| diff --git a/src/common/utilities.cpp b/src/common/utilities.cpp |
| index 274d8029521aa8e6d6fdae039a21931f00125fe6..5314733998768cdbda34126ed61d4247ef952541 100644 |
| --- a/src/common/utilities.cpp |
| +++ b/src/common/utilities.cpp |
| @@ -6,6 +6,12 @@ |
| |
| // utilities.cpp: Conversion functions and other utility routines. |
| |
| +// Older clang versions have a false positive on this warning here. |
| +// TODO(dino): Is this still necessary? |
| +#if defined(__clang__) |
| +# pragma clang diagnostic ignored "-Wglobal-constructors" |
| +#endif |
| + |
| #include "common/utilities.h" |
| #include "GLES3/gl3.h" |
| #include "common/mathutil.h" |
| diff --git a/src/compiler/preprocessor/preprocessor_tab_autogen.cpp b/src/compiler/preprocessor/preprocessor_tab_autogen.cpp |
| index 50822f2007292801b41be16ad956ae0bf06e7e88..3ece2105bcdaea98917001b33d1c6834de022d5e 100644 |
| --- a/src/compiler/preprocessor/preprocessor_tab_autogen.cpp |
| +++ b/src/compiler/preprocessor/preprocessor_tab_autogen.cpp |
| @@ -1,5 +1,8 @@ |
| /* A Bison parser, made by GNU Bison 3.3.2. */ |
| |
| +/* Apple Note: For the avoidance of doubt, Apple elects to distribute this file under the terms of |
| + * the BSD license. */ |
| + |
| /* Bison implementation for Yacc-like parsers in C |
| |
| Copyright (C) 1984, 1989-1990, 2000-2015, 2018-2019 Free Software Foundation, |
| diff --git a/src/compiler/translator/CodeGen.cpp b/src/compiler/translator/CodeGen.cpp |
| index 497ef881e2d9cf0efb669ca8c7c4f2669633baba..47213c90497c8c9d8f76baf0653fdbc3c7a2e19d 100644 |
| --- a/src/compiler/translator/CodeGen.cpp |
| +++ b/src/compiler/translator/CodeGen.cpp |
| @@ -23,6 +23,9 @@ |
| #ifdef ANGLE_ENABLE_METAL |
| # include "compiler/translator/TranslatorMetalDirect.h" |
| #endif // ANGLE_ENABLE_METAL |
| +#ifdef ANGLE_ENABLE_METAL_SPIRV |
| +# include "compiler/translator/TranslatorMetal.h" |
| +#endif // ANGLE_ENABLE_METAL_SPIRV |
| |
| #ifdef ANGLE_ENABLE_METAL_SPIRV |
| # include "compiler/translator/TranslatorMetal.h" |
| diff --git a/src/compiler/translator/IntermNode.cpp b/src/compiler/translator/IntermNode.cpp |
| index 11cb0f04409902a33f2c3677ac280d3d9b3f373c..791dd6dc4e3ea5bbb6289245ded8acffe27568c7 100644 |
| --- a/src/compiler/translator/IntermNode.cpp |
| +++ b/src/compiler/translator/IntermNode.cpp |
| @@ -279,7 +279,7 @@ bool TIntermLoop::replaceChildNode(TIntermNode *original, TIntermNode *replaceme |
| } |
| |
| TIntermBranch::TIntermBranch(const TIntermBranch &node) |
| - : TIntermBranch(node.mFlowOp, node.mExpression->deepCopy()) |
| + : TIntermBranch(node.mFlowOp, node.mExpression ? node.mExpression->deepCopy() : nullptr) |
| {} |
| |
| size_t TIntermBranch::getChildCount() const |
| @@ -1602,10 +1602,10 @@ TIntermLoop::TIntermLoop(TLoopType type, |
| |
| TIntermLoop::TIntermLoop(const TIntermLoop &node) |
| : TIntermLoop(node.mType, |
| - node.mInit->deepCopy(), |
| - node.mCond->deepCopy(), |
| - node.mExpr->deepCopy(), |
| - node.mBody->deepCopy()) |
| + node.mInit ? node.mInit->deepCopy() : nullptr, |
| + node.mCond ? node.mCond->deepCopy() : nullptr, |
| + node.mExpr ? node.mExpr->deepCopy() : nullptr, |
| + node.mBody ? node.mBody->deepCopy() : nullptr) |
| {} |
| |
| TIntermIfElse::TIntermIfElse(TIntermTyped *cond, TIntermBlock *trueB, TIntermBlock *falseB) |
| diff --git a/src/compiler/translator/TranslatorMetalDirect/IdGen.cpp b/src/compiler/translator/TranslatorMetalDirect/IdGen.cpp |
| index 81ff2722f4983df6bc508602ea4fe06d6da74dd2..9d2b43bc0f987e8b9cb37b2119778e8b5ceb886c 100644 |
| --- a/src/compiler/translator/TranslatorMetalDirect/IdGen.cpp |
| +++ b/src/compiler/translator/TranslatorMetalDirect/IdGen.cpp |
| @@ -25,7 +25,7 @@ Name IdGen::createNewName(size_t count, |
| { |
| const unsigned id = mNext++; |
| char idBuffer[std::numeric_limits<unsigned>::digits10 + 1]; |
| - sprintf(idBuffer, "%u", id); |
| + snprintf(idBuffer, sizeof(idBuffer), "%u", id); |
| |
| mNewNameBuffer.clear(); |
| mNewNameBuffer += '_'; |
| diff --git a/src/compiler/translator/TranslatorMetalDirect/ModifyStruct.cpp b/src/compiler/translator/TranslatorMetalDirect/ModifyStruct.cpp |
| index bcc5d65897a4ab90b2b64c5a5108b7e0d25a3e04..60758fbe8f45b609799af915943e166d2e697e8c 100644 |
| --- a/src/compiler/translator/TranslatorMetalDirect/ModifyStruct.cpp |
| +++ b/src/compiler/translator/TranslatorMetalDirect/ModifyStruct.cpp |
| @@ -601,7 +601,7 @@ class ConvertStructState : angle::NonCopyable |
| void pushNamePath(unsigned extra) |
| { |
