// // Copyright (c) 2012 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. // // InputLayoutCache.cpp: Defines InputLayoutCache, a class that builds and caches // D3D11 input layouts. #include "libANGLE/renderer/d3d/d3d11/InputLayoutCache.h" #include "common/utilities.h" #include "libANGLE/Program.h" #include "libANGLE/VertexAttribute.h" #include "libANGLE/renderer/d3d/IndexDataManager.h" #include "libANGLE/renderer/d3d/ProgramD3D.h" #include "libANGLE/renderer/d3d/VertexDataManager.h" #include "libANGLE/renderer/d3d/d3d11/Buffer11.h" #include "libANGLE/renderer/d3d/d3d11/ShaderExecutable11.h" #include "libANGLE/renderer/d3d/d3d11/VertexBuffer11.h" #include "libANGLE/renderer/d3d/d3d11/formatutils11.h" #include "third_party/murmurhash/MurmurHash3.h" namespace rx { namespace { size_t GetReservedBufferCount(bool usesPointSpriteEmulation) { return usesPointSpriteEmulation ? 1 : 0; } gl::InputLayout GetInputLayout(const SortedAttribArray &translatedAttributes, size_t attributeCount) { gl::InputLayout inputLayout(attributeCount, gl::VERTEX_FORMAT_INVALID); for (size_t attributeIndex = 0; attributeIndex < attributeCount; ++attributeIndex) { const TranslatedAttribute *translatedAttribute = translatedAttributes[attributeIndex]; if (translatedAttribute->active) { inputLayout[attributeIndex] = gl::GetVertexFormatType( *translatedAttribute->attribute, translatedAttribute->currentValueType); } } return inputLayout; } GLenum GetGLSLAttributeType(const std::vector &shaderAttributes, int index) { // Count matrices differently for (const sh::Attribute &attrib : shaderAttributes) { if (attrib.location == -1) { continue; } GLenum transposedType = gl::TransposeMatrixType(attrib.type); int rows = gl::VariableRowCount(transposedType); if (index >= attrib.location && index < attrib.location + rows) { return transposedType; } } UNREACHABLE(); return GL_NONE; } const unsigned int kDefaultCacheSize = 1024; struct PackedAttribute { uint8_t attribType; uint8_t semanticIndex; uint8_t vertexFormatType; uint8_t divisor; }; Optional FindFirstNonInstanced(const SortedAttribArray &sortedAttributes, size_t maxIndex) { for (size_t index = 0; index < maxIndex; ++index) { if (sortedAttributes[index]->divisor == 0) { return Optional(index); } } return Optional::Invalid(); } } // anonymous namespace void InputLayoutCache::PackedAttributeLayout::addAttributeData( GLenum glType, UINT semanticIndex, gl::VertexFormatType vertexFormatType, unsigned int divisor) { gl::AttributeType attribType = gl::GetAttributeType(glType); PackedAttribute packedAttrib; packedAttrib.attribType = static_cast(attribType); packedAttrib.semanticIndex = static_cast(semanticIndex); packedAttrib.vertexFormatType = static_cast(vertexFormatType); packedAttrib.divisor = static_cast(divisor); ASSERT(static_cast(packedAttrib.attribType) == attribType); ASSERT(static_cast(packedAttrib.semanticIndex) == semanticIndex); ASSERT(static_cast(packedAttrib.vertexFormatType) == vertexFormatType); ASSERT(static_cast(packedAttrib.divisor) == divisor); static_assert(sizeof(uint32_t) == sizeof(PackedAttribute), "PackedAttributes must be 32-bits exactly."); attributeData[numAttributes++] = gl::bitCast(packedAttrib); } bool InputLayoutCache::PackedAttributeLayout::operator<(const PackedAttributeLayout &other) const { if (numAttributes != other.numAttributes) { return numAttributes < other.numAttributes; } if (flags != other.flags) { return flags < other.flags; } return memcmp(attributeData, other.attributeData, sizeof(uint32_t) * numAttributes) < 0; } InputLayoutCache::InputLayoutCache() : mUnsortedAttributesCount(0), mCacheSize(kDefaultCacheSize) { mCounter = 0; mDevice = NULL; mDeviceContext = NULL; mCurrentIL = NULL; for (unsigned int i = 0; i < gl::MAX_VERTEX_ATTRIBS; i++) { mCurrentBuffers[i] = NULL; mCurrentVertexStrides[i] = static_cast(-1); mCurrentVertexOffsets[i] = static_cast(-1); } mPointSpriteVertexBuffer = NULL; mPointSpriteIndexBuffer = NULL; } InputLayoutCache::~InputLayoutCache() { clear(); } void InputLayoutCache::initialize(ID3D11Device *device, ID3D11DeviceContext *context) { clear(); mDevice = device; mDeviceContext = context; mFeatureLevel = device->GetFeatureLevel(); } void InputLayoutCache::clear() { for (auto &layout : mLayoutMap) { SafeRelease(layout.second); } mLayoutMap.clear(); SafeRelease(mPointSpriteVertexBuffer); SafeRelease(mPointSpriteIndexBuffer); markDirty(); } void InputLayoutCache::markDirty() { mCurrentIL = NULL; for (unsigned int i = 0; i < gl::MAX_VERTEX_ATTRIBS; i++) { mCurrentBuffers[i] = NULL; mCurrentVertexStrides[i] = static_cast(-1); mCurrentVertexOffsets[i] = static_cast(-1); } mUnsortedAttributesCount = 0; } gl::Error InputLayoutCache::applyVertexBuffers( const std::vector &unsortedAttributes, GLenum mode, gl::Program *program, TranslatedIndexData *indexInfo, GLsizei numIndicesPerInstance) { ASSERT(mDevice && mDeviceContext); ProgramD3D *programD3D = GetImplAs(program); bool programUsesInstancedPointSprites = programD3D->usesPointSize() && programD3D->usesInstancedPointSpriteEmulation(); bool instancedPointSpritesActive = programUsesInstancedPointSprites && (mode == GL_POINTS); SortedIndexArray sortedSemanticIndices; mSortedAttributes.fill(nullptr); mUnsortedAttributesCount = unsortedAttributes.size(); programD3D->sortAttributesByLayout(unsortedAttributes, sortedSemanticIndices.data(), mSortedAttributes.data()); // If we are using FL 9_3, make sure the first attribute is not instanced if (mFeatureLevel <= D3D_FEATURE_LEVEL_9_3 && !unsortedAttributes.empty()) { if (mSortedAttributes[0]->divisor > 0) { Optional firstNonInstancedIndex = FindFirstNonInstanced(mSortedAttributes, unsortedAttributes.size()); if (firstNonInstancedIndex.valid()) { size_t index = firstNonInstancedIndex.value(); std::swap(mSortedAttributes[0], mSortedAttributes[index]); std::swap(sortedSemanticIndices[0], sortedSemanticIndices[index]); } } } gl::Error error = updateInputLayout(program, mode, mSortedAttributes, sortedSemanticIndices, unsortedAttributes.size(), numIndicesPerInstance); if (error.isError()) { return error; } bool dirtyBuffers = false; size_t minDiff = gl::MAX_VERTEX_ATTRIBS; size_t maxDiff = 0; // Note that if we use instance emulation, we reserve the first buffer slot. size_t reservedBuffers = GetReservedBufferCount(programUsesInstancedPointSprites); for (size_t attribIndex = 0; attribIndex < (gl::MAX_VERTEX_ATTRIBS - reservedBuffers); ++attribIndex) { ID3D11Buffer *buffer = NULL; UINT vertexStride = 0; UINT vertexOffset = 0; const auto &attrib = *mSortedAttributes[attribIndex]; if (attribIndex < unsortedAttributes.size() && attrib.active) { VertexBuffer11 *vertexBuffer = GetAs(attrib.vertexBuffer); Buffer11 *bufferStorage = attrib.storage ? GetAs(attrib.storage) : nullptr; // If indexed pointsprite emulation is active, then we need to take a less efficent code path. // Emulated indexed pointsprite rendering requires that the vertex buffers match exactly to // the indices passed by the caller. This could expand or shrink the vertex buffer depending // on the number of points indicated by the index list or how many duplicates are found on the index list. if (bufferStorage == nullptr) { buffer = vertexBuffer->getBuffer(); } else if (instancedPointSpritesActive && (indexInfo != nullptr)) { if (indexInfo->srcIndexData.srcBuffer != nullptr) { const uint8_t *bufferData = nullptr; error = indexInfo->srcIndexData.srcBuffer->getData(&bufferData); if (error.isError()) { return error; } ASSERT(bufferData != nullptr); ptrdiff_t offset = reinterpret_cast(indexInfo->srcIndexData.srcIndices); indexInfo->srcIndexData.srcBuffer = nullptr; indexInfo->srcIndexData.srcIndices = bufferData + offset; } buffer = bufferStorage->getEmulatedIndexedBuffer(&indexInfo->srcIndexData, &attrib); } else { buffer = bufferStorage->getBuffer(BUFFER_USAGE_VERTEX_OR_TRANSFORM_FEEDBACK); } vertexStride = attrib.stride; vertexOffset = attrib.offset; } size_t bufferIndex = reservedBuffers + attribIndex; if (buffer != mCurrentBuffers[bufferIndex] || vertexStride != mCurrentVertexStrides[bufferIndex] || vertexOffset != mCurrentVertexOffsets[bufferIndex]) { dirtyBuffers = true; minDiff = std::min(minDiff, bufferIndex); maxDiff = std::max(maxDiff, bufferIndex); mCurrentBuffers[bufferIndex] = buffer; mCurrentVertexStrides[bufferIndex] = vertexStride; mCurrentVertexOffsets[bufferIndex] = vertexOffset; } } // Instanced PointSprite emulation requires two additional ID3D11Buffers. A vertex buffer needs // to be created and added to the list of current buffers, strides and offsets collections. // This buffer contains the vertices for a single PointSprite quad. // An index buffer also needs to be created and applied because rendering instanced data on // D3D11 FL9_3 requires DrawIndexedInstanced() to be used. Shaders that contain gl_PointSize and // used without the GL_POINTS rendering mode require a vertex buffer because some drivers cannot // handle missing vertex data and will TDR the system. if (programUsesInstancedPointSprites) { HRESULT result = S_OK; const UINT pointSpriteVertexStride = sizeof(float) * 5; if (!mPointSpriteVertexBuffer) { static const float pointSpriteVertices[] = { // Position // TexCoord -1.0f, -1.0f, 0.0f, 0.0f, 1.0f, -1.0f, 1.0f, 0.0f, 0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 1.0f, 0.0f, 1.0f, -1.0f, 0.0f, 1.0f, 1.0f, -1.0f, -1.0f, 0.0f, 0.0f, 1.0f, 1.0f, 1.0f, 0.0f, 1.0f, 0.0f, }; D3D11_SUBRESOURCE_DATA vertexBufferData = { pointSpriteVertices, 0, 0 }; D3D11_BUFFER_DESC vertexBufferDesc; vertexBufferDesc.ByteWidth = sizeof(pointSpriteVertices); vertexBufferDesc.BindFlags = D3D11_BIND_VERTEX_BUFFER; vertexBufferDesc.Usage = D3D11_USAGE_IMMUTABLE; vertexBufferDesc.CPUAccessFlags = 0; vertexBufferDesc.MiscFlags = 0; vertexBufferDesc.StructureByteStride = 0; result = mDevice->CreateBuffer(&vertexBufferDesc, &vertexBufferData, &mPointSpriteVertexBuffer); if (FAILED(result)) { return gl::Error(GL_OUT_OF_MEMORY, "Failed to create instanced pointsprite emulation vertex buffer, HRESULT: 0x%08x", result); } } mCurrentBuffers[0] = mPointSpriteVertexBuffer; // Set the stride to 0 if GL_POINTS mode is not being used to instruct the driver to avoid // indexing into the vertex buffer. mCurrentVertexStrides[0] = instancedPointSpritesActive ? pointSpriteVertexStride : 0; mCurrentVertexOffsets[0] = 0; // Update maxDiff to include the additional point sprite vertex buffer // to ensure that IASetVertexBuffers uses the correct buffer count. minDiff = 0; maxDiff = std::max(maxDiff, static_cast(0)); if (!mPointSpriteIndexBuffer) { // Create an index buffer and set it for pointsprite rendering static const unsigned short pointSpriteIndices[] = { 0, 1, 2, 3, 4, 5, }; D3D11_SUBRESOURCE_DATA indexBufferData = { pointSpriteIndices, 0, 0 }; D3D11_BUFFER_DESC indexBufferDesc; indexBufferDesc.ByteWidth = sizeof(pointSpriteIndices); indexBufferDesc.BindFlags = D3D11_BIND_INDEX_BUFFER; indexBufferDesc.Usage = D3D11_USAGE_IMMUTABLE; indexBufferDesc.CPUAccessFlags = 0; indexBufferDesc.MiscFlags = 0; indexBufferDesc.StructureByteStride = 0; result = mDevice->CreateBuffer(&indexBufferDesc, &indexBufferData, &mPointSpriteIndexBuffer); if (FAILED(result)) { SafeRelease(mPointSpriteVertexBuffer); return gl::Error(GL_OUT_OF_MEMORY, "Failed to create instanced pointsprite emulation index buffer, HRESULT: 0x%08x", result); } } if (instancedPointSpritesActive) { // The index buffer is applied here because Instanced PointSprite emulation uses the a // non-indexed rendering path in ANGLE (DrawArrays). This means that applyIndexBuffer() // on the renderer will not be called and setting this buffer here ensures that the // rendering path will contain the correct index buffers. mDeviceContext->IASetIndexBuffer(mPointSpriteIndexBuffer, DXGI_FORMAT_R16_UINT, 0); } } if (dirtyBuffers) { ASSERT(minDiff <= maxDiff && maxDiff < gl::MAX_VERTEX_ATTRIBS); mDeviceContext->IASetVertexBuffers( static_cast(minDiff), static_cast(maxDiff - minDiff + 1), mCurrentBuffers + minDiff, mCurrentVertexStrides + minDiff, mCurrentVertexOffsets + minDiff); } return gl::Error(GL_NO_ERROR); } gl::Error InputLayoutCache::updateVertexOffsetsForPointSpritesEmulation(GLsizei emulatedInstanceId) { size_t reservedBuffers = GetReservedBufferCount(true); for (size_t attribIndex = 0; attribIndex < mUnsortedAttributesCount; ++attribIndex) { const auto &attrib = *mSortedAttributes[attribIndex]; size_t bufferIndex = reservedBuffers + attribIndex; if (attrib.active && attrib.divisor > 0) { mCurrentVertexOffsets[bufferIndex] = attrib.offset + (attrib.stride * (emulatedInstanceId / attrib.divisor)); } } mDeviceContext->IASetVertexBuffers(0, gl::MAX_VERTEX_ATTRIBS, mCurrentBuffers, mCurrentVertexStrides, mCurrentVertexOffsets); return gl::Error(GL_NO_ERROR); } gl::Error InputLayoutCache::updateInputLayout(gl::Program *program, GLenum mode, const SortedAttribArray &sortedAttributes, const SortedIndexArray &sortedSemanticIndices, size_t attribCount, GLsizei numIndicesPerInstance) { const std::vector &shaderAttributes = program->getAttributes(); PackedAttributeLayout layout; ProgramD3D *programD3D = GetImplAs(program); bool