(brw, stage_state, - prog_data.base.base.total_scratch, - devinfo->max_tcs_threads); + prog_data.base.base.total_scratch); + /* The param and pull_param arrays will be freed by the shader cache. */ + ralloc_steal(NULL, prog_data.base.base.param); + ralloc_steal(NULL, prog_data.base.base.pull_param); brw_upload_cache(&brw->cache, BRW_CACHE_TCS_PROG, key, sizeof(*key), - program, program_size, + program, prog_data.base.base.program_size, &prog_data, sizeof(prog_data), &stage_state->prog_offset, &brw->tcs.base.prog_data); ralloc_free(mem_ctx); - if (!tcs) - ralloc_free(nir); return true; } @@ -316,8 +274,11 @@ void brw_tcs_populate_key(struct brw_context *brw, struct brw_tcs_prog_key *key) { - struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program; - struct brw_program *tep = (struct brw_program *) brw->tess_eval_program; + const struct gen_device_info *devinfo = &brw->screen->devinfo; + struct brw_program *tcp = + (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL]; + struct brw_program *tep = + (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL]; struct gl_program *tes_prog = &tep->program; uint64_t per_vertex_slots = tes_prog->info.inputs_read; @@ -331,7 +292,7 @@ brw_tcs_populate_key(struct brw_context *brw, per_patch_slots |= prog->info.patch_outputs_written; } - if (brw->gen < 8 || !tcp) + if (devinfo->gen < 8 || !tcp) key->input_vertices = brw->ctx.TessCtrlProgram.patch_vertices; key->outputs_written = per_vertex_slots; key->patch_outputs_written = per_patch_slots; @@ -339,31 +300,29 @@ brw_tcs_populate_key(struct brw_context *brw, /* We need to specialize our code generation for tessellation levels * based on the domain the DS is expecting to tessellate. */ - key->tes_primitive_mode = tep->program.info.tes.primitive_mode; - key->quads_workaround = brw->gen < 9 && - tep->program.info.tes.primitive_mode == GL_QUADS && - tep->program.info.tes.spacing == GL_EQUAL; + key->tes_primitive_mode = tep->program.info.tess.primitive_mode; + key->quads_workaround = devinfo->gen < 9 && + tep->program.info.tess.primitive_mode == GL_QUADS && + tep->program.info.tess.spacing == TESS_SPACING_EQUAL; if (tcp) { key->program_string_id = tcp->id; /* _NEW_TEXTURE */ brw_populate_sampler_prog_key_data(&brw->ctx, &tcp->program, &key->tex); - } else { - key->outputs_written = tes_prog->info.inputs_read; } } void brw_upload_tcs_prog(struct brw_context *brw) { - struct gl_shader_program **current = brw->ctx._Shader->CurrentProgram; struct brw_stage_state *stage_state = &brw->tcs.base; struct brw_tcs_prog_key key; /* BRW_NEW_TESS_PROGRAMS */ - struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program; + struct brw_program *tcp = + (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL]; MAYBE_UNUSED struct brw_program *tep = - (struct brw_program *) brw->tess_eval_program; + (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL]; assert(tep); if (!brw_state_dirty(brw, @@ -374,17 +333,53 @@ brw_upload_tcs_prog(struct brw_context *brw) brw_tcs_populate_key(brw, &key); - if (!brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG, - &key, sizeof(key), - &stage_state->prog_offset, - &brw->tcs.base.prog_data)) { - bool success = brw_codegen_tcs_prog(brw, current[MESA_SHADER_TESS_CTRL], - tcp, &key); - assert(success); - (void)success; - } + if (brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG, &key, sizeof(key), + &stage_state->prog_offset, &brw->tcs.base.prog_data, + true)) + return; + + if (brw_disk_cache_upload_program(brw, MESA_SHADER_TESS_CTRL)) + return; + + tcp = (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL]; + if (tcp) + tcp->id = key.program_string_id; + + MAYBE_UNUSED bool success = brw_codegen_tcs_prog(brw, tcp, tep, &key); + assert(success); } +void +brw_tcs_populate_default_key(const struct gen_device_info *devinfo, + struct brw_tcs_prog_key *key, + struct gl_shader_program *sh_prog, + struct gl_program *prog) +{ + struct brw_program *btcp = brw_program(prog); + const struct gl_linked_shader *tes = + sh_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL]; + + memset(key, 0, sizeof(*key)); + + key->program_string_id = btcp->id; + brw_setup_tex_for_precompile(devinfo, &key->tex, prog); + + /* Guess that the input and output patches have the same dimensionality. */ + if (devinfo->gen < 8) + key->input_vertices = prog->info.tess.tcs_vertices_out; + + if (tes) { + key->tes_primitive_mode = tes->Program->info.tess.primitive_mode; + key->quads_workaround = devinfo->gen < 9 && + tes->Program->info.tess.primitive_mode == GL_QUADS && + tes->Program->info.tess.spacing == TESS_SPACING_EQUAL; + } else { + key->tes_primitive_mode = GL_TRIANGLES; + } + + key->outputs_written = prog->nir->info.outputs_written; + key->patch_outputs_written = prog->nir->info.patch_outputs_written; +} bool brw_tcs_precompile(struct gl_context *ctx, @@ -400,31 +395,11 @@ brw_tcs_precompile(struct gl_context *ctx, struct brw_program *btcp = brw_program(prog); const struct gl_linked_shader *tes = shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL]; + struct brw_program *btep = tes ? brw_program(tes->Program) : NULL; - memset(&key, 0, sizeof(key)); - - key.program_string_id = btcp->id; - brw_setup_tex_for_precompile(brw, &key.tex, prog); - - /* Guess that the input and output patches have the same dimensionality. */ - if (brw->gen < 8) { - key.input_vertices = shader_prog-> - _LinkedShaders[MESA_SHADER_TESS_CTRL]->info.TessCtrl.VerticesOut; - } - - if (tes) { - key.tes_primitive_mode = tes->info.TessEval.PrimitiveMode; - key.quads_workaround = brw->gen < 9 && - tes->info.TessEval.PrimitiveMode == GL_QUADS && - tes->info.TessEval.Spacing == GL_EQUAL; - } else { - key.tes_primitive_mode = GL_TRIANGLES; - } - - key.outputs_written = prog->nir->info->outputs_written; - key.patch_outputs_written = prog->nir->info->patch_outputs_written; + brw_tcs_populate_default_key(&brw->screen->devinfo, &key, shader_prog, prog); - success = brw_codegen_tcs_prog(brw, shader_prog, btcp, &key); + success = brw_codegen_tcs_prog(brw, btcp, btep, &key); brw->tcs.base.prog_offset = old_prog_offset; brw->tcs.base.prog_data = old_prog_data;