/*
- Copyright (C) Intel Corp. 2006. All Rights Reserved.
- Intel funded Tungsten Graphics to
- develop this 3D driver.
-
- Permission is hereby granted, free of charge, to any person obtaining
- a copy of this software and associated documentation files (the
- "Software"), to deal in the Software without restriction, including
- without limitation the rights to use, copy, modify, merge, publish,
- distribute, sublicense, and/or sell copies of the Software, and to
- permit persons to whom the Software is furnished to do so, subject to
- the following conditions:
-
- The above copyright notice and this permission notice (including the
- next paragraph) shall be included in all copies or substantial
- portions of the Software.
-
- THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
- IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
- LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
- OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
- WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
- **********************************************************************/
- /*
- * Authors:
- * Keith Whitwell <keithw@vmware.com>
- */
-
+ * Copyright (C) Intel Corp. 2006. All Rights Reserved.
+ * Intel funded Tungsten Graphics to
+ * develop this 3D driver.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
#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 "main/samplerobj.h"
+#include "main/framebuffer.h"
#include "program/prog_parameter.h"
#include "program/program.h"
#include "intel_mipmap_tree.h"
+#include "brw_nir.h"
-#include "glsl/ralloc.h"
+#include "util/ralloc.h"
/**
* Return a bitfield where bit n is set if barycentric interpolation mode n
return barycentric_interp_modes;
}
-bool
-brw_wm_prog_data_compare(const void *in_a, const void *in_b)
+static uint8_t
+computed_depth_mode(struct gl_fragment_program *fp)
{
- const struct brw_wm_prog_data *a = in_a;
- const struct brw_wm_prog_data *b = in_b;
+ 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;
+}
- /* Compare the base structure. */
- if (!brw_stage_prog_data_compare(&a->base, &b->base))
- return false;
+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)
+{
+ uint32_t next_binding_table_offset = 0;
- /* 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;
+ /* If there are no color regions, we still perform an FB write to a null
+ * renderbuffer, which we place at surface index 0.
+ */
+ prog_data->binding_table.render_target_start = next_binding_table_offset;
+ next_binding_table_offset += MAX2(key->nr_color_regions, 1);
- return true;
+ brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
+ shader_prog, prog, &prog_data->base,
+ next_binding_table_offset);
}
/**
* Depending on the instructions used (i.e. flow control instructions)
* we'll use one of two code generators.
*/
-bool do_wm_prog(struct brw_context *brw,
- struct gl_shader_program *prog,
- struct brw_fragment_program *fp,
- struct brw_wm_prog_key *key)
+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 gl_context *ctx = &brw->ctx;
- struct brw_wm_compile *c;
+ 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->base.EarlyFragmentTests;
- c = rzalloc(NULL, struct brw_wm_compile);
+ /* 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;
- } else {
- param_count = fp->program.Base.Parameters->NumParameters * 4;
- }
+ 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;
- c->prog_data.base.param = rzalloc_array(NULL, const float *, param_count);
- c->prog_data.base.pull_param =
- rzalloc_array(NULL, const float *, param_count);
- c->prog_data.base.nr_params = param_count;
-
- memcpy(&c->key, key, sizeof(*key));
+ 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,
+ &prog_data.base, true);
+ } else {
+ brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
+ &prog_data.base);
+ }
- c->prog_data.barycentric_interp_modes =
- brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
- c->key.persample_shading,
+ prog_data.barycentric_interp_modes =
+ brw_compute_barycentric_interp_modes(brw, key->flat_shade,
+ key->persample_shading,
&fp->program);
- program = brw_wm_fs_emit(brw, c, &fp->program, prog, &program_size);
- if (program == NULL)
+ if (unlikely(brw->perf_debug)) {
+ start_busy = (brw->batch.last_bo &&
+ drm_intel_bo_busy(brw->batch.last_bo));
+ start_time = get_time();
+ }
+
+ program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
+ &fp->program, prog, &program_size);
+ if (program == NULL) {
+ ralloc_free(mem_ctx);
return false;
+ }
- /* Scratch space is used for register spilling */
- if (c->last_scratch) {
- perf_debug("Fragment shader triggered register spilling. "
- "Try reducing the number of live scalar values to "
- "improve performance.\n");
+ if (unlikely(brw->perf_debug) && fs) {
+ if (fs->compiled_once)
+ brw_wm_debug_recompile(brw, prog, key);
+ fs->compiled_once = true;
- c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
+ 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);
+ }
+ }
+ if (prog_data.base.total_scratch) {
brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
- c->prog_data.total_scratch * brw->max_wm_threads);
+ prog_data.base.total_scratch * brw->max_wm_threads);
}
if (unlikely(INTEL_DEBUG & DEBUG_WM))
fprintf(stderr, "\n");
- brw_upload_cache(&brw->cache, BRW_WM_PROG,
- &c->key, sizeof(c->key),
+ brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
+ key, sizeof(struct brw_wm_prog_key),
program, program_size,
- &c->prog_data, sizeof(c->prog_data),
+ &prog_data, sizeof(prog_data),
&brw->wm.base.prog_offset, &brw->wm.prog_data);
- ralloc_free(c);
+ ralloc_free(mem_ctx);
return true;
}
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);
+
+ 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;
}
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_WM_PROG) {
+ if (c->cache_id == BRW_CACHE_FS_PROG) {
old_key = c->key;
if (old_key->program_string_id == key->program_string_id)
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",
gen6_gather_workaround(GLenum internalformat)
{
switch (internalformat) {
- case GL_R8I: return WA_SIGN | WA_8BIT;
- case GL_R8UI: return WA_8BIT;
- case GL_R16I: return WA_SIGN | WA_16BIT;
- case GL_R16UI: return WA_16BIT;
- /* note that even though GL_R32I and GL_R32UI have format overrides
- * in the surface state, there is no shader w/a required */
- default: return 0;
+ case GL_R8I: return WA_SIGN | WA_8BIT;
+ case GL_R8UI: return WA_8BIT;
+ case GL_R16I: return WA_SIGN | WA_16BIT;
+ case GL_R16UI: return WA_16BIT;
+ default:
+ /* Note that even though GL_R32I and GL_R32UI have format overrides in
+ * the surface state, there is no shader w/a required.