programUsesInstancedPointSprites = programD3D->usesPointSize() && programD3D->usesInstancedPointSpriteEmulation(); bool instancedPointSpritesActive = programUsesInstancedPointSprites && (mode == GL_POINTS); if (programUsesInstancedPointSprites) { layout.flags |= PackedAttributeLayout::FLAG_USES_INSTANCED_SPRITES; } if (instancedPointSpritesActive) { layout.flags |= PackedAttributeLayout::FLAG_INSTANCED_SPRITES_ACTIVE; } if (numIndicesPerInstance > 0) { layout.flags |= PackedAttributeLayout::FLAG_INSTANCED_RENDERING_ACTIVE; } const auto &semanticToLocation = programD3D->getAttributesByLayout(); for (size_t attribIndex = 0; attribIndex < attribCount; ++attribIndex) { const auto &attrib = *sortedAttributes[attribIndex]; int sortedIndex = sortedSemanticIndices[attribIndex]; if (!attrib.active) continue; gl::VertexFormatType vertexFormatType = gl::GetVertexFormatType(*attrib.attribute, attrib.currentValueType); // Record the type of the associated vertex shader vector in our key // This will prevent mismatched vertex shaders from using the same input layout GLenum glslElementType = GetGLSLAttributeType(shaderAttributes, semanticToLocation[sortedIndex]); layout.addAttributeData(glslElementType, sortedIndex, vertexFormatType, attrib.divisor); } ID3D11InputLayout *inputLayout = nullptr; if (layout.numAttributes > 0 || layout.flags != 0) { auto layoutMapIt = mLayoutMap.find(layout); if (layoutMapIt != mLayoutMap.end()) { inputLayout = layoutMapIt->second; } else { gl::Error error = createInputLayout(sortedAttributes, sortedSemanticIndices, attribCount, mode, program, numIndicesPerInstance, &inputLayout); if (error.isError()) { return error; } if (mLayoutMap.size() >= mCacheSize) { TRACE("Overflowed the limit of %u input layouts, purging half the cache.", mCacheSize); // Randomly release every second element auto it = mLayoutMap.begin(); while (it != mLayoutMap.end()) { it++; if (it != mLayoutMap.end()) { // c++11 erase allows us to easily delete the current iterator. SafeRelease(it->second); it = mLayoutMap.erase(it); } } } mLayoutMap[layout] = inputLayout; } } if (inputLayout != mCurrentIL) { mDeviceContext->IASetInputLayout(inputLayout); mCurrentIL = inputLayout; } return gl::Error(GL_NO_ERROR); } gl::Error InputLayoutCache::createInputLayout(const SortedAttribArray &sortedAttributes, const SortedIndexArray &sortedSemanticIndices, size_t attribCount, GLenum mode, gl::Program *program, GLsizei numIndicesPerInstance, ID3D11InputLayout **inputLayoutOut) { ProgramD3D *programD3D = GetImplAs(program); bool programUsesInstancedPointSprites = programD3D->usesPointSize() && programD3D->usesInstancedPointSpriteEmulation(); unsigned int inputElementCount = 0; std::array inputElements; for (size_t attribIndex = 0; attribIndex < attribCount; ++attribIndex) { const auto &attrib = *sortedAttributes[attribIndex]; const int sortedIndex = sortedSemanticIndices[attribIndex]; if (!attrib.active) continue; D3D11_INPUT_CLASSIFICATION inputClass = attrib.divisor > 0 ? D3D11_INPUT_PER_INSTANCE_DATA : D3D11_INPUT_PER_VERTEX_DATA; const auto &vertexFormatType = gl::GetVertexFormatType(*attrib.attribute, attrib.currentValueType); const auto &vertexFormatInfo = d3d11::GetVertexFormatInfo(vertexFormatType, mFeatureLevel); auto *inputElement = &inputElements[inputElementCount]; inputElement->SemanticName = "TEXCOORD"; inputElement->SemanticIndex = sortedIndex; inputElement->Format = vertexFormatInfo.nativeFormat; inputElement->InputSlot = static_cast(attribIndex); inputElement->AlignedByteOffset = 0; inputElement->InputSlotClass = inputClass; inputElement->InstanceDataStepRate = attrib.divisor; inputElementCount++; } // Instanced PointSprite emulation requires additional entries in the // inputlayout to support the vertices that make up the pointsprite quad. // We do this even if mode != GL_POINTS, since the shader signature has these inputs, and the // input layout must match the shader if (programUsesInstancedPointSprites) { // On 9_3, we must ensure that slot 0 contains non-instanced data. // If slot 0 currently contains instanced data then we swap it with a non-instanced element. // Note that instancing is only available on 9_3 via ANGLE_instanced_arrays, since 9_3 // doesn't support OpenGL ES 3.0. // As per the spec for ANGLE_instanced_arrays, not all attributes can be instanced // simultaneously, so a non-instanced element must exist. for (size_t elementIndex = 0; elementIndex < inputElementCount; ++elementIndex) { if (sortedAttributes[elementIndex]->active) { // If rendering points and instanced pointsprite emulation is being used, the // inputClass is required to be configured as per instance data if (mode == GL_POINTS) { inputElements[elementIndex].InputSlotClass = D3D11_INPUT_PER_INSTANCE_DATA; inputElements[elementIndex].InstanceDataStepRate = 1; if (numIndicesPerInstance > 0 && sortedAttributes[elementIndex]->divisor > 0) { inputElements[elementIndex].InstanceDataStepRate = numIndicesPerInstance; } } inputElements[elementIndex].InputSlot++; } } inputElements[inputElementCount].SemanticName = "SPRITEPOSITION"; inputElements[inputElementCount].SemanticIndex = 0; inputElements[inputElementCount].Format = DXGI_FORMAT_R32G32B32_FLOAT; inputElements[inputElementCount].InputSlot = 0; inputElements[inputElementCount].AlignedByteOffset = 0; inputElements[inputElementCount].InputSlotClass = D3D11_INPUT_PER_VERTEX_DATA; inputElements[inputElementCount].InstanceDataStepRate = 0; inputElementCount++; inputElements[inputElementCount].SemanticName = "SPRITETEXCOORD"; inputElements[inputElementCount].SemanticIndex = 0; inputElements[inputElementCount].Format = DXGI_FORMAT_R32G32_FLOAT; inputElements[inputElementCount].InputSlot = 0; inputElements[inputElementCount].AlignedByteOffset = sizeof(float) * 3; inputElements[inputElementCount].InputSlotClass = D3D11_INPUT_PER_VERTEX_DATA; inputElements[inputElementCount].InstanceDataStepRate = 0; inputElementCount++; } const gl::InputLayout &shaderInputLayout = GetInputLayout(sortedAttributes, attribCount); ShaderExecutableD3D *shader = nullptr; gl::Error error = programD3D->getVertexExecutableForInputLayout(shaderInputLayout, &shader, nullptr); if (error.isError()) { return error; } ShaderExecutableD3D *shader11 = GetAs(shader); HRESULT result = mDevice->CreateInputLayout(inputElements.data(), inputElementCount, shader11->getFunction(), shader11->getLength(), inputLayoutOut); if (FAILED(result)) { return gl::Error(GL_OUT_OF_MEMORY, "Failed to create internal input layout, HRESULT: 0x%08x", result); } return gl::Error(GL_NO_ERROR); } } // namespace rx