From d8d72434e51896c79b117229624ffed341da89a6 Mon Sep 17 00:00:00 2001 From: Brandon Jones Date: Fri, 22 Aug 2014 15:11:23 -0700 Subject: Moved DynamicHLSL into renderer/d3d Obviously D3D-only class is obvious. :) BUG=angle:731 Change-Id: Ide54e76a8b9bacb9b6834e2cb0a801140eecfe46 Reviewed-on: https://chromium-review.googlesource.com/213862 Reviewed-by: Jamie Madill Tested-by: Brandon Jones --- src/libGLESv2/renderer/d3d/DynamicHLSL.cpp | 1125 ++++++++++++++++++++++++++++ src/libGLESv2/renderer/d3d/DynamicHLSL.h | 96 +++ src/libGLESv2/renderer/d3d/ShaderD3D.h | 11 +- 3 files changed, 1224 insertions(+), 8 deletions(-) create mode 100644 src/libGLESv2/renderer/d3d/DynamicHLSL.cpp create mode 100644 src/libGLESv2/renderer/d3d/DynamicHLSL.h (limited to 'src/libGLESv2/renderer/d3d') diff --git a/src/libGLESv2/renderer/d3d/DynamicHLSL.cpp b/src/libGLESv2/renderer/d3d/DynamicHLSL.cpp new file mode 100644 index 00000000..ba893fef --- /dev/null +++ b/src/libGLESv2/renderer/d3d/DynamicHLSL.cpp @@ -0,0 +1,1125 @@ +// +// Copyright (c) 2014 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. +// +// DynamicHLSL.cpp: Implementation for link and run-time HLSL generation +// + +#include "libGLESv2/common_includes.h" + +#include "libGLESv2/renderer/d3d/DynamicHLSL.h" +#include "libGLESv2/Shader.h" +#include "libGLESv2/Program.h" +#include "libGLESv2/renderer/d3d/ShaderD3D.h" +#include "libGLESv2/renderer/Renderer.h" +#include "common/utilities.h" +#include "libGLESv2/ProgramBinary.h" +#include "libGLESv2/formatutils.h" +#include "common/blocklayout.h" + +// For use with ArrayString, see angleutils.h +META_ASSERT(GL_INVALID_INDEX == UINT_MAX); + +using namespace gl; + +namespace gl_d3d +{ + +std::string HLSLComponentTypeString(GLenum componentType) +{ + switch (componentType) + { + case GL_UNSIGNED_INT: return "uint"; + case GL_INT: return "int"; + case GL_UNSIGNED_NORMALIZED: + case GL_SIGNED_NORMALIZED: + case GL_FLOAT: return "float"; + default: UNREACHABLE(); return "not-component-type"; + } +} + +std::string HLSLComponentTypeString(GLenum componentType, int componentCount) +{ + return HLSLComponentTypeString(componentType) + (componentCount > 1 ? Str(componentCount) : ""); +} + +std::string HLSLMatrixTypeString(GLenum type) +{ + switch (type) + { + case GL_FLOAT_MAT2: return "float2x2"; + case GL_FLOAT_MAT3: return "float3x3"; + case GL_FLOAT_MAT4: return "float4x4"; + case GL_FLOAT_MAT2x3: return "float2x3"; + case GL_FLOAT_MAT3x2: return "float3x2"; + case GL_FLOAT_MAT2x4: return "float2x4"; + case GL_FLOAT_MAT4x2: return "float4x2"; + case GL_FLOAT_MAT3x4: return "float3x4"; + case GL_FLOAT_MAT4x3: return "float4x3"; + default: UNREACHABLE(); return "not-matrix-type"; + } +} + +std::string HLSLTypeString(GLenum type) +{ + if (gl::IsMatrixType(type)) + { + return HLSLMatrixTypeString(type); + } + + return HLSLComponentTypeString(gl::VariableComponentType(type), gl::VariableComponentCount(type)); +} + +} + +namespace rx +{ + +const std::string VERTEX_ATTRIBUTE_STUB_STRING = "@@ VERTEX ATTRIBUTES @@"; +const std::string PIXEL_OUTPUT_STUB_STRING = "@@ PIXEL OUTPUT @@"; + +DynamicHLSL::DynamicHLSL(rx::Renderer *const renderer) + : mRenderer(renderer) +{ +} + +static bool packVarying(PackedVarying *varying, const int maxVaryingVectors, VaryingPacking packing) +{ + GLenum transposedType = TransposeMatrixType(varying->type); + + // matrices within varying structs are not transposed + int registers = (varying->isStruct() ? HLSLVariableRegisterCount(*varying) : VariableRowCount(transposedType)) * varying->elementCount(); + int elements = (varying->isStruct() ? 