aurora: Integrate Optick & hashing performance improvements

Runtime/CMain.cpp

@@ -351,8 +351,6 @@ public:
       }
     }
 
-    OPTICK_FRAME("MainThread");
-
     // Check if fullscreen has been toggled, if so set the fullscreen cvar accordingly
     if (m_fullscreenToggleRequested) {
       m_cvarCommons.m_fullscreen->fromBoolean(!m_cvarCommons.getFullscreen());

aurora/CMakeLists.txt

@@ -19,8 +19,12 @@ add_library(aurora STATIC
 target_compile_definitions(aurora PRIVATE IMGUI_USER_CONFIG="imconfig_user.h") # IMGUI_USE_WCHAR32
 target_include_directories(aurora PUBLIC include ../)
 target_include_directories(aurora PRIVATE ../imgui ../extern/imgui)
+target_include_directories(aurora PRIVATE
+    ../extern/dawn/src
+    ../extern/dawn/third_party/abseil-cpp
+    ${CMAKE_BINARY_DIR}/extern/dawn/gen/src) # for hacks :)
 target_link_libraries(aurora PRIVATE dawn_native dawncpp webgpu_dawn zeus logvisor SDL2-static xxhash
-    absl::btree absl::flat_hash_map)
+    absl::btree absl::flat_hash_map OptickCore)
 if (APPLE)
   target_compile_definitions(aurora PRIVATE DAWN_ENABLE_BACKEND_METAL)
   target_sources(aurora PRIVATE lib/dawn/MetalBinding.mm)

aurora/lib/aurora.cpp

@@ -280,12 +280,18 @@ void app_run(std::unique_ptr<AppDelegate> app, Icon icon, int argc, char** argv)
   g_AppDelegate->onAppWindowResized(size);
 
   while (poll_events()) {
+    OPTICK_FRAME("MainThread");
+
     imgui::new_frame(g_windowSize);
     if (!g_AppDelegate->onAppIdle(ImGui::GetIO().DeltaTime)) {
       break;
     }
 
-    const wgpu::TextureView view = g_swapChain.GetCurrentTextureView();
+    wgpu::TextureView view;
+    {
+      OPTICK_EVENT("SwapChain GetCurrentTextureView");
+      view = g_swapChain.GetCurrentTextureView();
+    }
     gfx::begin_frame();
     g_AppDelegate->onAppDraw();
 
@@ -293,8 +299,12 @@ void app_run(std::unique_ptr<AppDelegate> app, Icon icon, int argc, char** argv)
         .label = "Redraw encoder",
     };
     auto encoder = g_device.CreateCommandEncoder(&encoderDescriptor);
+#if USE_OPTICK
+    auto prevContext = gpu::begin_cmdlist();
+#endif
     gfx::end_frame(encoder);
     {
+      OPTICK_EVENT("Main Render Pass");
       const std::array attachments{
           wgpu::RenderPassColorAttachment{
               .view = view,
@@ -329,6 +339,7 @@ void app_run(std::unique_ptr<AppDelegate> app, Icon icon, int argc, char** argv)
       pass.End();
     }
     {
+      OPTICK_EVENT("ImGui Render Pass");
       const std::array attachments{
           wgpu::RenderPassColorAttachment{
               .view = view,
@@ -345,9 +356,19 @@ void app_run(std::unique_ptr<AppDelegate> app, Icon icon, int argc, char** argv)
       imgui::render(pass);
       pass.End();
     }
-    const auto buffer = encoder.Finish();
-    g_queue.Submit(1, &buffer);
-    g_swapChain.Present();
+#if USE_OPTICK
+    gpu::end_cmdlist(prevContext);
+#endif
+    {
+      OPTICK_EVENT("Queue Submit");
+      const auto buffer = encoder.Finish();
+      g_queue.Submit(1, &buffer);
+    }
+    {
+      OPTICK_GPU_FLIP(gpu::get_native_swapchain());
+      OPTICK_CATEGORY("Present", Optick::Category::Wait);
+      g_swapChain.Present();
+    }
 
     g_AppDelegate->onAppPostDraw();
 

aurora/lib/dawn/BackendBinding.cpp

@@ -8,22 +8,23 @@
 namespace aurora::gpu::utils {
 
 #if defined(DAWN_ENABLE_BACKEND_D3D12)
-BackendBinding* CreateD3D12Binding(SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateD3D12Binding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_METAL)
-BackendBinding* CreateMetalBinding(SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateMetalBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_NULL)
-BackendBinding* CreateNullBinding(SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateNullBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_OPENGL)
-BackendBinding* CreateOpenGLBinding(SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateOpenGLBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_VULKAN)
-BackendBinding* CreateVulkanBinding(SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateVulkanBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 #endif
 
-BackendBinding::BackendBinding(SDL_Window* window, WGPUDevice device) : m_window(window), m_device(device) {}
+BackendBinding::BackendBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device)
+: m_window(window), m_adapter(adapter), m_device(device) {}
 
 void DiscoverAdapter(dawn::native::Instance* instance, SDL_Window* window, wgpu::BackendType type) {
   if (type == wgpu::BackendType::OpenGL || type == wgpu::BackendType::OpenGLES) {
@@ -45,31 +46,31 @@ void DiscoverAdapter(dawn::native::Instance* instance, SDL_Window* window, wgpu:
   }
 }
 
-BackendBinding* CreateBinding(wgpu::BackendType type, SDL_Window* window, WGPUDevice device) {
+BackendBinding* CreateBinding(wgpu::BackendType type, SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) {
   switch (type) {
 #if defined(DAWN_ENABLE_BACKEND_D3D12)
   case wgpu::BackendType::D3D12:
-    return CreateD3D12Binding(window, device);
+    return CreateD3D12Binding(window, adapter, device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_METAL)
   case wgpu::BackendType::Metal:
-    return CreateMetalBinding(window, device);
+    return CreateMetalBinding(window, adapter, device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_NULL)
   case wgpu::BackendType::Null:
-    return CreateNullBinding(window, device);
+    return CreateNullBinding(window, adapter, device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_DESKTOP_GL)
   case wgpu::BackendType::OpenGL:
-    return CreateOpenGLBinding(window, device);
+    return CreateOpenGLBinding(window, adapter, device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_OPENGLES)
   case wgpu::BackendType::OpenGLES:
-    return CreateOpenGLBinding(window, device);
+    return CreateOpenGLBinding(window, adapter, device);
 #endif
 #if defined(DAWN_ENABLE_BACKEND_VULKAN)
   case wgpu::BackendType::Vulkan:
-    return CreateVulkanBinding(window, device);
+    return CreateVulkanBinding(window, adapter, device);
 #endif
   default:
     return nullptr;

aurora/lib/dawn/BackendBinding.hpp

@@ -2,6 +2,7 @@
 
 #include <dawn/native/DawnNative.h>
 #include <dawn/webgpu_cpp.h>
+#include <optick.h>
 
 struct SDL_Window;
 
