GPUDevice: Add API version field

Also tie shader caches to API version and device LUID. That way we don't
have tons of cache files, and they're regenerated if the GPU/driver
changes.

src/common/small_string.cpp

@@ -111,6 +111,18 @@ void SmallStringBase::shrink_to_fit()
   m_buffer_size = buffer_size;
 }
 
+void SmallStringBase::convert_to_lower_case()
+{
+  for (u32 i = 0; i < m_length; i++)
+    m_buffer[i] = static_cast<char>(std::tolower(m_buffer[i]));
+}
+
+void SmallStringBase::convert_to_upper_case()
+{
+  for (u32 i = 0; i < m_length; i++)
+    m_buffer[i] = static_cast<char>(std::toupper(m_buffer[i]));
+}
+
 std::string_view SmallStringBase::view() const
 {
   return (m_length == 0) ? std::string_view() : std::string_view(m_buffer, m_length);

src/common/small_string.h

@@ -188,6 +188,10 @@ public:
   ALWAYS_INLINE void push_back(value_type val) { append(val); }
   ALWAYS_INLINE void pop_back() { erase(-1); }
 
+  // case conversion
+  void convert_to_lower_case();
+  void convert_to_upper_case();
+
   // returns a string view for this string
   std::string_view view() const;
 

src/core/host.h

@@ -21,7 +21,7 @@ struct WindowInfo;
 enum class AudioBackend : u8;
 enum class AudioExpansionMode : u8;
 enum class AudioStretchMode : u8;
-enum class RenderAPI : u32;
+enum class RenderAPI : u8;
 class AudioStream;
 class CDImage;
 

src/core/settings.h

@@ -18,7 +18,7 @@
 #include <string_view>
 #include <vector>
 
-enum class RenderAPI : u32;
+enum class RenderAPI : u8;
 enum class MediaCaptureBackend : u8;
 
 struct SettingInfo

src/util/d3d11_device.cpp

@@ -53,11 +53,6 @@ D3D11Device::~D3D11Device()
   Assert(!m_device);
 }
 
-RenderAPI D3D11Device::GetRenderAPI() const
-{
-  return RenderAPI::D3D11;
-}
-
 bool D3D11Device::HasSurface() const
 {
   return static_cast<bool>(m_swap_chain);
@@ -127,7 +122,8 @@ bool D3D11Device::CreateDevice(std::string_view adapter, std::optional<bool> exc
     INFO_LOG("D3D Adapter: {}", D3DCommon::GetAdapterName(dxgi_adapter.Get()));
   else
     ERROR_LOG("Failed to obtain D3D adapter name.");
-  INFO_LOG("Max device feature level: {}", D3DCommon::GetFeatureLevelString(m_max_feature_level));
+  INFO_LOG("Max device feature level: {}",
+           D3DCommon::GetFeatureLevelString(D3DCommon::GetRenderAPIVersionForFeatureLevel(m_max_feature_level)));
 
   BOOL allow_tearing_supported = false;
   hr = m_dxgi_factory->CheckFeatureSupport(DXGI_FEATURE_PRESENT_ALLOW_TEARING, &allow_tearing_supported,
@@ -164,6 +160,8 @@ void D3D11Device::SetFeatures(FeatureMask disabled_features)
 {
   const D3D_FEATURE_LEVEL feature_level = m_device->GetFeatureLevel();
 
+  m_render_api = RenderAPI::D3D11;
+  m_render_api_version = D3DCommon::GetRenderAPIVersionForFeatureLevel(feature_level);
   m_max_texture_size = D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION;
   m_max_multisamples = 1;
   for (u32 multisamples = 2; multisamples < D3D11_MAX_MULTISAMPLE_SAMPLE_COUNT; multisamples++)
@@ -443,12 +441,11 @@ bool D3D11Device::SupportsExclusiveFullscreen() const
 
 std::string D3D11Device::GetDriverInfo() const
 {
-  const D3D_FEATURE_LEVEL fl = m_device->GetFeatureLevel();
+  std::string ret = fmt::format("{} (Shader Model {})\n", D3DCommon::GetFeatureLevelString(m_render_api_version),
-  std::string ret =
+                                D3DCommon::GetShaderModelForFeatureLevelNumber(m_render_api_version));
-    fmt::format("{} ({})\n", D3DCommon::GetFeatureLevelString(fl), D3DCommon::GetFeatureLevelShaderModelString(fl));
 
   ComPtr<IDXGIDevice> dxgi_dev;
-  if (m_device.As(&dxgi_dev))
+  if (SUCCEEDED(m_device.As(&dxgi_dev)))
   {
     ComPtr<IDXGIAdapter> dxgi_adapter;
     if (SUCCEEDED(dxgi_dev->GetAdapter(dxgi_adapter.GetAddressOf())))

src/util/d3d11_device.h

@@ -36,8 +36,6 @@ public:
   ALWAYS_INLINE static ID3D11DeviceContext1* GetD3DContext() { return GetInstance().m_context.Get(); }
   ALWAYS_INLINE static D3D_FEATURE_LEVEL GetMaxFeatureLevel() { return GetInstance().m_max_feature_level; }
 