4 : VariableColumnCount(transposedType)); + + if (elements >= 2 && elements <= 4) + { + for (int r = 0; r <= maxVaryingVectors - registers; r++) + { + bool available = true; + + for (int y = 0; y < registers && available; y++) + { + for (int x = 0; x < elements && available; x++) + { + if (packing[r + y][x]) + { + available = false; + } + } + } + + if (available) + { + varying->registerIndex = r; + + for (int y = 0; y < registers; y++) + { + for (int x = 0; x < elements; x++) + { + packing[r + y][x] = &*varying; + } + } + + return true; + } + } + + if (elements == 2) + { + for (int r = maxVaryingVectors - registers; r >= 0; r--) + { + bool available = true; + + for (int y = 0; y < registers && available; y++) + { + for (int x = 2; x < 4 && available; x++) + { + if (packing[r + y][x]) + { + available = false; + } + } + } + + if (available) + { + varying->registerIndex = r; + + for (int y = 0; y < registers; y++) + { + for (int x = 2; x < 4; x++) + { + packing[r + y][x] = &*varying; + } + } + + return true; + } + } + } + } + else if (elements == 1) + { + int space[4] = { 0 }; + + for (int y = 0; y < maxVaryingVectors; y++) + { + for (int x = 0; x < 4; x++) + { + space[x] += packing[y][x] ? 0 : 1; + } + } + + int column = 0; + + for (int x = 0; x < 4; x++) + { + if (space[x] >= registers && space[x] < space[column]) + { + column = x; + } + } + + if (space[column] >= registers) + { + for (int r = 0; r < maxVaryingVectors; r++) + { + if (!packing[r][column]) + { + varying->registerIndex = r; + + for (int y = r; y < r + registers; y++) + { + packing[y][column] = &*varying; + } + + break; + } + } + + return true; + } + } + else UNREACHABLE(); + + return false; +} + +// Packs varyings into generic varying registers, using the algorithm from [OpenGL ES Shading Language 1.00 rev. 17] appendix A section 7 page 111 +// Returns the number of used varying registers, or -1 if unsuccesful +int DynamicHLSL::packVaryings(InfoLog &infoLog, VaryingPacking packing, rx::FragmentShaderD3D *fragmentShader, + rx::VertexShaderD3D *vertexShader, const std::vector& transformFeedbackVaryings) +{ + const int maxVaryingVectors = mRenderer->getMaxVaryingVectors(); + + vertexShader->resetVaryingsRegisterAssignment(); + fragmentShader->resetVaryingsRegisterAssignment(); + + std::set packedVaryings; + + for (unsigned int varyingIndex = 0; varyingIndex < fragmentShader->mVaryings.size(); varyingIndex++) + { + PackedVarying *varying = &fragmentShader->mVaryings[varyingIndex]; + if (packVarying(varying, maxVaryingVectors, packing)) + { + packedVaryings.insert(varying->name); + } + else + { + infoLog.append("Could not pack varying %s", varying->name.c_str()); + return -1; + } + } + + for (unsigned int feedbackVaryingIndex = 0; feedbackVaryingIndex < transformFeedbackVaryings.size(); feedbackVaryingIndex++) + { + const std::string &transformFeedbackVarying = transformFeedbackVaryings[feedbackVaryingIndex]; + if (packedVaryings.find(transformFeedbackVarying) == packedVaryings.end()) + { + bool found = false; + for (unsigned int varyingIndex = 0; varyingIndex < vertexShader->mVaryings.size(); varyingIndex++) + { + PackedVarying *varying = &vertexShader->mVaryings[varyingIndex]; + if (transformFeedbackVarying == varying->name) + { + if (!packVarying(varying, maxVaryingVectors, packing)) + { + infoLog.append("Could not pack varying %s", varying->name.c_str()); + return -1; + } + + found = true; + break; + } + } + + if (!found && transformFeedbackVarying != "gl_Position" && transformFeedbackVarying != "gl_PointSize") + { + infoLog.append("Transform feedback varying %s does not exist in the vertex shader.", transformFeedbackVarying.c_str()); + return -1; + } + } + } + + // Return the number of used registers + int registers = 0; + + for (int r = 0; r < maxVaryingVectors; r++) + { + if (packing[r][0] || packing[r][1] || packing[r][2] || packing[r][3]) + { + registers++; + } + } + + return registers; +} + +std::string DynamicHLSL::generateVaryingHLSL(rx::VertexShaderD3D *shader) const +{ + std::string varyingSemantic = getVaryingSemantic(shader->mUsesPointSize); + std::string varyingHLSL; + + for (unsigned int varyingIndex = 0; varyingIndex < shader->mVaryings.size(); varyingIndex++) + { + const PackedVarying &varying = shader->mVaryings[varyingIndex]; + if (varying.registerAssigned()) + { + GLenum transposedType = TransposeMatrixType(varying.type); + int variableRows = (varying.isStruct() ? 