Merge pull request KhronosGroup#2512 from squidbus/maintenance8

billhollings · web-flow · commit 93a6739bea8a · 2025-05-04T14:25:07.000-04:00
Add support for VK_KHR_maintenance8
diff --git a/Docs/MoltenVK_Runtime_UserGuide.md b/Docs/MoltenVK_Runtime_UserGuide.md
@@ -266,6 +266,7 @@ In addition to core *Vulkan* functionality, **MoltenVK**  also supports the foll
 - `VK_KHR_maintenance5`
 - `VK_KHR_maintenance6`
 - `VK_KHR_maintenance7`
+- `VK_KHR_maintenance8`
 - `VK_KHR_map_memory2`
 - `VK_KHR_multiview`
 - `VK_KHR_portability_subset`
diff --git a/Docs/Whats_New.md b/Docs/Whats_New.md
@@ -21,6 +21,7 @@ Released TBD
 - Add support for extensions:
 	- `VK_KHR_global_priority`
 	- `VK_KHR_maintenance5`
+	- `VK_KHR_maintenance8`
 	- `VK_KHR_present_id`
 	- `VK_KHR_present_wait`
 	- `VK_KHR_shader_relaxed_extended_instruction`
diff --git a/ExternalRevisions/SPIRV-Cross_repo_revision b/ExternalRevisions/SPIRV-Cross_repo_revision
@@ -1 +1 @@
-e07cec7482ed583f58625d3f3c34874e53a003be
+f1cb5b6634c58be69c327450d0023fb061f0f650
diff --git a/MoltenVK/MoltenVK/Commands/MVKCmdTransfer.mm b/MoltenVK/MoltenVK/Commands/MVKCmdTransfer.mm
@@ -102,15 +102,31 @@ static inline MTLSize mvkClampMTLSize(MTLSize size, MTLOrigin origin, MTLSize ma
     return VK_SUCCESS;
 }
 
+static inline MTLPixelFormat getDepthStencilAspectFormat(const MTLPixelFormat format, const VkImageAspectFlags aspectMask) {
+    if (format == MTLPixelFormatDepth32Float_Stencil8) {
+        if (aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) return MTLPixelFormatDepth32Float;
+        if (aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) return MTLPixelFormatStencil8;
+    }
+#if MVK_MACOS
+    if (format == MTLPixelFormatDepth24Unorm_Stencil8 && (aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT))
+        return MTLPixelFormatStencil8;
+#endif
+    return format;
+}
+
 template <size_t N>
 inline VkResult MVKCmdCopyImage<N>::validate(MVKCommandBuffer* cmdBuff, const VkImageCopy2* region) {
     uint8_t srcPlaneIndex = MVKImage::getPlaneFromVkImageAspectFlags(region->srcSubresource.aspectMask);
     uint8_t dstPlaneIndex = MVKImage::getPlaneFromVkImageAspectFlags(region->dstSubresource.aspectMask);
 
+    // If the format is combined depth-stencil, use the format based on the aspect for the following checks.
+    auto srcFormat = getDepthStencilAspectFormat(_srcImage->getMTLPixelFormat(srcPlaneIndex), region->srcSubresource.aspectMask);
+    auto dstFormat = getDepthStencilAspectFormat(_dstImage->getMTLPixelFormat(dstPlaneIndex), region->dstSubresource.aspectMask);
+
     // Validate
     MVKPixelFormats* pixFmts = cmdBuff->getPixelFormats();
     if ((_dstImage->getSampleCount() != _srcImage->getSampleCount()) ||
-        (pixFmts->getBytesPerBlock(_dstImage->getMTLPixelFormat(dstPlaneIndex)) != pixFmts->getBytesPerBlock(_srcImage->getMTLPixelFormat(srcPlaneIndex)))) {
+        (pixFmts->getBytesPerBlock(srcFormat) != pixFmts->getBytesPerBlock(dstFormat))) {
         return cmdBuff->reportError(VK_ERROR_FEATURE_NOT_PRESENT, "vkCmdCopyImage(): Cannot copy between incompatible formats, such as formats of different pixel sizes, or between images with different sample counts.");
     }
     return VK_SUCCESS;
@@ -133,17 +149,20 @@ static inline MTLSize mvkClampMTLSize(MTLSize size, MTLOrigin origin, MTLSize ma
         
