Merge pull request #736 from billhollings/master

Updates to support Vulkan SDK 1.1.121.

Docs/Whats_New.md

@@ -16,7 +16,7 @@ For best results, use a Markdown reader.*
 MoltenVK 1.0.37
 ---------------
 
-Released TBD
+Released 2019/09/10
 
 - Add support for extensions:
 	- `VK_KHR_device_group`
@@ -25,17 +25,22 @@ Released TBD
 - `vkInvalidateMappedMemoryRanges()` synchronizes managed device memory to CPU.
 - Track supported instance and device extensions correctly.
 - Revert to supporting host-coherent memory for linear images on macOS.
+- Report limit of 64KB for constant buffer bindings on macOS.
 - Disable depth and/or stencil testing if corresponding attachment is missing.
 - Ensure Vulkan loader magic number is set every time before returning any dispatchable Vulkan handle.
 - Fix crash when `VkDeviceCreateInfo` specifies queue families out of numerical order.
 - Fix crash in `vkDestroyPipelineLayout()`.
 - Fix crash when signalling swapchain semaphore using `MTLEvent`.
+- Fix crash when determining alignment of invalid pixel formats.
 - `vkCmdBlitImage():` Support format component swizzling.
 - `vkCmdClearImage():` Set error if attempt made to clear 1D image, and fix validation of depth attachment formats.
 - `vkCreateRenderPass():` Return `VK_ERROR_FORMAT_NOT_SUPPORTED` if format not supported.
 - `vkCmdFillBuffer():` Improve performance 150x by using parallelism more effectively.
 - Support optional use of `MTLFence` for Vulkan semaphores via the `MVK_ALLOW_METAL_FENCES` environment variable.
 - Remove error logging on `VK_TIMEOUT` of `VkSemaphore` and `VkFence`.
+- Remove log message warning of obsolescence of `vkCreateMacOSSurfaceMVK()` and `vkCreateIOSSurfaceMVK()` functions.
+- Report error only on the first time a format substitution is made.
+- Streamline design and use of `MVKSemaphore`.
 - Consolidate the various linkable objects into a `MVKLinkableMixin` template base class.
 - Use `MVKVector` whenever possible in MoltenVK, especially within render loop.
 - No longer prefer dedicated allocations for buffer memory, including buffer-backed images.
@@ -44,26 +49,34 @@ Released TBD
 - Always submit surface presentations using `MTLCommandBuffer`. 
   `MVKConfiguration::presentWithCommandBuffer` is now obsolete.
 - Don't use `MTLCommandBuffer push/popDebugGroup` if not available.
+- `MVKSwapchain::signalWhenAvailable()` add autoreleasepool around `MTLCommandBuffer` use.
 - Add ability to automatically cause an *Xcode* GPU capture without developer intervention.
-- On macOS, limit uniform buffer bindings to 64k.
 - Update `VK_MVK_MOLTENVK_SPEC_VERSION` to version 22.
+- Update dependency libraries to match Vulkan SDK 1.1.121.
+- Update to renaming of `VK_INTEL_shader_integer_functions2` enums and structs in latest Vulkan headers.
 - Update to latest SPIRV-Cross version:
+	- Support the `SPV_EXT_fragment_shader_interlock` extension.
 	- MSL: Deal with array copies from and to threadgroup.
+	- MSL: Deal with `Modf/Frexp` where output is access chain to scalar.
 	- MSL: Inline all emitted functions.
 	- MSL: Inline all non-entry-point functions.
 	- MSL: Add `{Base,}{Vertex,Instance}{,Index}` to `bitcast_from_builtin_load`.
 	- MSL: Add support for sampler Y'CbCr conversion.
 	- MSL: Force storage images on iOS to use discrete descriptors.
 	- MSL: Support dynamic offsets for buffers in argument buffers.
-	- Support the `SPV_EXT_fragment_shader_interlock` extension.
+	- MSL: Cleanup temporary use with `emit_uninitialized_temporary`.
+	- MSL: Unify the `get_*_address_space()` methods.
+	- Assume image and sampler can be `RelaxedPrecision`.
+	- Fix post-depth coverage for ESSL.
 	- Fix variable scope when switch block exits multiple times.
+	- Fix severe performance issue with invariant expression invalidation.
+	- Fix `ParsedIR::mark_used_as_array_length(uint32_t id)`
 	- Deal correctly with sign on bitfield operations.
 	- Elide branches to continue block when continue block is also a merge.
 	- Move branchless analysis to CFG.
 	- Deal with `ldexp` taking `uint` input.
 	- Do not allow base expressions for non-native row-major matrices.
-	- GLSL: Assume image and sampler can be `RelaxedPrecision`.
-	- GLSL: Fix post-depth coverage for ESSL.
+	- Do not force temporary unless continue-only for loop dominates.
 	- Fix `ParsedIR::mark_used_as_array_length(uint32_t id)`.
 	- Refactor into stronger types in public API.
 

