EEVEE: Default Startup Speedup

Low hanging fruit optimizations for improving default
startup time.

Went from 7.2sec to 4sec on my system.

Pull Request: https://projects.blender.org/blender/blender/pulls/139278

source/blender/draw/engines/eevee/eevee_pipeline.cc

@@ -1275,7 +1275,7 @@ void DeferredProbePipeline::begin_sync()
 
 void DeferredProbePipeline::end_sync()
 {
-  {
+  if (!opaque_layer_.prepass_ps_.is_empty()) {
     PassSimple &pass = eval_light_ps_;
     pass.init();
     /* Use depth test to reject background pixels. */
@@ -1388,7 +1388,13 @@ void PlanarProbePipeline::begin_sync()
 
   this->gbuffer_pass_sync(inst_);
 
-  {
+  closure_bits_ = CLOSURE_NONE;
+  closure_count_ = 0;
+}
+
+void PlanarProbePipeline::end_sync()
+{
+  if (!prepass_ps_.is_empty()) {
     PassSimple &pass = eval_light_ps_;
     pass.init();
     pass.state_set(DRW_STATE_WRITE_COLOR | DRW_STATE_BLEND_ADD_FULL);
@@ -1405,14 +1411,6 @@ void PlanarProbePipeline::begin_sync()
     pass.barrier(GPU_BARRIER_TEXTURE_FETCH | GPU_BARRIER_SHADER_IMAGE_ACCESS);
     pass.draw_procedural(GPU_PRIM_TRIS, 1, 3);
   }
-
-  closure_bits_ = CLOSURE_NONE;
-  closure_count_ = 0;
-}
-
-void PlanarProbePipeline::end_sync()
-{
-  /* No-op for now. */
 }
 
 PassMain::Sub *PlanarProbePipeline::prepass_add(::Material *blender_mat, GPUMaterial *gpumat)

source/blender/draw/engines/eevee/eevee_pipeline.hh

@@ -252,6 +252,11 @@ struct DeferredLayerBase {
     return count;
   }
 
+  eClosureBits closure_bits_get() const
+  {
+    return closure_bits_;
+  }
+
   void gbuffer_pass_sync(Instance &inst);
 };
 
@@ -406,6 +411,11 @@ class DeferredPipeline {
     return opaque_layer_.is_empty() && refraction_layer_.is_empty();
   }
 
+  eClosureBits closure_bits_get() const
+  {
+    return opaque_layer_.closure_bits_get() | refraction_layer_.closure_bits_get();
+  }
+
  private:
   void debug_pass_sync();
 };

source/blender/draw/engines/eevee/eevee_subsurface.cc

@@ -21,6 +21,10 @@ void SubsurfaceModule::end_sync()
 {
   data_.sample_len = 16;
 
+  if (!(inst_.pipelines.deferred.closure_bits_get() & CLOSURE_SSS)) {
+    return;
+  }
+
   {
     PassSimple &pass = setup_ps_;
     pass.init();

