X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fbrw_tcs.c;h=1050850bb1c1e2912e7c504701a77215b5cb6814;hb=646924cfa1bb51c9091a8cf0079f00759cbea372;hp=037a2da76818ebf7ce4545d616be8a813d9fe33f;hpb=8498cb4a45e8ed53a2ee2b35d3c2cbb9963e1756;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/brw_tcs.c b/src/mesa/drivers/dri/i965/brw_tcs.c index 037a2da7681..1050850bb1c 100644 --- a/src/mesa/drivers/dri/i965/brw_tcs.c +++ b/src/mesa/drivers/dri/i965/brw_tcs.c @@ -28,116 +28,92 @@ */ #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" - -static void -brw_tcs_debug_recompile(struct brw_context *brw, - struct gl_shader_program *shader_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); - - 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; - } - - if (!c) { - perf_debug(" Didn't find previous compile in the shader cache for " - "debug\n"); - return; - } - - found |= key_debug(brw, "input vertices", old_key->input_vertices, - key->input_vertices); - found |= key_debug(brw, "TES primitive mode", old_key->tes_primitive_mode, - key->tes_primitive_mode); - found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex); - - if (!found) { - perf_debug(" Something else\n"); - } -} +#include "nir_builder.h" static bool -brw_codegen_tcs_prog(struct brw_context *brw, - struct gl_shader_program *shader_prog, - struct brw_tess_ctrl_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) { - const struct brw_compiler *compiler = brw->intelScreen->compiler; + struct gl_context *ctx = &brw->ctx; + const struct brw_compiler *compiler = brw->screen->compiler; + const struct gen_device_info *devinfo = compiler->devinfo; struct brw_stage_state *stage_state = &brw->tcs.base; - nir_shader *nir = tcp->program.Base.nir; + nir_shader *nir; struct brw_tcs_prog_data prog_data; bool start_busy = false; double start_time = 0; + void *mem_ctx = ralloc_context(NULL); + if (tcp) { + nir = nir_shader_clone(mem_ctx, tcp->program.nir); + } else { + const nir_shader_compiler_options *options = + ctx->Const.ShaderCompilerOptions[MESA_SHADER_TESS_CTRL].NirOptions; + nir = brw_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_shader *tcs = shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]; - int param_count = nir->num_uniforms; - if (!compiler->scalar_stage[MESA_SHADER_TESS_CTRL]) - param_count *= 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.image_param = - rzalloc_array(NULL, struct brw_image_param, tcs->NumImages); - prog_data.base.base.nr_params = param_count; - prog_data.base.base.nr_image_params = tcs->NumImages; - - brw_nir_setup_glsl_uniforms(nir, shader_prog, &tcp->program.Base, - &prog_data.base.base, false); - - if (unlikely(INTEL_DEBUG & DEBUG_TCS)) - brw_dump_ir("tessellation control", shader_prog, tcs, NULL); + if (tcp) { + brw_assign_common_binding_table_offsets(devinfo, &tcp->program, + &prog_data.base.base, 0); + + 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, 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. + */ + 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] = BRW_PARAM_BUILTIN_ZERO; + + if (key->tes_primitive_mode == GL_QUADS) { + for (int i = 0; i < 4; i++) + param[7 - i] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i; + + 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] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X + i; + + param[4] = BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X; + } else { + assert(key->tes_primitive_mode == GL_ISOLINES); + param[7] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_Y; + param[6] = BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X; + } + } 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); + 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: " @@ -148,85 +124,149 @@ brw_codegen_tcs_prog(struct brw_context *brw, } if (unlikely(brw->perf_debug)) { - struct brw_shader *btcs = (struct brw_shader *) tcs; - if (btcs->compiled_once) { - brw_tcs_debug_recompile(brw, shader_prog, key); + if (tcp) { + if (tcp->compiled_once) { + brw_debug_recompile(brw, MESA_SHADER_TESS_CTRL, tcp->program.Id, + key->program_string_id, key); + } + 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); } - btcs->compiled_once = true; } /* Scratch space is used for register spilling */ - if (prog_data.base.base.total_scratch) { - brw_get_scratch_bo(brw, &stage_state->scratch_bo, - prog_data.base.base.total_scratch * - brw->max_hs_threads); - } + brw_alloc_stage_scratch(brw, stage_state, + 