Pipeline strategy (#72)

* Attempt to aid the driver when creating ubershader pipelines.

* Second attempt.

* Missing include.

* Change the index in the condition.

src/render/rt64_raster_shader.cpp

@@ -531,9 +531,28 @@ namespace RT64 {
     }
 
     void RasterShaderUber::threadCreatePipelines(uint32_t threadIndex) {
+        // Delay the creation of all other pipelines until the first pipeline is created. This can help the
+        // driver reuse its shader cache between pipelines and achieve a much lower creation time than if
+        // all threads started at the same time.
+        if (threadIndex > 0) {
+            std::unique_lock<std::mutex> lock(firstPipelineMutex);
+            firstPipelineCondition.wait(lock, [this]() {
+                return (pipelines[0] != nullptr);
+            });
+        }
+
         for (const PipelineCreation &creation : pipelineThreadCreations[threadIndex]) {
             uint32_t pipelineIndex = pipelineStateIndex(creation.zCmp, creation.zUpd, creation.cvgAdd);
+
+            if (pipelineIndex == 0) {
+                firstPipelineMutex.lock();
                 pipelines[pipelineIndex] = RasterShader::createPipeline(creation);
+                firstPipelineMutex.unlock();
+                firstPipelineCondition.notify_all();
+            }
+            else {
+                pipelines[pipelineIndex] = RasterShader::createPipeline(creation);
+            }
         }
     }
 

src/render/rt64_raster_shader.h

@@ -6,6 +6,7 @@
 
 #include "rt64_shader_common.h"
 
+#include <condition_variable>
 #include <mutex>
 #include <thread>
 
@@ -78,7 +79,8 @@ namespace RT64 {
         std::unique_ptr<RenderPipeline> pipelines[8];
         std::unique_ptr<RenderPipeline> postBlendDitherNoiseAddPipeline;
         std::unique_ptr<RenderPipeline> postBlendDitherNoiseSubPipeline;
-        std::mutex pipelinesMutex;
+        std::mutex firstPipelineMutex;
+        std::condition_variable firstPipelineCondition;
         bool pipelinesCreated = false;
         std::unique_ptr<RenderPipelineLayout> pipelineLayout;
         std::vector<std::vector<PipelineCreation>> pipelineThreadCreations;
@@ -88,6 +90,7 @@ namespace RT64 {
 
         RasterShaderUber(RenderDevice *device, RenderShaderFormat shaderFormat, const RenderMultisampling &multisampling, const ShaderLibrary *shaderLibrary, uint32_t threadCount);
         ~RasterShaderUber();
+        void threadCreatePipeline(const PipelineCreation &creation);
         void threadCreatePipelines(uint32_t threadIndex);
         void waitForPipelineCreation();
         uint32_t pipelineStateIndex(bool zCmp, bool zUpd, bool cvgAdd) const;