source/blender/draw/engines/eevee/shaders/infos/eevee_deferred_info.hh

@@ -69,8 +69,8 @@ IMAGE_FREQ(4, DEFERRED_RADIANCE_FORMAT, write, uimage2D, direct_radiance_3_img,
 IMAGE_FREQ(5, RAYTRACE_RADIANCE_FORMAT, write, image2D, indirect_radiance_1_img, PASS)
 IMAGE_FREQ(6, RAYTRACE_RADIANCE_FORMAT, write, image2D, indirect_radiance_2_img, PASS)
 IMAGE_FREQ(7, RAYTRACE_RADIANCE_FORMAT, write, image2D, indirect_radiance_3_img, PASS)
-SPECIALIZATION_CONSTANT(bool, use_split_indirect, false)
-SPECIALIZATION_CONSTANT(bool, use_lightprobe_eval, false)
+SPECIALIZATION_CONSTANT(bool, use_split_indirect, true)
+SPECIALIZATION_CONSTANT(bool, use_lightprobe_eval, true)
 SPECIALIZATION_CONSTANT(bool, use_transmission, false)
 SPECIALIZATION_CONSTANT(int, render_pass_shadow_id, -1)
 DEFINE("SPECIALIZED_SHADOW_PARAMS")
@@ -127,11 +127,11 @@ ADDITIONAL_INFO(gpu_fullscreen)
 FRAGMENT_SOURCE("eevee_deferred_combine_frag.glsl")
 /* NOTE: Both light IDs have a valid specialized assignment of '-1' so only when default is
  * present will we instead dynamically look-up ID from the uniform buffer. */
-SPECIALIZATION_CONSTANT(bool, render_pass_diffuse_light_enabled, true)
-SPECIALIZATION_CONSTANT(bool, render_pass_specular_light_enabled, true)
-SPECIALIZATION_CONSTANT(bool, render_pass_normal_enabled, true)
+SPECIALIZATION_CONSTANT(bool, render_pass_diffuse_light_enabled, false)
+SPECIALIZATION_CONSTANT(bool, render_pass_specular_light_enabled, false)
+SPECIALIZATION_CONSTANT(bool, render_pass_normal_enabled, false)
 SPECIALIZATION_CONSTANT(bool, use_radiance_feedback, false)
-SPECIALIZATION_CONSTANT(bool, use_split_radiance, false)
+SPECIALIZATION_CONSTANT(bool, use_split_radiance, true)
 DO_STATIC_COMPILATION()
 GPU_SHADER_CREATE_END()
 

source/blender/draw/engines/eevee/shaders/infos/eevee_film_info.hh

@@ -27,9 +27,9 @@ SAMPLER(5, sampler2D, in_combined_tx)
 SAMPLER(6, sampler2D, cryptomatte_tx)
 IMAGE(0, GPU_R32F, read, image2DArray, in_weight_img)
 IMAGE(1, GPU_R32F, write, image2DArray, out_weight_img)
-SPECIALIZATION_CONSTANT(uint, enabled_categories, 0)
-SPECIALIZATION_CONSTANT(int, samples_len, 0)
-SPECIALIZATION_CONSTANT(bool, use_reprojection, false)
+SPECIALIZATION_CONSTANT(uint, enabled_categories, 1)
+SPECIALIZATION_CONSTANT(int, samples_len, 9)
+SPECIALIZATION_CONSTANT(bool, use_reprojection, true)
 SPECIALIZATION_CONSTANT(int, scaling_factor, 1)
 SPECIALIZATION_CONSTANT(int, combined_id, 0)
 SPECIALIZATION_CONSTANT(int, display_id, -1)

source/blender/gpu/GPU_debug.hh

@@ -62,6 +62,7 @@
 #pragma once
 
 #define GPU_DEBUG_SHADER_COMPILATION_GROUP "Shader Compilation"
+#define GPU_DEBUG_SHADER_SPECIALIZATION_GROUP "Shader Specialization"
 
 void GPU_debug_group_begin(const char *name);
 void GPU_debug_group_end();

source/blender/gpu/opengl/gl_debug.cc

@@ -92,7 +92,9 @@ static void APIENTRY debug_callback(GLenum /*source*/,
     GPU_debug_get_groups_names(sizeof(debug_groups), debug_groups);
     CLG_Severity clog_severity;
 
-    if (GPU_debug_group_match(GPU_DEBUG_SHADER_COMPILATION_GROUP)) {
+    if (GPU_debug_group_match(GPU_DEBUG_SHADER_COMPILATION_GROUP) ||
+        GPU_debug_group_match(GPU_DEBUG_SHADER_SPECIALIZATION_GROUP))
+    {
       /* Do not duplicate shader compilation error/warnings. */
       return;
     }

source/blender/gpu/opengl/gl_shader.cc

@@ -19,6 +19,7 @@
 #include BLI_SYSTEM_PID_H
 
 #include "GPU_capabilities.hh"
+#include "GPU_debug.hh"
 #include "GPU_platform.hh"
 #include "gpu_capabilities_private.hh"
 #include "gpu_shader_dependency_private.hh"
@@ -1617,8 +1618,14 @@ GLShader::GLProgram &GLShader::program_get(const shader::SpecializationConstants
     return program;
   }
 
+  GPU_debug_group_begin(GPU_DEBUG_SHADER_SPECIALIZATION_GROUP);
+  GPU_debug_group_begin(this->name);
+
   program.program_link(name);
 
+  GPU_debug_group_end();
+  GPU_debug_group_end();
+
   return program;
 }