| char buffer[std::numeric_limits<unsigned>::digits10 + 1]; |
| - sprintf(buffer, "%u", extra); |
| + snprintf(buffer, sizeof(buffer), "%u", extra); |
| pushNamePath(buffer); |
| } |
| |
| diff --git a/src/compiler/translator/glslang_tab_autogen.cpp b/src/compiler/translator/glslang_tab_autogen.cpp |
| index daa1c614f71f0788bc8b1ea1d0e4d8709f6c7b0a..e3a3ddba082f4629478be45068cc7c560276e7e6 100644 |
| --- a/src/compiler/translator/glslang_tab_autogen.cpp |
| +++ b/src/compiler/translator/glslang_tab_autogen.cpp |
| @@ -1,5 +1,8 @@ |
| /* A Bison parser, made by GNU Bison 3.3.2. */ |
| |
| +/* Apple Note: For the avoidance of doubt, Apple elects to distribute this file under the terms of |
| + * the BSD license. */ |
| + |
| /* Bison implementation for Yacc-like parsers in C |
| |
| Copyright (C) 1984, 1989-1990, 2000-2015, 2018-2019 Free Software Foundation, |
| diff --git a/src/compiler/translator/glslang_tab_autogen.h b/src/compiler/translator/glslang_tab_autogen.h |
| index d2eab0b7bc55675eb9ad5e210fb4d0cb869f35d1..2ec2d2c4495c1ed97a55173e49ea329d82d3f155 100644 |
| --- a/src/compiler/translator/glslang_tab_autogen.h |
| +++ b/src/compiler/translator/glslang_tab_autogen.h |
| @@ -1,5 +1,8 @@ |
| /* A Bison parser, made by GNU Bison 3.3.2. */ |
| |
| +/* Apple Note: For the avoidance of doubt, Apple elects to distribute this file under the terms of |
| + * the BSD license. */ |
| + |
| /* Bison interface for Yacc-like parsers in C |
| |
| Copyright (C) 1984, 1989-1990, 2000-2015, 2018-2019 Free Software Foundation, |
| diff --git a/src/compiler/translator/glslang_wrapper.cpp b/src/compiler/translator/glslang_wrapper.cpp |
| index 187f350014fd283b3f406f5c3144a753f02436cb..3de7597d17b8813b60a62ffb14dd29e71491f8af 100644 |
| --- a/src/compiler/translator/glslang_wrapper.cpp |
| +++ b/src/compiler/translator/glslang_wrapper.cpp |
| @@ -10,6 +10,7 @@ |
| // This file is separated as glslang's header contains conflicting macro definitions with ANGLE's. |
| // |
| |
| +#include "common/spirv/spirv_types.h" |
| #include "compiler/translator/glslang_wrapper.h" |
| |
| // glslang has issues with some specific warnings. |
| diff --git a/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.cpp b/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.cpp |
| index 55e2f8b9717ec73d2d52607d400352f5f4c0c191..4237558526274ea3e643d45ccf574b6b10eac785 100644 |
| --- a/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.cpp |
| +++ b/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.cpp |
| @@ -10,6 +10,7 @@ |
| #include "compiler/translator/tree_util/IntermNode_util.h" |
| #include "compiler/translator/tree_util/IntermTraverse.h" |
| |
| +#if defined(ANGLE_ENABLE_GLSL) && defined(ANGLE_ENABLE_APPLE_WORKAROUNDS) |
| namespace sh |
| { |
| |
| @@ -70,5 +71,5 @@ bool UnfoldShortCircuitAST(TCompiler *compiler, TIntermBlock *root) |
| root->traverse(&traverser); |
| return traverser.updateTree(compiler, root); |
| } |
| - |
| +#endif |
| } // namespace sh |
| diff --git a/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.h b/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.h |
| index ad3cc080de119748939e5aa893fe095b9dd4a2f3..75069c57284bb3b0cb796717191c7d55934c11af 100644 |
| --- a/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.h |
| +++ b/src/compiler/translator/tree_ops/apple/UnfoldShortCircuitAST.h |
| @@ -11,7 +11,7 @@ |
| #define COMPILER_TRANSLATOR_TREEOPS_APPLE_UNFOLDSHORTCIRCUITAST_H_ |
| |
| #include "common/angleutils.h" |
| - |
| +#include "common/debug.h" |
| namespace sh |
| { |
| |
| diff --git a/src/libANGLE/State.cpp b/src/libANGLE/State.cpp |
| index 1a5b0d8874f173fed2ee1f27d69b8ea4db6fd919..4459f29401d15fd2b3c553c10c71f1625dc2e2e1 100644 |
| --- a/src/libANGLE/State.cpp |
| +++ b/src/libANGLE/State.cpp |
| @@ -6,6 +6,9 @@ |
| |
| // State.cpp: Implements the State class, encapsulating raw GL state. |
| |
| +// Older clang versions have a false positive on this warning here. |
| +#pragma clang diagnostic ignored "-Wglobal-constructors" |
| + |
| #include "libANGLE/State.h" |
| |
| #include <string.h> |
| diff --git a/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.cpp b/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.cpp |
| index ae040dfc19960b7cc686d0c2136454eb246ca80d..9e69e2fa308675249c48aee6b44b09cd4a5d2dfa 100644 |
| --- a/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.cpp |
| +++ b/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.cpp |
| @@ -114,12 +114,14 @@ const ShaderInterfaceVariableInfo &ShaderInterfaceVariableInfoMap::getVariableBy |
