Update xgl from commit ee46f91a

* Add navi32 support
* Update PAL Version in XGL 819
* Vulkan Graphics pipeline library fast-link revise
* Bump LLPC client version to 66
* Refine graphics pipeline api hash
* Add shader tuning option EliminateFuncLoadStore
* Gpurt Version Bump to 39
* Switch to old PAL metadata for Detroit
* Change default for EnableUpdateParallel
* Add Ac01WaNotNeeded VkPanel.
* Uber fetch shader improvements
* Workaround double free in Baldur’s Gate
* Update Khronos Vulkan Headers to 1.3.264
* Simplify LLPC cache
* Fix dEQP-VK.info.device_mandatory_features  test failed on Gfx8
* Support dynamic sample info
* Tessellation' demo cannot be seen in Wireframe mode.
* Amdvlk Treat ErrorUnknown as OOM
* RasterizationSamples fix dynamic
* AppProfile Yuzu disable DccForColorAttachments
* Handle new compiler interface for shader handle
* Fixes for blend state and uber fetch shader
* White lines observed on edges in dark locations
* Rename ShaderStageTessEvaluation to ShaderStageTessEval

cmake/XglCompileDefinitions.cmake

@@ -67,6 +67,12 @@ macro(xgl_set_compile_definitions)
     endif()
 #endif
 
+#if VKI_BUILD_NAVI32
+    if(XGL_BUILD_NAVI32)
+        target_compile_definitions(xgl PRIVATE VKI_BUILD_NAVI32=1)
+    endif()
+#endif
+
 #if VKI_BUILD_NAVI33
     if(XGL_BUILD_NAVI33)
         target_compile_definitions(xgl PRIVATE VKI_BUILD_NAVI33=1)

cmake/XglOptions.cmake

@@ -53,6 +53,10 @@ macro(xgl_options)
     option(XGL_BUILD_NAVI31 "Build vulkan for Navi31" ON)
 #endif
 
+#if VKI_BUILD_NAVI32
+    option(XGL_BUILD_NAVI32 "Build vulkan for Navi32" ON)
+#endif
+
 #if VKI_BUILD_NAVI33
     option(XGL_BUILD_NAVI33 "Build vulkan for Navi33" ON)
 #endif

cmake/XglOverrides.cmake

@@ -129,6 +129,10 @@ macro(xgl_overrides_pal)
     set(PAL_BUILD_NAVI31 ${XGL_BUILD_NAVI31} CACHE BOOL "${PROJECT_NAME} override." FORCE)
 #endif
 
+#if VKI_BUILD_NAVI32
+    set(PAL_BUILD_NAVI32 ${XGL_BUILD_NAVI32} CACHE BOOL "${PROJECT_NAME} override." FORCE)
+#endif
+
 #if VKI_BUILD_NAVI33
     set(PAL_BUILD_NAVI33 ${XGL_BUILD_NAVI33} CACHE BOOL "${PROJECT_NAME} override." FORCE)
 #endif
@@ -175,6 +179,10 @@ macro(xgl_overrides_vkgc)
     set(LLPC_BUILD_NAVI31 ${XGL_BUILD_NAVI31} CACHE BOOL "${PROJECT_NAME} override." FORCE)
 #endif
 
+#if VKI_BUILD_NAVI32
+    set(LLPC_BUILD_NAVI32 ${XGL_BUILD_NAVI32} CACHE BOOL "${PROJECT_NAME} override." FORCE)
+#endif
+
 #if VKI_BUILD_NAVI33
     set(LLPC_BUILD_NAVI33 ${XGL_BUILD_NAVI33} CACHE BOOL "${PROJECT_NAME} override." FORCE)
 #endif

cmake/XglVersions.cmake

@@ -28,7 +28,7 @@ include_guard()
 # This will become the value of PAL_CLIENT_INTERFACE_MAJOR_VERSION.  It describes the version of the PAL interface
 # that the ICD supports.  PAL uses this value to enable backwards-compatibility for older interface versions.
 # It must be updated on each PAL promotion after handling all of the interface changes described in palLib.h.
-set(ICD_PAL_CLIENT_MAJOR_VERSION "816")
+set(ICD_PAL_CLIENT_MAJOR_VERSION "819")
 
