#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"
#include "program/program.h"
#include "intel_mipmap_tree.h"
#include "intel_image.h"
-#include "brw_nir.h"
+#include "intel_fbo.h"
+#include "compiler/brw_nir.h"
#include "brw_program.h"
#include "util/ralloc.h"
+#include "util/u_math.h"
static void
assign_fs_binding_table_offsets(const struct gen_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)
{
- uint32_t next_binding_table_offset = 0;
-
- /* If there are no color regions, we still perform an FB write to a null
- * renderbuffer, which we place at surface index 0.
+ /* Render targets implicitly start at surface index 0. Even if there are
+ * no color regions, we still perform an FB write to a null render target,
+ * which will be surface 0.
*/
- prog_data->binding_table.render_target_start = next_binding_table_offset;
- next_binding_table_offset += MAX2(key->nr_color_regions, 1);
+ uint32_t 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,
+ 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) {
+ 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)
-{
- 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->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_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 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, "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, "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");
- }
+ /* Update the binding table size */
+ prog_data->base.binding_table.size_bytes = next_binding_table_offset * 4;
}
-/**
- * All Mesa program -> GPU code generation goes through this function.
- * 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_program *fp,
struct brw_wm_prog_key *key,
struct brw_vue_map *vue_map)
{
const struct gen_device_info *devinfo = &brw->screen->devinfo;
- struct gl_context *ctx = &brw->ctx;
void *mem_ctx = ralloc_context(NULL);
struct brw_wm_prog_data prog_data;
const GLuint *program;
- GLuint program_size;
bool start_busy = false;
double start_time = 0;
+ nir_shader *nir = nir_shader_clone(mem_ctx, fp->program.nir);
+
memset(&prog_data, 0, sizeof(prog_data));
/* 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(devinfo, prog, &fp->program, 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.nir->num_uniforms / 4;
- prog_data.base.nr_image_params = fp->program.info.num_images;
- /* 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.nir, prog, &fp->program,
+ if (!fp->program.is_arb_asm) {
+ brw_nir_setup_glsl_uniforms(mem_ctx, nir, &fp->program,
&prog_data.base, true);
+ brw_nir_analyze_ubo_ranges(brw->screen->compiler, nir,
+ NULL, prog_data.base.ubo_ranges);
} else {
- brw_nir_setup_arb_uniforms(fp->program.nir, &fp->program,
- &prog_data.base);
+ brw_nir_setup_arb_uniforms(mem_ctx, nir, &fp->program, &prog_data.base);
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();
}
- int st_index8 = -1, st_index16 = -1;
+ int st_index8 = -1, st_index16 = -1, st_index32 = -1;
if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
- bool is_glsl_sh = prog != NULL;
st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
- is_glsl_sh);
+ !fp->program.is_arb_asm);
st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
- is_glsl_sh);
+ !fp->program.is_arb_asm);
+ st_index32 = brw_get_shader_time_index(brw, &fp->program, ST_FS32,
+ !fp->program.is_arb_asm);
}
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, brw->use_rep_send, vue_map,
- &program_size, &error_str);
+ key, &prog_data, nir,
+ st_index8, st_index16, st_index32,
+ true, false, vue_map,
+ NULL, &error_str);
if (program == NULL) {
- if (prog) {
- fp->program.sh.data->LinkStatus = false;
+ if (!fp->program.is_arb_asm) {
+ fp->program.sh.data->LinkStatus = LINKING_FAILURE;
ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
}
}
if (unlikely(brw->perf_debug)) {
- if (fp->compiled_once)
- brw_wm_debug_recompile(brw, &fp->program, key);
+ if (fp->compiled_once) {
+ brw_debug_recompile(brw, MESA_SHADER_FRAGMENT, fp->program.Id,
+ &key->base);
+ }
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);
}
}
- brw_alloc_stage_scratch(brw, &brw->wm.base,
- prog_data.base.total_scratch,
- devinfo->max_wm_threads);
+ brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
- if (unlikely((INTEL_DEBUG & DEBUG_WM) && !prog))
+ 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,
+ program, prog_data.base.program_size,
&prog_data, sizeof(prog_data),
&brw->wm.base.prog_offset, &brw->wm.base.prog_data);
return true;
}
-bool
-brw_debug_recompile_sampler_key(struct brw_context *brw,
- const struct brw_sampler_prog_key_data *old_key,
- const struct brw_sampler_prog_key_data *key)
-{
- bool found = false;
-
- for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