-  RenderAPI GetRenderAPI() const override;
-
   bool HasSurface() const override;
 
   bool UpdateWindow() override;

src/util/d3d11_pipeline.cpp

@@ -100,7 +100,7 @@ std::unique_ptr<GPUShader> D3D11Device::CreateShaderFromSource(GPUShaderStage st
                                                                std::string_view source, const char* entry_point,
                                                                DynamicHeapArray<u8>* out_binary, Error* error)
 {
-  const u32 shader_model = D3DCommon::GetShaderModelForFeatureLevel(m_device->GetFeatureLevel());
+  const u32 shader_model = D3DCommon::GetShaderModelForFeatureLevelNumber(m_render_api_version);
   if (language != GPUShaderLanguage::HLSL)
   {
     return TranspileAndCreateShaderFromSource(stage, language, source, entry_point, GPUShaderLanguage::HLSL,

src/util/d3d12_device.cpp

@@ -147,12 +147,13 @@ bool D3D12Device::CreateDevice(std::string_view adapter, std::optional<bool> exc
   }
 
   // Create the actual device.
+  D3D_FEATURE_LEVEL feature_level = D3D_FEATURE_LEVEL_1_0_CORE;
   for (D3D_FEATURE_LEVEL try_feature_level : {D3D_FEATURE_LEVEL_11_0})
   {
     hr = D3D12CreateDevice(m_adapter.Get(), try_feature_level, IID_PPV_ARGS(&m_device));
     if (SUCCEEDED(hr))
     {
-      m_feature_level = try_feature_level;
+      feature_level = try_feature_level;
       break;
     }
   }
@@ -232,7 +233,7 @@ bool D3D12Device::CreateDevice(std::string_view adapter, std::optional<bool> exc
     return false;
   }
 
-  SetFeatures(disabled_features);
+  SetFeatures(feature_level, disabled_features);
 
   if (!CreateCommandLists(error) || !CreateDescriptorHeaps(error))
     return false;
@@ -282,10 +283,28 @@ void D3D12Device::DestroyDevice()
   m_dxgi_factory.Reset();
 }
 
+void D3D12Device::GetPipelineCacheHeader(PIPELINE_CACHE_HEADER* hdr)
+{
+  const LUID adapter_luid = m_device->GetAdapterLuid();
+  std::memcpy(&hdr->adapter_luid, &adapter_luid, sizeof(hdr->adapter_luid));
+  hdr->render_api_version = m_render_api_version;
+  hdr->unused = 0;
+}
+
 bool D3D12Device::ReadPipelineCache(std::optional<DynamicHeapArray<u8>> data)
 {
+  PIPELINE_CACHE_HEADER expected_header;
+  GetPipelineCacheHeader(&expected_header);
+  if (data.has_value() && (data->size() < sizeof(PIPELINE_CACHE_HEADER) ||
+                           std::memcmp(data->data(), &expected_header, sizeof(PIPELINE_CACHE_HEADER)) != 0))
+  {
+    WARNING_LOG("Pipeline cache header does not match current device, ignoring.");
+    data.reset();
+  }
+
   HRESULT hr =
-    m_device->CreatePipelineLibrary(data.has_value() ? data->data() : nullptr, data.has_value() ? data->size() : 0,
+    m_device->CreatePipelineLibrary(data.has_value() ? (data->data() + sizeof(PIPELINE_CACHE_HEADER)) : nullptr,
+                                    data.has_value() ? (data->size() - sizeof(PIPELINE_CACHE_HEADER)) : 0,
                                     IID_PPV_ARGS(m_pipeline_library.ReleaseAndGetAddressOf()));
   if (SUCCEEDED(hr))
   {
@@ -328,8 +347,13 @@ bool D3D12Device::GetPipelineCacheData(DynamicHeapArray<u8>* data)
     return true;
   }
 
-  data->resize(size);
+  PIPELINE_CACHE_HEADER header;
-  const HRESULT hr = m_pipeline_library->Serialize(data->data(), data->size());
+  GetPipelineCacheHeader(&header);
+
+  data->resize(sizeof(PIPELINE_CACHE_HEADER) + size);
+  std::memcpy(data->data(), &header, sizeof(PIPELINE_CACHE_HEADER));
+
+  const HRESULT hr = m_pipeline_library->Serialize(data->data() + sizeof(PIPELINE_CACHE_HEADER), size);
   if (FAILED(hr))
   {
     ERROR_LOG("Serialize() failed with HRESULT {:08X}", static_cast<unsigned>(hr));
@@ -771,11 +795,6 @@ void D3D12Device::DestroyDeferredObjects(u64 fence_value)
   }
 }
 
-RenderAPI D3D12Device::GetRenderAPI() const
-{
-  return RenderAPI::D3D12;
-}
-
 bool D3D12Device::HasSurface() const
 {
   return static_cast<bool>(m_swap_chain);
@@ -1070,8 +1089,8 @@ bool D3D12Device::SupportsTextureFormat(GPUTexture::Format format) const
 
 std::string D3D12Device::GetDriverInfo() const
 {
-  std::string ret = fmt::format("{} ({})\n", D3DCommon::GetFeatureLevelString(m_feature_level),
+  std::string ret = fmt::format("{} (Shader Model {})\n", D3DCommon::GetFeatureLevelString(m_render_api_version),
-                                D3DCommon::GetFeatureLevelShaderModelString(m_feature_level));
+                                D3DCommon::GetShaderModelForFeatureLevelNumber(m_render_api_version));
 
   DXGI_ADAPTER_DESC desc;
   if (m_adapter && SUCCEEDED(m_adapter->GetDesc(&desc)))
@@ -1234,8 +1253,10 @@ void D3D12Device::InsertDebugMessage(const char* msg)
 #endif
 }
 