@@ -13,15 +14,20 @@ public:
 
   virtual uint64_t GetSwapChainImplementation() = 0;
   virtual WGPUTextureFormat GetPreferredSwapChainTextureFormat() = 0;
+#if USE_OPTICK
+  virtual void* GetNativeSwapChain() { return nullptr; };
+  virtual Optick::GPUContext OptickSetGpuContext() { return {}; };
+#endif
 
 protected:
-  BackendBinding(SDL_Window* window, WGPUDevice device);
+  BackendBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 
   SDL_Window* m_window = nullptr;
+  WGPUAdapter m_adapter = nullptr;
   WGPUDevice m_device = nullptr;
 };
 
 void DiscoverAdapter(dawn::native::Instance* instance, SDL_Window* window, wgpu::BackendType type);
-BackendBinding* CreateBinding(wgpu::BackendType type, SDL_Window* window, WGPUDevice device);
+BackendBinding* CreateBinding(wgpu::BackendType type, SDL_Window* window, WGPUAdapter adapter, WGPUDevice device);
 
 } // namespace aurora::gpu::utils

aurora/lib/dawn/D3D12Binding.cpp

@@ -2,11 +2,21 @@
 
 #include <SDL_syswm.h>
 #include <dawn/native/D3D12Backend.h>
+#include <optick.h>
+
+#if USE_OPTICK
+// Internal headers
+#include <dawn/native/d3d12/CommandBufferD3D12.h>
+#include <dawn/native/d3d12/DeviceD3D12.h>
+#define private public
+#include <dawn/native/d3d12/NativeSwapChainImplD3D12.h>
+#undef private
+#endif
 
 namespace aurora::gpu::utils {
 class D3D12Binding : public BackendBinding {
 public:
-  D3D12Binding(SDL_Window* window, WGPUDevice device) : BackendBinding(window, device) {}
+  D3D12Binding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) : BackendBinding(window, adapter, device) {}
 
   uint64_t GetSwapChainImplementation() override {
     if (m_swapChainImpl.userData == nullptr) {
@@ -22,6 +32,19 @@ public:
     return dawn::native::d3d12::GetNativeSwapChainPreferredFormat(&m_swapChainImpl);
   }
 
+#if USE_OPTICK
+  void* GetNativeSwapChain() override {
+    auto* impl = static_cast<dawn::native::d3d12::NativeSwapChainImpl*>(m_swapChainImpl.userData);
+    return impl->mSwapChain.Get();
+  }
+
+  Optick::GPUContext OptickSetGpuContext() override {
+    auto* device = dawn::native::d3d12::ToBackend(static_cast<dawn::native::DeviceBase*>(static_cast<void*>(m_device)));
+    auto* commandList = device->GetPendingCommandContext().AcquireSuccess()->GetCommandList();
+    return Optick::SetGpuContext({commandList});
+  }
+#endif
+
 private:
   DawnSwapChainImplementation m_swapChainImpl{};
 
@@ -30,8 +53,16 @@ private:
     SDL_VERSION(&wmInfo.version);
     SDL_GetWindowWMInfo(m_window, &wmInfo);
     m_swapChainImpl = dawn::native::d3d12::CreateNativeSwapChainImpl(m_device, wmInfo.info.win.window);
+#if USE_OPTICK
+    auto* device = dawn::native::d3d12::ToBackend(static_cast<dawn::native::DeviceBase*>(static_cast<void*>(m_device)));
+    auto* d3d12Device = device->GetD3D12Device();
+    auto* d3d12CommandQueue = device->GetCommandQueue().Get();
+    OPTICK_GPU_INIT_D3D12(d3d12Device, &d3d12CommandQueue, 1);
+#endif
   }
 };
 
-BackendBinding* CreateD3D12Binding(SDL_Window* window, WGPUDevice device) { return new D3D12Binding(window, device); }
+BackendBinding* CreateD3D12Binding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) {
+  return new D3D12Binding(window, adapter, device);
+}
 } // namespace aurora::gpu::utils

aurora/lib/dawn/MetalBinding.mm

@@ -89,7 +89,7 @@ private:
 
 class MetalBinding : public BackendBinding {
 public:
-  MetalBinding(SDL_Window *window, WGPUDevice device) : BackendBinding(window, device) {}
+  MetalBinding(SDL_Window *window, WGPUAdapter adapter, WGPUDevice device) : BackendBinding(window, adapter, device) {}
 
   uint64_t GetSwapChainImplementation() override {
     if (m_swapChainImpl.userData == nullptr) {
@@ -104,5 +104,7 @@ private:
   DawnSwapChainImplementation m_swapChainImpl{};
 };
 
-BackendBinding *CreateMetalBinding(SDL_Window *window, WGPUDevice device) { return new MetalBinding(window, device); }
+BackendBinding *CreateMetalBinding(SDL_Window *window, WGPUAdapter adapter, WGPUDevice device) {
+  return new MetalBinding(window, adapter, device);
+}
 } // namespace aurora::gpu::utils

aurora/lib/dawn/OpenGLBinding.cpp

@@ -6,7 +6,7 @@
 namespace aurora::gpu::utils {
 class OpenGLBinding : public BackendBinding {
 public:
-  OpenGLBinding(SDL_Window* window, WGPUDevice device) : BackendBinding(window, device) {}
+  OpenGLBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) : BackendBinding(window, adapter, device) {}
 
   uint64_t GetSwapChainImplementation() override {
     if (m_swapChainImpl.userData == nullptr) {
@@ -31,5 +31,7 @@ private:
   }
 };
 
-BackendBinding* CreateOpenGLBinding(SDL_Window* window, WGPUDevice device) { return new OpenGLBinding(window, device); }
+BackendBinding* CreateOpenGLBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) {
+  return new OpenGLBinding(window, adapter, device);
+}
 } // namespace aurora::gpu::utils

aurora/lib/dawn/VulkanBinding.cpp

@@ -1,13 +1,24 @@
 #include "BackendBinding.hpp"
 
-#include <SDL_vulkan.h>
 #include <cassert>
+#include <optick.h>
+
+#if USE_OPTICK
+// Internal headers
+#include <dawn/native/vulkan/AdapterVk.h>
+#include <dawn/native/vulkan/DeviceVk.h>
+#define private public
+#include <dawn/native/vulkan/NativeSwapChainImplVk.h>
+#undef private
+#endif
+
+#include <SDL_vulkan.h>
 #include <dawn/native/VulkanBackend.h>
 
 namespace aurora::gpu::utils {
 class VulkanBinding : public BackendBinding {
 public:
-  VulkanBinding(SDL_Window* window, WGPUDevice device) : BackendBinding(window, device) {}
+  VulkanBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) : BackendBinding(window, adapter, device) {}
 
   uint64_t GetSwapChainImplementation() override {
     if (m_swapChainImpl.userData == nullptr) {
@@ -23,6 +34,19 @@ public:
     return dawn::native::vulkan::GetNativeSwapChainPreferredFormat(&m_swapChainImpl);
   }
 
+#if USE_OPTICK
+  void* GetNativeSwapChain() override {
+    auto* impl = static_cast<dawn::native::vulkan::NativeSwapChainImpl*>(m_swapChainImpl.userData);
+    return impl->mSwapChain;
+  }
+
+  Optick::GPUContext OptickSetGpuContext() override {
+    auto* device =
+        dawn::native::vulkan::ToBackend(static_cast<dawn::native::DeviceBase*>(static_cast<void*>(m_device)));
+    return Optick::SetGpuContext({device->GetPendingRecordingContext()->commandBuffer});
+  }
+#endif
+
 private:
   DawnSwapChainImplementation m_swapChainImpl{};
 