| return mData[shaderType][typeAndIndex.variableType][typeAndIndex.index]; |
| } |
| |
| +#if ANGLE_ENABLE_METAL_SPIRV |
| bool ShaderInterfaceVariableInfoMap::hasTransformFeedbackInfo(gl::ShaderType shaderType, |
| uint32_t bufferIndex) const |
| { |
| std::string bufferName = rx::GetXfbBufferName(bufferIndex); |
| return hasVariable(shaderType, bufferName); |
| } |
| +#endif |
| |
| void ShaderInterfaceVariableInfoMap::mapIndexedResourceByName(gl::ShaderType shaderType, |
| ShaderVariableType variableType, |
| diff --git a/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.h b/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.h |
| index ce9c26e401ffd4fa59eb8308a5210998970bec8c..dcaa142fc2a5e0d45ddf7e8fcaac14bb7f30ff2d 100644 |
| --- a/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.h |
| +++ b/src/libANGLE/renderer/ShaderInterfaceVariableInfoMap.h |
| @@ -89,7 +89,9 @@ class ShaderInterfaceVariableInfoMap final : angle::NonCopyable |
| bool hasAtomicCounterInfo(gl::ShaderType shaderType) const; |
| const ShaderInterfaceVariableInfo &getAtomicCounterInfo(gl::ShaderType shaderType) const; |
| const ShaderInterfaceVariableInfo &getFramebufferFetchInfo(gl::ShaderType shaderType) const; |
| +#if ANGLE_ENABLE_METAL_SPIRV |
| bool hasTransformFeedbackInfo(gl::ShaderType shaderType, uint32_t bufferIndex) const; |
| +#endif |
| const ShaderInterfaceVariableInfo &getTransformFeedbackInfo(gl::ShaderType shaderType, |
| uint32_t bufferIndex) const; |
| |
| diff --git a/src/libANGLE/renderer/driver_utils.cpp b/src/libANGLE/renderer/driver_utils.cpp |
| index e1555cff0ce083f309e845eab2678be57839122e..ad08da0ed2befa72cf2a5291a6d91c62538939b4 100644 |
| --- a/src/libANGLE/renderer/driver_utils.cpp |
| +++ b/src/libANGLE/renderer/driver_utils.cpp |
| @@ -268,6 +268,10 @@ const char *GetVendorString(uint32_t vendorId) |
| return "Qualcomm"; |
| case VENDOR_ID_SAMSUNG: |
| return "Samsung Electronics Co., Ltd."; |
| + case VENDOR_ID_VMWARE: |
| + return "VMware"; |
| + case VENDOR_ID_VIVANTE: |
| + return "Vivante"; |
| case 0xba5eba11: // Mock vendor ID used for tests. |
| return "Test"; |
| case 0: |
| diff --git a/src/libANGLE/renderer/glslang_wrapper_utils.h b/src/libANGLE/renderer/glslang_wrapper_utils.h |
| index 6eb21d9e204cd958448603938053f9c72bb91846..e21456e19ce5112a5535080193979d0858b7d58b 100644 |
| --- a/src/libANGLE/renderer/glslang_wrapper_utils.h |
| +++ b/src/libANGLE/renderer/glslang_wrapper_utils.h |
| @@ -157,7 +157,7 @@ void GlslangAssignTransformFeedbackLocations(gl::ShaderType shaderType, |
| bool isTransformFeedbackStage, |
| GlslangProgramInterfaceInfo *programInterfaceInfo, |
| ShaderInterfaceVariableInfoMap *variableInfoMapOut); |
| - |
| +#if ANGLE_ENABLE_METAL_SPIRV |
| // Retrieves the compiled SPIR-V code for each shader stage, and calls |GlslangAssignLocations|. |
| void GlslangGetShaderSpirvCode(const GlslangSourceOptions &options, |
| const gl::ProgramState &programState, |
| @@ -170,6 +170,7 @@ angle::Result GlslangTransformSpirvCode(const GlslangSpirvOptions &options, |
| const ShaderInterfaceVariableInfoMap &variableInfoMap, |
| const angle::spirv::Blob &initialSpirvBlob, |
| angle::spirv::Blob *spirvBlobOut); |
| +#endif |
| |
| } // namespace rx |
| |
| diff --git a/src/libANGLE/renderer/metal/BufferMtl.mm b/src/libANGLE/renderer/metal/BufferMtl.mm |
| index c8769efa38736be06a1c7fc12a4f21ef578c5f01..1766256930914b030119724f01d2adb2f54269a2 100644 |
| --- a/src/libANGLE/renderer/metal/BufferMtl.mm |
| +++ b/src/libANGLE/renderer/metal/BufferMtl.mm |
| @@ -365,7 +365,7 @@ |
| { |
| if (buffer.offset == offset) |
| { |
| - return &buffer; |
| + return static_cast<ConversionBufferMtl *>(&buffer); |
| } |
| } |
| |
| @@ -520,7 +520,7 @@ |
| default: |
| // dynamic buffer, allow up to 10 update per frame/encoding without |
| // waiting for GPU. |
| - if (adjustedSize <= mtl::kSharedMemBufferMaxBufSizeHint) |
| + if (adjustedSize <= kConvertedElementArrayBufferInitialSize) |
| { |
| maxBuffers = 10; |
| mBufferPool.setAlwaysUseSharedMem(); |
| diff --git a/src/libANGLE/renderer/metal/ContextMtl.h b/src/libANGLE/renderer/metal/ContextMtl.h |
| index 038515414371ccde7b55e50d41465438dc163b00..67e137dc22a12cf56394162754090f7d102bf48b 100644 |
| --- a/src/libANGLE/renderer/metal/ContextMtl.h |
| +++ b/src/libANGLE/renderer/metal/ContextMtl.h |
| @@ -577,6 +577,8 @@ class ContextMtl : public ContextImpl, public mtl::Context |
| gl::AttributesMask mDirtyDefaultAttribsMask; |
| DirtyBits mDirtyBits; |