+ */
+ return 0;
}
}
if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
- if (sampler->MinFilter != GL_NEAREST &&
+ if (brw->gen < 8 &&
+ sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
key->gl_clamp_mask[0] |= 1 << s;
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. */
+ /* 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;
}
}
-static void brw_wm_populate_key( struct brw_context *brw,
- struct brw_wm_prog_key *key )
+static bool
+brw_wm_state_dirty (struct brw_context *brw)
+{
+ return brw_state_dirty(brw,
+ _NEW_BUFFERS |
+ _NEW_COLOR |
+ _NEW_DEPTH |
+ _NEW_FRAG_CLAMP |
+ _NEW_HINT |
+ _NEW_LIGHT |
+ _NEW_LINE |
+ _NEW_MULTISAMPLE |
+ _NEW_POLYGON |
+ _NEW_STENCIL |
+ _NEW_TEXTURE,
+ BRW_NEW_FRAGMENT_PROGRAM |
+ BRW_NEW_REDUCED_PRIMITIVE |
+ BRW_NEW_STATS_WM |
+ BRW_NEW_VUE_MAP_GEOM_OUT);
+}
+
+static void
+brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_FRAGMENT_PROGRAM */
const struct brw_fragment_program *fp =
- (struct brw_fragment_program *)brw->fragment_program;
+ (struct brw_fragment_program *) brw->fragment_program;
const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
GLuint lookup = 0;
GLuint line_aa;
bool program_uses_dfdy = fp->program.UsesDFdy;
- bool multisample_fbo = ctx->DrawBuffer->Visual.samples > 1;
+ const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
memset(key, 0, sizeof(*key));
key->line_aa = line_aa;
/* _NEW_HINT */
- if (brw->disable_derivative_optimization) {
- key->high_quality_derivatives =
- ctx->Hint.FragmentShaderDerivative != GL_FASTEST;
- } else {
- key->high_quality_derivatives =
- ctx->Hint.FragmentShaderDerivative == GL_NICEST;
- }
+ key->high_quality_derivatives =
+ ctx->Hint.FragmentShaderDerivative == GL_NICEST;
if (brw->gen < 6)
key->stats_wm = brw->stats_wm;
* drawable height in order to invert the Y axis.
*/
if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
- key->drawable_height = ctx->DrawBuffer->Height;
+ key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
}
if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
/* 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 &&
* 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 && ctx->Color.AlphaEnabled) {
+ if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
+ ctx->Color.AlphaEnabled) {
key->alpha_test_func = ctx->Color.AlphaFunc;
key->alpha_test_ref = ctx->Color.AlphaRef;
}
key->program_string_id = fp->id;
}
-
-static void
+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;
+ if (!brw_wm_state_dirty(brw))
+ return;
+
brw_wm_populate_key(brw, &key);
- if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
+ if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
&key, sizeof(key),
&brw->wm.base.prog_offset, &brw->wm.prog_data)) {
- bool success = do_wm_prog(brw, ctx->_Shader->_CurrentFragmentProgram, fp,
- &key);
+ bool success = brw_codegen_wm_prog(brw, current, fp, &key);
(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)
+{
+ 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);
+ bool program_uses_dfdy = fp->UsesDFdy;
+
+ memset(&key, 0, sizeof(key));
-const struct brw_tracked_state brw_wm_prog = {
- .dirty = {
- .mesa = (_NEW_COLOR |
- _NEW_DEPTH |
- _NEW_STENCIL |
- _NEW_POLYGON |
- _NEW_LINE |
- _NEW_HINT |
- _NEW_LIGHT |
- _NEW_FRAG_CLAMP |
- _NEW_BUFFERS |
- _NEW_TEXTURE |
- _NEW_MULTISAMPLE),
- .brw = (BRW_NEW_FRAGMENT_PROGRAM |
- BRW_NEW_REDUCED_PRIMITIVE |
- BRW_NEW_VUE_MAP_GEOM_OUT |
- BRW_NEW_STATS_WM)
- },
- .emit = brw_upload_wm_prog
-};
+ 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);
+
+ if (fp->Base.InputsRead & VARYING_BIT_POS) {
+ key.drawable_height = ctx->DrawBuffer->Height;
+ }
+ key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
+ ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
+ BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
+
+ 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.program_string_id = bfp->id;
+
+ 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;
+}