X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fbrw_wm.c;h=ad1c8aed79e1056462681095cf8567a2abfe6a00;hb=f970e4d481bb459822e8f3b6a9599e661c5ceb0d;hp=4d5e7f67bd608b93478ee38875b1bb83b4cc7526;hpb=fc2a66cfcddea34af0e93dd2221ae1fd3fdd9e87;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 4d5e7f67bd6..ad1c8aed79e 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -26,7 +26,6 @@ #include "brw_context.h" #include "brw_wm.h" #include "brw_state.h" -#include "brw_shader.h" #include "main/enums.h" #include "main/formats.h" #include "main/fbobject.h" @@ -35,96 +34,14 @@ #include "program/prog_parameter.h" #include "program/program.h" #include "intel_mipmap_tree.h" -#include "brw_nir.h" +#include "intel_image.h" +#include "compiler/brw_nir.h" +#include "brw_program.h" #include "util/ralloc.h" -/** - * Return a bitfield where bit n is set if barycentric interpolation mode n - * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader. - */ -static unsigned -brw_compute_barycentric_interp_modes(const struct brw_device_info *devinfo, - bool shade_model_flat, - bool persample_shading, - nir_shader *shader) -{ - unsigned barycentric_interp_modes = 0; - - nir_foreach_variable(var, &shader->inputs) { - enum glsl_interp_qualifier interp_qualifier = var->data.interpolation; - bool is_centroid = var->data.centroid && !persample_shading; - bool is_sample = var->data.sample || persample_shading; - bool is_gl_Color = (var->data.location == VARYING_SLOT_COL0) || - (var->data.location == VARYING_SLOT_COL1); - - /* Ignore WPOS and FACE, because they don't require interpolation. */ - if (var->data.location == VARYING_SLOT_POS || - var->data.location == VARYING_SLOT_FACE) - continue; - - /* Determine the set (or sets) of barycentric coordinates needed to - * interpolate this variable. Note that when - * brw->needs_unlit_centroid_workaround is set, centroid interpolation - * uses PIXEL interpolation for unlit pixels and CENTROID interpolation - * for lit pixels, so we need both sets of barycentric coordinates. - */ - if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) { - if (is_centroid) { - barycentric_interp_modes |= - 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC; - } else if (is_sample) { - barycentric_interp_modes |= - 1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC; - } - if ((!is_centroid && !is_sample) || - devinfo->needs_unlit_centroid_workaround) { - barycentric_interp_modes |= - 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC; - } - } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH || - (!(shade_model_flat && is_gl_Color) && - interp_qualifier == INTERP_QUALIFIER_NONE)) { - if (is_centroid) { - barycentric_interp_modes |= - 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC; - } else if (is_sample) { - barycentric_interp_modes |= - 1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC; - } - if ((!is_centroid && !is_sample) || - devinfo->needs_unlit_centroid_workaround) { - barycentric_interp_modes |= - 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC; - } - } - } - - return barycentric_interp_modes; -} - -static uint8_t -computed_depth_mode(struct gl_fragment_program *fp) -{ - if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { - switch (fp->FragDepthLayout) { - case FRAG_DEPTH_LAYOUT_NONE: - case FRAG_DEPTH_LAYOUT_ANY: - return BRW_PSCDEPTH_ON; - case FRAG_DEPTH_LAYOUT_GREATER: - return BRW_PSCDEPTH_ON_GE; - case FRAG_DEPTH_LAYOUT_LESS: - return BRW_PSCDEPTH_ON_LE; - case FRAG_DEPTH_LAYOUT_UNCHANGED: - return BRW_PSCDEPTH_OFF; - } - } - return BRW_PSCDEPTH_OFF; -} - static void -assign_fs_binding_table_offsets(const struct brw_device_info *devinfo, - const struct gl_shader_program *shader_prog, +assign_fs_binding_table_offsets(const struct gen_device_info *devinfo, const struct gl_program *prog, const struct brw_wm_prog_key *key, struct brw_wm_prog_data *prog_data) @@ -137,9 +54,76 @@ assign_fs_binding_table_offsets(const struct brw_device_info *devinfo, prog_data->binding_table.render_target_start = next_binding_table_offset; next_binding_table_offset += MAX2(key->nr_color_regions, 1); - brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo, - shader_prog, prog, &prog_data->base, - next_binding_table_offset); + next_binding_table_offset = + brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base, + next_binding_table_offset); + + if (prog->nir->info.outputs_read && !key->coherent_fb_fetch) { + prog_data->binding_table.render_target_read_start = + next_binding_table_offset; + next_binding_table_offset += key->nr_color_regions; + } +} + +static void +brw_wm_debug_recompile(struct brw_context *brw, struct gl_program *prog, + const struct brw_wm_prog_key *key) +{ + perf_debug("Recompiling fragment shader for program %d\n", prog->Id); + + bool found = false; + const struct brw_wm_prog_key *old_key = + brw_find_previous_compile(&brw->cache, BRW_CACHE_FS_PROG, + key->program_string_id); + + if (!old_key) { + perf_debug(" Didn't find previous compile in the shader cache for debug\n"); + return; + } + + found |= key_debug(brw, "alphatest, computed depth, depth test, or " + "depth write", + old_key->iz_lookup, key->iz_lookup); + found |= key_debug(brw, "depth statistics", + old_key->stats_wm, key->stats_wm); + found |= key_debug(brw, "flat shading", + old_key->flat_shade, key->flat_shade); + found |= key_debug(brw, "number of color buffers", + old_key->nr_color_regions, key->nr_color_regions); + found |= key_debug(brw, "MRT alpha test or alpha-to-coverage", + old_key->replicate_alpha, key->replicate_alpha); + found |= key_debug(brw, "fragment color clamping", + old_key->clamp_fragment_color, key->clamp_fragment_color); + found |= key_debug(brw, "per-sample interpolation", + old_key->persample_interp, key->persample_interp); + found |= key_debug(brw, "multisampled FBO", + old_key->multisample_fbo, key->multisample_fbo); + found |= key_debug(brw, "frag coord adds sample pos", + old_key->frag_coord_adds_sample_pos, + key->frag_coord_adds_sample_pos); + found |= key_debug(brw, "line smoothing", + old_key->line_aa, key->line_aa); + found |= key_debug(brw, "high quality derivatives", + old_key->high_quality_derivatives, + key->high_quality_derivatives); + found |= key_debug(brw, "force dual color blending", + old_key->force_dual_color_blend, + key->force_dual_color_blend); + found |= key_debug(brw, "coherent fb fetch", + old_key->coherent_fb_fetch, key->coherent_fb_fetch); + + found |= key_debug(brw, "input slots valid", + old_key->input_slots_valid, key->input_slots_valid); + found |= key_debug(brw, "mrt alpha test function", + old_key->alpha_test_func, key->alpha_test_func); + found |= key_debug(brw, "mrt alpha test reference value", + old_key->alpha_test_ref, key->alpha_test_ref); + + found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex); + + if (!found) { + perf_debug(" Something else\n"); + } } /** @@ -147,137 +131,106 @@ assign_fs_binding_table_offsets(const struct brw_device_info *devinfo, * Depending on the instructions used (i.e. flow control instructions) * we'll use one of two code generators. */ -bool +static bool brw_codegen_wm_prog(struct brw_context *brw, - struct gl_shader_program *prog, - struct brw_fragment_program *fp, - struct brw_wm_prog_key *key) + struct brw_program *fp, + struct brw_wm_prog_key *key, + struct brw_vue_map *vue_map) { - struct gl_context *ctx = &brw->ctx; + const struct gen_device_info *devinfo = &brw->screen->devinfo; void *mem_ctx = ralloc_context(NULL); struct brw_wm_prog_data prog_data; const GLuint *program; - struct brw_shader *fs = NULL; GLuint program_size; bool start_busy = false; double start_time = 0; - if (prog) - fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; - memset(&prog_data, 0, sizeof(prog_data)); - /* key->alpha_test_func means simulating alpha testing via discards, - * so the shader definitely kills pixels. - */ - prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func; - prog_data.uses_omask = - fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK); - prog_data.computed_depth_mode = computed_depth_mode(&fp->program); - - prog_data.early_fragment_tests = fs && fs->base.EarlyFragmentTests; /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */ - if (!prog) + if (fp->program.is_arb_asm) prog_data.base.use_alt_mode = true; - assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog, - &fp->program.Base, key, &prog_data); + assign_fs_binding_table_offsets(devinfo, &fp->program, key, &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. - */ - int param_count = fp->program.Base.nir->num_uniforms; - if (fs) - prog_data.base.nr_image_params = fs->base.NumImages; - /* The backend also sometimes adds params for texture size. */ - param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits; - prog_data.base.param = - rzalloc_array(NULL, const gl_constant_value *, param_count); - prog_data.base.pull_param = - rzalloc_array(NULL, const gl_constant_value *, param_count); - prog_data.base.image_param = - rzalloc_array(NULL, struct brw_image_param, - prog_data.base.nr_image_params); - prog_data.base.nr_params = param_count; - - if (prog) { - brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base, + if (!fp->program.is_arb_asm) { + brw_nir_setup_glsl_uniforms(mem_ctx, fp->program.nir, &fp->program, &prog_data.base, true); + brw_nir_analyze_ubo_ranges(brw->screen->compiler, fp->program.nir, + prog_data.base.ubo_ranges); } else { - brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base, + brw_nir_setup_arb_uniforms(mem_ctx, fp->program.nir, &fp->program, &prog_data.base); - } - prog_data.barycentric_interp_modes = - brw_compute_barycentric_interp_modes(brw->intelScreen->devinfo, - key->flat_shade, - key->persample_shading, - fp->program.Base.nir); + if (unlikely(INTEL_DEBUG & DEBUG_WM)) + brw_dump_arb_asm("fragment", &fp->program); + } if (unlikely(brw->perf_debug)) { start_busy = (brw->batch.last_bo && - drm_intel_bo_busy(brw->batch.last_bo)); + brw_bo_busy(brw->batch.last_bo)); start_time = get_time(); } - if (unlikely(INTEL_DEBUG & DEBUG_WM)) - brw_dump_ir("fragment", prog, &fs->base, &fp->program.Base); - int st_index8 = -1, st_index16 = -1; if (INTEL_DEBUG & DEBUG_SHADER_TIME) { - st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8); - st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16); + st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8, + !fp->program.is_arb_asm); + st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16, + !fp->program.is_arb_asm); } - program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data, - &fp->program, prog, st_index8, st_index16, &program_size); + char *error_str = NULL; + program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx, + key, &prog_data, fp->program.nir, + &fp->program, st_index8, st_index16, + true, false, vue_map, + &program_size, &error_str); + if (program == NULL) { + if (!fp->program.is_arb_asm) { + fp->program.sh.data->LinkStatus = linking_failure; + ralloc_strcat(&fp->program.sh.data->InfoLog, error_str); + } + + _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str); + ralloc_free(mem_ctx); return false; } - if (unlikely(brw->perf_debug) && fs) { - if (fs->compiled_once) - brw_wm_debug_recompile(brw, prog, key); - fs->compiled_once = true; + if (unlikely(brw->perf_debug)) { + if (fp->compiled_once) + brw_wm_debug_recompile(brw, &fp->program, key); + fp->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("FS compile took %.03f ms and stalled the GPU\n", (get_time() - start_time) * 1000); } } - if (prog_data.base.total_scratch) { - brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo, - prog_data.base.total_scratch * brw->max_wm_threads); - } + brw_alloc_stage_scratch(brw, &brw->wm.base, + prog_data.base.total_scratch, + devinfo->max_wm_threads); - if (unlikely(INTEL_DEBUG & DEBUG_WM)) + if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm)) fprintf(stderr, "\n"); + /* The param and pull_param arrays will be freed by the shader cache. */ + ralloc_steal(NULL, prog_data.base.param); + ralloc_steal(NULL, prog_data.base.pull_param); brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG, - key, sizeof(struct brw_wm_prog_key), - program, program_size, - &prog_data, sizeof(prog_data), - &brw->wm.base.prog_offset, &brw->wm.prog_data); + key, sizeof(struct