| |
| + uint32_t mRenderPassesSinceFlush = 0; |
| + |
| // State |
| mtl::RenderPipelineDesc mRenderPipelineDesc; |
| mtl::DepthStencilDesc mDepthStencilDesc; |
| diff --git a/src/libANGLE/renderer/metal/ContextMtl.mm b/src/libANGLE/renderer/metal/ContextMtl.mm |
| index 89bb8b23a4f712adfa3f9b72967d36d6cfe14527..dc2f5fe8fbd3c8bc2c11f9982f76f7a290be0517 100644 |
| --- a/src/libANGLE/renderer/metal/ContextMtl.mm |
| +++ b/src/libANGLE/renderer/metal/ContextMtl.mm |
| @@ -1693,10 +1693,19 @@ GLint GetOwnershipIdentity(const egl::AttributeMap &attribs) |
| |
| endEncoding(true); |
| mCmdBuffer.commit(operation); |
| + mRenderPassesSinceFlush = 0; |
| } |
| |
| void ContextMtl::flushCommandBufferIfNeeded() |
| { |
| + if (mRenderPassesSinceFlush >= mtl::kMaxRenderPassesPerCommandBuffer) |
| + { |
| + // WaitUntilScheduled here is intended to help the CPU-GPU pipeline and |
| + // helps to keep the number of inflight render passes in the system to a |
| + // minimum. |
| + flushCommandBuffer(mtl::WaitUntilScheduled); |
| + } |
| + |
| if (mCmdBuffer.needsFlushForDrawCallLimits()) |
| { |
| flushCommandBuffer(mtl::NoWait); |
| @@ -1717,6 +1726,7 @@ GLint GetOwnershipIdentity(const egl::AttributeMap &attribs) |
| endEncoding(false); |
| mCmdBuffer.present(presentationDrawable); |
| mCmdBuffer.commit(mtl::NoWait); |
| + mRenderPassesSinceFlush = 0; |
| } |
| |
| angle::Result ContextMtl::finishCommandBuffer() |
| @@ -1752,6 +1762,7 @@ GLint GetOwnershipIdentity(const egl::AttributeMap &attribs) |
| endEncoding(false); |
| |
| ensureCommandBufferReady(); |
| + mRenderPassesSinceFlush++; |
| |
| // Need to re-apply everything on next draw call. |
| mDirtyBits.set(); |
| diff --git a/src/libANGLE/renderer/metal/DisplayMtl.mm b/src/libANGLE/renderer/metal/DisplayMtl.mm |
| index 13aec00fac3e4ea8932a86883dcc6eff7f30f59c..295a3788aa012838660233608df48f10bd269831 100644 |
| --- a/src/libANGLE/renderer/metal/DisplayMtl.mm |
| +++ b/src/libANGLE/renderer/metal/DisplayMtl.mm |
| @@ -1210,8 +1210,7 @@ bool IsMetalDisplayAvailable() |
| bool DisplayMtl::supports32BitFloatFiltering() const |
| { |
| #if (defined(__MAC_11_0) && __MAC_OS_X_VERSION_MIN_REQUIRED >= __MAC_11_0) || \ |
| - (defined(__IPHONE_14_0) && __IPHONE_OS_VERSION_MIN_REQUIRED >= __IPHONE_14_0) || \ |
| - (defined(__TVOS_14_0) && __TV_OS_VERSION_MIN_REQUIRED >= __TVOS_14_0) |
| + (defined(__IPHONE_14_0) && __IPHONE_OS_VERSION_MIN_REQUIRED >= __IPHONE_14_0) |
| if (@available(ios 14.0, macOS 11.0, *)) |
| { |
| return [mMetalDevice supports32BitFloatFiltering]; |
| diff --git a/src/libANGLE/renderer/metal/mtl_buffer_pool.h b/src/libANGLE/renderer/metal/mtl_buffer_pool.h |
| index c3be11f78156c7fab4ee8a90733de2f3b049de7e..204ebb949873112c010242b2b16b0860cd40c00a 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_buffer_pool.h |
| +++ b/src/libANGLE/renderer/metal/mtl_buffer_pool.h |
| @@ -126,7 +126,7 @@ class BufferPool |
| |
| size_t mBuffersAllocated; |
| size_t mMaxBuffers; |
| - BufferPoolMemPolicy mMemPolicy; |
| + BufferPoolMemPolicy mMemPolicy = BufferPoolMemPolicy::Auto; |
| bool mAlwaysAllocateNewBuffer; |
| }; |
| |
| diff --git a/src/libANGLE/renderer/metal/mtl_common.h b/src/libANGLE/renderer/metal/mtl_common.h |
| index 3f3086c36d3ae3886af05a044bbdc834e63d2239..c6618ecc2259f22756b61738d041e679420ce8bf 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_common.h |
| +++ b/src/libANGLE/renderer/metal/mtl_common.h |
| @@ -147,6 +147,11 @@ constexpr uint32_t kMaxViewports = 1; |
| // will be flushed next time |
| constexpr const size_t kMaximumResidentMemorySizeInBytes = 400 * 1024 * 1024; |
| |
| +// Restrict in-flight render passes per command buffer to 16. |
| +// The goal is to reduce the number of active render passes on the system at |
| +// anyone time and this value was determined through experimentation. |
| +constexpr uint32_t kMaxRenderPassesPerCommandBuffer = 16; |
| + |
| constexpr uint32_t kVertexAttribBufferStrideAlignment = 4; |
| // Alignment requirement for offset passed to setVertex|FragmentBuffer |
| #if TARGET_OS_OSX || TARGET_OS_MACCATALYST |
| diff --git a/src/libANGLE/renderer/metal/mtl_render_utils.mm b/src/libANGLE/renderer/metal/mtl_render_utils.mm |
| index 4b854d589c5c7042ba8ab2e32960425c4070eb2a..1e2b6888ef205c355b2220fb36d3b889a614fa25 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_render_utils.mm |
| +++ b/src/libANGLE/renderer/metal/mtl_render_utils.mm |
| @@ -2540,45 +2540,13 @@ ANGLE_INLINE void SetPipelineState(ComputeCommandEncoder *encoder, |