@@ -32,8 +56,21 @@ private:
       assert(false);
     }
     m_swapChainImpl = dawn::native::vulkan::CreateNativeSwapChainImpl(m_device, surface);
+#if USE_OPTICK
+    auto* adapter =
+        dawn::native::vulkan::ToBackend(static_cast<dawn::native::AdapterBase*>(static_cast<void*>(m_adapter)));
+    auto* device =
+        dawn::native::vulkan::ToBackend(static_cast<dawn::native::DeviceBase*>(static_cast<void*>(m_device)));
+    VkDevice vkDevice = device->GetVkDevice();
+    VkPhysicalDevice vkPhysicalDevice = adapter->GetPhysicalDevice();
+    VkQueue vkQueue = device->GetQueue();
+    uint32_t queueFamily = device->GetGraphicsQueueFamily();
+    OPTICK_GPU_INIT_VULKAN(&vkDevice, &vkPhysicalDevice, &vkQueue, &queueFamily, 1, nullptr);
+#endif
   }
 };
 
-BackendBinding* CreateVulkanBinding(SDL_Window* window, WGPUDevice device) { return new VulkanBinding(window, device); }
+BackendBinding* CreateVulkanBinding(SDL_Window* window, WGPUAdapter adapter, WGPUDevice device) {
+  return new VulkanBinding(window, adapter, device);
+}
 } // namespace aurora::gpu::utils

aurora/lib/gfx/colored_quad/shader.cpp

@@ -298,6 +298,7 @@ DrawData make_draw_data_verts(const State& state, CameraFilterType filter_type,
 }
 
 void render(const State& state, const DrawData& data, const wgpu::RenderPassEncoder& pass) {
+  OPTICK_EVENT();
   if (!bind_pipeline(data.pipeline, pass)) {
     return;
   }

aurora/lib/gfx/common.cpp

@@ -84,107 +84,44 @@ struct Command {
 
 namespace aurora {
 template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::colored_quad::PipelineConfig& input) {
-  XXH3_64bits_update(&state, &input.filterType, sizeof(gfx::colored_quad::PipelineConfig::filterType));
-  XXH3_64bits_update(&state, &input.zComparison, sizeof(gfx::colored_quad::PipelineConfig::zComparison));
-  XXH3_64bits_update(&state, &input.zTest, sizeof(gfx::colored_quad::PipelineConfig::zTest));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::textured_quad::PipelineConfig& input) {
-  XXH3_64bits_update(&state, &input.filterType, sizeof(gfx::textured_quad::PipelineConfig::filterType));
-  XXH3_64bits_update(&state, &input.zComparison, sizeof(gfx::textured_quad::PipelineConfig::zComparison));
-  XXH3_64bits_update(&state, &input.zTest, sizeof(gfx::textured_quad::PipelineConfig::zTest));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::movie_player::PipelineConfig& input) {
-  // no-op
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::PipelineConfig& input) {
-  xxh3_update(state, input.shaderConfig);
-  XXH3_64bits_update(&state, &input.primitive, sizeof(gfx::gx::PipelineConfig::primitive));
-  XXH3_64bits_update(&state, &input.depthFunc, sizeof(gfx::gx::PipelineConfig::depthFunc));
-  XXH3_64bits_update(&state, &input.cullMode, sizeof(gfx::gx::PipelineConfig::cullMode));
-  XXH3_64bits_update(&state, &input.blendMode, sizeof(gfx::gx::PipelineConfig::blendMode));
-  XXH3_64bits_update(&state, &input.blendFacSrc, sizeof(gfx::gx::PipelineConfig::blendFacSrc));
-  XXH3_64bits_update(&state, &input.blendFacDst, sizeof(gfx::gx::PipelineConfig::blendFacDst));
-  XXH3_64bits_update(&state, &input.blendOp, sizeof(gfx::gx::PipelineConfig::blendOp));
-  if (input.dstAlpha) {
-    XXH3_64bits_update(&state, &*input.dstAlpha, sizeof(float));
-  }
-  XXH3_64bits_update(&state, &input.depthCompare, sizeof(gfx::gx::PipelineConfig::depthCompare));
-  XXH3_64bits_update(&state, &input.depthUpdate, sizeof(gfx::gx::PipelineConfig::depthUpdate));
-  XXH3_64bits_update(&state, &input.alphaUpdate, sizeof(gfx::gx::PipelineConfig::alphaUpdate));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::stream::PipelineConfig& input) {
-  xxh3_update<gfx::gx::PipelineConfig>(state, input);
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::model::PipelineConfig& input) {
-  xxh3_update<gfx::gx::PipelineConfig>(state, input);
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::PipelineCreateCommand& input) {
-  XXH3_64bits_update(&state, &input.type, sizeof(gfx::PipelineCreateCommand::type));
+inline XXH64_hash_t xxh3_hash(const gfx::PipelineCreateCommand& input, XXH64_hash_t seed) {
+  constexpr auto typeSize = sizeof(gfx::PipelineCreateCommand::type);
   switch (input.type) {
   case gfx::ShaderType::Aabb:
     // TODO
     break;
   case gfx::ShaderType::ColoredQuad:
-    xxh3_update(state, input.coloredQuad);
-    break;
+    return xxh3_hash_s(&input, typeSize + sizeof(gfx::PipelineCreateCommand::coloredQuad), seed);
   case gfx::ShaderType::TexturedQuad:
-    xxh3_update(state, input.texturedQuad);
-    break;
+    return xxh3_hash_s(&input, typeSize + sizeof(gfx::PipelineCreateCommand::texturedQuad), seed);
   case gfx::ShaderType::MoviePlayer:
-    xxh3_update(state, input.moviePlayer);
-    break;
+    return xxh3_hash_s(&input, typeSize + sizeof(gfx::PipelineCreateCommand::moviePlayer), seed);
   case gfx::ShaderType::Stream:
-    xxh3_update(state, input.stream);
-    break;
+    return xxh3_hash_s(&input, typeSize + sizeof(gfx::PipelineCreateCommand::stream), seed);
   case gfx::ShaderType::Model:
-    xxh3_update(state, input.model);
-    break;
+    return xxh3_hash_s(&input, typeSize + sizeof(gfx::PipelineCreateCommand::model), seed);
   }
+  return 0;
 }
 template <>
-inline void xxh3_update(XXH3_state_t& state, const wgpu::BindGroupEntry& input) {
-  XXH3_64bits_update(&state, &input.binding, sizeof(wgpu::BindGroupEntry::binding));
-  XXH3_64bits_update(&state, &input.buffer, sizeof(wgpu::BindGroupEntry::buffer));
-  XXH3_64bits_update(&state, &input.offset, sizeof(wgpu::BindGroupEntry::offset));
-  if (input.buffer != nullptr) {
-    XXH3_64bits_update(&state, &input.size, sizeof(wgpu::BindGroupEntry::size));
-  }
-  XXH3_64bits_update(&state, &input.sampler, sizeof(wgpu::BindGroupEntry::sampler));
-  XXH3_64bits_update(&state, &input.textureView, sizeof(wgpu::BindGroupEntry::textureView));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const wgpu::BindGroupDescriptor& input) {
-  if (input.label != nullptr) {
-    XXH3_64bits_update(&state, input.label, strlen(input.label));
-  }
-  XXH3_64bits_update(&state, &input.layout, sizeof(wgpu::BindGroupDescriptor::layout));
-  XXH3_64bits_update(&state, &input.entryCount, sizeof(wgpu::BindGroupDescriptor::entryCount));
-  for (int i = 0; i < input.entryCount; ++i) {
-    xxh3_update(state, input.entries[i]);
-  }
+inline XXH64_hash_t xxh3_hash(const wgpu::BindGroupDescriptor& input, XXH64_hash_t seed) {
+  constexpr auto offset = sizeof(void*) * 2; // skip nextInChain, label
+  const auto hash = xxh3_hash_s(reinterpret_cast<const u8*>(&input) + offset,
+                                sizeof(wgpu::BindGroupDescriptor) - offset - sizeof(void*) /* skip entries */, seed);
+  //  // TODO ensure size is zeroed elsewhere
+  //  for (int i = 0; i < input.entryCount; ++i) {
+  //    const wgpu::BindGroupEntry& entry = input.entries[i];
+  //    if (!entry.buffer && entry.size != 0) {
+  //      gfx::Log.report(logvisor::Fatal, FMT_STRING("Size != 0! {}"), entry.size);
+  //    }
+  //  }
+  return xxh3_hash_s(input.entries, sizeof(wgpu::BindGroupEntry) * input.entryCount, hash);
 }
 template <>
-inline void xxh3_update(XXH3_state_t& state, const wgpu::SamplerDescriptor& input) {
-  if (input.label != nullptr) {
-    XXH3_64bits_update(&state, input.label, strlen(input.label));
-  }
-  XXH3_64bits_update(&state, &input.addressModeU, sizeof(wgpu::SamplerDescriptor::addressModeU));
-  XXH3_64bits_update(&state, &input.addressModeV, sizeof(wgpu::SamplerDescriptor::addressModeV));
-  XXH3_64bits_update(&state, &input.addressModeW, sizeof(wgpu::SamplerDescriptor::addressModeW));
-  XXH3_64bits_update(&state, &input.magFilter, sizeof(wgpu::SamplerDescriptor::magFilter));
-  XXH3_64bits_update(&state, &input.minFilter, sizeof(wgpu::SamplerDescriptor::minFilter));
-  XXH3_64bits_update(&state, &input.mipmapFilter, sizeof(wgpu::SamplerDescriptor::mipmapFilter));
-  XXH3_64bits_update(&state, &input.lodMinClamp, sizeof(wgpu::SamplerDescriptor::lodMinClamp));
-  XXH3_64bits_update(&state, &input.lodMaxClamp, sizeof(wgpu::SamplerDescriptor::lodMaxClamp));
-  XXH3_64bits_update(&state, &input.compare, sizeof(wgpu::SamplerDescriptor::compare));
-  XXH3_64bits_update(&state, &input.maxAnisotropy, sizeof(wgpu::SamplerDescriptor::maxAnisotropy));
+inline XXH64_hash_t xxh3_hash(const wgpu::SamplerDescriptor& input, XXH64_hash_t seed) {
+  constexpr auto offset = sizeof(void*) * 2; // skip nextInChain, label
+  return xxh3_hash_s(reinterpret_cast<const u8*>(&input) + offset,
+                     sizeof(wgpu::SamplerDescriptor) - offset - 2 /* skip padding */, seed);
 }
 } // namespace aurora
 