-void D3D12Device::SetFeatures(FeatureMask disabled_features)
+void D3D12Device::SetFeatures(D3D_FEATURE_LEVEL feature_level, FeatureMask disabled_features)
 {
+  m_render_api = RenderAPI::D3D12;
+  m_render_api_version = D3DCommon::GetRenderAPIVersionForFeatureLevel(feature_level);
   m_max_texture_size = D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION;
   m_max_multisamples = 1;
   for (u32 multisamples = 2; multisamples < D3D12_MAX_MULTISAMPLE_SAMPLE_COUNT; multisamples++)

src/util/d3d12_device.h

@@ -58,8 +58,6 @@ public:
   D3D12Device();
   ~D3D12Device() override;
 
-  RenderAPI GetRenderAPI() const override;
-
   bool HasSurface() const override;
 
   bool UpdateWindow() override;
@@ -220,9 +218,18 @@ private:
     bool has_timestamp_query = false;
   };
 
+  struct PIPELINE_CACHE_HEADER
+  {
+    u64 adapter_luid;
+    u32 render_api_version;
+    u32 unused;
+  };
+  static_assert(sizeof(PIPELINE_CACHE_HEADER) == 16);
+
   using SamplerMap = std::unordered_map<u64, D3D12DescriptorHandle>;
 
-  void SetFeatures(FeatureMask disabled_features);
+  void GetPipelineCacheHeader(PIPELINE_CACHE_HEADER* hdr);
+  void SetFeatures(D3D_FEATURE_LEVEL feature_level, FeatureMask disabled_features);
 
   u32 GetSwapChainBufferCount() const;
   bool CreateSwapChain(Error* error);
@@ -291,7 +298,6 @@ private:
 
   std::array<CommandList, NUM_COMMAND_LISTS> m_command_lists;
   u32 m_current_command_list = NUM_COMMAND_LISTS - 1;
-  D3D_FEATURE_LEVEL m_feature_level = D3D_FEATURE_LEVEL_11_0;
 
   ComPtr<IDXGIFactory5> m_dxgi_factory;
   ComPtr<IDXGISwapChain1> m_swap_chain;

src/util/d3d12_pipeline.cpp

@@ -38,7 +38,7 @@ std::unique_ptr<GPUShader> D3D12Device::CreateShaderFromSource(GPUShaderStage st
                                                                std::string_view source, const char* entry_point,
                                                                DynamicHeapArray<u8>* out_binary, Error* error)
 {
-  const u32 shader_model = D3DCommon::GetShaderModelForFeatureLevel(m_feature_level);
+  const u32 shader_model = D3DCommon::GetShaderModelForFeatureLevelNumber(m_render_api_version);
   if (language != GPUShaderLanguage::HLSL)
   {
     return TranspileAndCreateShaderFromSource(stage, language, source, entry_point, GPUShaderLanguage::HLSL,

src/util/d3d_common.cpp

@@ -19,47 +19,65 @@
 
 Log_SetChannel(D3DCommon);
 
-const char* D3DCommon::GetFeatureLevelString(D3D_FEATURE_LEVEL feature_level)
+namespace D3DCommon {
+namespace {
+struct FeatureLevelTableEntry
 {
-  static constexpr std::array<std::tuple<D3D_FEATURE_LEVEL, const char*>, 11> feature_level_names = {{
+  D3D_FEATURE_LEVEL d3d_feature_level;
-    {D3D_FEATURE_LEVEL_1_0_CORE, "D3D_FEATURE_LEVEL_1_0_CORE"},
+  u16 render_api_version;
-    {D3D_FEATURE_LEVEL_9_1, "D3D_FEATURE_LEVEL_9_1"},
+  u16 shader_model_number;
-    {D3D_FEATURE_LEVEL_9_2, "D3D_FEATURE_LEVEL_9_2"},
+  const char* feature_level_str;
-    {D3D_FEATURE_LEVEL_9_3, "D3D_FEATURE_LEVEL_9_3"},
+};
-    {D3D_FEATURE_LEVEL_10_0, "D3D_FEATURE_LEVEL_10_0"},
+} // namespace
-    {D3D_FEATURE_LEVEL_10_1, "D3D_FEATURE_LEVEL_10_1"},
-    {D3D_FEATURE_LEVEL_11_0, "D3D_FEATURE_LEVEL_11_0"},
-    {D3D_FEATURE_LEVEL_11_1, "D3D_FEATURE_LEVEL_11_1"},
-    {D3D_FEATURE_LEVEL_12_0, "D3D_FEATURE_LEVEL_12_0"},
-    {D3D_FEATURE_LEVEL_12_1, "D3D_FEATURE_LEVEL_12_1"},
-    {D3D_FEATURE_LEVEL_12_2, "D3D_FEATURE_LEVEL_12_2"},
-  }};
 
