Update xgl from commit daeb5249

* Update PAL Version in XGL 841 * Update Khronos Vulkan Headers to 1.3.271 * Update to new DevDriver Settings * Bump Gpurt Version to 41 * Bump LLPC version to 69 * Generate debug file amdvlk64.so.debug to debug with release driver * Power Optimization for ML workloads * Alert, not assert, when ML power opt fails * Add ImageType PAL DMA constraint in XGL * Add setting for debug printf buffer creation * Rename OverrideNumVGPRsAvailable to camelCase * Simplify VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT * Bypass MALL when device_coherent_memory extension is used * Export "NV_COMPUTE_SHADER_DERIVATIVES" for vkd3d engine to run DX12 games on steam * Cleanup VKI_BUILD_NAVI31 and VKI_BUILD_GFX11 * Cleanup VariableName field in settings_xgl.json * Remove ICD_BUILD_{LLPC,SPV}ONLY * Remove the option 'enableRobustUberFetchShader' * Use core names for dynamic rendering instead of KHR names * Pass option cpsFlag to LLPC * Force PwsMode::NoLateAcquirePoint for GfxIp11 and BG3 * Improve CPU performance in QueuePresent * Move ac01WaNotNeeded to UpdatePalSettings * CS2 - App detect for waDisableAc01 on NV3x * CS2 optimizations * Fix incorrect attachment clears for dynamic rendering * Fix pipeline creation failures for mesh shader pipelines during renderdoc replay * Fix graphics pipeline library * Fix RenderPassClearUpfront OOM handling * Fix memory leak and delay BIL conversion again * Fix crash when image is blit via gpu_decode_layer * Fix warning about ignored 'const' qualifier * Fix compilation failure due to designated initializers * Fix RT Apps fail with Wave32
GPUOpen-Drivers · Dec 21, 2023 · efc2d75 · efc2d75
1 parent 565dce2
commit efc2d75
Show file tree

Hide file tree

Showing 44 changed files with 732 additions and 623 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -103,13 +103,6 @@ endif()
 # for intellisense used by VScode, YouCompleteMe, etc
 set(CMAKE_EXPORT_COMPILE_COMMANDS 1)
 
-# Strip the symbols for release build
-if(UNIX)
-    if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
-        string(APPEND CMAKE_SHARED_LINKER_FLAGS_RELEASE " -s")
-    endif()
-endif()
-
 ### Set Compiler ######################################################################################################
 xgl_set_compiler()
 
@@ -309,9 +302,7 @@ else()
 endif()
 
 # icd
-if(NOT ICD_BUILD_LLPCONLY)
 add_subdirectory(icd)
-endif()
 
 # VKGC compiler
 add_subdirectory(${XGL_VKGC_PATH} ${CMAKE_BINARY_DIR}/compiler)

diff --git a/cmake/XglOptions.cmake b/cmake/XglOptions.cmake
@@ -78,7 +78,6 @@ macro(xgl_options)
 #endif
 
         option(ICD_BUILD_LLPC "Build LLPC?" ON)
-    option(ICD_BUILD_LLPCONLY "Build LLPC Only?" OFF)
 
     option(XGL_LLVM_UPSTREAM "Build with upstreamed LLVM?" OFF)
 

diff --git a/cmake/XglOverrides.cmake b/cmake/XglOverrides.cmake
@@ -206,11 +206,6 @@ macro(xgl_overrides)
         set(XGL_BUILD_TOOLS ON CACHE BOOL "XGL_BUILD_TOOLS override by XGL_BUILD_TESTS." FORCE)
     endif()
 
-    if(ICD_BUILD_LLPCONLY)
-        set(ICD_BUILD_LLPC ON CACHE BOOL "ICD_BUILD_LLPC override." FORCE)
-        set(XGL_BUILD_TOOLS ON CACHE BOOL "XGL_BUILD_TOOLS override by ICD_BUILD_LLPCONLY." FORCE)
-    endif()
-
     if(NOT ICD_BUILD_LLPC)
         set(XGL_LLVM_UPSTREAM OFF CACHE BOOL "XGL_LLVM_UPSTREAM is overrided to false." FORCE)
     endif()

diff --git a/cmake/XglVersions.cmake b/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 "834")
+set(ICD_PAL_CLIENT_MAJOR_VERSION "841")
 