ExternalRevisions/Vulkan-Headers_repo_revision

@@ -1 +1 @@
-e3f96a9ccab9397481eb81c4d9bce4ea7590dc33
+5671d014bc1792a116d374792f8a6cfba2e6f512

ExternalRevisions/Vulkan-Tools_repo_revision

@@ -1 +1 @@
-fb1fd1d63eb20cc6f922228f04d8b96cec452b46
+bcabc204a27a93d6b686920a6d46c17f9aaa3ce0

ExternalRevisions/VulkanSamples_repo_revision

@@ -1 +1 @@
-5a52ebd648b5baf94f2346819b5cf3fdc14b4127
+2403b94ed6463062acff8cfff563ad7ec5cf2b58

ExternalRevisions/glslang_repo_revision

@@ -1 +1 @@
-333d1c95792692205472c457d7bec915a94c8000
+34953810a62c5230cc059cfdf42dc44c5aab2246

MoltenVK/MoltenVK/Commands/MVKCmdTransfer.mm

@@ -1177,8 +1177,10 @@
 	NSUInteger dstMTLBuffOffset = _dstBuffer->getMTLBufferOffset() + _dstOffset;
 
 	// Determine the number of full threadgroups we can dispatch to cover the buffer content efficiently.
+	// Some GPU's report different values for max threadgroup width between the pipeline state and device,
+	// so conservatively use the minimum of these two reported values.
 	id<MTLComputePipelineState> cps = getCommandEncodingPool()->getCmdFillBufferMTLComputePipelineState();
-	NSUInteger tgWidth = cps.maxTotalThreadsPerThreadgroup;
+	NSUInteger tgWidth = std::min(cps.maxTotalThreadsPerThreadgroup, getMTLDevice().maxThreadsPerThreadgroup.width);
 	NSUInteger tgCount = _wordCount / tgWidth;
 
 	id<MTLComputeCommandEncoder> mtlComputeEnc = cmdEncoder->getMTLComputeEncoder(kMVKCommandUseFillBuffer);

MoltenVK/MoltenVK/GPUObjects/MVKDevice.mm

@@ -139,8 +139,8 @@
                     portabilityFeatures->samplerMipLodBias = false;
                     break;
                 }
-                case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_INTEGER_FUNCTIONS2_FEATURES_INTEL: {
-                    auto* shaderIntFuncsFeatures = (VkPhysicalDeviceShaderIntegerFunctions2INTEL*)next;
+                case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_INTEGER_FUNCTIONS_2_FEATURES_INTEL: {
+                    auto* shaderIntFuncsFeatures = (VkPhysicalDeviceShaderIntegerFunctions2FeaturesINTEL*)next;
                     shaderIntFuncsFeatures->shaderIntegerFunctions2 = true;
                     break;
                 }
@@ -1164,22 +1164,20 @@
     _properties.limits.nonCoherentAtomSize = _metalFeatures.mtlBufferAlignment;
 