         MTLPixelFormat srcMTLPixFmt = _srcImage->getMTLPixelFormat(srcPlaneIndex);
         bool isSrcCompressed = _srcImage->getIsCompressed();
+        bool isSrcCombinedDepthStencilAspect = getDepthStencilAspectFormat(srcMTLPixFmt, vkIC.srcSubresource.aspectMask) != srcMTLPixFmt;
         bool canReinterpretSrc = _srcImage->hasPixelFormatView(srcPlaneIndex);
 
         MTLPixelFormat dstMTLPixFmt = _dstImage->getMTLPixelFormat(dstPlaneIndex);
         bool isDstCompressed = _dstImage->getIsCompressed();
+        bool isDstCombinedDepthStencilAspect = getDepthStencilAspectFormat(dstMTLPixFmt, vkIC.dstSubresource.aspectMask) != dstMTLPixFmt;
         bool canReinterpretDst = _dstImage->hasPixelFormatView(dstPlaneIndex);
 
         bool isEitherCompressed = isSrcCompressed || isDstCompressed;
+        bool isOneCombinedDepthStencilAspect = isSrcCombinedDepthStencilAspect != isDstCombinedDepthStencilAspect;
         bool canReinterpret = canReinterpretSrc || canReinterpretDst;
 
         // If source and destination can't be reinterpreted to matching formats use a temporary intermediary buffer
-        bool useTempBuffer = (srcMTLPixFmt != dstMTLPixFmt) && (isEitherCompressed || !canReinterpret);
+        bool useTempBuffer = (srcMTLPixFmt != dstMTLPixFmt) && (isEitherCompressed || isOneCombinedDepthStencilAspect || !canReinterpret);
 
         if (useTempBuffer) {
             // Add copy from source image to temp buffer.
@@ -587,11 +606,23 @@ static inline MTLSize mvkClampMTLSize(MTLSize size, MTLOrigin origin, MTLSize ma
 
             bool isLayeredBlit = blitKey.dstSampleCount > 1 ? mtlFeats.multisampleLayeredRendering : mtlFeats.layeredRendering;
 