 # 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 "40")
+set(ICD_GPURT_CLIENT_MAJOR_VERSION "41")
 #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 "68")
+set(ICD_LLPC_CLIENT_MAJOR_VERSION "69")
diff --git a/icd/CMakeLists.txt b/icd/CMakeLists.txt
@@ -261,19 +261,26 @@ target_sources(xgl PRIVATE api/strings/strings.cpp)
 
 ### ICD Auto-generated Settings Files #########################################
 # ICD settings code generation main script
-set(ICD_GEN_SETTINGS ${ICD_GENDIR}/genSettingsCode.py)
 
-set(ICD_GEN_SETTINGS_FILES ${ICD_GEN_SETTINGS} ${ICD_GENDIR}/vulkanSettingsCodeTemplates.py)
+set(XGL_DEVDRIVER_PATH ${XGL_PAL_PATH}/shared/devdriver)
+set(ICD_DD_GENDIR ${XGL_DEVDRIVER_PATH}/apis/settings/codegen)
 
-set(ICD_SETTINGS_DIR ${CMAKE_CURRENT_SOURCE_DIR}/settings)
+set(ICD_GEN_SETTINGS ${ICD_DD_GENDIR}/settings_codegen.py)
 
-set(PAL_GENDIR ${XGL_PAL_PATH}/tools/generate)
+set(ICD_GEN_SETTINGS_FILES ${ICD_GEN_SETTINGS})
 
-set(COMPONENT_NAME vulkan)
+set(ICD_SETTINGS_DIR ${CMAKE_CURRENT_SOURCE_DIR}/settings)
 
 add_custom_command(
     OUTPUT ${ICD_SETTINGS_DIR}/g_settings.cpp ${ICD_SETTINGS_DIR}/g_settings.h
-    COMMAND ${PYTHON_CMD} ${ICD_GEN_SETTINGS} ${XGL_PAL_PATH} ${COMPONENT_NAME}
+    COMMAND ${PYTHON_CMD} ${ICD_GEN_SETTINGS}
+        -i ${ICD_SETTINGS_DIR}/settings_xgl.json
+        -o ${ICD_SETTINGS_DIR}
+        -g settings
+        -s settings/settings.h
+        --namespaces vk
+        --settings-struct-name RuntimeSettings
+        --include-headers pal.h palImage.h
     DEPENDS ${ICD_GEN_SETTINGS_FILES} ${ICD_SETTINGS_DIR}/settings_xgl.json
     COMMENT "Generating Vulkan settings code from settings_xgl.json"
 )
@@ -381,6 +388,18 @@ target_link_libraries(xgl_cache_support INTERFACE pal)
 
 target_link_libraries(xgl PRIVATE xgl_cache_support)
 
+# Strip the symbols for release build, generate debug file
+if(UNIX)
+        if("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+            add_custom_command(
+                TARGET xgl POST_BUILD
+                COMMAND objcopy --only-keep-debug ${CMAKE_CURRENT_BINARY_DIR}/${ICD_TARGET}.so ${CMAKE_CURRENT_BINARY_DIR}/${ICD_TARGET}.so.debug
+                COMMAND strip  ${CMAKE_CURRENT_BINARY_DIR}/${ICD_TARGET}.so
+                COMMAND objcopy --add-gnu-debuglink=${CMAKE_CURRENT_BINARY_DIR}/${ICD_TARGET}.so.debug ${CMAKE_CURRENT_BINARY_DIR}/${ICD_TARGET}.so
+            )
+        endif()
+endif()
+
 ### ICD loader configuration ###########################################################################################
 if(UNIX)
     include(GNUInstallDirs)

diff --git a/icd/Loader/LunarG/Lnx/amd-icd.json b/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": "1.3.269"
+    "api_version": "1.3.271"
   },
   "layer": {
     "name": "VK_LAYER_AMD_switchable_graphics_@ISABITS@",
     "type": "GLOBAL",
     "library_path": "@AMDVLK_INSTALL_PATH@/amdvlk@[email protected]",
-    "api_version": "1.3.269",
+    "api_version": "1.3.271",
     "implementation_version": "1",
     "description": "AMD switchable graphics layer",
     "functions": {

diff --git a/icd/api/app_profile.cpp b/icd/api/app_profile.cpp
@@ -406,18 +406,13 @@ constexpr AppProfilePatternEntry AppNameSpidermanRemastered =
     "spider-man.exe"
 };
 
-#if VKI_RAY_TRACING
 constexpr AppProfilePatternEntry AppEngineVKD3D =
 {
     PatternEngineNameLower,
-    Util::MetroHash::Hash{{{
-        0x32778d0a,
-        0x05b56a84,
-        0x8f0c25bc,
-        0x1d75f3eb
-    }}}
+    "vkd3d"
 };
 