     if ([_mtlDevice respondsToSelector: @selector(minimumLinearTextureAlignmentForPixelFormat:)]) {
-        // Figure out the greatest alignment required by all supported formats, and
-        // whether or not they only require alignment to a single texel. We'll use this
-        // information to fill out the VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT
-        // struct.
+        // Figure out the greatest alignment required by all supported formats, and whether
+		// or not they only require alignment to a single texel. We'll use this information
+		// to fill out the VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT struct.
         uint32_t maxStorage = 0, maxUniform = 0;
         bool singleTexelStorage = true, singleTexelUniform = true;
         mvkEnumerateSupportedFormats({0, 0, VK_FORMAT_FEATURE_UNIFORM_TEXEL_BUFFER_BIT | VK_FORMAT_FEATURE_STORAGE_TEXEL_BUFFER_BIT}, true, [&](VkFormat vk) {
-            // MTLDevice minimumLinearTextureAlignmentForPixelFormat with MTLPixelFormatInvalid will cause as a failed assertion on iOS 13.x and iPad OS 13.x
-            if ( vk == VK_FORMAT_UNDEFINED )
-              return false;
-
+			MTLPixelFormat mtlFmt = mvkMTLPixelFormatFromVkFormat(vk);
+			if ( !mtlFmt ) { return false; }	// If format is invalid, avoid validation errors on MTLDevice format alignment calls
+
             NSUInteger alignment;
             if ([_mtlDevice respondsToSelector: @selector(minimumTextureBufferAlignmentForPixelFormat:)]) {
-                alignment = [_mtlDevice minimumTextureBufferAlignmentForPixelFormat: mvkMTLPixelFormatFromVkFormat(vk)];
+                alignment = [_mtlDevice minimumTextureBufferAlignmentForPixelFormat: mtlFmt];
             } else {
-                alignment = [_mtlDevice minimumLinearTextureAlignmentForPixelFormat: mvkMTLPixelFormatFromVkFormat(vk)];
+                alignment = [_mtlDevice minimumLinearTextureAlignmentForPixelFormat: mtlFmt];
             }
             VkFormatProperties props = mvkVkFormatProperties(vk, getFormatIsSupported(vk));
             // For uncompressed formats, this is the size of a single texel.

MoltenVK/MoltenVK/GPUObjects/MVKSurface.mm

@@ -43,7 +43,7 @@
 					   const Vk_PLATFORM_SurfaceCreateInfoMVK* pCreateInfo,
 					   const VkAllocationCallbacks* pAllocator) : _mvkInstance(mvkInstance) {
 
-	MVKLogInfo("%s(): This function is obsolete. Consider using the vkCreateMetalSurfaceEXT() function from the VK_EXT_metal_surface extension instead.", STR(vkCreate_PLATFORM_SurfaceMVK));
+//	MVKLogInfo("%s(): This function is obsolete. Consider using the vkCreateMetalSurfaceEXT() function from the VK_EXT_metal_surface extension instead.", STR(vkCreate_PLATFORM_SurfaceMVK));
 
 	// Get the platform object contained in pView
 	id<NSObject> obj = (id<NSObject>)pCreateInfo->pView;

MoltenVK/MoltenVK/Layers/MVKExtensions.def

@@ -87,7 +87,7 @@ MVK_EXTENSION(AMD_negative_viewport_height, AMD_NEGATIVE_VIEWPORT_HEIGHT, MVK_EX
 MVK_EXTENSION(AMD_shader_image_load_store_lod, AMD_SHADER_IMAGE_LOAD_STORE_LOD, MVK_EXTENSION_DEVICE)
 MVK_EXTENSION(AMD_shader_trinary_minmax, AMD_SHADER_TRINARY_MINMAX, MVK_EXTENSION_DEVICE)
 MVK_EXTENSION(IMG_format_pvrtc, IMG_FORMAT_PVRTC, MVK_EXTENSION_DEVICE)
-MVK_EXTENSION(INTEL_shader_integer_functions2, INTEL_SHADER_INTEGER_FUNCTIONS2, MVK_EXTENSION_DEVICE)
+MVK_EXTENSION(INTEL_shader_integer_functions2, INTEL_SHADER_INTEGER_FUNCTIONS_2, MVK_EXTENSION_DEVICE)
 MVK_EXTENSION_LAST(NV_glsl_shader, NV_GLSL_SHADER, MVK_EXTENSION_DEVICE)
 
 #undef MVK_EXTENSION

MoltenVK/MoltenVK/Vulkan/mvk_datatypes.mm

@@ -94,7 +94,7 @@
 #define MVK_FMT_NO_FEATS			0
 
 #define MVK_MAKE_FMT_STRUCT(VK_FMT, MTL_FMT, MTL_FMT_ALT, IOS_SINCE, MACOS_SINCE, BLK_W, BLK_H, BLK_BYTE_CNT, MTL_VTX_FMT, MTL_VTX_FMT_ALT, VTX_IOS_SINCE, VTX_MACOS_SINCE, CLR_TYPE, PIXEL_FEATS, BUFFER_FEATS)  \
-        { VK_FMT, MTL_FMT, MTL_FMT_ALT, IOS_SINCE, MACOS_SINCE, { BLK_W, BLK_H }, BLK_BYTE_CNT, MTL_VTX_FMT, MTL_VTX_FMT_ALT, VTX_IOS_SINCE, VTX_MACOS_SINCE, CLR_TYPE, { (PIXEL_FEATS & MVK_FMT_LINEAR_TILING_FEATS), PIXEL_FEATS, BUFFER_FEATS }, #VK_FMT, #MTL_FMT }
+        { VK_FMT, MTL_FMT, MTL_FMT_ALT, IOS_SINCE, MACOS_SINCE, { BLK_W, BLK_H }, BLK_BYTE_CNT, MTL_VTX_FMT, MTL_VTX_FMT_ALT, VTX_IOS_SINCE, VTX_MACOS_SINCE, CLR_TYPE, { (PIXEL_FEATS & MVK_FMT_LINEAR_TILING_FEATS), PIXEL_FEATS, BUFFER_FEATS }, #VK_FMT, #MTL_FMT, false }
 
