GPU Vulkan Display: Use in-memory pipeline cache

Author: davidrohr

GPU/GPUTracking/display/backend/GPUDisplayBackendVulkan.cxx

@@ -575,11 +575,13 @@ void GPUDisplayBackendVulkan::createDevice()
   poolInfo.flags = vk::CommandPoolCreateFlagBits::eResetCommandBuffer;
   poolInfo.queueFamilyIndex = mGraphicsFamily;
   mCommandPool = mDevice.createCommandPool(poolInfo, nullptr);
+  mPipelineCache = mDevice.createPipelineCache(vk::PipelineCacheCreateInfo{}, nullptr);
 }
 
 void GPUDisplayBackendVulkan::clearDevice()
 {
   mDevice.destroyCommandPool(mCommandPool, nullptr);
+  mDevice.destroyPipelineCache(mPipelineCache, nullptr);
   mDevice.destroy(nullptr);
   mInstance.destroySurfaceKHR(mSurface, nullptr);
   if (mEnableValidationLayers) {
@@ -1078,7 +1080,7 @@ void GPUDisplayBackendVulkan::createPipeline()
       multisampling.rasterizationSamples = mMSAASampleCount;
     }
 
-    CHKERR(mDevice.createGraphicsPipelines(VkPipelineCache(VK_NULL_HANDLE), 1, &pipelineInfo, nullptr, &mPipelines[i])); // TODO: multiple at once + cache?
+    CHKERR(mDevice.createGraphicsPipelines(mPipelineCache, 1, &pipelineInfo, nullptr, &mPipelines[i]));
   }
 }
 

GPU/GPUTracking/display/backend/GPUDisplayBackendVulkan.h

@@ -163,6 +163,7 @@ class GPUDisplayBackendVulkan : public GPUDisplayBackend
   vk::PipelineLayout mPipelineLayoutTexture;
   std::vector<vk::Pipeline> mPipelines;
   vk::CommandPool mCommandPool;
+  vk::PipelineCache mPipelineCache;
 
   bool mCommandInfrastructureCreated = false;
   uint32_t mImageCount = 0;
@@ -211,7 +212,7 @@ class GPUDisplayBackendVulkan : public GPUDisplayBackend
   bool mZActive = false;
   bool mSmoothLinesSupported = false; // Optional Vulkan 1.4 feature smoothLines, enabled if available
   bool mSmoothLines = false;
-  bool mStrictLines = false; // Wide lines are rectangles, not parallelograms
+  bool mStrictLines = false;
   bool mZSupported = false;
   bool mStencilSupported = false;
   bool mDownsampleFSAA = false;