+#if VKI_RAY_TRACING
 constexpr AppProfilePatternEntry AppNameControlDX12 =
 {
     PatternAppNameLower,
@@ -1435,6 +1430,14 @@ AppProfilePattern AppPatternTable[] =
             AppNameEnscape,
             PatternEnd
         }
+    },
+
+    {
+        AppProfile::Vkd3dEngine,
+        {
+            AppEngineVKD3D,
+            PatternEnd
+        }
     }
 };
 

diff --git a/icd/api/app_shader_optimizer.cpp b/icd/api/app_shader_optimizer.cpp
@@ -903,10 +903,10 @@ void ShaderOptimizer::BuildTuningProfile()
             pPattern->codeHash.lower = m_settings.overrideShaderHashLower;
             pPattern->codeHash.upper = m_settings.overrideShaderHashUpper;
 
-            if (m_settings.overrideNumVGPRsAvailable != 0)
+            if (m_settings.overrideNumVgprsAvailable != 0)
             {
                 pAction->shaderCreate.apply.vgprLimit         = true;
-                pAction->shaderCreate.tuningOptions.vgprLimit = m_settings.overrideNumVGPRsAvailable;
+                pAction->shaderCreate.tuningOptions.vgprLimit = m_settings.overrideNumVgprsAvailable;
             }
 
             if (m_settings.overrideMaxLdsSpillDwords != 0)
@@ -1180,15 +1180,6 @@ void ShaderOptimizer::BuildAppProfileLlpc()
         pEntry->action.shaders[ShaderStage::ShaderStageVertex].shaderCreate.tuningOptions.disableFastMathFlags = 8u | 32u;
     }
 
-    if (appProfile == AppProfile::CSGO)
-    {
-        i = m_appProfile.entryCount++;
-        PipelineProfileEntry *pEntry = &m_appProfile.pEntries[i];
-        pEntry->pattern.match.always = true;
-        pEntry->action.shaders[ShaderStage::ShaderStageFragment].shaderCreate.apply.disableFastMathFlags = true;
-        pEntry->action.shaders[ShaderStage::ShaderStageFragment].shaderCreate.tuningOptions.disableFastMathFlags = 32u;
-    }
-
     if (appProfile == AppProfile::WarHammerIII)
     {
         i = m_appProfile.entryCount++;

diff --git a/icd/api/appopt/gpu_decode_layer.cpp b/icd/api/appopt/gpu_decode_layer.cpp
@@ -545,7 +545,7 @@ static VkResult gpuBlitImage(
     uint32_t maxObj                 = pCmdBuffer->EstimateMaxObjectsOnVirtualStack(sizeof(Pal::ImageCopyRegion));
     VkFormat dstFormat              = pDstImage->GetFormat();
 
-    if(settings.enableBC3Encoder)
+    if(settings.enableBc3Encoder)
     {
         if (type == GpuTexDecoder::InternalTexConvertCsType::ConvertRGBA8ToBc3)
         {
@@ -661,7 +661,7 @@ static VkResult gpuBlitBuffer(
     GpuTexDecoder::CompileTimeConstants constInfo = {};
     Pal::SwizzledFormat sourceViewFormat          = {};
 
-    if(settings.enableBC3Encoder)
+    if(settings.enableBc3Encoder)
     {
         if (type != GpuTexDecoder::InternalTexConvertCsType::ConvertETC2ToBc3)
         {
@@ -722,8 +722,8 @@ static VkResult gpuBlitBuffer(
             pPalRegions[i] = VkToPalMemoryImageCopyRegion(
                 pRegions[regionIdx + i],
                 dstSwzFormat.format,
-                pDstImage->GetArraySize(),
                 plane,
+                pDstImage->GetArraySize(),
                 pSrcBuffer->MemOffset());
         }
 