-  for (const auto& [fl, name] : feature_level_names)
+static constexpr std::array<FeatureLevelTableEntry, 11> s_feature_levels = {{
-  {
+  {D3D_FEATURE_LEVEL_1_0_CORE, 100, 40, "D3D_FEATURE_LEVEL_1_0_CORE"},
-    if (fl == feature_level)
+  {D3D_FEATURE_LEVEL_9_1, 910, 40, "D3D_FEATURE_LEVEL_9_1"},
-      return name;
+  {D3D_FEATURE_LEVEL_9_2, 920, 40, "D3D_FEATURE_LEVEL_9_2"},
-  }
+  {D3D_FEATURE_LEVEL_9_3, 930, 40, "D3D_FEATURE_LEVEL_9_3"},
+  {D3D_FEATURE_LEVEL_10_0, 1000, 40, "D3D_FEATURE_LEVEL_10_0"},
+  {D3D_FEATURE_LEVEL_10_1, 1010, 41, "D3D_FEATURE_LEVEL_10_1"},
+  {D3D_FEATURE_LEVEL_11_0, 1100, 50, "D3D_FEATURE_LEVEL_11_0"},
+  {D3D_FEATURE_LEVEL_11_1, 1110, 50, "D3D_FEATURE_LEVEL_11_1"},
+  {D3D_FEATURE_LEVEL_12_0, 1200, 50, "D3D_FEATURE_LEVEL_12_0"},
+  {D3D_FEATURE_LEVEL_12_1, 1210, 50, "D3D_FEATURE_LEVEL_12_1"},
+  {D3D_FEATURE_LEVEL_12_2, 1220, 50, "D3D_FEATURE_LEVEL_12_2"},
+}};
+} // namespace D3DCommon
 
-  return "D3D_FEATURE_LEVEL_UNKNOWN";
+const char* D3DCommon::GetFeatureLevelString(u32 render_api_version)
+{
+  const auto iter =
+    std::find_if(s_feature_levels.begin(), s_feature_levels.end(),
+                 [&render_api_version](const auto& it) { return it.render_api_version == render_api_version; });
+  return (iter != s_feature_levels.end()) ? iter->feature_level_str : "D3D_FEATURE_LEVEL_UNKNOWN";
 }
 
-const char* D3DCommon::GetFeatureLevelShaderModelString(D3D_FEATURE_LEVEL feature_level)
+u32 D3DCommon::GetRenderAPIVersionForFeatureLevel(D3D_FEATURE_LEVEL feature_level)
 {
-  static constexpr std::array<std::tuple<D3D_FEATURE_LEVEL, const char*>, 4> feature_level_names = {{
+  const FeatureLevelTableEntry* highest_entry = nullptr;
-    {D3D_FEATURE_LEVEL_10_0, "sm40"},
+  for (const FeatureLevelTableEntry& entry : s_feature_levels)
-    {D3D_FEATURE_LEVEL_10_1, "sm41"},
-    {D3D_FEATURE_LEVEL_11_0, "sm50"},
-    {D3D_FEATURE_LEVEL_11_1, "sm50"},
-  }};
-
-  for (const auto& [fl, name] : feature_level_names)
   {
-    if (fl == feature_level)
+    if (feature_level >= entry.d3d_feature_level)
-      return name;
+      highest_entry = &entry;
   }
+  return highest_entry ? highest_entry->render_api_version : 0;
+}
 
-  return "unk";
+D3D_FEATURE_LEVEL D3DCommon::GetFeatureLevelForNumber(u32 render_api_version)
+{
+  const auto iter =
+    std::find_if(s_feature_levels.begin(), s_feature_levels.end(),
+                 [&render_api_version](const auto& it) { return it.render_api_version == render_api_version; });
+  return (iter != s_feature_levels.end()) ? iter->d3d_feature_level : D3D_FEATURE_LEVEL_1_0_CORE;
+}
+
+u32 D3DCommon::GetShaderModelForFeatureLevelNumber(u32 render_api_version)
+{
+  const auto iter =
+    std::find_if(s_feature_levels.begin(), s_feature_levels.end(),
+                 [&render_api_version](const auto& it) { return it.render_api_version == render_api_version; });
+  return (iter != s_feature_levels.end()) ? iter->shader_model_number : 40;
 }
 
 D3D_FEATURE_LEVEL D3DCommon::GetDeviceMaxFeatureLevel(IDXGIAdapter1* adapter)
@@ -379,23 +397,6 @@ std::string D3DCommon::GetDriverVersionFromLUID(const LUID& luid)
   return ret;
 }
 
-u32 D3DCommon::GetShaderModelForFeatureLevel(D3D_FEATURE_LEVEL feature_level)
-{
-  switch (feature_level)
-  {
-    case D3D_FEATURE_LEVEL_10_0:
-      return 40;
-
-    case D3D_FEATURE_LEVEL_10_1:
-      return 41;
-
-    case D3D_FEATURE_LEVEL_11_0:
-    case D3D_FEATURE_LEVEL_11_1:
-    default:
-      return 50;
-  }
-}
-
 std::optional<DynamicHeapArray<u8>> D3DCommon::CompileShader(u32 shader_model, bool debug_device, GPUShaderStage stage,
                                                              std::string_view source, const char* entry_point,
                                                              Error* error)

src/util/d3d_common.h

@@ -25,8 +25,10 @@ struct DXGI_MODE_DESC;
 
 namespace D3DCommon {
 // returns string representation of feature level
-const char* GetFeatureLevelString(D3D_FEATURE_LEVEL feature_level);
+const char* GetFeatureLevelString(u32 render_api_version);
-const char* GetFeatureLevelShaderModelString(D3D_FEATURE_LEVEL feature_level);
+u32 GetRenderAPIVersionForFeatureLevel(D3D_FEATURE_LEVEL feature_level);
+D3D_FEATURE_LEVEL GetFeatureLevelForNumber(u32 render_api_version);
+u32 GetShaderModelForFeatureLevelNumber(u32 render_api_version);
 