| bool sRGBMipmap, |
| NativeTexLevelArray *mipmapOutputViews) |
| { |
| - // Only support 3D texture for now. |
| - ASSERT(srcTexture->textureType() == MTLTextureType3D); |
| - |
| MTLSize threadGroupSize; |
| uint32_t slices = 1; |
| id<MTLComputePipelineState> computePipeline = nil; |
| - |
| - ensure3DMipGeneratorPipelineInitialized(contextMtl); |
| - computePipeline = m3DMipGeneratorPipeline; |
| - threadGroupSize = |
| - MTLSizeMake(kGenerateMipThreadGroupSizePerDim, kGenerateMipThreadGroupSizePerDim, |
| - kGenerateMipThreadGroupSizePerDim); |
| - |
| // The compute shader supports up to 4 mipmaps generated per pass. |
| // See shaders/gen_mipmap.metal |
| uint32_t maxMipsPerBatch = 4; |
| |
| - if (threadGroupSize.width * threadGroupSize.height * threadGroupSize.depth > |
| - computePipeline.maxTotalThreadsPerThreadgroup || |
| - ANGLE_UNLIKELY( |
| - !contextMtl->getDisplay()->getFeatures().allowGenMultipleMipsPerPass.enabled)) |
| - { |
| - // Multiple mipmaps generation is not supported due to hardware's thread group size limits. |
| - // Fallback to generate one mip per pass and reduce thread group size. |
| - if (ANGLE_UNLIKELY(threadGroupSize.width * threadGroupSize.height > |
| - computePipeline.maxTotalThreadsPerThreadgroup)) |
| - { |
| - // Even with reduced thread group size, we cannot proceed. |
| - // HACK: use blit command encoder to generate mipmaps if it is not possible |
| - // to use compute shader due to hardware limits. |
| - BlitCommandEncoder *blitEncoder = contextMtl->getBlitCommandEncoder(); |
| - blitEncoder->generateMipmapsForTexture(srcTexture); |
| - return angle::Result::Continue; |
| - } |
| - |
| - threadGroupSize.depth = 1; |
| - maxMipsPerBatch = 1; |
| - } |
| - |
| ComputeCommandEncoder *cmdEncoder = contextMtl->getComputeCommandEncoder(); |
| ASSERT(cmdEncoder); |
| |
| @@ -2615,8 +2583,29 @@ ANGLE_INLINE void SetPipelineState(ComputeCommandEncoder *encoder, |
| UNREACHABLE(); |
| } |
| |
| - Generate3DMipmapUniform options; |
| + if (threadGroupSize.width * threadGroupSize.height * threadGroupSize.depth > |
| + computePipeline.maxTotalThreadsPerThreadgroup || |
| + ANGLE_UNLIKELY( |
| + !contextMtl->getDisplay()->getFeatures().allowGenMultipleMipsPerPass.enabled)) |
| + { |
| + // Multiple mipmaps generation is not supported due to hardware's thread group size limits. |
| + // Fallback to generate one mip per pass and reduce thread group size. |
| + if (ANGLE_UNLIKELY(threadGroupSize.width * threadGroupSize.height > |
| + computePipeline.maxTotalThreadsPerThreadgroup)) |
| + { |
| + // Even with reduced thread group size, we cannot proceed. |
| + // HACK: use blit command encoder to generate mipmaps if it is not possible |
| + // to use compute shader due to hardware limits. |
| + BlitCommandEncoder *blitEncoder = contextMtl->getBlitCommandEncoder(); |
| + blitEncoder->generateMipmapsForTexture(srcTexture); |
| + return angle::Result::Continue; |
| + } |
| + |
| + threadGroupSize.depth = 1; |
| + maxMipsPerBatch = 1; |
| + } |
| |
| + Generate3DMipmapUniform options; |
| uint32_t remainMips = srcTexture->mipmapLevels() - 1; |
| MipmapNativeLevel batchSrcLevel = kZeroNativeMipLevel; |
| options.srcLevel = batchSrcLevel.get(); |
| diff --git a/src/libANGLE/renderer/metal/mtl_resource_spi.h b/src/libANGLE/renderer/metal/mtl_resource_spi.h |
| index 3eaa9a64e79af26faa786102ece3f0bdea84d090..2ecf7b40bd5a2edc3fd9d69841e266b6c3cee685 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_resource_spi.h |
| +++ b/src/libANGLE/renderer/metal/mtl_resource_spi.h |
| @@ -1,9 +1,63 @@ |
| -// |
| -// Copyright 2021 The ANGLE Project Authors. All rights reserved. |
| -// Use of this source code is governed by a BSD-style license that can be |
| -// found in the LICENSE file. |
| +/* |
| + * Copyright (C) 2021 Apple Inc. All rights reserved. |
| + * |
| + * Redistribution and use in source and binary forms, with or without |
| + * modification, are permitted provided that the following conditions |
| + * are met: |
| + * 1. Redistributions of source code must retain the above copyright |
| + * notice, this list of conditions and the following disclaimer. |
| + * 2. Redistributions in binary form must reproduce the above copyright |
| + * notice, this list of conditions and the following disclaimer in the |
| + * documentation and/or other materials provided with the distribution. |
| + * |
| + * THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS'' |
| + * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, |
| + * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS |
| + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF |
| + * THE POSSIBILITY OF SUCH DAMAGE. |
| + */ |
| + |
| // |
| // mtl_resource_spi.h: |
| -// Used to set Metal resource ownership identity with SPI. |
| -// Purposefully empty header file. Actual implementation will be hosted in WebKit. |
| +// Used to set Metal resource ownership identity with SPI |
| // |
| + |
| +#ifndef LIBANGLE_RENDERER_METAL_RESOURCE_SPI_H_ |
| +#define LIBANGLE_RENDERER_METAL_RESOURCE_SPI_H_ |
| + |
| +#import "common/apple/apple_platform.h" |
| + |
| +#if ANGLE_USE_METAL_OWNERSHIP_IDENTITY |
| + |
| +# import <Metal/MTLResource_Private.h> |
| +# import <Metal/Metal.h> |
| +# import <mach/mach_types.h> |
| + |
| +namespace rx |
| +{ |
| +namespace mtl |
| +{ |
| +inline void setOwnerWithIdentity(id<MTLResource> resource, task_id_token_t identityToken) |
| +{ |
| + if (identityToken != TASK_ID_TOKEN_NULL) |
| + { |
| + kern_return_t kr = [(id<MTLResourceSPI>)resource setOwnerWithIdentity:identityToken]; |
| + if (ANGLE_UNLIKELY(kr != KERN_SUCCESS)) |
| + { |
| + ERR() << "setOwnerWithIdentity failed with: %s (%x)" << mach_error_string(kr) << kr; |
| + ASSERT(false); |
| + } |
| + } |
| + return; |
| +} |
| +} |
| +} |
| +#endif |
| + |
| +#endif /* LIBANGLE_RENDERER_METAL_RESOURCE_SPI_H_ */ |
| diff --git a/src/libANGLE/renderer/metal/mtl_resources.h b/src/libANGLE/renderer/metal/mtl_resources.h |
| index afbc985eb92dfbd5af964bd8ac207ac1db6ac97e..9805aa705a259a97e92418387f12385171b2fa59 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_resources.h |
| +++ b/src/libANGLE/renderer/metal/mtl_resources.h |
| @@ -339,7 +339,9 @@ class Texture final : public Resource, |
| size_t mEstimatedByteSize = 0; |
| }; |
| |
| -class Buffer final : public Resource, public WrappedObject<id<MTLBuffer>> |
| +class Buffer final : public Resource, |
| + public WrappedObject<id<MTLBuffer>>, |
| + public std::enable_shared_from_this<Buffer> |
| { |
| public: |
| static angle::Result MakeBuffer(ContextMtl *context, |
| diff --git a/src/libANGLE/renderer/metal/mtl_resources.mm b/src/libANGLE/renderer/metal/mtl_resources.mm |
| index 6bcb9cd8d3e2087dd269b4c5fb8b8214c33e2f0e..414a3d9dbb7ab125172ee99aa9df636383c3d4f5 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_resources.mm |
| +++ b/src/libANGLE/renderer/metal/mtl_resources.mm |
| @@ -35,12 +35,14 @@ inline NSUInteger GetMipSize(NSUInteger baseSize, const MipmapNativeLevel level) |
| // Asynchronously synchronize the content of a resource between GPU memory and its CPU cache. |
| // NOTE: This operation doesn't finish immediately upon function's return. |
| template <class T> |
| -void InvokeCPUMemSync(ContextMtl *context, mtl::BlitCommandEncoder *blitEncoder, T *resource) |
| +void InvokeCPUMemSync(ContextMtl *context, |
| + mtl::BlitCommandEncoder *blitEncoder, |
| + const std::shared_ptr<T> &resource) |
| { |
| #if TARGET_OS_OSX || TARGET_OS_MACCATALYST |
| if (blitEncoder) |
| { |
| - blitEncoder->synchronizeResource(resource); |
| + blitEncoder->synchronizeResource(resource.get()); |
| |
| resource->resetCPUReadMemNeedSync(); |
| resource->setCPUReadMemSyncPending(true); |
| @@ -49,7 +51,7 @@ void InvokeCPUMemSync(ContextMtl *context, mtl::BlitCommandEncoder *blitEncoder, |
| } |
| |
| template <class T> |
| -void EnsureCPUMemWillBeSynced(ContextMtl *context, T *resource) |
| +void EnsureCPUMemWillBeSynced(ContextMtl *context, const std::shared_ptr<T> &resource) |
| { |
| #if TARGET_OS_OSX || TARGET_OS_MACCATALYST |
| // Make sure GPU & CPU contents are synchronized. |
| @@ -271,11 +273,10 @@ void EnsureCPUMemWillBeSynced(ContextMtl *context, T *resource) |
| { |
| return angle::Result::Stop; |
| } |
| - refOut->reset(new Texture(context, desc, mips, renderTargetOnly, allowFormatView, memoryLess)); |
| - if (!refOut || !refOut->get()) |
| - { |
| - ANGLE_MTL_CHECK(context, false, GL_OUT_OF_MEMORY); |
| - } |
| + ASSERT(refOut); |
| + Texture *newTexture = new Texture(context, desc, mips, renderTargetOnly, allowFormatView, memoryLess); |
| + ANGLE_MTL_CHECK(context, newTexture->valid(), GL_OUT_OF_MEMORY); |
| + refOut->reset(newTexture); |
| if (!mtlFormat.hasDepthAndStencilBits()) |
| { |
| refOut->get()->setColorWritableMask(GetEmulatedColorWriteMask(mtlFormat)); |
| @@ -299,13 +300,10 @@ void EnsureCPUMemWillBeSynced(ContextMtl *context, T *resource) |
| bool renderTargetOnly, |
| TextureRef *refOut) |
| { |
| - |