 # This will become the value of GPUOPEN_CLIENT_INTERFACE_MAJOR_VERSION if ICD_GPUOPEN_DEVMODE_BUILD=1.
 # It describes the interface version of the gpuopen shared module (part of PAL) that the ICD supports.
@@ -37,9 +37,9 @@ set(ICD_GPUOPEN_CLIENT_MAJOR_VERSION "42")
 #if VKI_RAY_TRACING
 # This will become the value of GPURT_CLIENT_INTERFACE_MAJOR_VERSION if VKI_RAY_TRACING=1.
 # It describes the interface version of the GpuRT shared module that the ICD supports.
-set(ICD_GPURT_CLIENT_MAJOR_VERSION "33")
+set(ICD_GPURT_CLIENT_MAJOR_VERSION "39")
 #endif
 
 # This will become the value of LLPC_CLIENT_INTERFACE_MAJOR_VERSION if ICD_BUILD_LLPC=1.
 # It describes the version of the interface version of LLPC that the ICD supports.
-set(ICD_LLPC_CLIENT_MAJOR_VERSION "63")
+set(ICD_LLPC_CLIENT_MAJOR_VERSION "66")

icd/Loader/LunarG/Lnx/amd-icd.json

@@ -2,13 +2,13 @@
   "file_format_version": "1.0.0",
   "ICD": {
     "library_path": "@AMDVLK_INSTALL_PATH@/amdvlk@[email protected]",
-    "api_version": "1.3.261"
+    "api_version": "1.3.264"
   },
   "layer": {
     "name": "VK_LAYER_AMD_switchable_graphics_@ISABITS@",
     "type": "GLOBAL",
     "library_path": "@AMDVLK_INSTALL_PATH@/amdvlk@[email protected]",
-    "api_version": "1.3.261",
+    "api_version": "1.3.264",
     "implementation_version": "1",
     "description": "AMD switchable graphics layer",
     "functions": {

icd/api/app_resource_optimizer.cpp

@@ -481,26 +481,6 @@ void ResourceOptimizer::BuildAppProfile()
             m_appProfile.entries[i].action.resourceCreate.dccMode = DccMode::DccDisableMode;
         }
     }
-#if PAL_BUILD_GFX11
-    else if (appProfile == AppProfile::Yuzu)
-    {
-        if (gfxIpLevel == Pal::GfxIpLevel::GfxIp11_0)
-        {
-            // 0x0000064000000384f83db7fb629521bd
-            i = m_appProfile.entryCount++;
-            m_appProfile.entries[i].pattern.match.apiHash = true;
-            m_appProfile.entries[i].pattern.targetKey.apiHash = 0xf83db7fb629521bd;
-            m_appProfile.entries[i].action.resourceCreate.apply.dccMode = true;
-            m_appProfile.entries[i].action.resourceCreate.dccMode = DccMode::DccDisableMode;
-            // 0x00000640000003845a18ecbb795c8bed
-            i = m_appProfile.entryCount++;
-            m_appProfile.entries[i].pattern.match.apiHash = true;
-            m_appProfile.entries[i].pattern.targetKey.apiHash = 0x5a18ecbb795c8bed;
-            m_appProfile.entries[i].action.resourceCreate.apply.dccMode = true;
-            m_appProfile.entries[i].action.resourceCreate.dccMode = DccMode::DccDisableMode;
-        }
-    }
-#endif
 }
 