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.prog_data); + &stage_state->prog_offset, &brw->tcs.base.prog_data); ralloc_free(mem_ctx); return true; } +void +brw_tcs_populate_key(struct brw_context *brw, + struct brw_tcs_prog_key *key) +{ + const struct gen_device_info *devinfo = &brw->screen->devinfo; + const struct brw_compiler *compiler = brw->screen->compiler; + 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; + uint32_t per_patch_slots = tes_prog->info.patch_inputs_read; + + memset(key, 0, sizeof(*key)); + + if (tcp) { + struct gl_program *prog = &tcp->program; + per_vertex_slots |= prog->info.outputs_written; + per_patch_slots |= prog->info.patch_outputs_written; + } + + if (devinfo->gen < 8 || !tcp || compiler->use_tcs_8_patch) + key->input_vertices = brw->ctx.TessCtrlProgram.patch_vertices; + key->outputs_written = per_vertex_slots; + key->patch_outputs_written = per_patch_slots; + + /* 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.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); + } +} void brw_upload_tcs_prog(struct brw_context *brw) { - struct gl_context *ctx = &brw->ctx; - struct gl_shader_program **current = ctx->_Shader->CurrentProgram; struct brw_stage_state *stage_state = &brw->tcs.base; struct brw_tcs_prog_key key; - /* BRW_NEW_TESS_CTRL_PROGRAM */ - struct brw_tess_ctrl_program *tcp = - (struct brw_tess_ctrl_program *) brw->tess_ctrl_program; - /* BRW_NEW_TESS_EVAL_PROGRAM */ - struct brw_tess_eval_program *tep = - (struct brw_tess_eval_program *) brw->tess_eval_program; - assert(tcp && tep); + /* BRW_NEW_TESS_PROGRAMS */ + struct brw_program *tcp = + (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL]; + MAYBE_UNUSED struct brw_program *tep = + (struct brw_program *) brw->programs[MESA_SHADER_TESS_EVAL]; + assert(tep); if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_PATCH_PRIMITIVE | - BRW_NEW_TESS_CTRL_PROGRAM | - BRW_NEW_TESS_EVAL_PROGRAM)) + BRW_NEW_TESS_PROGRAMS)) return; - struct gl_program *prog = &tcp->program.Base; + brw_tcs_populate_key(brw, &key); - memset(&key, 0, sizeof(key)); + if (brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG, &key, sizeof(key), + &stage_state->prog_offset, &brw->tcs.base.prog_data, + true)) + return; - key.program_string_id = tcp->id; + if (brw_disk_cache_upload_program(brw, MESA_SHADER_TESS_CTRL)) + return; - key.input_vertices = ctx->TessCtrlProgram.patch_vertices; + tcp = (struct brw_program *) brw->programs[MESA_SHADER_TESS_CTRL]; + if (tcp) + tcp->id = key.program_string_id; - /* _NEW_TEXTURE */ - brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count, - &key.tex); + MAYBE_UNUSED bool success = brw_codegen_tcs_prog(brw, tcp, tep, &key); + assert(success); +} - /* 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.PrimitiveMode; - - if (!brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG, - &key, sizeof(key), - &stage_state->prog_offset, &brw->tcs.prog_data)) { - bool success = brw_codegen_tcs_prog(brw, current[MESA_SHADER_TESS_CTRL], - tcp, &key); - assert(success); - (void)success; +void +brw_tcs_populate_default_key(const struct brw_compiler *compiler, + struct brw_tcs_prog_key *key, + struct gl_shader_program *sh_prog, + struct gl_program *prog) +{ + const struct gen_device_info *devinfo = compiler->devinfo; + 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 || compiler->use_tcs_8_patch) + 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; } - brw->tcs.base.prog_data = &brw->tcs.prog_data->base.base; -} + 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, @@ -234,28 +274,23 @@ brw_tcs_precompile(struct gl_context *ctx, struct gl_program *prog) { struct brw_context *brw = brw_context(ctx); + const struct brw_compiler *compiler = brw->screen->compiler; struct brw_tcs_prog_key key; uint32_t old_prog_offset = brw->tcs.base.prog_offset; - struct brw_tcs_prog_data *old_prog_data = brw->tcs.prog_data; + struct brw_stage_prog_data *old_prog_data = brw->tcs.base.prog_data; bool success; - struct gl_tess_ctrl_program *tcp = (struct gl_tess_ctrl_program *)prog; - struct brw_tess_ctrl_program *btcp = brw_tess_ctrl_program(tcp); - - 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. */ - key.input_vertices = shader_prog->TessCtrl.VerticesOut; + 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; - key.tes_primitive_mode = GL_TRIANGLES; + brw_tcs_populate_default_key(compiler, &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.prog_data = old_prog_data; + brw->tcs.base.prog_data = old_prog_data; return success; }