-            uint32_t layCnt = mvkIBR.region.srcSubresource.layerCount == VK_REMAINING_ARRAY_LAYERS ?
+            uint32_t srcLayCnt = mvkIBR.region.srcSubresource.layerCount == VK_REMAINING_ARRAY_LAYERS ?
                 _srcImage->getLayerCount() - mvkIBR.region.srcSubresource.baseArrayLayer :
                  mvkIBR.region.srcSubresource.layerCount;
+            uint32_t dstLayCnt = mvkIBR.region.dstSubresource.layerCount == VK_REMAINING_ARRAY_LAYERS ?
+                _dstImage->getLayerCount() - mvkIBR.region.dstSubresource.baseArrayLayer :
+                 mvkIBR.region.dstSubresource.layerCount;
+            uint32_t layCnt;
+            // If either image is 3D, the difference in z offset must:
+            // - Equal the difference in z offset of the other subresource, if it is also 3D.
+            // - Equal the number of layers in the other subresource, if it is not 3D.
+            // Otherwise, the number of layers should be the same.
             if (_dstImage->getMTLTextureType() == MTLTextureType3D) {
                 layCnt = mvkAbsDiff(mvkIBR.region.dstOffsets[1].z, mvkIBR.region.dstOffsets[0].z);
+            } else if (blitKey.srcMTLTextureType == MTLTextureType3D) {
+                layCnt = mvkAbsDiff(mvkIBR.region.srcOffsets[1].z, mvkIBR.region.srcOffsets[0].z);
+            } else {
+                layCnt = srcLayCnt;
             }
             if (isLayeredBlit) {
                 // In this case, I can blit all layers at once with a layered draw.
@@ -629,14 +660,19 @@ static inline MTLSize mvkClampMTLSize(MTLSize size, MTLOrigin origin, MTLSize ma
                     // In this case, I need to interpolate along the third dimension manually.
                     VkExtent3D srcExtent = _srcImage->getExtent3D(srcPlaneIndex, mvkIBR.region.dstSubresource.mipLevel);
                     VkOffset3D so0 = mvkIBR.region.srcOffsets[0], so1 = mvkIBR.region.srcOffsets[1];
-                    VkOffset3D do0 = mvkIBR.region.dstOffsets[0], do1 = mvkIBR.region.dstOffsets[1];
+                    // If the dst is also 3D use the z offsets, otherwise use the layers.
+                    float do0z = mvkIBR.region.dstOffsets[0].z, do1z = mvkIBR.region.dstOffsets[1].z;
+                    if (_dstImage->getMTLTextureType() != MTLTextureType3D) {
+                        do0z = mvkIBR.region.dstSubresource.baseArrayLayer;
+                        do1z = do0z + dstLayCnt;
+                    }
                     float startZ = (float)so0.z / (float)srcExtent.depth;
                     float endZ = (float)so1.z / (float)srcExtent.depth;
-                    if (isLayeredBlit && do0.z > do1.z) {
+                    if (isLayeredBlit && do0z > do1z) {
                         // Swap start and end points so interpolation moves in the right direction.
                         std::swap(startZ, endZ);
                     }
-                    zIncr = (endZ - startZ) / mvkAbsDiff(do1.z, do0.z);
+                    zIncr = (endZ - startZ) / mvkAbsDiff(do1z, do0z);
                     float z = startZ + (isLayeredBlit ? 0.0 : (layIdx + 0.5)) * zIncr;
                     for (uint32_t i = 0; i < kMVKBlitVertexCount; ++i) {
                         mvkIBR.vertices[i].texCoord.z = z;
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm b/MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm
@@ -515,6 +515,11 @@
 				maintenance7Features->maintenance7 = true;
 				break;
 			}
+			case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_8_FEATURES_KHR: {
+				auto* maintenance8Features = (VkPhysicalDeviceMaintenance8FeaturesKHR*)next;
+				maintenance8Features->maintenance8 = true;
+				break;
+			}
 			case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PORTABILITY_SUBSET_FEATURES_KHR: {
 				auto* portabilityFeatures = (VkPhysicalDevicePortabilitySubsetFeaturesKHR*)next;
 				portabilityFeatures->constantAlphaColorBlendFactors = true;
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKDeviceFeatureStructs.def b/MoltenVK/MoltenVK/GPUObjects/MVKDeviceFeatureStructs.def
@@ -73,6 +73,7 @@ MVK_DEVICE_FEATURE_EXTN(FragmentShaderBarycentric,        FRAGMENT_SHADER_BARYCE
 MVK_DEVICE_FEATURE_EXTN(Maintenance5,                     MAINTENANCE_5,                       KHR,   1)
 MVK_DEVICE_FEATURE_EXTN(Maintenance6,                     MAINTENANCE_6,                       KHR,   1)
 MVK_DEVICE_FEATURE_EXTN(Maintenance7,                     MAINTENANCE_7,                       KHR,   1)
+MVK_DEVICE_FEATURE_EXTN(Maintenance8,                     MAINTENANCE_8,                       KHR,   1)
 MVK_DEVICE_FEATURE_EXTN(PortabilitySubset,                PORTABILITY_SUBSET,                  KHR,  15)
 MVK_DEVICE_FEATURE_EXTN(PresentId,                        PRESENT_ID,                          KHR,   1)
 MVK_DEVICE_FEATURE_EXTN(PresentWait,                      PRESENT_WAIT,                        KHR,   1)
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKPipeline.h b/MoltenVK/MoltenVK/GPUObjects/MVKPipeline.h
@@ -576,6 +576,7 @@ class MVKPipelineCache : public MVKVulkanAPIDeviceObject {
 	size_t _dataSize = 0;
 	std::mutex _shaderCacheLock;
 	bool _isExternallySynchronized = false;
+	bool _isMergeInternallySynchronized = false;
 };
 