 #if ICD_RUNTIME_APP_PROFILE

icd/api/app_shader_optimizer.cpp

@@ -628,7 +628,7 @@ void ShaderOptimizer::ApplyProfileToGraphicsPipelineCreateInfo(
                     case ShaderStage::ShaderStageTessControl:
                         ApplyProfileToDynamicGraphicsShaderInfo(shaders[vkgcStage], &pGraphicsShaderInfos->hs);
                         break;
-                    case ShaderStage::ShaderStageTessEvaluation:
+                    case ShaderStage::ShaderStageTessEval:
                         ApplyProfileToDynamicGraphicsShaderInfo(shaders[vkgcStage], &pGraphicsShaderInfos->ds);
                         break;
                     case ShaderStage::ShaderStageGeometry:
@@ -1300,7 +1300,7 @@ void ShaderOptimizer::PrintProfileEntryMatch(
             case ShaderStage::ShaderStageTessControl:
                 pStage = "HS";
                 break;
-            case ShaderStage::ShaderStageTessEvaluation:
+            case ShaderStage::ShaderStageTessEval:
                 pStage = "DS";
                 break;
             case ShaderStage::ShaderStageGeometry:

icd/api/compiler_solution.cpp

@@ -114,7 +114,8 @@ const char* CompilerSolution::GetGraphicsLibraryName(
     static const char* GraphicsLibraryTypeNames[] =
     {
         "PreRasterLib",
-        "FragmentLib"
+        "FragmentLib",
+        "ColorExportLib"
     };
 
     pName = GraphicsLibraryTypeNames[static_cast<uint32_t>(libraryType)];
@@ -204,17 +205,8 @@ void CompilerSolution::UpdateRayTracingFunctionNames(
 
         VK_ASSERT(result == VK_SUCCESS);
 
-        if (settings.rtUseRayQueryForTraceRays)
-        {
-            SetRayTracingFunctionName(
-                entryFuncTable.traceRay.pTraceRayUsingRayQuery, pTable->pFunc[Vkgc::RT_ENTRY_TRACE_RAY]);
-        }
-        else
-        {
-            SetRayTracingFunctionName(
-                entryFuncTable.traceRay.pTraceRay, pTable->pFunc[Vkgc::RT_ENTRY_TRACE_RAY]);
-        }
-
+        SetRayTracingFunctionName(
+            entryFuncTable.traceRay.pTraceRay, pTable->pFunc[Vkgc::RT_ENTRY_TRACE_RAY]);
         SetRayTracingFunctionName(
             entryFuncTable.traceRay.pTraceRayUsingHitToken, pTable->pFunc[Vkgc::RT_ENTRY_TRACE_RAY_HIT_TOKEN]);
         SetRayTracingFunctionName(

icd/api/compiler_solution_llpc.cpp

@@ -224,10 +224,6 @@ VkResult CompilerSolutionLlpc::CreateGraphicsPipelineBinary(
     pPipelineBuildInfo->pInstance      = pInstance;
     pPipelineBuildInfo->pfnOutputAlloc = AllocateShaderOutput;
     pPipelineBuildInfo->iaState.deviceIndex = deviceIdx;
-    if ((pPipelineCache != nullptr) && (settings.shaderCacheMode != ShaderCacheDisable))
-    {
-        pPipelineBuildInfo->cache = pPipelineCache->GetCacheAdapter();
-    }
 
     // By default the client hash provided to PAL is more accurate than the one used by pipeline
     // profiles.
@@ -411,6 +407,47 @@ VkResult CompilerSolutionLlpc::CreateGraphicsPipelineBinary(
     return result;
 }
 