| - refOut->reset(new Texture(context, desc, surfaceRef, slice, renderTargetOnly)); |
| - |
| - if (!(*refOut) || !(*refOut)->get()) |
| - { |
| - ANGLE_MTL_CHECK(context, false, GL_OUT_OF_MEMORY); |
| - } |
| + ASSERT(refOut); |
| + Texture *newTexture = new Texture(context, desc, surfaceRef, slice, renderTargetOnly); |
| + ANGLE_MTL_CHECK(context, newTexture->valid(), GL_OUT_OF_MEMORY); |
| + refOut->reset(newTexture); |
| if (!mtlFormat.hasDepthAndStencilBits()) |
| { |
| refOut->get()->setColorWritableMask(GetEmulatedColorWriteMask(mtlFormat)); |
| @@ -504,12 +502,12 @@ bool needMultisampleColorFormatShaderReadWorkaround(ContextMtl *context, MTLText |
| |
| void Texture::syncContent(ContextMtl *context, mtl::BlitCommandEncoder *blitEncoder) |
| { |
| - InvokeCPUMemSync(context, blitEncoder, this); |
| + InvokeCPUMemSync(context, blitEncoder, shared_from_this()); |
| } |
| |
| void Texture::syncContentIfNeeded(ContextMtl *context) |
| { |
| - EnsureCPUMemWillBeSynced(context, this); |
| + EnsureCPUMemWillBeSynced(context, shared_from_this()); |
| } |
| |
| bool Texture::isCPUAccessible() const |
| @@ -988,7 +986,7 @@ bool needMultisampleColorFormatShaderReadWorkaround(ContextMtl *context, MTLText |
| |
| void Buffer::syncContent(ContextMtl *context, mtl::BlitCommandEncoder *blitEncoder) |
| { |
| - InvokeCPUMemSync(context, blitEncoder, this); |
| + InvokeCPUMemSync(context, blitEncoder, shared_from_this()); |
| } |
| |
| const uint8_t *Buffer::mapReadOnly(ContextMtl *context) |
| @@ -1009,7 +1007,7 @@ bool needMultisampleColorFormatShaderReadWorkaround(ContextMtl *context, MTLText |
| { |
| CommandQueue &cmdQueue = context->cmdQueue(); |
| |
| - EnsureCPUMemWillBeSynced(context, this); |
| + EnsureCPUMemWillBeSynced(context, shared_from_this()); |
| |
| if (this->isBeingUsedByGPU(context)) |
| { |
| diff --git a/src/libANGLE/renderer/metal/mtl_state_cache.mm b/src/libANGLE/renderer/metal/mtl_state_cache.mm |
| index 7ef96c443d7b780ad045dfeac423ef0a86889750..625416a3ba8f1f819ecbe63e81371a5b13949b32 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_state_cache.mm |
| +++ b/src/libANGLE/renderer/metal/mtl_state_cache.mm |
| @@ -943,6 +943,62 @@ void ToObjC(const RenderPassStencilAttachmentDesc &desc, |
| return re.first->second; |
| } |
| |
| +static bool ValidateRenderPipelineState(const MTLRenderPipelineDescriptor *descriptor, |
| + ContextMtl *context, |
| + const mtl::ContextDevice &device) |
| +{ |
| + // Ensure there is at least one valid render target. |
| + bool hasValidRenderTarget = false; |
| + |
| + const NSUInteger maxColorRenderTargets = GetMaxNumberOfRenderTargetsForDevice(device); |
| + for (NSUInteger i = 0; i < maxColorRenderTargets; ++i) |
| + { |
| + auto colorAttachment = descriptor.colorAttachments[i]; |
| + if (colorAttachment && colorAttachment.pixelFormat != MTLPixelFormatInvalid) |
| + { |
| + hasValidRenderTarget = true; |
| + break; |
| + } |
| + } |
| + |
| + if (!hasValidRenderTarget && descriptor.depthAttachmentPixelFormat != MTLPixelFormatInvalid) |
| + { |
| + hasValidRenderTarget = true; |
| + } |
| + |
| + if (!hasValidRenderTarget && descriptor.stencilAttachmentPixelFormat != MTLPixelFormatInvalid) |
| + { |
| + hasValidRenderTarget = true; |
| + } |
| + |
| + if (!hasValidRenderTarget) |
| + { |
| + UNREACHABLE(); |
| + ANGLE_MTL_HANDLE_ERROR(context, "Render pipeline requires at least one render target.", |
| + GL_INVALID_OPERATION); |
| + return false; |
| + } |
| + |
| + // Ensure the device can support the storage requirement for render targets. |
| + if (DeviceHasMaximumRenderTargetSize(device)) |
| + { |
| + // TODO: Is the use of NSUInteger in 32 bit systems ok without any overflow checking? |
| + NSUInteger maxSize = GetMaxRenderTargetSizeForDeviceInBytes(device); |
| + NSUInteger renderTargetSize = |
| + ComputeTotalSizeUsedForMTLRenderPipelineDescriptor(descriptor, context, device); |
| + if (renderTargetSize > maxSize) |
| + { |
| + std::stringstream errorStream; |
| + errorStream << "This set of render targets requires " << renderTargetSize |
| + << " bytes of pixel storage. This device supports " << maxSize << " bytes."; |
| + ANGLE_MTL_HANDLE_ERROR(context, errorStream.str().c_str(), GL_INVALID_OPERATION); |
| + return false; |
| + } |
| + } |
| + |
| + return true; |
| +} |
| + |
| AutoObjCPtr<id<MTLRenderPipelineState>> RenderPipelineCache::createRenderPipelineState( |
| ContextMtl *context, |
| const RenderPipelineDesc &originalDesc, |
| @@ -1005,23 +1061,10 @@ void ToObjC(const RenderPassStencilAttachmentDesc &desc, |