- found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
- old_key->swizzles[i], key->swizzles[i]);
- }
- found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
- old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
- found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
- old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
- found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
- old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
- found |= key_debug(brw, "gather channel quirk on any texture unit",
- old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
- 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",
- old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
- }
-
- return found;
-}
-
static uint8_t
gen6_gather_workaround(GLenum internalformat)
{
}
}
-void
+static void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
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;
while (mask) {
const int s = u_bit_scan(&mask);
key->swizzles[s] = SWIZZLE_NOOP;
+ key->scale_factors[s] = 0.0f;
int unit_id = prog->SamplerUnits[s];
const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
/* 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[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->nir->info->uses_texture_gather) {
- if (img->InternalFormat == GL_RG32F)
- key->gather_channel_quirk_mask |= 1 << s;
+ /* gather4 for RG32* is broken in multiple ways on Gen7. */
+ if (devinfo->gen == 7 && prog->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->nir->info->uses_texture_gather) {
+ if (devinfo->gen == 6 && prog->info.uses_texture_gather) {
key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
}
/* 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->num_samples > 1) {
+ 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->aux_buf);
+ assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
key->compressed_multisample_layout_mask |= 1 << s;
- if (intel_tex->mt->num_samples >= 16) {
- assert(brw->gen >= 9);
+ 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) {
+
+ /* Setup possible scaling factor. */
+ key->scale_factors[s] = intel_tex->planar_format->scaling_factor;
+
switch (intel_tex->planar_format->components) {
case __DRI_IMAGE_COMPONENTS_Y_UV:
key->y_uv_image_mask |= 1 << s;
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;
+ case __DRI_IMAGE_COMPONENTS_AYUV:
+ key->ayuv_image_mask |= 1 << s;
+ break;
+ case __DRI_IMAGE_COMPONENTS_XYUV:
+ key->xyuv_image_mask |= 1 << s;
+ break;
default:
break;
}
+
+ switch (intel_tex->yuv_color_space) {
+ case __DRI_YUV_COLOR_SPACE_ITU_REC709:
+ key->bt709_mask |= 1 << s;
+ break;
+ case __DRI_YUV_COLOR_SPACE_ITU_REC2020:
+ key->bt2020_mask |= 1 << s;
+ break;
+ default:
+ break;
+ }
}
}
}
}
+void
+brw_populate_base_prog_key(struct gl_context *ctx,
+ const struct brw_program *prog,
+ struct brw_base_prog_key *key)
+{
+ key->program_string_id = prog->id;
+ key->subgroup_size_type = BRW_SUBGROUP_SIZE_UNIFORM;
+ brw_populate_sampler_prog_key_data(ctx, &prog->program, &key->tex);
+}
+
+void
+brw_populate_default_base_prog_key(const struct gen_device_info *devinfo,
+ const struct brw_program *prog,
+ struct brw_base_prog_key *key)
+{
+ key->program_string_id = prog->id;
+ key->subgroup_size_type = BRW_SUBGROUP_SIZE_UNIFORM;
+ brw_setup_tex_for_precompile(devinfo, &key->tex, &prog->program);
+}
+
static bool
brw_wm_state_dirty(const struct brw_context *brw)
{
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_program *fp = brw_program_const(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;
/* Build the index for table lookup
*/
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
+ struct intel_renderbuffer *depth_irb =
+ intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
+
/* _NEW_COLOR */
if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
- lookup |= IZ_PS_KILL_ALPHATEST_BIT;
+ lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
}
if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
- lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
+ lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
}
/* _NEW_DEPTH */
- if (ctx->Depth.Test)
- lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
+ if (depth_irb && ctx->Depth.Test) {
+ lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
- if (brw_depth_writes_enabled(brw))
- 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;
+ 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;
+ line_aa = BRW_WM_AA_SOMETIMES;
if (ctx->Polygon.BackMode == GL_LINE ||
(ctx->Polygon.CullFlag &&
ctx->Polygon.CullFaceMode == GL_BACK))
- line_aa = AA_ALWAYS;
+ line_aa = BRW_WM_AA_ALWAYS;
}
else if (ctx->Polygon.BackMode == GL_LINE) {
- line_aa = AA_SOMETIMES;
+ line_aa = BRW_WM_AA_SOMETIMES;