1 : VariableRowCount(transposedType)); + + for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++) + { + for (int row = 0; row < variableRows; row++) + { + switch (varying.interpolation) + { + case sh::INTERPOLATION_SMOOTH: varyingHLSL += " "; break; + case sh::INTERPOLATION_FLAT: varyingHLSL += " nointerpolation "; break; + case sh::INTERPOLATION_CENTROID: varyingHLSL += " centroid "; break; + default: UNREACHABLE(); + } + + unsigned int semanticIndex = elementIndex * variableRows + varying.registerIndex + row; + std::string n = Str(semanticIndex); + + std::string typeString; + + if (varying.isStruct()) + { + // matrices within structs are not transposed, so + // do not use the special struct prefix "rm" + typeString = decorateVariable(varying.structName); + } + else + { + GLenum componentType = VariableComponentType(transposedType); + int columnCount = VariableColumnCount(transposedType); + typeString = gl_d3d::HLSLComponentTypeString(componentType, columnCount); + } + varyingHLSL += typeString + " v" + n + " : " + varyingSemantic + n + ";\n"; + } + } + } + } + + return varyingHLSL; +} + +std::string DynamicHLSL::generateVertexShaderForInputLayout(const std::string &sourceShader, + const VertexFormat inputLayout[], + const sh::Attribute shaderAttributes[]) const +{ + std::string structHLSL, initHLSL; + + int semanticIndex = 0; + unsigned int inputIndex = 0; + + for (unsigned int attributeIndex = 0; attributeIndex < MAX_VERTEX_ATTRIBS; attributeIndex++) + { + ASSERT(inputIndex < MAX_VERTEX_ATTRIBS); + + const VertexFormat &vertexFormat = inputLayout[inputIndex]; + const sh::Attribute &shaderAttribute = shaderAttributes[attributeIndex]; + + if (!shaderAttribute.name.empty()) + { + // HLSL code for input structure + if (IsMatrixType(shaderAttribute.type)) + { + // Matrix types are always transposed + structHLSL += " " + gl_d3d::HLSLMatrixTypeString(TransposeMatrixType(shaderAttribute.type)); + } + else + { + GLenum componentType = mRenderer->getVertexComponentType(vertexFormat); + structHLSL += " " + gl_d3d::HLSLComponentTypeString(componentType, VariableComponentCount(shaderAttribute.type)); + } + + structHLSL += " " + decorateVariable(shaderAttribute.name) + " : TEXCOORD" + Str(semanticIndex) + ";\n"; + semanticIndex += VariableRegisterCount(shaderAttribute.type); + + // HLSL code for initialization + initHLSL += " " + decorateVariable(shaderAttribute.name) + " = "; + + // Mismatched vertex attribute to vertex input may result in an undefined + // data reinterpretation (eg for pure integer->float, float->pure integer) + // TODO: issue warning with gl debug info extension, when supported + if (IsMatrixType(shaderAttribute.type) || + (mRenderer->getVertexConversionType(vertexFormat) & rx::VERTEX_CONVERT_GPU) != 0) + { + initHLSL += generateAttributeConversionHLSL(vertexFormat, shaderAttribute); + } + else + { + initHLSL += "input." + decorateVariable(shaderAttribute.name); + } + + initHLSL += ";\n"; + + inputIndex += VariableRowCount(TransposeMatrixType(shaderAttribute.type)); + } + } + + std::string replacementHLSL = "struct VS_INPUT\n" + "{\n" + + structHLSL + + "};\n" + "\n" + "void initAttributes(VS_INPUT input)\n" + "{\n" + + initHLSL + + "}\n"; + + std::string vertexHLSL(sourceShader); + + size_t copyInsertionPos = vertexHLSL.find(VERTEX_ATTRIBUTE_STUB_STRING); + vertexHLSL.replace(copyInsertionPos, VERTEX_ATTRIBUTE_STUB_STRING.length(), replacementHLSL); + + return vertexHLSL; +} + +std::string DynamicHLSL::generatePixelShaderForOutputSignature(const std::string &sourceShader, const std::vector &outputVariables, + bool usesFragDepth, const std::vector &outputLayout) const +{ + const int shaderModel = mRenderer->getMajorShaderModel(); + std::string targetSemantic = (shaderModel >= 4) ? "SV_TARGET" : "COLOR"; + std::string depthSemantic = (shaderModel >= 4) ? "SV_Depth" : "DEPTH"; + + std::string declarationHLSL; + std::string copyHLSL; + for (size_t i = 0; i < outputVariables.size(); i++) + { + const PixelShaderOuputVariable& outputVariable = outputVariables[i]; + ASSERT(outputLayout.size() > outputVariable.outputIndex); + + // FIXME(geofflang): Work around NVIDIA driver bug by repacking buffers + bool outputIndexEnabled = true; // outputLayout[outputVariable.outputIndex] != GL_NONE + if (outputIndexEnabled) + { + declarationHLSL += " " + gl_d3d::HLSLTypeString(outputVariable.type) + " " + outputVariable.name + + " : " + targetSemantic + Str(outputVariable.outputIndex) + ";\n"; + + copyHLSL += " output." + outputVariable.name + " = " + outputVariable.source + ";\n"; + } + } + + if (usesFragDepth) + { + declarationHLSL += " float gl_Depth : " + depthSemantic + ";\n"; + copyHLSL += " output.gl_Depth = gl_Depth; \n"; + } + + std::string replacementHLSL = "struct PS_OUTPUT\n" + "{\n" + + declarationHLSL + + "};\n" + "\n" + "PS_OUTPUT generateOutput()\n" + "{\n" + " PS_OUTPUT output;\n" + + copyHLSL + + " return output;\n" + "}\n"; + + std::string pixelHLSL(sourceShader); + + size_t outputInsertionPos = pixelHLSL.find(PIXEL_OUTPUT_STUB_STRING); + pixelHLSL.replace(outputInsertionPos, PIXEL_OUTPUT_STUB_STRING.length(), replacementHLSL); + + return pixelHLSL; +} + +std::string DynamicHLSL::getVaryingSemantic(bool pointSize) const +{ + // SM3 reserves the TEXCOORD semantic for point sprite texcoords (gl_PointCoord) + // In D3D11 we manually compute gl_PointCoord in the GS. + int shaderModel = mRenderer->getMajorShaderModel(); + return ((pointSize && shaderModel < 4) ? "COLOR" : "TEXCOORD"); +} + +struct DynamicHLSL::SemanticInfo +{ + struct BuiltinInfo + { + BuiltinInfo() + : enabled(false), + index(0), + systemValue(false) + {} + + bool enabled; + std::string semantic; + unsigned int index; + bool systemValue; + + std::string str() const + { + return (systemValue ? semantic : (semantic + Str(index))); + } + + void enableSystem(const std::string &systemValueSemantic) + { + enabled = true; + semantic = systemValueSemantic; + systemValue = true; + } + + void enable(const std::string &semanticVal, unsigned int indexVal) + { + enabled = true; + semantic = semanticVal; + index = indexVal; + } + }; + + BuiltinInfo dxPosition; + BuiltinInfo glPosition; + BuiltinInfo glFragCoord; + BuiltinInfo glPointCoord; + BuiltinInfo glPointSize; +}; + +DynamicHLSL::SemanticInfo DynamicHLSL::getSemanticInfo(int startRegisters, bool fragCoord, bool pointCoord, + bool pointSize, bool pixelShader) const +{ + SemanticInfo info; + bool hlsl4 = (mRenderer->getMajorShaderModel() >= 4); + const std::string &varyingSemantic = getVaryingSemantic(pointSize); + + int reservedRegisterIndex = startRegisters; + + if (hlsl4) + { + info.dxPosition.enableSystem("SV_Position"); + } + else if (pixelShader) + { + info.dxPosition.enableSystem("VPOS"); + } + else + { + info.dxPosition.enableSystem("POSITION"); + } + + info.glPosition.enable(varyingSemantic, reservedRegisterIndex++); + + if (fragCoord) + { + info.glFragCoord.enable(varyingSemantic, reservedRegisterIndex++); + } + + if (pointCoord) + { + // SM3 reserves the TEXCOORD semantic for point sprite texcoords (gl_PointCoord) + // In D3D11 we manually compute gl_PointCoord in the GS. + if (hlsl4) + { + info.glPointCoord.enable(varyingSemantic, reservedRegisterIndex++); + } + else + { + info.glPointCoord.enable("TEXCOORD", 0); + } + } + + // Special case: do not include PSIZE semantic in HLSL 3 pixel shaders + if (pointSize && (!pixelShader || hlsl4)) + { + info.glPointSize.enableSystem("PSIZE"); + } + + return info; +} + +std::string DynamicHLSL::generateVaryingLinkHLSL(const SemanticInfo &info, const std::string &varyingHLSL) const +{ + std::string linkHLSL = "{\n"; + + ASSERT(info.dxPosition.enabled && info.glPosition.enabled); + + linkHLSL += " float4 dx_Position : " + info.dxPosition.str() + ";\n"; + linkHLSL += " float4 gl_Position : " + info.glPosition.str() + ";\n"; + + if (info.glFragCoord.enabled) + { + linkHLSL += " float4 gl_FragCoord : " + info.glFragCoord.str() + ";\n"; + } + + if (info.glPointCoord.enabled) + { + linkHLSL += " float2 gl_PointCoord : " + info.glPointCoord.str() + ";\n"; + } + + linkHLSL += varyingHLSL; + + if (info.glPointSize.enabled) + { + linkHLSL += " float gl_PointSize : " + info.glPointSize.str() + ";\n"; + } + + linkHLSL += "};\n"; + + return linkHLSL; +} + +void DynamicHLSL::storeBuiltinLinkedVaryings(const