X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fbrw_wm.c;h=3f929c46cf5cd18d250a254be9d35e0547f747e3;hb=40241d40d0a481f9f1cd112d2d48dadf1ec6ad0f;hp=41266f57560cb660accb37f29da6008ff745be67;hpb=6350c97412d993de1ae876312bab9689a7f34839;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 41266f57560..3f929c46cf5 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -26,6 +26,7 @@ #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" @@ -34,119 +35,37 @@ #include "program/prog_parameter.h" #include "program/program.h" #include "intel_mipmap_tree.h" +#include "intel_image.h" +#include "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. - */ -unsigned -brw_compute_barycentric_interp_modes(struct brw_context *brw, - bool shade_model_flat, - bool persample_shading, - const struct gl_fragment_program *fprog) +static void +assign_fs_binding_table_offsets(const struct brw_device_info *devinfo, + const struct gl_shader_program *shader_prog, + const struct gl_program *prog, + const struct brw_wm_prog_key *key, + struct brw_wm_prog_data *prog_data) { - unsigned barycentric_interp_modes = 0; - int attr; + uint32_t next_binding_table_offset = 0; - /* Loop through all fragment shader inputs to figure out what interpolation - * modes are in use, and set the appropriate bits in - * barycentric_interp_modes. + /* If there are no color regions, we still perform an FB write to a null + * renderbuffer, which we place at surface index 0. */ - for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) { - enum glsl_interp_qualifier interp_qualifier = - fprog->InterpQualifier[attr]; - bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) && - !persample_shading; - bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) || - persample_shading; - bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1; - - /* Ignore unused inputs. */ - if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr))) - continue; - - /* Ignore WPOS and FACE, because they don't require interpolation. */ - if (attr == VARYING_SLOT_POS || attr == 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) || - brw->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) || - brw->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; - } + prog_data->binding_table.render_target_start = next_binding_table_offset; + next_binding_table_offset += MAX2(key->nr_color_regions, 1); + + next_binding_table_offset = + brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo, + shader_prog, 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; } - return BRW_PSCDEPTH_OFF; -} - -bool -brw_wm_prog_data_compare(const void *in_a, const void *in_b) -{ - const struct brw_wm_prog_data *a = in_a; - const struct brw_wm_prog_data *b = in_b; - - /* Compare the base structure. */ - if (!brw_stage_prog_data_compare(&a->base, &b->base)) - return false; - - /* Compare the rest of the structure. */ - const unsigned offset = sizeof(struct brw_stage_prog_data); - if (memcmp(((char *) a) + offset, ((char *) b) + offset, - sizeof(struct brw_wm_prog_data) - offset)) - return false; - - return true; } /** @@ -164,39 +83,30 @@ brw_codegen_wm_prog(struct brw_context *brw, void *mem_ctx = ralloc_context(NULL); struct brw_wm_prog_data prog_data; const GLuint *program; - struct gl_shader *fs = NULL; + struct brw_shader *fs = NULL; GLuint program_size; + bool start_busy = false; + double start_time = 0; if (prog) - fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; + 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->EarlyFragmentTests; /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */ if (!prog) prog_data.base.use_alt_mode = true; + assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog, + &fp->program.Base, 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; - if (fs) { - param_count = fs->num_uniform_components + - fs->NumImages * BRW_IMAGE_PARAM_SIZE; - prog_data.base.nr_image_params = fs->NumImages; - } else { - param_count = fp->program.Base.Parameters->NumParameters * 4; - } + int param_count = fp->program.Base.nir->num_uniforms / 4; + 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 = @@ -208,23 +118,62 @@ brw_codegen_wm_prog(struct brw_context *brw, prog_data.base.nr_image_params); prog_data.base.nr_params = param_count; - prog_data.barycentric_interp_modes = - brw_compute_barycentric_interp_modes(brw, key->flat_shade, - key->persample_shading, - &fp->program); + if (prog) { + brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base, + &prog_data.base, true); + } else { + brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base, + &prog_data.base); + } + + if (unlikely(brw->perf_debug)) { + start_busy = (brw->batch.last_bo && + drm_intel_bo_busy(brw->batch.last_bo)); + start_time = get_time(); + } + + if (unlikely(INTEL_DEBUG & DEBUG_WM)) + brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &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); + } - program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data, - &fp->program, prog, &program_size); + char *error_str = NULL; + program = brw_compile_fs(brw->intelScreen->compiler, brw, mem_ctx, + key, &prog_data, fp->program.Base.nir, + &fp->program.Base, st_index8, st_index16, + true, brw->use_rep_send, + &program_size, &error_str); if (program == NULL) { + if (prog) { + prog->LinkStatus = false; + ralloc_strcat(&prog->InfoLog, error_str); + } + + _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str); + ralloc_free(mem_ctx); return false; } - 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); + if (unlikely(brw->perf_debug) && fs) { + if (fs->compiled_once) + brw_wm_debug_recompile(brw, prog, key); + fs->compiled_once = true; + + if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) { + perf_debug("FS compile took %.03f ms and stalled the GPU\n", + (get_time() - start_time) * 1000); + } } + brw_alloc_stage_scratch(brw, &brw->wm.base, + prog_data.base.total_scratch, + brw->max_wm_threads); + if (unlikely(INTEL_DEBUG & DEBUG_WM)) fprintf(stderr, "\n"); @@ -239,17 +188,6 @@ brw_codegen_wm_prog(struct brw_context *brw, 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, @@ -272,6 +210,19 @@ 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); for (unsigned int i = 0; i < MAX_SAMPLERS; i++) { found |= key_debug(brw, "textureGather workarounds", @@ -317,22 +268,18 @@ brw_wm_debug_recompile(struct brw_context *brw, 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, "per-sample interpolation", + old_key->persample_interp, key->persample_interp); 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, "multisampled FBO", + old_key->multisample_fbo, key->multisample_fbo); 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", @@ -366,16 +313,15 @@ 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) { struct brw_context *brw = brw_context(ctx); + 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]; @@ -428,16 +374,42 @@ brw_populate_sampler_prog_key_data(struct gl_context *ctx, struct intel_texture_object *intel_tex = intel_texture_object((struct gl_texture_object *)t); + /* From gen9 onwards some single sampled buffers can also be + * compressed. These don't need ld2dms sampling along with mcs fetch. + */ if (brw->gen >= 7 && - intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) { + intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS && + intel_tex->mt->num_samples > 1) { key->compressed_multisample_layout_mask |= 1 << s; + + if (intel_tex->mt->num_samples >= 16) { + assert(brw->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; + 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 | @@ -467,8 +439,6 @@ brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) const struct gl_program *prog = (struct gl_program *) brw->fragment_program; 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)); @@ -542,60 +512,29 @@ brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key) 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); /* _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 & @@ -617,6 +556,9 @@ 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 @@ -642,3 +584,55 @@ brw_upload_wm_prog(struct brw_context *brw) } 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) +{ + struct brw_context *brw = brw_context(ctx); + 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); + + memset(&key, 0, sizeof(key)); + + if (brw->gen < 6) { + if (fp->UsesKill) + key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT; + + if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) + key.iz_lookup |= 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; + } + + 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); + + key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten & + ~(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; + + bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key); + + brw->wm.base.prog_offset = old_prog_offset; + brw->wm.prog_data = old_prog_data; + + return success; +}