brw_wm_prog_key), + program, program_size, + &prog_data, sizeof(prog_data), + &brw->wm.base.prog_offset, &brw->wm.base.prog_data); ralloc_free(mem_ctx); return true; } -static bool -key_debug(struct brw_context *brw, const char *name, int a, int b) -{ - if (a != b) { - perf_debug(" %s %d->%d\n", name, a, b); - return true; - } else { - return false; - } -} - bool brw_debug_recompile_sampler_key(struct brw_context *brw, const struct brw_sampler_prog_key_data *old_key, @@ -300,6 +253,23 @@ brw_debug_recompile_sampler_key(struct brw_context *brw, found |= key_debug(brw, "compressed multisample layout", old_key->compressed_multisample_layout_mask, key->compressed_multisample_layout_mask); + found |= key_debug(brw, "16x msaa", + old_key->msaa_16, + key->msaa_16); + + found |= key_debug(brw, "y_uv image bound", + old_key->y_uv_image_mask, + key->y_uv_image_mask); + found |= key_debug(brw, "y_u_v image bound", + old_key->y_u_v_image_mask, + key->y_u_v_image_mask); + found |= key_debug(brw, "yx_xuxv image bound", + old_key->yx_xuxv_image_mask, + key->yx_xuxv_image_mask); + found |= key_debug(brw, "xy_uxvx image bound", + old_key->xy_uxvx_image_mask, + key->xy_uxvx_image_mask); + for (unsigned int i = 0; i < MAX_SAMPLERS; i++) { found |= key_debug(brw, "textureGather workarounds", @@ -309,72 +279,6 @@ brw_debug_recompile_sampler_key(struct brw_context *brw, return found; } -void -brw_wm_debug_recompile(struct brw_context *brw, - struct gl_shader_program *prog, - const struct brw_wm_prog_key *key) -{ - struct brw_cache_item *c = NULL; - const struct brw_wm_prog_key *old_key = NULL; - bool found = false; - - perf_debug("Recompiling fragment shader for program %d\n", 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_FS_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, "alphatest, computed depth, depth test, or " - "depth write", - old_key->iz_lookup, key->iz_lookup); - found |= key_debug(brw, "depth statistics", - old_key->stats_wm, key->stats_wm); - found |= key_debug(brw, "flat shading", - old_key->flat_shade, key->flat_shade); - found |= key_debug(brw, "per-sample shading", - old_key->persample_shading, key->persample_shading); - found |= key_debug(brw, "per-sample shading and 2x MSAA", - old_key->persample_2x, key->persample_2x); - found |= key_debug(brw, "number of color buffers", - old_key->nr_color_regions, key->nr_color_regions); - found |= key_debug(brw, "MRT alpha test or alpha-to-coverage", - old_key->replicate_alpha, key->replicate_alpha); - found |= key_debug(brw, "rendering to FBO", - old_key->render_to_fbo, key->render_to_fbo); - found |= key_debug(brw, "fragment color clamping", - old_key->clamp_fragment_color, key->clamp_fragment_color); - found |= key_debug(brw, "line smoothing", - old_key->line_aa, key->line_aa); - found |= key_debug(brw, "renderbuffer height", - old_key->drawable_height, key->drawable_height); - found |= key_debug(brw, "input slots valid", - old_key->input_slots_valid, key->input_slots_valid); - found |= key_debug(brw, "mrt alpha test function", - old_key->alpha_test_func, key->alpha_test_func); - found |= key_debug(brw, "mrt alpha test reference value", - old_key->alpha_test_ref, key->alpha_test_ref); - - found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex); - - if (!found) { - perf_debug(" Something else\n"); - } -} - static uint8_t gen6_gather_workaround(GLenum internalformat) { @@ -393,25 +297,25 @@ gen6_gather_workaround(GLenum internalformat) void brw_populate_sampler_prog_key_data(struct gl_context *ctx, - const struct gl_program *prog, - unsigned sampler_count, - struct brw_sampler_prog_key_data *key) + const struct gl_program *prog, + struct brw_sampler_prog_key_data *key) { struct brw_context *brw = brw_context(ctx); + const struct gen_device_info *devinfo = &brw->screen->devinfo; + GLbitfield mask = prog->SamplersUsed; - for (int s = 0; s < sampler_count; s++) { - key->swizzles[s] = SWIZZLE_NOOP; + while (mask) { + const int s = u_bit_scan(&mask); - if (!