SemanticInfo &info, + std::vector *linkedVaryings) const +{ + ASSERT(info.glPosition.enabled); + + linkedVaryings->push_back(LinkedVarying("gl_Position", GL_FLOAT_VEC4, 1, info.glPosition.semantic, + info.glPosition.index, 1)); + + if (info.glFragCoord.enabled) + { + linkedVaryings->push_back(LinkedVarying("gl_FragCoord", GL_FLOAT_VEC4, 1, info.glFragCoord.semantic, + info.glFragCoord.index, 1)); + } + + if (info.glPointSize.enabled) + { + linkedVaryings->push_back(LinkedVarying("gl_PointSize", GL_FLOAT, 1, "PSIZE", 0, 1)); + } +} + +void DynamicHLSL::storeUserLinkedVaryings(const rx::VertexShaderD3D *vertexShader, + std::vector *linkedVaryings) const +{ + const std::string &varyingSemantic = getVaryingSemantic(vertexShader->mUsesPointSize); + const std::vector &varyings = vertexShader->mVaryings; + + for (unsigned int varyingIndex = 0; varyingIndex < varyings.size(); varyingIndex++) + { + const PackedVarying &varying = varyings[varyingIndex]; + if (varying.registerAssigned()) + { + GLenum transposedType = TransposeMatrixType(varying.type); + int variableRows = (varying.isStruct() ? 1 : VariableRowCount(transposedType)); + + linkedVaryings->push_back(LinkedVarying(varying.name, varying.type, varying.elementCount(), + varyingSemantic, varying.registerIndex, + variableRows * varying.elementCount())); + } + } +} + +bool DynamicHLSL::generateShaderLinkHLSL(InfoLog &infoLog, int registers, const VaryingPacking packing, + std::string& pixelHLSL, std::string& vertexHLSL, + rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader, + const std::vector& transformFeedbackVaryings, + std::vector *linkedVaryings, + std::map *programOutputVars, + std::vector *outPixelShaderKey, + bool *outUsesFragDepth) const +{ + if (pixelHLSL.empty() || vertexHLSL.empty()) + { + return false; + } + + bool usesMRT = fragmentShader->mUsesMultipleRenderTargets; + bool usesFragColor = fragmentShader->mUsesFragColor; + bool usesFragData = fragmentShader->mUsesFragData; + bool usesFragCoord = fragmentShader->mUsesFragCoord; + bool usesPointCoord = fragmentShader->mUsesPointCoord; + bool usesPointSize = vertexShader->mUsesPointSize; + + if (usesFragColor && usesFragData) + { + infoLog.append("Cannot use both gl_FragColor and gl_FragData in the same fragment shader."); + return false; + } + + // Write the HLSL input/output declarations + const int shaderModel = mRenderer->getMajorShaderModel(); + const int maxVaryingVectors = mRenderer->getMaxVaryingVectors(); + + const int registersNeeded = registers + (usesFragCoord ? 1 : 0) + (usesPointCoord ? 1 : 0); + + // Two cases when writing to gl_FragColor and using ESSL 1.0: + // - with a 3.0 context, the output color is copied to channel 0 + // - with a 2.0 context, the output color is broadcast to all channels + const bool broadcast = (fragmentShader->mUsesFragColor && mRenderer->getCurrentClientVersion() < 3); + const unsigned int numRenderTargets = (broadcast || usesMRT ? mRenderer->getRendererCaps().maxDrawBuffers : 1); + + int shaderVersion = vertexShader->getShaderVersion(); + + if (registersNeeded > maxVaryingVectors) + { + infoLog.append("No varying registers left to support gl_FragCoord/gl_PointCoord"); + return false; + } + + const std::string &varyingHLSL = generateVaryingHLSL(vertexShader); + const SemanticInfo &vertexSemantics = getSemanticInfo(registers, usesFragCoord, + false, usesPointSize, false); + + storeUserLinkedVaryings(vertexShader, linkedVaryings); + storeBuiltinLinkedVaryings(vertexSemantics, linkedVaryings); + + // Add stub string to be replaced when shader is dynamically defined by its layout + vertexHLSL += "\n" + VERTEX_ATTRIBUTE_STUB_STRING + "\n" + "struct VS_OUTPUT\n" + generateVaryingLinkHLSL(vertexSemantics, varyingHLSL) + "\n" + "VS_OUTPUT main(VS_INPUT input)\n" + "{\n" + " initAttributes(input);\n"; + + if (shaderModel >= 4) + { + vertexHLSL += "\n" + " gl_main();\n" + "\n" + " VS_OUTPUT output;\n" + " output.gl_Position = gl_Position;\n" + " output.dx_Position.x = gl_Position.x;\n" + " output.dx_Position.y = -gl_Position.y;\n" + " output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n" + " output.dx_Position.w = gl_Position.w;\n"; + } + else + { + vertexHLSL += "\n" + " gl_main();\n" + "\n" + " VS_OUTPUT output;\n" + " output.gl_Position = gl_Position;\n" + " output.dx_Position.x = gl_Position.x * dx_ViewAdjust.z + dx_ViewAdjust.x * gl_Position.w;\n" + " output.dx_Position.y = -(gl_Position.y * dx_ViewAdjust.w + dx_ViewAdjust.y * gl_Position.w);\n" + " output.dx_Position.z = (gl_Position.z + gl_Position.w) * 0.5;\n" + " output.dx_Position.w = gl_Position.w;\n"; + } + + if (usesPointSize && shaderModel >= 3) + { + vertexHLSL += " output.gl_PointSize = gl_PointSize;\n"; + } + + if (usesFragCoord) + { + vertexHLSL += " output.gl_FragCoord = gl_Position;\n"; + } + + for (unsigned int vertVaryingIndex = 0; vertVaryingIndex < vertexShader->mVaryings.size(); vertVaryingIndex++) + { + const PackedVarying &varying = vertexShader->mVaryings[vertVaryingIndex]; + if (varying.registerAssigned()) + { + for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++) + { + int variableRows = (varying.isStruct() ? 1 : VariableRowCount(TransposeMatrixType(varying.type))); + + for (int row = 0; row < variableRows; row++) + { + int r = varying.registerIndex + elementIndex * variableRows + row; + vertexHLSL += " output.v" + Str(r); + + bool sharedRegister = false; // Register used by multiple varyings + + for (int x = 0; x < 4; x++) + { + if (packing[r][x] && packing[r][x] != packing[r][0]) + { + sharedRegister = true; + break; + } + } + + if(sharedRegister) + { + vertexHLSL += "."; + + for (int x = 0; x < 4; x++) + { + if (packing[r][x] == &varying) + { + switch(x) + { + case 0: vertexHLSL += "x"; break; + case 1: vertexHLSL += "y"; break; + case 2: vertexHLSL += "z"; break; + case 3: vertexHLSL += "w"; break; + } + } + } + } + + vertexHLSL += " = _" + varying.name; + + if (varying.isArray()) + { + vertexHLSL += ArrayString(elementIndex); + } + + if (variableRows > 1) + { + vertexHLSL += ArrayString(row); + } + + vertexHLSL += ";\n"; + } + } + } + } + + vertexHLSL += "\n" + " return output;\n" + "}\n"; + + const SemanticInfo &pixelSemantics = getSemanticInfo(registers, usesFragCoord, usesPointCoord, + usesPointSize, true); + + pixelHLSL += "struct PS_INPUT\n" + generateVaryingLinkHLSL(pixelSemantics, varyingHLSL) + "\n"; + + if (shaderVersion < 300) + { + for (unsigned int renderTargetIndex = 0; renderTargetIndex < numRenderTargets; renderTargetIndex++) + { + PixelShaderOuputVariable outputKeyVariable; + outputKeyVariable.type = GL_FLOAT_VEC4; + outputKeyVariable.name = "gl_Color" + Str(renderTargetIndex); + outputKeyVariable.source = broadcast ? "gl_Color[0]" : "gl_Color[" + Str(renderTargetIndex) + "]"; + outputKeyVariable.outputIndex = renderTargetIndex; + + outPixelShaderKey->push_back(outputKeyVariable); + } + + *outUsesFragDepth = fragmentShader->mUsesFragDepth; + } + else + { + defineOutputVariables(fragmentShader, programOutputVars); + + const std::vector &shaderOutputVars = fragmentShader->getOutputVariables(); + for (auto locationIt = programOutputVars->begin(); locationIt != programOutputVars->end(); locationIt++) + { + const VariableLocation &outputLocation = locationIt->second; + const sh::ShaderVariable &outputVariable = shaderOutputVars[outputLocation.index]; + const std::string &variableName = "out_" + outputLocation.name; + const std::string &elementString = (outputLocation.element == GL_INVALID_INDEX ? "" : Str(outputLocation.element)); + + PixelShaderOuputVariable outputKeyVariable; + outputKeyVariable.type = outputVariable.type; + outputKeyVariable.name = variableName + elementString; + outputKeyVariable.source = variableName + ArrayString(outputLocation.element); + outputKeyVariable.outputIndex = locationIt->first; + + outPixelShaderKey->push_back(outputKeyVariable); + } + + *outUsesFragDepth = false; + } + + pixelHLSL += PIXEL_OUTPUT_STUB_STRING + "\n"; + + if (fragmentShader->mUsesFrontFacing) + { + if (shaderModel >= 4) + { + pixelHLSL += "PS_OUTPUT main(PS_INPUT input, bool isFrontFace : SV_IsFrontFace)\n" + "{\n"; + } + else + { + pixelHLSL += "PS_OUTPUT