if ((ctx->Polygon.CullFlag &&
ctx->Polygon.CullFaceMode == GL_FRONT))
- line_aa = AA_ALWAYS;
+ line_aa = BRW_WM_AA_ALWAYS;
}
}
}
/* _NEW_HINT */
key->high_quality_derivatives =
+ prog->info.uses_fddx_fddy &&
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, &key->tex);
+ brw_populate_base_prog_key(ctx, fp, &key->base);
/* _NEW_BUFFERS */
key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
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 &&
- (_mesa_is_alpha_test_enabled(ctx) ||
- _mesa_is_alpha_to_coverage_enabled(ctx));
+ /* _NEW_MULTISAMPLE, _NEW_BUFFERS */
+ key->alpha_to_coverage = _mesa_is_alpha_to_coverage_enabled(ctx);
+
+ /* _NEW_COLOR, _NEW_BUFFERS */
+ key->alpha_test_replicate_alpha =
+ ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
+ _mesa_is_alpha_test_enabled(ctx);
/* _NEW_BUFFERS _NEW_MULTISAMPLE */
/* Ignore sample qualifier while computing this flag. */
key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
}
+ key->ignore_sample_mask_out = !key->multisample_fbo;
+
/* BRW_NEW_VUE_MAP_GEOM_OUT */
- if (brw->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
- BRW_FS_VARYING_INPUT_MASK) > 16) {
+ if (devinfo->gen < 6 || util_bitcount64(prog->info.inputs_read &
+ BRW_FS_VARYING_INPUT_MASK) > 16) {
key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
}
* 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;
}
- /* 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;
+ key->coherent_fb_fetch = ctx->Extensions.EXT_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_program *fp = (struct brw_program *) brw->fragment_program;
+ struct brw_program *fp =
+ (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
if (!brw_wm_state_dirty(brw))
return;
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.base.prog_data)) {
- bool success = brw_codegen_wm_prog(brw, current, fp, &key,
- &brw->vue_map_geom_out);
- (void) success;
- assert(success);
- }
+ if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG, &key, sizeof(key),
+ &brw->wm.base.prog_offset, &brw->wm.base.prog_data,
+ true))
+ return;
+
+ if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
+ return;
+
+ fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
+ fp->id = key.base.program_string_id;
+
+ ASSERTED bool success = brw_codegen_wm_prog(brw, fp, &key,
+ &brw->vue_map_geom_out);
+ assert(success);
}
-bool
-brw_fs_precompile(struct gl_context *ctx,
- struct gl_shader_program *shader_prog,
- struct gl_program *prog)
+void
+brw_wm_populate_default_key(const struct brw_compiler *compiler,
+ struct brw_wm_prog_key *key,
+ struct gl_program *prog)
{
- struct brw_context *brw = brw_context(ctx);
- struct brw_wm_prog_key key;
+ const struct gen_device_info *devinfo = compiler->devinfo;
- struct brw_program *bfp = brw_program(prog);
+ memset(key, 0, sizeof(*key));
- memset(&key, 0, sizeof(key));
+ brw_populate_default_base_prog_key(devinfo, brw_program(prog),
+ &key->base);
uint64_t outputs_written = prog->info.outputs_written;
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
if (prog->info.fs.uses_discard)
- key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
+ key->iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
- key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
+ 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(prog->info.inputs_read &
- BRW_FS_VARYING_INPUT_MASK) > 16) {
- key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
+ if (devinfo->gen < 6 || util_bitcount64(prog->info.inputs_read &
+ BRW_FS_VARYING_INPUT_MASK) > 16) {
+ key->input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
}
- brw_setup_tex_for_precompile(brw, &key.tex, prog);
-
- key.nr_color_regions = _mesa_bitcount_64(outputs_written &
+ key->nr_color_regions = util_bitcount64(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;
+ key->coherent_fb_fetch = devinfo->gen >= 9;
+}
+
+bool
+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 brw_program *bfp = brw_program(prog);
+
+ brw_wm_populate_default_key(brw->screen->compiler, &key, prog);
+
+ /* check brw_wm_populate_default_key coherent_fb_fetch setting */
+ assert(key.coherent_fb_fetch ==
+ ctx->Extensions.EXT_shader_framebuffer_fetch);
uint32_t old_prog_offset = brw->wm.base.prog_offset;
struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
struct brw_vue_map vue_map;
- if (brw->gen < 6) {
+ if (devinfo->gen < 6) {
brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
prog->info.inputs_read | VARYING_BIT_POS,
- false);
+ false, 1);
}
- bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key, &vue_map);
+ bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
brw->wm.base.prog_offset = old_prog_offset;
brw->wm.base.prog_data = old_prog_data;
projects / mesa.git / blobdiff