@@ -765,7 +765,7 @@ VKAPI_ATTR void VKAPI_CALL vkCmdCopyImage(
 
     if (Formats::IsASTCFormat(pDstImage->GetFormat()))
     {
-        if (settings.enableBC3Encoder != 0)
+        if (settings.enableBc3Encoder != 0)
         {
             // ASTC one step convert to BC3 haven't implemented, so force two steps if Bc3 encoder is enabled
             twoStepsOp = true;
@@ -774,7 +774,7 @@ VKAPI_ATTR void VKAPI_CALL vkCmdCopyImage(
     }
     else if (Formats::IsEtc2Format(pDstImage->GetFormat()))
     {
-        switch (settings.enableBC3Encoder)
+        switch (settings.enableBc3Encoder)
         {
             case 0:
                 convType   = GpuTexDecoder::InternalTexConvertCsType::ConvertETC2ToRGBA8;
@@ -896,7 +896,7 @@ VKAPI_ATTR void VKAPI_CALL vkCmdCopyBufferToImage(
 
     if (Formats::IsASTCFormat(pDstImage->GetFormat()))
     {
-        if (settings.enableBC3Encoder != 0)
+        if (settings.enableBc3Encoder != 0)
         {
             // ASTC one step convert to BC3 haven't implemented, so force two steps if Bc3 encoder is enabled
             twoStepsOp = true;
@@ -905,7 +905,7 @@ VKAPI_ATTR void VKAPI_CALL vkCmdCopyBufferToImage(
     }
     else if (Formats::IsEtc2Format(pDstImage->GetFormat()))
     {
-        switch (settings.enableBC3Encoder)
+        switch (settings.enableBc3Encoder)
         {
             case 0:
                 convType   = GpuTexDecoder::InternalTexConvertCsType::ConvertETC2ToRGBA8;
@@ -1101,7 +1101,7 @@ VKAPI_ATTR void VKAPI_CALL vkDestroyImage(
     GpuDecoderLayer* pDecodeWrapper = pDevice->GetGpuDecoderLayer();
     const RuntimeSettings& settings = pDevice->GetRuntimeSettings();
 
-    if (settings.enableBC3Encoder)
+    if (settings.enableBc3Encoder)
     {
         pDevice->GetGpuDecoderLayer()->ClearStagingResources(image);
     }

diff --git a/icd/api/compiler_solution_llpc.cpp b/icd/api/compiler_solution_llpc.cpp
@@ -192,6 +192,7 @@ void CompilerSolutionLlpc::FreeShaderModule(ShaderModuleHandle* pShaderModule)
     auto pInstance = m_pPhysicalDevice->Manager()->VkInstance();
 
     pInstance->FreeMem(pShaderModule->pLlpcShaderModule);
+    pShaderModule->pLlpcShaderModule = nullptr;
 }
 
 // =====================================================================================================================

diff --git a/icd/api/debug_printf.cpp b/icd/api/debug_printf.cpp
@@ -98,7 +98,10 @@ void DebugPrintf::BindPipeline(
         allocInfo.pal.size = Util::Pow2Align(settings.debugPrintfBufferSize, PAL_PAGE_BYTES);
         allocInfo.pal.alignment = PAL_PAGE_BYTES;
         allocInfo.pal.priority = Pal::GpuMemPriority::Normal;
-        pDevice->MemMgr()->GetCommonPool(InternalPoolCpuCacheableGpuUncached, &allocInfo);
+        InternalSubAllocPool bufferType = settings.enableHangOutput ?
+            InternalPoolCpuCacheableGpuUncached : InternalPoolDebugCpuRead;
+
+        pDevice->MemMgr()->GetCommonPool(bufferType, &allocInfo);
         VkResult result = pDevice->MemMgr()->AllocGpuMem(allocInfo,
                                                          &m_printfMemory,
                                                          pDevice->GetPalDeviceMask(),

diff --git a/icd/api/entry.cpp b/icd/api/entry.cpp
@@ -1219,8 +1219,8 @@ VKAPI_ATTR void VKAPI_CALL vkCmdWriteBufferMarker2AMD(
 
 // =====================================================================================================================
 VKAPI_ATTR void VKAPI_CALL vkCmdBeginRendering(
-    VkCommandBuffer           commandBuffer,
-    const VkRenderingInfoKHR* pRenderingInfo)
+    VkCommandBuffer        commandBuffer,
+    const VkRenderingInfo* pRenderingInfo)
 {
     ApiCmdBuffer::ObjectFromHandle(commandBuffer)->BeginRendering(pRenderingInfo);
 }
-Original file line number
+Diff line change
@@ Expand Up / @@ -78,7 +78,6 @@ macro(xgl_options) @@
     #endif
             option(ICD_BUILD_LLPC "Build LLPC?" ON)
-        option(ICD_BUILD_LLPCONLY "Build LLPC Only?" OFF)
         option(XGL_LLVM_UPSTREAM "Build with upstreamed LLVM?" OFF)
@@ Expand Down @@