@@ -219,6 +156,7 @@ static PipelineRef g_currentPipeline;
 static std::vector<Command> g_commands;
 
 static PipelineRef find_pipeline(PipelineCreateCommand command, NewPipelineCallback&& cb) {
+  OPTICK_EVENT();
   const auto hash = xxh3_hash(command);
   bool found = false;
   {
@@ -357,6 +295,7 @@ PipelineRef pipeline_ref(model::PipelineConfig config) {
 }
 
 static void pipeline_worker() {
+  OPTICK_THREAD("Pipeline Worker");
   bool hasMore = false;
   while (true) {
     std::pair<PipelineRef, NewPipelineCallback> cb;
@@ -387,6 +326,8 @@ static void pipeline_worker() {
 }
 
 void initialize() {
+  gx::initialize();
+
   g_pipelineThread = std::thread(pipeline_worker);
 
   const auto createBuffer = [](wgpu::Buffer& out, wgpu::BufferUsage usage, uint64_t size, const char* label) {
@@ -440,6 +381,7 @@ void shutdown() {
 static size_t currentStagingBuffer = 0;
 static bool bufferMapped = false;
 void map_staging_buffer() {
+  OPTICK_EVENT();
   bufferMapped = false;
   g_stagingBuffers[currentStagingBuffer].MapAsync(
       wgpu::MapMode::Write, 0, StagingBufferSize,
@@ -456,9 +398,13 @@ void map_staging_buffer() {
 }
 
 void begin_frame() {
-  while (!bufferMapped) {
-    g_device.Tick();
+  {
+    OPTICK_EVENT("Wait for buffer mapping");
+    while (!bufferMapped) {
+      g_device.Tick();
+    }
   }
+  OPTICK_EVENT();
   size_t bufferOffset = 0;
   auto& stagingBuf = g_stagingBuffers[currentStagingBuffer];
   const auto mapBuffer = [&](ByteBuffer& buf, uint64_t size) {
@@ -472,6 +418,7 @@ void begin_frame() {
 }
 
 void end_frame(const wgpu::CommandEncoder& cmd) {
+  OPTICK_EVENT();
   uint64_t bufferOffset = 0;
   const auto writeBuffer = [&](ByteBuffer& buf, wgpu::Buffer& out, uint64_t size, std::string_view label) {
     const auto writeSize = buf.size(); // Only need to copy this many bytes
@@ -491,6 +438,7 @@ void end_frame(const wgpu::CommandEncoder& cmd) {
 }
 
 void render(const wgpu::RenderPassEncoder& pass) {
+  OPTICK_EVENT();
   g_currentPipeline = UINT64_MAX;
 
   for (const auto& cmd : g_commands) {
@@ -533,6 +481,7 @@ void render(const wgpu::RenderPassEncoder& pass) {
 }
 
 bool bind_pipeline(PipelineRef ref, const wgpu::RenderPassEncoder& pass) {
+  OPTICK_EVENT();
   if (ref == g_currentPipeline) {
     return true;
   }
@@ -547,6 +496,7 @@ bool bind_pipeline(PipelineRef ref, const wgpu::RenderPassEncoder& pass) {
 }
 