 // returns max feature level of a device
 D3D_FEATURE_LEVEL GetDeviceMaxFeatureLevel(IDXGIAdapter1* adapter);
@@ -57,8 +59,6 @@ std::string GetAdapterName(IDXGIAdapter1* adapter);
 // returns the driver version from the registry as a string
 std::string GetDriverVersionFromLUID(const LUID& luid);
 
-u32 GetShaderModelForFeatureLevel(D3D_FEATURE_LEVEL feature_level);
-
 std::optional<DynamicHeapArray<u8>> CompileShader(u32 shader_model, bool debug_device, GPUShaderStage stage,
                                                   std::string_view source, const char* entry_point, Error* error);
 

src/util/gpu_device.cpp

@@ -367,6 +367,7 @@ bool GPUDevice::Create(std::string_view adapter, std::string_view shader_cache_p
     return false;
   }
 
+  INFO_LOG("Render API: {} Version {}", RenderAPIToString(m_render_api), m_render_api_version);
   INFO_LOG("Graphics Driver Info:\n{}", GetDriverInfo());
 
   OpenShaderCache(shader_cache_path, shader_cache_version);
@@ -401,7 +402,7 @@ void GPUDevice::OpenShaderCache(std::string_view base_path, u32 version)
   {
     const std::string basename = GetShaderCacheBaseName("shaders");
     const std::string filename = Path::Combine(base_path, basename);
-    if (!m_shader_cache.Open(filename.c_str(), version))
+    if (!m_shader_cache.Open(filename.c_str(), m_render_api_version, version))
     {
       WARNING_LOG("Failed to open shader cache. Creating new cache.");
       if (!m_shader_cache.Create())
@@ -423,7 +424,7 @@ void GPUDevice::OpenShaderCache(std::string_view base_path, u32 version)
   else
   {
     // Still need to set the version - GL needs it.
-    m_shader_cache.Open(std::string_view(), version);
+    m_shader_cache.Open(std::string_view(), m_render_api_version, version);
   }
 
   s_pipeline_cache_path = {};
@@ -473,52 +474,23 @@ std::string GPUDevice::GetShaderCacheBaseName(std::string_view type) const
 {
   const std::string_view debug_suffix = m_debug_device ? "_debug" : "";
 
-  std::string ret;
+  TinyString lower_api_name(RenderAPIToString(m_render_api));
-  switch (GetRenderAPI())
+  lower_api_name.convert_to_lower_case();
-  {
-#ifdef _WIN32
-    case RenderAPI::D3D11:
-      ret = fmt::format(
-        "d3d11_{}_{}{}", type,
-        D3DCommon::GetFeatureLevelShaderModelString(D3D11Device::GetInstance().GetD3DDevice()->GetFeatureLevel()),
-        debug_suffix);
-      break;
-    case RenderAPI::D3D12:
-      ret = fmt::format("d3d12_{}{}", type, debug_suffix);
-      break;
-#endif
-#ifdef ENABLE_VULKAN
-    case RenderAPI::Vulkan:
-      ret = fmt::format("vulkan_{}{}", type, debug_suffix);
-      break;
-#endif
-#ifdef ENABLE_OPENGL
-    case RenderAPI::OpenGL:
-      ret = fmt::format("opengl_{}{}", type, debug_suffix);
-      break;
-    case RenderAPI::OpenGLES:
-      ret = fmt::format("opengles_{}{}", type, debug_suffix);
-      break;
-#endif
-#ifdef __APPLE__
-    case RenderAPI::Metal:
-      ret = fmt::format("metal_{}{}", type, debug_suffix);
-      break;
-#endif
-    default:
-      UnreachableCode();
-      break;
-  }
 
-  return ret;
+  return fmt::format("{}_{}{}", lower_api_name, type, debug_suffix);
 }
 
 bool GPUDevice::OpenPipelineCache(const std::string& filename)
 {
-  Error error;
+  s_pipeline_cache_size = 0;
-  CompressHelpers::OptionalByteBuffer data =
+  s_pipeline_cache_hash = {};
-    CompressHelpers::DecompressFile(CompressHelpers::CompressType::Zstandard, filename.c_str(), std::nullopt, &error);
 
+  Error error;
+  CompressHelpers::OptionalByteBuffer data;
+  if (FileSystem::FileExists(filename.c_str()))
+  {
+    data =
+      CompressHelpers::DecompressFile(CompressHelpers::CompressType::Zstandard, filename.c_str(), std::nullopt, &error);
     if (data.has_value())
     {
       s_pipeline_cache_size = data->size();
@@ -527,20 +499,22 @@ bool GPUDevice::OpenPipelineCache(const std::string& filename)
     else
     {
       ERROR_LOG("Failed to load pipeline cache from '{}': {}", Path::GetFileName(filename), error.GetDescription());
-    s_pipeline_cache_size = 0;
+    }
-    s_pipeline_cache_hash = {};
   }
 
-  if (!ReadPipelineCache(std::move(data)))
+  INFO_LOG("Loading {} byte pipeline cache with hash {}", s_pipeline_cache_size,
+           SHA1Digest::DigestToString(s_pipeline_cache_hash));
+
+  if (ReadPipelineCache(std::move(data)))
+  {
+    return true;
+  }
+  else
   {
     s_pipeline_cache_size = 0;
     s_pipeline_cache_hash = {};
     return false;
   }
-
-  INFO_LOG("Loaded pipeline cache: {} bytes with hash: {}", s_pipeline_cache_size,
-           SHA1Digest::DigestToString(s_pipeline_cache_hash));
-  return true;
 }
 
 bool GPUDevice::ReadPipelineCache(std::optional<DynamicHeapArray<u8>> data)

src/util/gpu_device.h

@@ -25,7 +25,7 @@
 
 class Error;
 