+// =====================================================================================================================
+// Creates color export binary.
+VkResult CompilerSolutionLlpc::CreateColorExportBinary(
+    GraphicsPipelineBinaryCreateInfo* pCreateInfo,
+    void*                             pPipelineDumpHandle,
+    Vkgc::BinaryData*                 pOutputPackage)
+{
+    const RuntimeSettings& settings  = m_pPhysicalDevice->GetRuntimeSettings();
+    auto pInstance                   = m_pPhysicalDevice->Manager()->VkInstance();
+
+    VkResult result                            = VK_SUCCESS;
+    Llpc::GraphicsPipelineBuildOut pipelineOut = {};
+
+    auto pPipelineBuildInfo                 = &pCreateInfo->pipelineInfo;
+    pPipelineBuildInfo->pInstance           = pInstance;
+    pPipelineBuildInfo->pfnOutputAlloc      = AllocateShaderOutput;
+
+    VK_ASSERT(pCreateInfo->pBinaryMetadata->pFsOutputMetaData != nullptr);
+    Vkgc::Result llpcResult = m_pLlpc->BuildColorExportShader(pPipelineBuildInfo,
+                                                              pCreateInfo->pBinaryMetadata->pFsOutputMetaData,
+                                                              &pipelineOut,
+                                                              pPipelineDumpHandle);
+
+    if (llpcResult == Vkgc::Result::Success)
+    {
+        pOutputPackage->pCode    = pipelineOut.pipelineBin.pCode;
+        pOutputPackage->codeSize = pipelineOut.pipelineBin.codeSize;
+        if (settings.enablePipelineDump && (pPipelineDumpHandle != nullptr))
+        {
+            Vkgc::IPipelineDumper::DumpPipelineBinary(pPipelineDumpHandle, m_gfxIp, &pipelineOut.pipelineBin);
+        }
+    }
+    else
+    {
+        result = (llpcResult == Vkgc::Result::RequireFullPipeline) ? VK_PIPELINE_COMPILE_REQUIRED :
+                                                                     VK_ERROR_INITIALIZATION_FAILED;
+    }
+
+    return result;
+}
+
 // =====================================================================================================================
 // Build ElfPackage for a specific shader module based on pipeline information
 VkResult CompilerSolutionLlpc::CreateGraphicsShaderBinary(
@@ -469,38 +506,43 @@ VkResult CompilerSolutionLlpc::CreateGraphicsShaderBinary(
                 pPipelineDumpHandle);
             if (llpcResult == Vkgc::Result::Success)
             {
+                if (stage == ShaderStage::ShaderStageFragment)
+                {
+                    pCreateInfo->pBinaryMetadata->pFsOutputMetaData    = pipelineOut.fsOutputMetaData;
+                    pCreateInfo->pBinaryMetadata->fsOutputMetaDataSize = pipelineOut.fsOutputMetaDataSize;
+                    pipelineOut.pipelineBin.codeSize += pipelineOut.fsOutputMetaDataSize;
+                }
+
                 pShaderModule->elfPackage = pipelineOut.pipelineBin;
                 if ((pPipelineCache != nullptr) && (pPipelineCache->GetPipelineCache() != nullptr))
                 {
                     pPipelineCache->GetPipelineCache()->StorePipelineBinary(
                         &cacheId, pipelineOut.pipelineBin.codeSize, pipelineOut.pipelineBin.pCode);
                 }
             }
-            else
+            else if (llpcResult != Vkgc::Result::RequireFullPipeline)
             {
-
-                result = (llpcResult == Vkgc::Result::ErrorOutOfMemory) ?
-                    VK_ERROR_OUT_OF_HOST_MEMORY : VK_ERROR_INITIALIZATION_FAILED;
-
+                result = (llpcResult == Vkgc::Result::ErrorOutOfMemory) ? VK_ERROR_OUT_OF_HOST_MEMORY :
+                                                                          VK_ERROR_INITIALIZATION_FAILED;
             }
-        }
 
-        if (result == VK_SUCCESS)
-        {
-            pCreateInfo->earlyElfPackage[gplType]     = pShaderModule->elfPackage;
-            pCreateInfo->earlyElfPackageHash[gplType] = cacheId;
-
-            if (stage == ShaderStage::ShaderStageFragment)
+            if (llpcResult == Vkgc::Result::Success)
             {
-                const auto* pModuleData =
-                    reinterpret_cast<const Vkgc::ShaderModuleData*>(pCreateInfo->pipelineInfo.fs.pModuleData);
-                pCreateInfo->pBinaryMetadata->needsSampleInfo = pModuleData->usage.useSampleInfo;
-            }
+                pCreateInfo->earlyElfPackage[gplType]     = pShaderModule->elfPackage;
+                pCreateInfo->earlyElfPackageHash[gplType] = cacheId;
 