 static inline Range push(ByteBuffer& target, const uint8_t* data, size_t length, size_t alignment) {
+  OPTICK_EVENT();
   size_t padding = 0;
   if (alignment != 0) {
     padding = alignment - length % alignment;
@@ -565,6 +515,7 @@ static inline Range push(ByteBuffer& target, const uint8_t* data, size_t length,
   return {static_cast<uint32_t>(begin), static_cast<uint32_t>(length + padding)};
 }
 static inline Range map(ByteBuffer& target, size_t length, size_t alignment) {
+  OPTICK_EVENT();
   size_t padding = 0;
   if (alignment != 0) {
     padding = alignment - length % alignment;
@@ -576,19 +527,28 @@ static inline Range map(ByteBuffer& target, size_t length, size_t alignment) {
   target.append_zeroes(length + padding);
   return {static_cast<uint32_t>(begin), static_cast<uint32_t>(length + padding)};
 }
-Range push_verts(const uint8_t* data, size_t length) { return push(g_verts, data, length, 0 /* TODO? */); }
-Range push_indices(const uint8_t* data, size_t length) { return push(g_indices, data, length, 0 /* TODO? */); }
+Range push_verts(const uint8_t* data, size_t length) {
+  OPTICK_EVENT();
+  return push(g_verts, data, length, 0 /* TODO? */);
+}
+Range push_indices(const uint8_t* data, size_t length) {
+  OPTICK_EVENT();
+  return push(g_indices, data, length, 0 /* TODO? */);
+}
 Range push_uniform(const uint8_t* data, size_t length) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits);
   return push(g_uniforms, data, length, limits.limits.minUniformBufferOffsetAlignment);
 }
 Range push_storage(const uint8_t* data, size_t length) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits);
   return push(g_storage, data, length, limits.limits.minStorageBufferOffsetAlignment);
 }
 Range push_static_storage(const uint8_t* data, size_t length) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits);
   auto range = push(g_staticStorage, data, length, limits.limits.minStorageBufferOffsetAlignment);
@@ -596,20 +556,24 @@ Range push_static_storage(const uint8_t* data, size_t length) {
   return range;
 }
 std::pair<ByteBuffer, Range> map_verts(size_t length) {
+  OPTICK_EVENT();
   const auto range = map(g_verts, length, 0 /* TODO? */);
   return {ByteBuffer{g_verts.data() + range.offset, range.size}, range};
 }
 std::pair<ByteBuffer, Range> map_indices(size_t length) {
+  OPTICK_EVENT();
   const auto range = map(g_indices, length, 0 /* TODO? */);
   return {ByteBuffer{g_indices.data() + range.offset, range.size}, range};
 }
 std::pair<ByteBuffer, Range> map_uniform(size_t length) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits);
   const auto range = map(g_uniforms, length, limits.limits.minUniformBufferOffsetAlignment);
   return {ByteBuffer{g_uniforms.data() + range.offset, range.size}, range};
 }
 std::pair<ByteBuffer, Range> map_storage(size_t length) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits);
   const auto range = map(g_storage, length, limits.limits.minStorageBufferOffsetAlignment);
@@ -617,6 +581,7 @@ std::pair<ByteBuffer, Range> map_storage(size_t length) {
 }
 
 BindGroupRef bind_group_ref(const wgpu::BindGroupDescriptor& descriptor) {
+  OPTICK_EVENT();
   const auto id = xxh3_hash(descriptor);
   if (!g_cachedBindGroups.contains(id)) {
     g_cachedBindGroups.try_emplace(id, g_device.CreateBindGroup(&descriptor));
@@ -624,6 +589,7 @@ BindGroupRef bind_group_ref(const wgpu::BindGroupDescriptor& descriptor) {
   return id;
 }
 const wgpu::BindGroup& find_bind_group(BindGroupRef id) {
+  OPTICK_EVENT();
   const auto it = g_cachedBindGroups.find(id);
   if (it == g_cachedBindGroups.end()) {
     Log.report(logvisor::Fatal, FMT_STRING("get_bind_group: failed to locate {}"), id);
@@ -633,6 +599,7 @@ const wgpu::BindGroup& find_bind_group(BindGroupRef id) {
 }
 
 const wgpu::Sampler& sampler_ref(const wgpu::SamplerDescriptor& descriptor) {
+  OPTICK_EVENT();
   const auto id = xxh3_hash(descriptor);
   auto it = g_cachedSamplers.find(id);
   if (it == g_cachedSamplers.end()) {
@@ -642,6 +609,7 @@ const wgpu::Sampler& sampler_ref(const wgpu::SamplerDescriptor& descriptor) {
 }
 
 uint32_t align_uniform(uint32_t value) {
+  OPTICK_EVENT();
   wgpu::SupportedLimits limits;
   g_device.GetLimits(&limits); // TODO cache
   const auto uniform_alignment = limits.limits.minUniformBufferOffsetAlignment;

aurora/lib/gfx/common.hpp

@@ -4,20 +4,32 @@
 #include <utility>
 
 #include <dawn/webgpu_cpp.h>
-#include <xxhash_impl.h>
+#define XXH_INLINE_ALL
+#define XXH_STATIC_LINKING_ONLY
+#define XXH_IMPLEMENTATION
+#include <xxhash.h>
+#include <optick.h>
 
 #ifndef ALIGN
 #define ALIGN(x, a) (((x) + ((a)-1)) & ~((a)-1))
 #endif
 
+#ifdef __GNUC__
+#define PACK(...) __VA_ARGS__  __attribute__((__packed__))
+#endif
+#ifdef _MSC_VER
+#define PACK(...) __pragma(pack(push, 1)); __VA_ARGS__; __pragma(pack(pop))
+#endif
+
 namespace aurora {
 template <typename T>
 static inline void xxh3_update(XXH3_state_t& state, const T& input);
-static inline XXH64_hash_t xxh3_hash(const void* input, size_t len, XXH64_hash_t seed = 0) {
+static inline XXH64_hash_t xxh3_hash_s(const void* input, size_t len, XXH64_hash_t seed = 0) {
   return XXH3_64bits_withSeed(input, len, seed);
 }
 template <typename T>
 static inline XXH64_hash_t xxh3_hash(const T& input, XXH64_hash_t seed = 0) {
+  OPTICK_EVENT();
   XXH3_state_t state;
   memset(&state, 0, sizeof(XXH3_state_t));
   XXH3_64bits_reset_withSeed(&state, seed);

aurora/lib/gfx/gx.cpp

@@ -335,6 +335,7 @@ static inline wgpu::PrimitiveState to_primitive_state(GX::Primitive gx_prim, GX:
 wgpu::RenderPipeline build_pipeline(const PipelineConfig& config, const ShaderInfo& info,
                                     ArrayRef<wgpu::VertexBufferLayout> vtxBuffers, wgpu::ShaderModule shader,
                                     zstring_view label) noexcept {
+  OPTICK_EVENT();
   const auto depthStencil = wgpu::DepthStencilState{
       .format = g_graphicsConfig.depthFormat,
       .depthWriteEnabled = config.depthUpdate,
@@ -387,6 +388,7 @@ wgpu::RenderPipeline build_pipeline(const PipelineConfig& config, const ShaderIn
 