main(PS_INPUT input, float vFace : VFACE)\n" + "{\n"; + } + } + else + { + pixelHLSL += "PS_OUTPUT main(PS_INPUT input)\n" + "{\n"; + } + + if (usesFragCoord) + { + pixelHLSL += " float rhw = 1.0 / input.gl_FragCoord.w;\n"; + + if (shaderModel >= 4) + { + pixelHLSL += " gl_FragCoord.x = input.dx_Position.x;\n" + " gl_FragCoord.y = input.dx_Position.y;\n"; + } + else if (shaderModel >= 3) + { + pixelHLSL += " gl_FragCoord.x = input.dx_Position.x + 0.5;\n" + " gl_FragCoord.y = input.dx_Position.y + 0.5;\n"; + } + else + { + // dx_ViewCoords contains the viewport width/2, height/2, center.x and center.y. See Renderer::setViewport() + pixelHLSL += " gl_FragCoord.x = (input.gl_FragCoord.x * rhw) * dx_ViewCoords.x + dx_ViewCoords.z;\n" + " gl_FragCoord.y = (input.gl_FragCoord.y * rhw) * dx_ViewCoords.y + dx_ViewCoords.w;\n"; + } + + pixelHLSL += " gl_FragCoord.z = (input.gl_FragCoord.z * rhw) * dx_DepthFront.x + dx_DepthFront.y;\n" + " gl_FragCoord.w = rhw;\n"; + } + + if (usesPointCoord && shaderModel >= 3) + { + pixelHLSL += " gl_PointCoord.x = input.gl_PointCoord.x;\n"; + pixelHLSL += " gl_PointCoord.y = 1.0 - input.gl_PointCoord.y;\n"; + } + + if (fragmentShader->mUsesFrontFacing) + { + if (shaderModel <= 3) + { + pixelHLSL += " gl_FrontFacing = (vFace * dx_DepthFront.z >= 0.0);\n"; + } + else + { + pixelHLSL += " gl_FrontFacing = isFrontFace;\n"; + } + } + + for (unsigned int varyingIndex = 0; varyingIndex < fragmentShader->mVaryings.size(); varyingIndex++) + { + const PackedVarying &varying = fragmentShader->mVaryings[varyingIndex]; + if (varying.registerAssigned()) + { + for (unsigned int elementIndex = 0; elementIndex < varying.elementCount(); elementIndex++) + { + GLenum transposedType = TransposeMatrixType(varying.type); + int variableRows = (varying.isStruct() ? 1 : VariableRowCount(transposedType)); + for (int row = 0; row < variableRows; row++) + { + std::string n = Str(varying.registerIndex + elementIndex * variableRows + row); + pixelHLSL += " _" + varying.name; + + if (varying.isArray()) + { + pixelHLSL += ArrayString(elementIndex); + } + + if (variableRows > 1) + { + pixelHLSL += ArrayString(row); + } + + if (varying.isStruct()) + { + pixelHLSL += " = input.v" + n + ";\n"; break; + } + else + { + switch (VariableColumnCount(transposedType)) + { + case 1: pixelHLSL += " = input.v" + n + ".x;\n"; break; + case 2: pixelHLSL += " = input.v" + n + ".xy;\n"; break; + case 3: pixelHLSL += " = input.v" + n + ".xyz;\n"; break; + case 4: pixelHLSL += " = input.v" + n + ";\n"; break; + default: UNREACHABLE(); + } + } + } + } + } + else UNREACHABLE(); + } + + pixelHLSL += "\n" + " gl_main();\n" + "\n" + " return generateOutput();\n" + "}\n"; + + return true; +} + +void DynamicHLSL::defineOutputVariables(rx::FragmentShaderD3D *fragmentShader, std::map *programOutputVars) const +{ + const std::vector &shaderOutputVars = fragmentShader->getOutputVariables(); + + for (unsigned int outputVariableIndex = 0; outputVariableIndex < shaderOutputVars.size(); outputVariableIndex++) + { + const sh::Attribute &outputVariable = shaderOutputVars[outputVariableIndex]; + const int baseLocation = outputVariable.location == -1 ? 0 : outputVariable.location; + + if (outputVariable.arraySize > 0) + { + for (unsigned int elementIndex = 0; elementIndex < outputVariable.arraySize; elementIndex++) + { + const int location = baseLocation + elementIndex; + ASSERT(programOutputVars->count(location) == 0); + (*programOutputVars)[location] = VariableLocation(outputVariable.name, elementIndex, outputVariableIndex); + } + } + else + { + ASSERT(programOutputVars->count(baseLocation) == 0); + (*programOutputVars)[baseLocation] = VariableLocation(outputVariable.name, GL_INVALID_INDEX, outputVariableIndex); + } + } +} + +std::string DynamicHLSL::generateGeometryShaderHLSL(int registers, rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader) const +{ + // for now we only handle point sprite emulation + ASSERT(vertexShader->mUsesPointSize && mRenderer->getMajorShaderModel() >= 4); + return generatePointSpriteHLSL(registers, fragmentShader, vertexShader); +} + +std::string