[d3d11,dxbc] Rework check for TypedUAVLoadAdditionalFormats

src/d3d11/d3d11_device.cpp

@@ -1681,10 +1681,9 @@ namespace dxvk {
           return E_INVALIDARG;
 
         const auto& extensions = m_dxvkDevice->extensions();
-        const auto& features = m_dxvkDevice->features();
 
         info->PSSpecifiedStencilRefSupported = extensions.extShaderStencilExport;
-        info->TypedUAVLoadAdditionalFormats  = features.core.features.shaderStorageImageReadWithoutFormat;
+        info->TypedUAVLoadAdditionalFormats  = m_dxbcOptions.supportsTypedUavLoadExtended;
         info->ROVsSupported                  = FALSE;
         info->ConservativeRasterizationTier  = D3D11_CONSERVATIVE_RASTERIZATION_NOT_SUPPORTED;
         info->MapOnDefaultTextures           = TRUE;
@@ -1977,7 +1976,6 @@ namespace dxvk {
       enabled.core.features.multiDrawIndirect                     = VK_TRUE;
       enabled.core.features.shaderFloat64                         = supported.core.features.shaderFloat64;
       enabled.core.features.shaderInt64                           = supported.core.features.shaderInt64;
-      enabled.core.features.shaderStorageImageReadWithoutFormat   = supported.core.features.shaderStorageImageReadWithoutFormat;
       enabled.core.features.tessellationShader                    = VK_TRUE;
     }
     

src/d3d11/d3d11_device.h

@@ -454,7 +454,7 @@ namespace dxvk {
     D3D11StateObjectSet<D3D11RasterizerState>   m_rsStateObjects;
     D3D11StateObjectSet<D3D11SamplerState>      m_samplerObjects;
     D3D11ShaderModuleSet                        m_shaderModules;
-    
+
     HRESULT CreateShaderModule(
             D3D11CommonShader*      pShaderModule,
             DxvkShaderKey           ShaderKey,

src/dxbc/dxbc_compiler.cpp

@@ -876,7 +876,7 @@ namespace dxvk {
     const bool isUav = ins.op == DxbcOpcode::DclUavTyped;
     
     if (isUav) {
-      if (m_moduleInfo.options.useStorageImageReadWithoutFormat)
+      if (m_moduleInfo.options.supportsTypedUavLoadR32)
         m_module.enableCapability(spv::CapabilityStorageImageReadWithoutFormat);
       m_module.enableCapability(spv::CapabilityStorageImageWriteWithoutFormat);
     }
@@ -951,7 +951,7 @@ namespace dxvk {
     if (isUav) {
       if ((m_analysis->uavInfos[registerId].accessAtomicOp)
        || (m_analysis->uavInfos[registerId].accessTypedLoad
-        && !m_moduleInfo.options.useStorageImageReadWithoutFormat))
+        && !m_moduleInfo.options.supportsTypedUavLoadR32))
         imageFormat = getScalarImageFormat(sampledType);
     }
     

src/dxbc/dxbc_options.cpp

@@ -17,8 +17,6 @@ namespace dxvk {
 
     useDepthClipWorkaround
       = !devFeatures.extDepthClipEnable.depthClipEnable;
-    useStorageImageReadWithoutFormat
-      = devFeatures.core.features.shaderStorageImageReadWithoutFormat;
     useSubgroupOpsForAtomicCounters
       = (devInfo.vk11.subgroupSupportedStages     & VK_SHADER_STAGE_COMPUTE_BIT)
      && (devInfo.vk11.subgroupSupportedOperations & VK_SUBGROUP_FEATURE_BALLOT_BIT);
@@ -28,7 +26,41 @@ namespace dxvk {
       = (devInfo.vk11.subgroupSize >= 4)
      && (devInfo.vk11.subgroupSupportedStages     & VK_SHADER_STAGE_FRAGMENT_BIT)
      && (devInfo.vk11.subgroupSupportedOperations & VK_SUBGROUP_FEATURE_BALLOT_BIT);
-    
+
+    supportsTypedUavLoadR32 = true;
+    supportsTypedUavLoadExtended = true;
+
+    static const std::array<std::pair<VkFormat, bool>, 18> s_typedUavFormats = {
+      std::make_pair(VK_FORMAT_R32_SFLOAT, false),
+      std::make_pair(VK_FORMAT_R32_UINT, false),
+      std::make_pair(VK_FORMAT_R32_SINT, false),
+      std::make_pair(VK_FORMAT_R32G32B32A32_SFLOAT, true),
+      std::make_pair(VK_FORMAT_R32G32B32A32_UINT, true),
+      std::make_pair(VK_FORMAT_R32G32B32A32_SINT, true),
+      std::make_pair(VK_FORMAT_R16G16B16A16_SFLOAT, true),
+      std::make_pair(VK_FORMAT_R16G16B16A16_UINT, true),
+      std::make_pair(VK_FORMAT_R16G16B16A16_SINT, true),
+      std::make_pair(VK_FORMAT_R8G8B8A8_UNORM, true),
+      std::make_pair(VK_FORMAT_R8G8B8A8_UINT, true),
+      std::make_pair(VK_FORMAT_R8G8B8A8_SINT, true),
+      std::make_pair(VK_FORMAT_R16_SFLOAT, true),
+      std::make_pair(VK_FORMAT_R16_UINT, true),
+      std::make_pair(VK_FORMAT_R16_SINT, true),
+      std::make_pair(VK_FORMAT_R8_UNORM, true),
+      std::make_pair(VK_FORMAT_R8_UINT, true),
+      std::make_pair(VK_FORMAT_R8_SINT, true),
+    };
+
+    for (const auto& f : s_typedUavFormats) {
+      DxvkFormatFeatures features = device->getFormatFeatures(f.first);
+      VkFormatFeatureFlags2 imgFeatures = features.optimal | features.linear;
+
+      if (!(imgFeatures & VK_FORMAT_FEATURE_2_STORAGE_READ_WITHOUT_FORMAT_BIT)) {
+        supportsTypedUavLoadR32 &= f.second;
+        supportsTypedUavLoadExtended = false;
+      }
+    }
+
     switch (device->config().useRawSsbo) {
       case Tristate::Auto:  minSsboAlignment = devInfo.core.properties.limits.minStorageBufferOffsetAlignment; break;
       case Tristate::True:  minSsboAlignment =  4u; break;

src/dxbc/dxbc_options.h

@@ -23,8 +23,10 @@ namespace dxvk {
     // clip device feature is not supported
     bool useDepthClipWorkaround = false;
 
-    /// Use the ShaderImageReadWithoutFormat capability.
+    /// Determines whether format qualifiers
-    bool useStorageImageReadWithoutFormat = false;
+    /// on typed UAV loads are required
+    bool supportsTypedUavLoadR32 = false;
+    bool supportsTypedUavLoadExtended = false;
 
     /// Use subgroup operations to reduce the number of
     /// atomic operations for append/consume buffers.