 ShaderInfo populate_pipeline_config(PipelineConfig& config, GX::Primitive primitive,
                                     const BindGroupRanges& ranges) noexcept {
+  OPTICK_EVENT();
   for (u8 i = 0; i < g_gxState.numTevStages; ++i) {
     config.shaderConfig.tevStages[i] = g_gxState.tevStages[i];
   }
@@ -396,7 +398,9 @@ ShaderInfo populate_pipeline_config(PipelineConfig& config, GX::Primitive primit
   for (u8 i = 0; i < g_gxState.numTexGens; ++i) {
     config.shaderConfig.tcgs[i] = g_gxState.tcgs[i];
   }
-  config.shaderConfig.alphaDiscard = g_gxState.alphaDiscard;
+  if (g_gxState.alphaDiscard) {
+    config.shaderConfig.alphaDiscard = g_gxState.alphaDiscard;
+  }
   config.shaderConfig.fogType = g_gxState.fog.type;
   config = {
       .shaderConfig = config.shaderConfig,
@@ -414,6 +418,7 @@ ShaderInfo populate_pipeline_config(PipelineConfig& config, GX::Primitive primit
   };
   // TODO separate shader info from build_shader for async
   {
+    OPTICK_EVENT("Shader info & bind groups");
     std::lock_guard lk{g_pipelineMutex};
     auto [_, info] = build_shader(config.shaderConfig);
     info.bindGroups = build_bind_groups(info, config.shaderConfig, ranges); // TODO this is hack
@@ -542,6 +547,7 @@ static absl::flat_hash_map<u32, std::pair<wgpu::BindGroupLayout, wgpu::BindGroup
 
 GXBindGroups build_bind_groups(const ShaderInfo& info, const ShaderConfig& config,
                                const BindGroupRanges& ranges) noexcept {
+  OPTICK_EVENT();
   const auto layouts = build_bind_group_layouts(info, config);
   u32 textureCount = info.sampledTextures.count();
 
@@ -578,6 +584,9 @@ GXBindGroups build_bind_groups(const ShaderInfo& info, const ShaderConfig& confi
   };
   std::array<wgpu::BindGroupEntry, MaxTextures> samplerEntries;
   std::array<wgpu::BindGroupEntry, MaxTextures> textureEntries;
+  {
+    OPTICK_EVENT("Build texture entries");
+
   for (u32 texIdx = 0, i = 0; texIdx < info.sampledTextures.size(); ++texIdx) {
     if (!info.sampledTextures.test(texIdx)) {
       continue;
@@ -597,6 +606,7 @@ GXBindGroups build_bind_groups(const ShaderInfo& info, const ShaderConfig& confi
     };
     i++;
   }
+  }
   return {
       .uniformBindGroup = bind_group_ref(wgpu::BindGroupDescriptor{
           .label = "GX Uniform Bind Group",
@@ -620,6 +630,7 @@ GXBindGroups build_bind_groups(const ShaderInfo& info, const ShaderConfig& confi
 }
 
 GXBindGroupLayouts build_bind_group_layouts(const ShaderInfo& info, const ShaderConfig& config) noexcept {
+  OPTICK_EVENT();
   GXBindGroupLayouts out;
   u32 uniformSizeKey = info.uniformSize + (config.denormalizedVertexAttributes ? 0 : 1);
   const auto uniformIt = sUniformBindGroupLayouts.find(uniformSizeKey);
@@ -729,6 +740,10 @@ GXBindGroupLayouts build_bind_group_layouts(const ShaderInfo& info, const Shader
   return out;
 }
 
+void initialize() noexcept {
+  memset(&g_gxState, 0, sizeof(GXState));
+}
+
 // TODO this is awkward
 extern absl::flat_hash_map<ShaderRef, std::pair<wgpu::ShaderModule, gx::ShaderInfo>> g_gxCachedShaders;
 void shutdown() noexcept {

aurora/lib/gfx/gx.hpp

@@ -14,23 +14,23 @@ constexpr u32 MaxTexMtx = 10;
 constexpr u32 MaxPTTexMtx = 20;
 constexpr u32 MaxTexCoord = GX::MAX_TEXCOORD;
 
-template <typename Arg, Arg Default>
+PACK(template <typename Arg, Arg Default>
 struct TevPass {
   Arg a = Default;
   Arg b = Default;
   Arg c = Default;
   Arg d = Default;
   bool operator==(const TevPass&) const = default;
-};
-struct TevOp {
+});
+PACK(struct TevOp {
   GX::TevOp op = GX::TevOp::TEV_ADD;
   GX::TevBias bias = GX::TevBias::TB_ZERO;
   GX::TevScale scale = GX::TevScale::CS_SCALE_1;
   GX::TevRegID outReg = GX::TevRegID::TEVPREV;
   bool clamp = true;
   bool operator==(const TevOp&) const = default;
-};
-struct TevStage {
+});
+PACK(struct TevStage {
   TevPass<GX::TevColorArg, GX::CC_ZERO> colorPass;
   TevPass<GX::TevAlphaArg, GX::CA_ZERO> alphaPass;
   TevOp colorOp;
@@ -41,7 +41,7 @@ struct TevStage {
   GX::TexMapID texMapId = GX::TEXMAP_NULL;
   GX::ChannelID channelId = GX::COLOR_NULL;
   bool operator==(const TevStage&) const = default;
-};
+});
 struct TextureBind {
   aurora::gfx::TextureHandle handle;
   metaforce::EClampMode clampMode;
@@ -55,37 +55,37 @@ struct TextureBind {
   operator bool() const noexcept { return handle; }
 };
 // For shader generation
-struct ColorChannelConfig {
+PACK(struct ColorChannelConfig {
   GX::ColorSrc matSrc = GX::SRC_REG;
   GX::ColorSrc ambSrc = GX::SRC_REG;
   bool lightingEnabled = false;
   bool operator==(const ColorChannelConfig&) const = default;
-};
+});
 // For uniform generation
-struct ColorChannelState {
+PACK(struct ColorChannelState {
   zeus::CColor matColor = zeus::skClear;
   zeus::CColor ambColor = zeus::skClear;
   GX::LightMask lightState;
-};
+});
 using LightVariant = std::variant<std::monostate, Light, zeus::CColor>;
 // Mat4x4 used instead of Mat4x3 for padding purposes
 using TexMtxVariant = std::variant<std::monostate, Mat4x2<float>, Mat4x4<float>>;
-struct TcgConfig {
+PACK(struct TcgConfig {
   GX::TexGenType type = GX::TG_MTX2x4;
   GX::TexGenSrc src = GX::MAX_TEXGENSRC;
   GX::TexMtx mtx = GX::IDENTITY;
   GX::PTTexMtx postMtx = GX::PTIDENTITY;
   bool normalize = false;
   bool operator==(const TcgConfig&) const = default;
-};
-struct FogState {
+});
+PACK(struct FogState {
   GX::FogType type = GX::FOG_NONE;
   float startZ = 0.f;
   float endZ = 0.f;
   float nearZ = 0.f;
   float farZ = 0.f;
   zeus::CColor color;
-};
+});
 
 struct GXState {
   zeus::CMatrix4f mv;
@@ -123,10 +123,11 @@ extern GXState g_gxState;
 
 static inline Mat4x4<float> get_combined_matrix() noexcept { return g_gxState.proj * g_gxState.mv; }
 
+void initialize() noexcept;
 void shutdown() noexcept;
 const TextureBind& get_texture(GX::TexMapID id) noexcept;
 