(prog->SamplersUsed & (1 << s))) - continue; + key->swizzles[s] = SWIZZLE_NOOP; int unit_id = prog->SamplerUnits[s]; const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id]; if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) { - const struct gl_texture_object *t = unit->_Current; - const struct gl_texture_image *img = t->Image[0][t->BaseLevel]; - struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id); + const struct gl_texture_object *t = unit->_Current; + const struct gl_texture_image *img = t->Image[0][t->BaseLevel]; + struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id); const bool alpha_depth = t->DepthMode == GL_ALPHA && (img->_BaseFormat == GL_DEPTH_COMPONENT || @@ -420,32 +324,60 @@ brw_populate_sampler_prog_key_data(struct gl_context *ctx, /* Haswell handles texture swizzling as surface format overrides * (except for GL_ALPHA); all other platforms need MOVs in the shader. */ - if (alpha_depth || (brw->gen < 8 && !brw->is_haswell)) + if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell)) key->swizzles[s] = brw_get_texture_swizzle(ctx, t); - if (brw->gen < 8 && + if (devinfo->gen < 8 && sampler->MinFilter != GL_NEAREST && - sampler->MagFilter != GL_NEAREST) { - if (sampler->WrapS == GL_CLAMP) - key->gl_clamp_mask[0] |= 1 << s; - if (sampler->WrapT == GL_CLAMP) - key->gl_clamp_mask[1] |= 1 << s; - if (sampler->WrapR == GL_CLAMP) - key->gl_clamp_mask[2] |= 1 << s; - } - - /* gather4's channel select for green from RG32F is broken; requires - * a shader w/a on IVB; fixable with just SCS on HSW. - */ - if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) { - if (img->InternalFormat == GL_RG32F) - key->gather_channel_quirk_mask |= 1 << s; + sampler->MagFilter != GL_NEAREST) { + if (sampler->WrapS == GL_CLAMP) + key->gl_clamp_mask[0] |= 1 << s; + if (sampler->WrapT == GL_CLAMP) + key->gl_clamp_mask[1] |= 1 << s; + if (sampler->WrapR == GL_CLAMP) + key->gl_clamp_mask[2] |= 1 << s; + } + + /* gather4 for RG32* is broken in multiple ways on Gen7. */ + if (devinfo->gen == 7 && prog->nir->info.uses_texture_gather) { + switch (img->InternalFormat) { + case GL_RG32I: + case GL_RG32UI: { + /* We have to override the format to R32G32_FLOAT_LD. + * This means that SCS_ALPHA and SCS_ONE will return 0x3f8 + * (1.0) rather than integer 1. This needs shader hacks. + * + * On Ivybridge, we whack W (alpha) to ONE in our key's + * swizzle. On Haswell, we look at the original texture + * swizzle, and use XYZW with channels overridden to ONE, + * leaving normal texture swizzling to SCS. + */ + unsigned src_swizzle = + devinfo->is_haswell ? t->_Swizzle : key->swizzles[s]; + for (int i = 0; i < 4; i++) { + unsigned src_comp = GET_SWZ(src_swizzle, i); + if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) { + key->swizzles[i] &= ~(0x7 << (3 * i)); + key->swizzles[i] |= SWIZZLE_ONE << (3 * i); + } + } + /* fallthrough */ + } + case GL_RG32F: + /* The channel select for green doesn't work - we have to + * request blue. Haswell can use SCS for this, but Ivybridge + * needs a shader workaround. + */ + if (!devinfo->is_haswell) + key->gather_channel_quirk_mask |= 1 << s; + break; + } } /* Gen6's gather4 is broken for UINT/SINT; we treat them as * UNORM/FLOAT instead and fix it in the shader. */ - if (brw->gen == 6 && prog->UsesGather) { + if (devinfo->gen == 6 && prog->nir->info.uses_texture_gather) { key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat); } @@ -456,16 +388,47 @@ brw_populate_sampler_prog_key_data(struct gl_context *ctx, struct intel_texture_object *intel_tex = intel_texture_object((struct gl_texture_object *)t); - if (brw->gen >= 7 && - intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) { + /* From gen9 onwards some single sampled buffers can also be + * compressed. These don't need ld2dms sampling along with mcs fetch. + */ + if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) { + assert(devinfo->gen >= 7); + assert(intel_tex->mt->surf.samples > 1); + assert(intel_tex->mt->mcs_buf); + assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY); key->compressed_multisample_layout_mask |= 1 << s; + + if (intel_tex->mt->surf.samples >= 16) { + assert(devinfo->gen >= 9); + key->msaa_16 |= 1 << s; + } + } + + if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) { + switch (intel_tex->planar_format->components) { + case __DRI_IMAGE_COMPONENTS_Y_UV: + key->y_uv_image_mask |= 1 << s; + break; + case __DRI_IMAGE_COMPONENTS_Y_U_V: + key->y_u_v_image_mask |= 1 << s; + break; + case __DRI_IMAGE_COMPONENTS_Y_XUXV: + key->yx_xuxv_image_mask |= 1 << s; + break; + case __DRI_IMAGE_COMPONENTS_Y_UXVX: + key->xy_uxvx_image_mask |= 1 << s; + break; + default: + break; + } } + } } } static bool -brw_wm_state_dirty (struct brw_context *brw) +brw_wm_state_dirty(const struct brw_context *brw) { return brw_state_dirty(brw, _NEW_BUFFERS | @@ -485,72 +448,72 @@ brw_wm_state_dirty (struct brw_context *brw) BRW_NEW_VUE_MAP_GEOM_OUT); } -static void +void brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) { + const struct gen_device_info *devinfo = &brw->screen->devinfo; struct gl_context *ctx = &brw->ctx; /* BRW_NEW_FRAGMENT_PROGRAM */ - const struct brw_fragment_program *fp = - (struct brw_fragment_program *) brw->fragment_program; - const struct gl_program *prog = (struct gl_program *) brw->fragment_program; + const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT]; + const struct brw_program *fp = brw_program_const(prog); GLuint lookup = 0; GLuint line_aa; - bool program_uses_dfdy = fp->program.UsesDFdy; - const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1; memset(key, 0, sizeof(*key)); /* Build the index for table lookup */ - if (brw->gen < 6) { + if (devinfo->gen < 6) { /* _NEW_COLOR */ - if (fp->program.UsesKill || ctx->Color.AlphaEnabled) - lookup |= IZ_PS_KILL_ALPHATEST_BIT; + if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) { + lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT; + } - if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) - lookup |= IZ_PS_COMPUTES_DEPTH_BIT; + if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) { + lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT; + } /* _NEW_DEPTH */ if (ctx->Depth.Test) - lookup |= IZ_DEPTH_TEST_ENABLE_BIT; + lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT; - if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */ - lookup |= IZ_DEPTH_WRITE_ENABLE_BIT; + if (brw_depth_writes_enabled(brw)) + lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT; /* _NEW_STENCIL | _NEW_BUFFERS */ - if (ctx->Stencil._Enabled) { - lookup |= IZ_STENCIL_TEST_ENABLE_BIT; + if (brw->stencil_enabled) { + lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT; - if (ctx->Stencil.WriteMask[0] || - ctx->Stencil.WriteMask[ctx->Stencil._BackFace]) - lookup |= IZ_STENCIL_WRITE_ENABLE_BIT; + if (ctx->Stencil.WriteMask[0] || + ctx->Stencil.WriteMask[ctx->Stencil._BackFace]) + lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT; } key->iz_lookup = lookup; } - line_aa = AA_NEVER; + line_aa = BRW_WM_AA_NEVER; /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */ if (ctx->Line.SmoothFlag) { if (brw->reduced_primitive == GL_LINES) { - line_aa = AA_ALWAYS; + line_aa = BRW_WM_AA_ALWAYS; } else if (brw->reduced_primitive == GL_TRIANGLES) { - if (ctx->Polygon.FrontMode == GL_LINE) { - line_aa = AA_SOMETIMES; - - if (ctx->Polygon.BackMode == GL_LINE || - (ctx->Polygon.CullFlag && - ctx->Polygon.CullFaceMode == GL_BACK)) - line_aa = AA_ALWAYS; - } - else if (ctx->Polygon.BackMode == GL_LINE) { - line_aa = AA_SOMETIMES; - - if ((ctx->Polygon.CullFlag && - ctx->Polygon.CullFaceMode == GL_FRONT)) - line_aa = AA_ALWAYS; - } + if (ctx->Polygon.FrontMode == GL_LINE) { + line_aa = BRW_WM_AA_SOMETIMES; + + if (ctx->Polygon.BackMode == GL_LINE || + (ctx->Polygon.CullFlag && + ctx->Polygon.CullFaceMode == GL_BACK)) + line_aa = BRW_WM_AA_ALWAYS; + } + else if (ctx->Polygon.BackMode == GL_LINE) { + line_aa = BRW_WM_AA_SOMETIMES; + + if ((ctx->Polygon.CullFlag && + ctx->Polygon.CullFaceMode == GL_FRONT)) + line_aa = BRW_WM_AA_ALWAYS; + } } } @@ -560,76 +523,48 @@ brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) key->high_quality_derivatives = ctx->Hint.FragmentShaderDerivative == GL_NICEST; - if (brw->gen < 6) + if (devinfo->gen < 6) key->stats_wm = brw->stats_wm; /* _NEW_LIGHT */ - key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT); + key->flat_shade = + (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) && + (ctx->Light.ShadeModel == GL_FLAT); /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */ key->clamp_fragment_color = ctx->Color._ClampFragmentColor; /* _NEW_TEXTURE */ - brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count, - &key->tex); - - /* _NEW_BUFFERS */ - /* - * Include the draw buffer origin and height so that we can calculate - * fragment position values relative to the bottom left of the drawable, - * from the incoming screen origin relative position we get as part of our - * payload. - * - * This is only needed for the WM_WPOSXY opcode when the fragment program - * uses the gl_FragCoord input. - * - * We could avoid recompiling by including this as a constant referenced by - * our program, but if we were to do that it would also be nice to handle - * getting that constant updated at batchbuffer submit time (when we - * hold the lock and know where the buffer really is) rather than at emit - * time when we don't hold the lock and are just guessing. We could also - * just avoid using this as key data if the program doesn't use - * fragment.position. - * - * For DRI2 the origin_x/y will always be (0,0) but we still need the - * drawable height in order to invert the Y axis. - */ - if (fp->program.Base.InputsRead & VARYING_BIT_POS) { - key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer); - } - - if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) { - key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer); - } + brw_populate_sampler_prog_key_data(ctx, prog, &key->tex); /* _NEW_BUFFERS */ key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers; + /* _NEW_COLOR */ + key->force_dual_color_blend = brw->dual_color_blend_by_location && + (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc; + /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */ key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 && - (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled); + (_mesa_is_alpha_test_enabled(ctx) || + _mesa_is_alpha_to_coverage_enabled(ctx)); /* _NEW_BUFFERS _NEW_MULTISAMPLE */ /* Ignore sample qualifier while computing this flag. */ - key->persample_shading = - _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1; - if (key->persample_shading) - key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2; - - key->compute_pos_offset = - _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 && - fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS; + if (ctx->Multisample.Enabled) { + key->persample_interp = + ctx->Multisample.SampleShading && + (ctx->Multisample.MinSampleShadingValue * + _mesa_geometric_samples(ctx->DrawBuffer) > 1); - key->compute_sample_id = - multisample_fbo && - ctx->Multisample.Enabled && - (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID); + key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1; + } /* BRW_NEW_VUE_MAP_GEOM_OUT */ - if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead & - BRW_FS_VARYING_INPUT_MASK) > 16) + if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read & + BRW_FS_VARYING_INPUT_MASK) > 16) { key->input_slots_valid = brw->vue_map_geom_out.slots_valid; - + } /* _NEW_COLOR | _NEW_BUFFERS */ /* Pre-gen6, the hardware alpha test always used each render @@ -637,7 +572,7 @@ brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) * like GL requires. Fix that by building the alpha test into the * shader, and we'll skip enabling the fixed function alpha test. */ - if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && + if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && ctx->Color.AlphaEnabled) { key->alpha_test_func = ctx->Color.AlphaFunc; key->alpha_test_ref = ctx->Color.AlphaRef; @@ -645,16 +580,17 @@ brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) /* The unique fragment program ID */ key->program_string_id = fp->id; + + /* Whether reads from the framebuffer should behave coherently. */ + key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch; } void brw_upload_wm_prog(struct brw_context *brw) { - struct gl_context *ctx = &brw->ctx; - struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram; struct brw_wm_prog_key key; - struct brw_fragment_program *fp = (struct brw_fragment_program *) - brw->fragment_program; + struct brw_program *fp = + (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT]; if (!brw_wm_state_dirty(brw)) return; @@ -662,69 +598,72 @@ brw_upload_wm_prog(struct brw_context *brw) brw_wm_populate_key(brw, &key); if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG, - &key, sizeof(key), - &brw->wm.base.prog_offset, &brw->wm.prog_data)) { - bool success = brw_codegen_wm_prog(brw, current, fp, &key); + &key, sizeof(key), + &brw->wm.base.prog_offset, + &brw->wm.base.prog_data)) { + bool success = brw_codegen_wm_prog(brw, fp, &key, + &brw->vue_map_geom_out); (void) success; assert(success); } - brw->wm.base.prog_data = &brw->wm.prog_data->base; } bool -brw_fs_precompile(struct gl_context *ctx, - struct gl_shader_program *shader_prog, - struct gl_program *prog) +brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog) { struct brw_context *brw = brw_context(ctx); + const struct gen_device_info *devinfo = &brw->screen->devinfo; struct brw_wm_prog_key key; - struct gl_fragment_program *fp = (struct gl_fragment_program *) prog; - struct brw_fragment_program *bfp = brw_fragment_program(fp); - bool program_uses_dfdy = fp->UsesDFdy; + struct brw_program *bfp = brw_program(prog); memset(&key, 0, sizeof(key)); - if (brw->gen < 6) { - if (fp->UsesKill) - key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT; + uint64_t outputs_written = prog->info.outputs_written; + + if (devinfo->gen < 6) { + if (prog->info.fs.uses_discard) + key.iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT; - if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) - key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT; + if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) + key.iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT; /* Just assume depth testing. */ - key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT; - key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT; + key.iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT; + key.iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT; } - if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead & - BRW_FS_VARYING_INPUT_MASK) > 16) - key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS; - - brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base); - - if (fp->Base.InputsRead & VARYING_BIT_POS) { - key.drawable_height = ctx->DrawBuffer->Height; + if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read & + BRW_FS_VARYING_INPUT_MASK) > 16) { + key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS; } - key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten & - ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) | - BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK))); + brw_setup_tex_for_precompile(brw, &key.tex, prog); - if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) { - key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) || - key.nr_color_regions > 1; - } + key.nr_color_regions = _mesa_bitcount_64(outputs_written & + ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) | + BITFIELD64_BIT(FRAG_RESULT_STENCIL) | + BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK))); key.program_string_id = bfp->id; + /* Whether reads from the framebuffer should behave coherently. */ + key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch; + uint32_t old_prog_offset = brw->wm.base.prog_offset; - struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data; + struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data; + + struct brw_vue_map vue_map; + if (devinfo->gen < 6) { + brw_compute_vue_map(&brw->screen->devinfo, &vue_map, + prog->info.inputs_read | VARYING_BIT_POS, + false); + } - bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key); + bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map); brw->wm.base.prog_offset = old_prog_offset; - brw->wm.prog_data = old_prog_data; + brw->wm.base.prog_data = old_prog_data; return success; }