 #pragma mark Texture formats
 
@@ -117,6 +117,7 @@
 	VkFormatProperties properties;
     const char* vkName;
     const char* mtlName;
+	bool hasReportedSubstitution;
 
     inline double bytesPerTexel() const { return (double)bytesPerBlock / (double)(blockTexelSize.width * blockTexelSize.height); };
 
@@ -569,31 +570,35 @@ MVK_PUBLIC_SYMBOL MTLPixelFormat mvkMTLPixelFormatFromVkFormat(VkFormat vkFormat
 }
 
 MTLPixelFormat mvkMTLPixelFormatFromVkFormatInObj(VkFormat vkFormat, MVKBaseObject* mvkObj) {
-    MTLPixelFormat mtlPixFmt = MTLPixelFormatInvalid;
+	MTLPixelFormat mtlPixFmt = MTLPixelFormatInvalid;
 
-    const MVKFormatDesc& fmtDesc = formatDescForVkFormat(vkFormat);
-    if (fmtDesc.isSupported()) {
-        mtlPixFmt = fmtDesc.mtl;
-    } else if (vkFormat != VK_FORMAT_UNDEFINED) {
-        // If the MTLPixelFormat is not supported but VkFormat is valid,
-        // report an error, and possibly substitute a different MTLPixelFormat.
-        string errMsg;
-        errMsg += "VkFormat ";
-        errMsg += (fmtDesc.vkName) ? fmtDesc.vkName : to_string(fmtDesc.vk);
-        errMsg += " is not supported on this device.";
-
-        if (fmtDesc.isSupportedOrSubstitutable()) {
-            mtlPixFmt = fmtDesc.mtlSubstitute;
-
-            const MVKFormatDesc& fmtDescSubs = formatDescForMTLPixelFormat(mtlPixFmt);
-            errMsg += " Using VkFormat ";
-            errMsg += (fmtDescSubs.vkName) ? fmtDescSubs.vkName : to_string(fmtDescSubs.vk);
-            errMsg += " instead.";
-        }
-		MVKBaseObject::reportError(mvkObj, VK_ERROR_FORMAT_NOT_SUPPORTED, "%s", errMsg.c_str());
-    }
+	const MVKFormatDesc& fmtDesc = formatDescForVkFormat(vkFormat);
+	if (fmtDesc.isSupported()) {
+		mtlPixFmt = fmtDesc.mtl;
+	} else if (vkFormat != VK_FORMAT_UNDEFINED) {
+		// If the MTLPixelFormat is not supported but VkFormat is valid, attempt to substitute a different format.
+		mtlPixFmt = fmtDesc.mtlSubstitute;
+
+		// Report an error if there is no substitute, or the first time a substitution is made.
+		if ( !mtlPixFmt || !fmtDesc.hasReportedSubstitution ) {
+			string errMsg;
+			errMsg += "VkFormat ";
+			errMsg += (fmtDesc.vkName) ? fmtDesc.vkName : to_string(fmtDesc.vk);
+			errMsg += " is not supported on this device.";
+
+			if (mtlPixFmt) {
+				((MVKFormatDesc*)&fmtDesc)->hasReportedSubstitution = true;
+
+				const MVKFormatDesc& fmtDescSubs = formatDescForMTLPixelFormat(mtlPixFmt);
+				errMsg += " Using VkFormat ";
+				errMsg += (fmtDescSubs.vkName) ? fmtDescSubs.vkName : to_string(fmtDescSubs.vk);
+				errMsg += " instead.";
+			}
+			MVKBaseObject::reportError(mvkObj, VK_ERROR_FORMAT_NOT_SUPPORTED, "%s", errMsg.c_str());
+		}
+	}
 
-    return mtlPixFmt;
+	return mtlPixFmt;
 }
 
 MVK_PUBLIC_SYMBOL VkFormat mvkVkFormatFromMTLPixelFormat(MTLPixelFormat mtlFormat) {