-struct ShaderConfig {
+PACK(struct ShaderConfig {
   GX::FogType fogType;
   std::array<std::optional<TevStage>, MaxTevStages> tevStages;
   std::array<ColorChannelConfig, MaxColorChannels> colorChannels;
@@ -135,8 +136,8 @@ struct ShaderConfig {
   bool denormalizedVertexAttributes = false;
   bool denormalizedHasNrm = false; // TODO this is a hack
   bool operator==(const ShaderConfig&) const = default;
-};
-struct PipelineConfig {
+});
+PACK(struct PipelineConfig {
   ShaderConfig shaderConfig;
   GX::Primitive primitive;
   GX::Compare depthFunc;
@@ -146,7 +147,7 @@ struct PipelineConfig {
   GX::LogicOp blendOp;
   std::optional<float> dstAlpha;
   bool depthCompare, depthUpdate, alphaUpdate;
-};
+});
 struct GXBindGroupLayouts {
   wgpu::BindGroupLayout uniformLayout;
   wgpu::BindGroupLayout samplerLayout;
@@ -202,70 +203,71 @@ struct DlVert {
 } // namespace aurora::gfx::gx
 
 namespace aurora {
-template <typename Arg, Arg Default>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevPass<Arg, Default>& input) {
-  XXH3_64bits_update(&state, &input.a, sizeof(Arg));
-  XXH3_64bits_update(&state, &input.b, sizeof(Arg));
-  XXH3_64bits_update(&state, &input.c, sizeof(Arg));
-  XXH3_64bits_update(&state, &input.d, sizeof(Arg));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevOp& input) {
-  XXH3_64bits_update(&state, &input.op, sizeof(gfx::gx::TevOp::op));
-  XXH3_64bits_update(&state, &input.bias, sizeof(gfx::gx::TevOp::bias));
-  XXH3_64bits_update(&state, &input.scale, sizeof(gfx::gx::TevOp::scale));
-  XXH3_64bits_update(&state, &input.outReg, sizeof(gfx::gx::TevOp::outReg));
-  XXH3_64bits_update(&state, &input.clamp, sizeof(bool));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevStage& input) {
-  xxh3_update(state, input.colorPass);
-  xxh3_update(state, input.alphaPass);
-  xxh3_update(state, input.colorOp);
-  xxh3_update(state, input.alphaOp);
-  XXH3_64bits_update(&state, &input.kcSel, sizeof(gfx::gx::TevStage::kcSel));
-  XXH3_64bits_update(&state, &input.kaSel, sizeof(gfx::gx::TevStage::kaSel));
-  XXH3_64bits_update(&state, &input.texCoordId, sizeof(gfx::gx::TevStage::texCoordId));
-  XXH3_64bits_update(&state, &input.texMapId, sizeof(gfx::gx::TevStage::texMapId));
-  XXH3_64bits_update(&state, &input.channelId, sizeof(gfx::gx::TevStage::channelId));
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::ColorChannelConfig& input) {
-  XXH3_64bits_update(&state, &input.lightingEnabled, sizeof(gfx::gx::ColorChannelConfig::lightingEnabled));
-  XXH3_64bits_update(&state, &input.matSrc, sizeof(gfx::gx::ColorChannelConfig::matSrc));
-  if (input.lightingEnabled) {
-    // Unused when lighting is disabled
-    XXH3_64bits_update(&state, &input.ambSrc, sizeof(gfx::gx::ColorChannelConfig::ambSrc));
-  }
-}
-template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TcgConfig& input) {
-  XXH3_64bits_update(&state, &input.type, sizeof(gfx::gx::TcgConfig::type));
-  XXH3_64bits_update(&state, &input.src, sizeof(gfx::gx::TcgConfig::src));
-  XXH3_64bits_update(&state, &input.mtx, sizeof(gfx::gx::TcgConfig::mtx));
-  XXH3_64bits_update(&state, &input.postMtx, sizeof(gfx::gx::TcgConfig::postMtx));
-  XXH3_64bits_update(&state, &input.normalize, sizeof(gfx::gx::TcgConfig::normalize));
-}
+//template <typename Arg, Arg Default>
+//inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevPass<Arg, Default>& input) {
+//  XXH3_64bits_update(&state, &input.a, sizeof(Arg));
+//  XXH3_64bits_update(&state, &input.b, sizeof(Arg));
+//  XXH3_64bits_update(&state, &input.c, sizeof(Arg));
+//  XXH3_64bits_update(&state, &input.d, sizeof(Arg));
+//}
+//template <>
+//inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevOp& input) {
+//  XXH3_64bits_update(&state, &input.op, sizeof(gfx::gx::TevOp::op));
+//  XXH3_64bits_update(&state, &input.bias, sizeof(gfx::gx::TevOp::bias));
+//  XXH3_64bits_update(&state, &input.scale, sizeof(gfx::gx::TevOp::scale));
+//  XXH3_64bits_update(&state, &input.outReg, sizeof(gfx::gx::TevOp::outReg));
+//  XXH3_64bits_update(&state, &input.clamp, sizeof(bool));
+//}
+//template <>
+//inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TevStage& input) {
+//  xxh3_update(state, input.colorPass);
+//  xxh3_update(state, input.alphaPass);
+//  xxh3_update(state, input.colorOp);
+//  xxh3_update(state, input.alphaOp);
+//  XXH3_64bits_update(&state, &input.kcSel, sizeof(gfx::gx::TevStage::kcSel));
+//  XXH3_64bits_update(&state, &input.kaSel, sizeof(gfx::gx::TevStage::kaSel));
+//  XXH3_64bits_update(&state, &input.texCoordId, sizeof(gfx::gx::TevStage::texCoordId));
+//  XXH3_64bits_update(&state, &input.texMapId, sizeof(gfx::gx::TevStage::texMapId));
+//  XXH3_64bits_update(&state, &input.channelId, sizeof(gfx::gx::TevStage::channelId));
+//}
+//template <>
+//inline void xxh3_update(XXH3_state_t& state, const gfx::gx::ColorChannelConfig& input) {
+//  XXH3_64bits_update(&state, &input.lightingEnabled, sizeof(gfx::gx::ColorChannelConfig::lightingEnabled));
+//  XXH3_64bits_update(&state, &input.matSrc, sizeof(gfx::gx::ColorChannelConfig::matSrc));
+//  if (input.lightingEnabled) {
+//    // Unused when lighting is disabled
+//    XXH3_64bits_update(&state, &input.ambSrc, sizeof(gfx::gx::ColorChannelConfig::ambSrc));
+//  }
+//}
+//template <>
+//inline void xxh3_update(XXH3_state_t& state, const gfx::gx::TcgConfig& input) {
+//  XXH3_64bits_update(&state, &input.type, sizeof(gfx::gx::TcgConfig::type));
+//  XXH3_64bits_update(&state, &input.src, sizeof(gfx::gx::TcgConfig::src));
+//  XXH3_64bits_update(&state, &input.mtx, sizeof(gfx::gx::TcgConfig::mtx));
+//  XXH3_64bits_update(&state, &input.postMtx, sizeof(gfx::gx::TcgConfig::postMtx));
+//  XXH3_64bits_update(&state, &input.normalize, sizeof(gfx::gx::TcgConfig::normalize));
+//}
 template <>
-inline void xxh3_update(XXH3_state_t& state, const gfx::gx::ShaderConfig& input) {
-  for (const auto& item : input.tevStages) {
-    if (!item) {
-      break;
-    }
-    xxh3_update(state, *item);
-  }
-  for (const auto& item : input.colorChannels) {
-    xxh3_update(state, item);
-  }
-  for (const auto& item : input.tcgs) {
-    xxh3_update(state, item);
-  }
-  if (input.alphaDiscard) {
-    XXH3_64bits_update(&state, &*input.alphaDiscard, sizeof(float));
-  }
-  XXH3_64bits_update(&state, &input.denormalizedVertexAttributes,
-                     sizeof(gfx::gx::ShaderConfig::denormalizedVertexAttributes));
-  XXH3_64bits_update(&state, &input.denormalizedHasNrm, sizeof(gfx::gx::ShaderConfig::denormalizedHasNrm));
-  XXH3_64bits_update(&state, &input.fogType, sizeof(gfx::gx::ShaderConfig::fogType));
+inline XXH64_hash_t xxh3_hash(const gfx::gx::ShaderConfig& input, XXH64_hash_t seed) {
+  return xxh3_hash_s(&input, sizeof(gfx::gx::ShaderConfig), seed);
+//  for (const auto& item : input.tevStages) {
+//    if (!item) {
+//      break;
+//    }
+//    xxh3_update(state, *item);
+//  }
+//  for (const auto& item : input.colorChannels) {
+//    xxh3_update(state, item);
+//  }
+//  for (const auto& item : input.tcgs) {
+//    xxh3_update(state, item);
+//  }
+//  if (input.alphaDiscard) {
+//    XXH3_64bits_update(&state, &*input.alphaDiscard, sizeof(float));
+//  }
+//  XXH3_64bits_update(&state, &input.denormalizedVertexAttributes,
+//                     sizeof(gfx::gx::ShaderConfig::denormalizedVertexAttributes));
+//  XXH3_64bits_update(&state, &input.denormalizedHasNrm, sizeof(gfx::gx::ShaderConfig::denormalizedHasNrm));
+//  XXH3_64bits_update(&state, &input.fogType, sizeof(gfx::gx::ShaderConfig::fogType));
 }
 } // namespace aurora