-            if (pPipelineDumpHandle != nullptr)
-            {
-                Vkgc::IPipelineDumper::DumpPipelineBinary(
-                    pPipelineDumpHandle, m_gfxIp, &pCreateInfo->earlyElfPackage[gplType]);
+                if (stage == ShaderStage::ShaderStageFragment)
+                {
+                    const auto* pModuleData =
+                        reinterpret_cast<const Vkgc::ShaderModuleData*>(pCreateInfo->pipelineInfo.fs.pModuleData);
+                    pCreateInfo->pBinaryMetadata->needsSampleInfo = pModuleData->usage.useSampleInfo;
+                }
+
+                if (pPipelineDumpHandle != nullptr)
+                {
+                    Vkgc::IPipelineDumper::DumpPipelineBinary(
+                        pPipelineDumpHandle, m_gfxIp, &pCreateInfo->earlyElfPackage[gplType]);
+                }
             }
         }
     }
@@ -540,10 +582,6 @@ VkResult CompilerSolutionLlpc::CreateComputePipelineBinary(
     // Fill pipeline create info for LLPC
     pPipelineBuildInfo->pInstance      = pInstance;
     pPipelineBuildInfo->pfnOutputAlloc = AllocateShaderOutput;
-    if ((pPipelineCache != nullptr) && (settings.shaderCacheMode != ShaderCacheDisable))
-    {
-        pPipelineBuildInfo->cache = pPipelineCache->GetCacheAdapter();
-    }
 
     // Force enable automatic workgroup reconfigure.
     if (appProfile == AppProfile::DawnOfWarIII)
@@ -798,10 +836,6 @@ VkResult CompilerSolutionLlpc::CreateRayTracingPipelineBinary(
     pPipelineBuildInfo->pfnOutputAlloc = AllocateShaderOutput;
     pPipelineBuildInfo->deviceIndex = deviceIdx;
     pPipelineBuildInfo->deviceCount = m_pPhysicalDevice->Manager()->GetDeviceCount();
-    if ((pPipelineCache != nullptr) && (settings.shaderCacheMode != ShaderCacheDisable))
-    {
-        pPipelineBuildInfo->cache = pPipelineCache->GetCacheAdapter();
-    }
 
     auto llpcResult = m_pLlpc->BuildRayTracingPipeline(pPipelineBuildInfo,
                                                        &pipelineOut,
@@ -966,13 +1000,15 @@ VkResult CompilerSolutionLlpc::CreateLlpcCompiler(
         llpcOptions[numOptions++] = "-amdgpu-max-memory-clause=1";
     }
 
+#if LLPC_CLIENT_INTERFACE_MAJOR_VERSION < 66
     optionLength = Util::Snprintf(pOptionBuffer, bufSize, "-executable-name=%s", pExecutablePtr);
     ++optionLength;
     llpcOptions[numOptions++] = pOptionBuffer;
     pOptionBuffer += optionLength;
     bufSize -= optionLength;
 
-    optionLength = Util::Snprintf(pOptionBuffer, bufSize, "-shader-cache-file-dir=%s", m_pPhysicalDevice->PalDevice()->GetCacheFilePath());
+    optionLength = Util::Snprintf(pOptionBuffer, bufSize, "-shader-cache-file-dir=%s",
+        m_pPhysicalDevice->PalDevice()->GetCacheFilePath());
     ++optionLength;
     llpcOptions[numOptions++] = pOptionBuffer;
     pOptionBuffer += optionLength;
@@ -983,6 +1019,9 @@ VkResult CompilerSolutionLlpc::CreateLlpcCompiler(
     llpcOptions[numOptions++] = pOptionBuffer;
     pOptionBuffer += optionLength;
     bufSize -= optionLength;
+#endif
+
+    llpcOptions[numOptions++] = "-cache-full-pipelines=0";
 
     optionLength = Util::Snprintf(pOptionBuffer, bufSize, "-subgroup-size=%d", m_pPhysicalDevice->GetSubgroupSize());
     ++optionLength;