 
diff --git a/MoltenVK/MoltenVK/GPUObjects/MVKPipeline.mm b/MoltenVK/MoltenVK/GPUObjects/MVKPipeline.mm
@@ -2645,7 +2645,7 @@ static MTLVertexFormat mvkAdjustFormatVectorToSize(MTLVertexFormat format, uint3
 }
 
 VkResult MVKPipelineCache::mergePipelineCaches(uint32_t srcCacheCount, const VkPipelineCache* pSrcCaches) {
-	if (_isExternallySynchronized) {
+	if (!_isMergeInternallySynchronized) {
 		return mergePipelineCachesImpl(srcCacheCount, pSrcCaches);
 	} else {
 		lock_guard<mutex> lock(_shaderCacheLock);
@@ -2894,7 +2894,9 @@ void serialize(Archive & archive, MVKCompressor<C>& comp) {
 MVKPipelineCache::MVKPipelineCache(MVKDevice* device, const VkPipelineCacheCreateInfo* pCreateInfo) :
 	MVKVulkanAPIDeviceObject(device),
 	_isExternallySynchronized(getEnabledPipelineCreationCacheControlFeatures().pipelineCreationCacheControl &&
-							  mvkIsAnyFlagEnabled(pCreateInfo->flags, VK_PIPELINE_CACHE_CREATE_EXTERNALLY_SYNCHRONIZED_BIT)) {
+							  mvkIsAnyFlagEnabled(pCreateInfo->flags, VK_PIPELINE_CACHE_CREATE_EXTERNALLY_SYNCHRONIZED_BIT)),
+	_isMergeInternallySynchronized(getEnabledPipelineCreationCacheControlFeatures().pipelineCreationCacheControl &&
+								   mvkIsAnyFlagEnabled(pCreateInfo->flags, VK_PIPELINE_CACHE_CREATE_INTERNALLY_SYNCHRONIZED_MERGE_BIT_KHR)) {
 
 	readData(pCreateInfo);
 }
diff --git a/MoltenVK/MoltenVK/Layers/MVKExtensions.def b/MoltenVK/MoltenVK/Layers/MVKExtensions.def
@@ -81,6 +81,7 @@ MVK_EXTENSION(KHR_maintenance4,                        KHR_MAINTENANCE_4,
 MVK_EXTENSION(KHR_maintenance5,                        KHR_MAINTENANCE_5,                       DEVICE,   10.11,  8.0,  1.0)
 MVK_EXTENSION(KHR_maintenance6,                        KHR_MAINTENANCE_6,                       DEVICE,   10.11,  8.0,  1.0)
 MVK_EXTENSION(KHR_maintenance7,                        KHR_MAINTENANCE_7,                       DEVICE,   10.11,  8.0,  1.0)
+MVK_EXTENSION(KHR_maintenance8,                        KHR_MAINTENANCE_8,                       DEVICE,   10.11,  8.0,  1.0)
 MVK_EXTENSION(KHR_map_memory2,                         KHR_MAP_MEMORY_2,                        DEVICE,   10.11,  8.0,  1.0)
 MVK_EXTENSION(KHR_multiview,                           KHR_MULTIVIEW,                           DEVICE,   10.11,  8.0,  1.0)
 MVK_EXTENSION(KHR_portability_subset,                  KHR_PORTABILITY_SUBSET,                  DEVICE,   10.11,  8.0,  1.0)

Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-e07cec7482ed583f58625d3f3c34874e53a003be`
	`1`	`+f1cb5b6634c58be69c327450d0023fb061f0f650`