| // Convert to Objective-C desc: |
| AutoObjCObj<MTLRenderPipelineDescriptor> objCDesc = ToObjC(vertShader, fragShader, desc); |
| |
| - // Validate Render Pipeline State: |
| - if (DeviceHasMaximumRenderTargetSize(metalDevice)) |
| - { |
| - // TODO: Is the use of NSUInteger in 32 bit systems ok without any overflow checking? |
| - NSUInteger maxSize = GetMaxRenderTargetSizeForDeviceInBytes(metalDevice); |
| - NSUInteger renderTargetSize = |
| - ComputeTotalSizeUsedForMTLRenderPipelineDescriptor(objCDesc, context, metalDevice); |
| - if (renderTargetSize > maxSize) |
| + if (!ValidateRenderPipelineState(objCDesc, context, metalDevice)) |
| { |
| - std::stringstream errorStream; |
| - errorStream << "This set of render targets requires " << renderTargetSize |
| - << " bytes of pixel storage. This device supports " << maxSize |
| - << " bytes."; |
| - ANGLE_MTL_HANDLE_ERROR(context, errorStream.str().c_str(), GL_INVALID_OPERATION); |
| return nil; |
| } |
| - } |
| |
| // Special attribute slot for default attribute |
| if (insertDefaultAttribLayout) |
| diff --git a/src/libANGLE/renderer/metal/mtl_utils.h b/src/libANGLE/renderer/metal/mtl_utils.h |
| index d218c13ed14be57f514cbf3c9dbeedefcdd49621..b236d92655d0307bb0b61ce51bf6406e9e934e96 100644 |
| --- a/src/libANGLE/renderer/metal/mtl_utils.h |
| +++ b/src/libANGLE/renderer/metal/mtl_utils.h |
| @@ -105,11 +105,9 @@ AutoObjCPtr<id<MTLLibrary>> CreateShaderLibrary( |
| bool enableFastMath, |
| AutoObjCPtr<NSError *> *error); |
| |
| -AutoObjCPtr<id<MTLLibrary>> CreateShaderLibraryFromBinary( |
| - id<MTLDevice> metalDevice, |
| +AutoObjCPtr<id<MTLLibrary>> CreateShaderLibraryFromBinary(id<MTLDevice> metalDevice, |
| const uint8_t *binarySource, |
| size_t binarySourceLen, |
| - NSDictionary<NSString *, NSObject *> *substitutionDictionary, |
| AutoObjCPtr<NSError *> *error); |
| |
| bool SupportsAppleGPUFamily(id<MTLDevice> device, uint8_t appleFamily); |
| diff --git a/src/tests/deqp_support/deqp_egl_test_expectations.txt b/src/tests/deqp_support/deqp_egl_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_egl_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_egl_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_gles2_test_expectations.txt b/src/tests/deqp_support/deqp_gles2_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_gles2_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_gles2_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_gles31_test_expectations.txt b/src/tests/deqp_support/deqp_gles31_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_gles31_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_gles31_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_gles31_rotate_test_expectations.txt b/src/tests/deqp_support/deqp_gles31_rotate_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_gles31_rotate_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_gles31_rotate_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_gles3_test_expectations.txt b/src/tests/deqp_support/deqp_gles3_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_gles3_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_gles3_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_gles3_rotate_test_expectations.txt b/src/tests/deqp_support/deqp_gles3_rotate_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_gles3_rotate_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_gles3_rotate_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_khr_gles2_test_expectations.txt b/src/tests/deqp_support/deqp_khr_gles2_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_khr_gles2_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_khr_gles2_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_khr_gles31_test_expectations.txt b/src/tests/deqp_support/deqp_khr_gles31_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_khr_gles31_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_khr_gles31_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_khr_gles32_test_expectations.txt b/src/tests/deqp_support/deqp_khr_gles32_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_khr_gles32_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_khr_gles32_TestExpectations |
| diff --git a/src/tests/deqp_support/deqp_khr_gles3_test_expectations.txt b/src/tests/deqp_support/deqp_khr_gles3_TestExpectations |
| similarity index 100% |
| rename from src/tests/deqp_support/deqp_khr_gles3_test_expectations.txt |
| rename to src/tests/deqp_support/deqp_khr_gles3_TestExpectations |