aurora/lib/gfx/gx_shader.cpp

@@ -359,6 +359,7 @@ std::pair<wgpu::ShaderModule, ShaderInfo> build_shader(const ShaderConfig& confi
 #endif
     return it->second;
   }
+  OPTICK_EVENT();
 
   Log.report(logvisor::Info, FMT_STRING("Shader config (hash {:x}):"), hash);
   ShaderInfo info{
@@ -636,7 +637,8 @@ var<storage, read> v_packed_uvs: Vec2Block;
     } else {
       vtxOutAttrs += fmt::format(FMT_STRING("\n    @location({}) tex{}_uv: vec2<f32>;"), locIdx, i);
       if (tcg.src >= GX::TG_TEX0 && tcg.src <= GX::TG_TEX7) {
-        vtxXfrAttrs += fmt::format(FMT_STRING("\n    var tc{} = vec4<f32>({}, 0.0, 1.0);"), i, in_uv(tcg.src - GX::TG_TEX0));
+        vtxXfrAttrs +=
+            fmt::format(FMT_STRING("\n    var tc{} = vec4<f32>({}, 0.0, 1.0);"), i, in_uv(tcg.src - GX::TG_TEX0));
       } else if (tcg.src == GX::TG_POS) {
         vtxXfrAttrs += fmt::format(FMT_STRING("\n    var tc{} = vec4<f32>(obj_pos.xyz, 1.0);"), i);
       } else if (tcg.src == GX::TG_NRM) {
@@ -663,7 +665,8 @@ var<storage, read> v_packed_uvs: Vec2Block;
     } else {
       u32 postMtxIdx = (tcg.postMtx - GX::PTTEXMTX0) / 3;
       info.usesPTTexMtx.set(postMtxIdx);
-      vtxXfrAttrs += fmt::format(FMT_STRING("\n    var tc{0}_proj = ubuf.postmtx{1} * vec4<f32>(tc{0}_tmp.xyz, 1.0);"), i, postMtxIdx);
+      vtxXfrAttrs += fmt::format(FMT_STRING("\n    var tc{0}_proj = ubuf.postmtx{1} * vec4<f32>(tc{0}_tmp.xyz, 1.0);"),
+                                 i, postMtxIdx);
     }
     vtxXfrAttrs += fmt::format(FMT_STRING("\n    out.tex{0}_uv = tc{0}_proj.xy;"), i);
     fragmentFnPre += fmt::format(
@@ -698,7 +701,8 @@ var<storage, read> v_packed_uvs: Vec2Block;
   if (config.fogType != GX::FOG_NONE) {
     info.usesFog = true;
 
-    uniformPre += "\n"
+    uniformPre +=
+        "\n"
         "struct Fog {\n"
         "    color: vec4<f32>;\n"
         "    a: f32;\n"
@@ -729,7 +733,8 @@ var<storage, read> v_packed_uvs: Vec2Block;
       break;
     case GX::FOG_PERSP_REVEXP2:
     case GX::FOG_ORTHO_REVEXP2:
-      fragmentFn += "\n    fogF = 1.0 - fogF;"
+      fragmentFn +=
+          "\n    fogF = 1.0 - fogF;"
           "\n    var fogZ = exp2(-8.0 * fogF * fogF);";
       break;
     default:
@@ -796,7 +801,11 @@ fn fs_main(in: VertexOutput) -> @location(0) vec4<f32> {{
       .nextInChain = &wgslDescriptor,
       .label = label.c_str(),
   };
-  auto shader = gpu::g_device.CreateShaderModule(&shaderDescriptor);
+  wgpu::ShaderModule shader;
+  {
+    OPTICK_EVENT("Device CreateShaderModule");
+    shader = gpu::g_device.CreateShaderModule(&shaderDescriptor);
+  }
 
   info.uniformSize = align_uniform(info.uniformSize);
   auto pair = std::make_pair(std::move(shader), info);

aurora/lib/gfx/model/shader.cpp

@@ -113,7 +113,8 @@ static inline std::pair<gx::DlVert, size_t> readVert(const u8* data) noexcept {
 static absl::flat_hash_map<XXH64_hash_t, std::pair<std::vector<gx::DlVert>, std::vector<u32>>> sCachedDisplayLists;
 
 void queue_surface(const u8* dlStart, u32 dlSize) noexcept {
-  const auto hash = xxh3_hash(dlStart, dlSize, 0);
+  OPTICK_EVENT();
+  const auto hash = xxh3_hash_s(dlStart, dlSize, 0);
   Range vertRange, idxRange;
   uint32_t numIndices;
   auto it = sCachedDisplayLists.find(hash);
@@ -123,6 +124,7 @@ void queue_surface(const u8* dlStart, u32 dlSize) noexcept {
     vertRange = push_verts(ArrayRef{verts});
     idxRange = push_indices(ArrayRef{indices});
   } else {
+    OPTICK_EVENT("Display list translation");
     std::vector<gx::DlVert