-enum class RenderAPI : u32
+enum class RenderAPI : u8
 {
   None,
   D3D11,
@@ -594,6 +594,8 @@ public:
   }
 
   ALWAYS_INLINE const Features& GetFeatures() const { return m_features; }
+  ALWAYS_INLINE RenderAPI GetRenderAPI() const { return m_render_api; }
+  ALWAYS_INLINE u32 GetRenderAPIVersion() const { return m_render_api_version; }
   ALWAYS_INLINE u32 GetMaxTextureSize() const { return m_max_texture_size; }
   ALWAYS_INLINE u32 GetMaxMultisamples() const { return m_max_multisamples; }
 
@@ -608,8 +610,6 @@ public:
 
   ALWAYS_INLINE bool IsGPUTimingEnabled() const { return m_gpu_timing_enabled; }
 
-  virtual RenderAPI GetRenderAPI() const = 0;
-
   bool Create(std::string_view adapter, std::string_view shader_cache_path, u32 shader_cache_version, bool debug_device,
               GPUVSyncMode vsync, bool allow_present_throttle, std::optional<bool> exclusive_fullscreen_control,
               FeatureMask disabled_features, Error* error);
@@ -776,6 +776,8 @@ protected:
   static std::optional<DynamicHeapArray<u8>> OptimizeVulkanSpv(const std::span<const u8> spirv, Error* error);
 
   Features m_features = {};
+  RenderAPI m_render_api = RenderAPI::None;
+  u32 m_render_api_version = 0;
   u32 m_max_texture_size = 0;
   u32 m_max_multisamples = 0;
 

src/util/gpu_shader_cache.cpp

@@ -18,6 +18,12 @@
 Log_SetChannel(GPUShaderCache);
 
 #pragma pack(push, 1)
+struct CacheFileHeader
+{
+  u32 signature;
+  u32 render_api_version;
+  u32 cache_version;
+};
 struct CacheIndexEntry
 {
   u8 shader_type;
@@ -34,6 +40,8 @@ struct CacheIndexEntry
 };
 #pragma pack(pop)
 
+static constexpr u32 EXPECTED_SIGNATURE = 0x434B5544; // DUKC
+
 GPUShaderCache::GPUShaderCache() = default;
 
 GPUShaderCache::~GPUShaderCache()
@@ -59,10 +67,11 @@ std::size_t GPUShaderCache::CacheIndexEntryHash::operator()(const CacheIndexKey&
   return h;
 }
 
-bool GPUShaderCache::Open(std::string_view base_filename, u32 version)
+bool GPUShaderCache::Open(std::string_view base_filename, u32 render_api_version, u32 cache_version)
 {
   m_base_filename = base_filename;
-  m_version = version;
+  m_render_api_version = render_api_version;
+  m_version = cache_version;
 
   if (base_filename.empty())
     return true;
@@ -127,7 +136,9 @@ bool GPUShaderCache::CreateNew(const std::string& index_filename, const std::str
     return false;
   }
 
-  if (std::fwrite(&m_version, sizeof(m_version), 1, m_index_file) != 1) [[unlikely]]
+  const CacheFileHeader file_header = {
+    .signature = EXPECTED_SIGNATURE, .render_api_version = m_render_api_version, .cache_version = m_version};
+  if (std::fwrite(&file_header, sizeof(file_header), 1, m_index_file) != 1) [[unlikely]]
   {
     ERROR_LOG("Failed to write version to index file '{}'", Path::GetFileName(index_filename));
     std::fclose(m_index_file);
@@ -165,8 +176,10 @@ bool GPUShaderCache::ReadExisting(const std::string& index_filename, const std::
     return false;
   }
 
-  u32 file_version = 0;
+  CacheFileHeader file_header;
-  if (std::fread(&file_version, sizeof(file_version), 1, m_index_file) != 1 || file_version != m_version) [[unlikely]]
+  if (std::fread(&file_header, sizeof(file_header), 1, m_index_file) != 1 ||
+      file_header.signature != EXPECTED_SIGNATURE || file_header.render_api_version != m_render_api_version ||
+      file_header.cache_version != m_version) [[unlikely]]
   {
     ERROR_LOG("Bad file/data version in '{}'", Path::GetFileName(index_filename));
     std::fclose(m_index_file);

src/util/gpu_shader_cache.h

@@ -50,7 +50,7 @@ public:
 
   bool IsOpen() const { return (m_index_file != nullptr); }
 
-  bool Open(std::string_view base_filename, u32 version);
+  bool Open(std::string_view base_filename, u32 render_api_version, u32 cache_version);
   bool Create();
   void Close();
 
@@ -77,6 +77,7 @@ private:
   CacheIndex m_index;
 
   std::string m_base_filename;
+  u32 m_render_api_version = 0;
   u32 m_version = 0;
 
   std::FILE* m_index_file = nullptr;

src/util/metal_device.h

@@ -198,8 +198,6 @@ public:
   MetalDevice();
   ~MetalDevice();
 
-  RenderAPI GetRenderAPI() const override;
-
   bool HasSurface() const override;
 
   bool UpdateWindow() override;

src/util/metal_device.mm

@@ -137,11 +137,6 @@ MetalDevice::~MetalDevice()
   Assert(m_device == nil);
 }
 
-RenderAPI MetalDevice::GetRenderAPI() const
-{
-  return RenderAPI::Metal;
-}
-
 bool MetalDevice::HasSurface() const
 {
   return (m_layer != nil);
@@ -234,6 +229,8 @@ bool MetalDevice::CreateDevice(std::string_view adapter, std::optional<bool> exc
 
 void MetalDevice::SetFeatures(FeatureMask disabled_features)
 {
+  m_render_api = RenderAPI::Metal;
+  m_render_api_version = 100; // TODO: Make this more meaningful.
   m_max_texture_size = GetMetalMaxTextureSize(m_device);
   m_max_multisamples = GetMetalMaxMultisamples(m_device);
 

src/util/opengl_device.cpp

@@ -55,11 +55,6 @@ void OpenGLDevice::SetErrorObject(Error* errptr, std::string_view prefix, GLenum
   Error::SetStringFmt(errptr, "{}GL Error 0x{:04X}", prefix, static_cast<unsigned>(glerr));
 }
 
-RenderAPI OpenGLDevice::GetRenderAPI() const
-{
-  return m_gl_context->IsGLES() ? RenderAPI::OpenGLES : RenderAPI::OpenGL;
-}
-
 std::unique_ptr<GPUTexture> OpenGLDevice::CreateTexture(u32 width, u32 height, u32 layers, u32 levels, u32 samples,
                                                         GPUTexture::Type type, GPUTexture::Format format,
                                                         const void* data, u32 data_stride)
@@ -347,6 +342,13 @@ bool OpenGLDevice::CheckFeatures(FeatureMask disabled_features)
 {
   const bool is_gles = m_gl_context->IsGLES();
 
+  m_render_api = is_gles ? RenderAPI::OpenGLES : RenderAPI::OpenGL;
+
+  GLint major_version = 0, minor_version = 0;
+  glGetIntegerv(GL_MAJOR_VERSION, &major_version);
+  glGetIntegerv(GL_MINOR_VERSION, &minor_version);
+  m_render_api_version = (static_cast<u32>(major_version) * 100u) + (static_cast<u32>(minor_version) * 10u);
+
   bool vendor_id_amd = false;
   // bool vendor_id_nvidia = false;
   bool vendor_id_intel = false;

src/util/opengl_device.h

@@ -40,8 +40,6 @@ public:
   static bool ShouldUsePBOsForDownloads();
   static void SetErrorObject(Error* errptr, std::string_view prefix, GLenum glerr);
 
-  RenderAPI GetRenderAPI() const override;
-
   bool HasSurface() const override;
   void DestroySurface() override;
 

src/util/postprocessing_shader_fx.cpp

@@ -30,6 +30,7 @@
 #include <cmath>
 #include <cstring>
 #include <sstream>
+#include <tuple>
 
 Log_SetChannel(ReShadeFXShader);
 
@@ -38,7 +39,12 @@ static constexpr s32 DEFAULT_BUFFER_HEIGHT = 2160;
 
 static RenderAPI GetRenderAPI()
 {
-  return g_gpu_device ? g_gpu_device->GetRenderAPI() : RenderAPI::D3D11;
+#ifdef _WIN32
+  static constexpr RenderAPI DEFAULT_RENDER_API = RenderAPI::D3D11;
+#else
+  static constexpr RenderAPI DEFAULT_RENDER_API = RenderAPI::D3D12;
+#endif
+  return g_gpu_device ? g_gpu_device->GetRenderAPI() : DEFAULT_RENDER_API;
 }
 
 static bool PreprocessorFileExistsCallback(const std::string& path)
@@ -63,37 +69,44 @@ static bool PreprocessorReadFileCallback(const std::string& path, std::string& d
   return true;
 }
 
-static std::unique_ptr<reshadefx::codegen> CreateRFXCodegen()
+static std::tuple<std::unique_ptr<reshadefx::codegen>, GPUShaderLanguage> CreateRFXCodegen()
 {
   const bool debug_info = g_gpu_device ? g_gpu_device->IsDebugDevice() : false;
   const bool uniforms_to_spec_constants = false;
   const RenderAPI rapi = GetRenderAPI();
+  [[maybe_unused]] const u32 rapi_version = g_gpu_device ? g_gpu_device->GetRenderAPIVersion() : 0;
 
   switch (rapi)
   {
-    case RenderAPI::None:
+#ifdef _WIN32
     case RenderAPI::D3D11:
     case RenderAPI::D3D12:
     {
-      return std::unique_ptr<reshadefx::codegen>(
+      return std::make_tuple(std::unique_ptr<reshadefx::codegen>(reshadefx::create_codegen_hlsl(
-        reshadefx::create_codegen_hlsl(50, debug_info, uniforms_to_spec_constants));
+                               (rapi_version < 1100) ? 40 : 50, debug_info, uniforms_to_spec_constants)),
+                             GPUShaderLanguage::HLSL);
     }
+    break;
+#endif
 
     case RenderAPI::Vulkan:
     case RenderAPI::Metal:
     {
-      return std::unique_ptr<reshadefx::codegen>(reshadefx::create_codegen_spirv(
+      return std::make_tuple(std::unique_ptr<reshadefx::codegen>(reshadefx::create_codegen_spirv(
-        true, debug_info, uniforms_to_spec_constants, false, (rapi == RenderAPI::Vulkan)));
+                               true, debug_info, uniforms_to_spec_constants, false, (rapi == RenderAPI::Vulkan))),
+                             GPUShaderLanguage::SPV);
     }
 
     case RenderAPI::OpenGL:
     case RenderAPI::OpenGLES:
     default:
     {
-      return std::unique_ptr<reshadefx::codegen>(
+      return std::make_tuple(std::unique_ptr<reshadefx::codegen>(reshadefx::create_codegen_glsl(
-        reshadefx::create_codegen_glsl(ShaderGen::GetGLSLVersion(rapi), (rapi == RenderAPI::OpenGLES), false,
+                               g_gpu_device ? ShaderGen::GetGLSLVersion(rapi) : 460, (rapi == RenderAPI::OpenGLES),
-                                       debug_info, uniforms_to_spec_constants, false, true));
+                               false, debug_info, uniforms_to_spec_constants, false, true)),
+                             (rapi == RenderAPI::OpenGLES) ? GPUShaderLanguage::GLSLES : GPUShaderLanguage::GLSL);
     }
+    break;
   }
 }
 
@@ -308,9 +321,7 @@ bool PostProcessing::ReShadeFXShader::LoadFromString(std::string name, std::stri
     code.push_back('\n');
 
   // TODO: This could use spv, it's probably fastest.
-  std::unique_ptr<reshadefx::codegen> cg = CreateRFXCodegen();
+  const auto& [cg, cg_language] = CreateRFXCodegen();
-  if (!cg)
-    return false;
 
   if (!CreateModule(only_config ? DEFAULT_BUFFER_WIDTH : g_gpu_device->GetWindowWidth(),
                     only_config ? DEFAULT_BUFFER_HEIGHT : g_gpu_device->GetWindowHeight(), cg.get(), std::move(code),
@@ -1319,9 +1330,7 @@ bool PostProcessing::ReShadeFXShader::CompilePipeline(GPUTexture::Format format,
   if (fxcode.empty() || fxcode.back() != '\n')
     fxcode.push_back('\n');
 
-  std::unique_ptr<reshadefx::codegen> cg = CreateRFXCodegen();
+  const auto& [cg, cg_language] = CreateRFXCodegen();
-  if (!cg)
-    return false;
 
   Error error;
   if (!CreateModule(width, height, cg.get(), std::move(fxcode), &error))
@@ -1339,16 +1348,13 @@ bool PostProcessing::ReShadeFXShader::CompilePipeline(GPUTexture::Format format,
     return false;
   }
 
-  const RenderAPI api = g_gpu_device->GetRenderAPI();
+  auto get_shader = [cg_language, &cg](const std::string& name, const std::span<Sampler> samplers,
-  auto get_shader = [api, &cg](const std::string& name, const std::span<Sampler> samplers, GPUShaderStage stage) {
+                                       GPUShaderStage stage) {
     const std::string real_code = cg->finalize_code_for_entry_point(name);
-    const GPUShaderLanguage lang = (api == RenderAPI::Vulkan || api == RenderAPI::Metal) ?
+    const char* entry_point = (cg_language == GPUShaderLanguage::HLSL) ? name.c_str() : "main";
-                                     GPUShaderLanguage::SPV :
-                                     ShaderGen::GetShaderLanguageForAPI(api);
-    const char* entry_point = (lang == GPUShaderLanguage::HLSL) ? name.c_str() : "main";
 
     Error error;
-    std::unique_ptr<GPUShader> sshader = g_gpu_device->CreateShader(stage, lang, real_code, &error, entry_point);
+    std::unique_ptr<GPUShader> sshader = g_gpu_device->CreateShader(stage, cg_language, real_code, &error, entry_point);
     if (!sshader)
       ERROR_LOG("Failed to compile function '{}': {}", name, error.GetDescription());
 

src/util/vulkan_device.cpp

@@ -1889,11 +1889,6 @@ bool VulkanDevice::IsSuitableDefaultRenderer()
 #endif
 }
 
-RenderAPI VulkanDevice::GetRenderAPI() const
-{
-  return RenderAPI::Vulkan;
-}
-
 bool VulkanDevice::HasSurface() const
 {
   return static_cast<bool>(m_swap_chain);
@@ -2507,6 +2502,10 @@ u32 VulkanDevice::GetMaxMultisamples(VkPhysicalDevice physical_device, const VkP
 
 void VulkanDevice::SetFeatures(FeatureMask disabled_features, const VkPhysicalDeviceFeatures& vk_features)
 {
+  const u32 store_api_version = std::min(m_device_properties.apiVersion, VK_API_VERSION_1_1);
+  m_render_api = RenderAPI::Vulkan;
+  m_render_api_version = (VK_API_VERSION_MAJOR(store_api_version) * 100u) +
+                         (VK_API_VERSION_MINOR(store_api_version) * 10u) + (VK_API_VERSION_PATCH(store_api_version));
   m_max_texture_size =
     std::min(m_device_properties.limits.maxImageDimension2D, m_device_properties.limits.maxFramebufferWidth);
   m_max_multisamples = GetMaxMultisamples(m_physical_device, m_device_properties);

src/util/vulkan_device.h

@@ -75,8 +75,6 @@ public:
   static GPUList EnumerateGPUs();
   static AdapterInfoList GetAdapterList();
 
-  RenderAPI GetRenderAPI() const override;
-
   bool HasSurface() const override;
 
   bool UpdateWindow() override;