DynamicHLSL::generatePointSpriteHLSL(int registers, rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader) const +{ + ASSERT(registers >= 0); + ASSERT(vertexShader->mUsesPointSize); + ASSERT(mRenderer->getMajorShaderModel() >= 4); + + std::string geomHLSL; + + const SemanticInfo &inSemantics = getSemanticInfo(registers, fragmentShader->mUsesFragCoord, + false, true, false); + const SemanticInfo &outSemantics = getSemanticInfo(registers, fragmentShader->mUsesFragCoord, + fragmentShader->mUsesPointCoord, true, false); + + std::string varyingHLSL = generateVaryingHLSL(vertexShader); + std::string inLinkHLSL = generateVaryingLinkHLSL(inSemantics, varyingHLSL); + std::string outLinkHLSL = generateVaryingLinkHLSL(outSemantics, varyingHLSL); + + // TODO(geofflang): use context's caps + geomHLSL += "uniform float4 dx_ViewCoords : register(c1);\n" + "\n" + "struct GS_INPUT\n" + inLinkHLSL + "\n" + + "struct GS_OUTPUT\n" + outLinkHLSL + "\n" + + "\n" + "static float2 pointSpriteCorners[] = \n" + "{\n" + " float2( 0.5f, -0.5f),\n" + " float2( 0.5f, 0.5f),\n" + " float2(-0.5f, -0.5f),\n" + " float2(-0.5f, 0.5f)\n" + "};\n" + "\n" + "static float2 pointSpriteTexcoords[] = \n" + "{\n" + " float2(1.0f, 1.0f),\n" + " float2(1.0f, 0.0f),\n" + " float2(0.0f, 1.0f),\n" + " float2(0.0f, 0.0f)\n" + "};\n" + "\n" + "static float minPointSize = " + Str(mRenderer->getRendererCaps().minAliasedPointSize) + ".0f;\n" + "static float maxPointSize = " + Str(mRenderer->getRendererCaps().maxAliasedPointSize) + ".0f;\n" + "\n" + "[maxvertexcount(4)]\n" + "void main(point GS_INPUT input[1], inout TriangleStream outStream)\n" + "{\n" + " GS_OUTPUT output = (GS_OUTPUT)0;\n" + " output.gl_Position = input[0].gl_Position;\n"; + " output.gl_PointSize = input[0].gl_PointSize;\n"; + + for (int r = 0; r < registers; r++) + { + geomHLSL += " output.v" + Str(r) + " = input[0].v" + Str(r) + ";\n"; + } + + if (fragmentShader->mUsesFragCoord) + { + geomHLSL += " output.gl_FragCoord = input[0].gl_FragCoord;\n"; + } + + geomHLSL += " \n" + " float gl_PointSize = clamp(input[0].gl_PointSize, minPointSize, maxPointSize);\n" + " float4 dx_Position = input[0].dx_Position;\n" + " float2 viewportScale = float2(1.0f / dx_ViewCoords.x, 1.0f / dx_ViewCoords.y) * dx_Position.w;\n"; + + for (int corner = 0; corner < 4; corner++) + { + geomHLSL += " \n" + " output.dx_Position = dx_Position + float4(pointSpriteCorners[" + Str(corner) + "] * viewportScale * gl_PointSize, 0.0f, 0.0f);\n"; + + if (fragmentShader->mUsesPointCoord) + { + geomHLSL += " output.gl_PointCoord = pointSpriteTexcoords[" + Str(corner) + "];\n"; + } + + geomHLSL += " outStream.Append(output);\n"; + } + + geomHLSL += " \n" + " outStream.RestartStrip();\n" + "}\n"; + + return geomHLSL; +} + +// This method needs to match OutputHLSL::decorate +std::string DynamicHLSL::decorateVariable(const std::string &name) +{ + if (name.compare(0, 3, "gl_") != 0) + { + return "_" + name; + } + + return name; +} + +std::string DynamicHLSL::generateAttributeConversionHLSL(const VertexFormat &vertexFormat, const sh::ShaderVariable &shaderAttrib) const +{ + std::string attribString = "input." + decorateVariable(shaderAttrib.name); + + // Matrix + if (IsMatrixType(shaderAttrib.type)) + { + return "transpose(" + attribString + ")"; + } + + GLenum shaderComponentType = VariableComponentType(shaderAttrib.type); + int shaderComponentCount = VariableComponentCount(shaderAttrib.type); + + // Perform integer to float conversion (if necessary) + bool requiresTypeConversion = (shaderComponentType == GL_FLOAT && vertexFormat.mType != GL_FLOAT); + + if (requiresTypeConversion) + { + // TODO: normalization for 32-bit integer formats + ASSERT(!vertexFormat.mNormalized && !vertexFormat.mPureInteger); + return "float" + Str(shaderComponentCount) + "(" + attribString + ")"; + } + + // No conversion necessary + return attribString; +} + +void DynamicHLSL::getInputLayoutSignature(const VertexFormat inputLayout[], GLenum signature[]) const +{ + for (size_t inputIndex = 0; inputIndex < MAX_VERTEX_ATTRIBS; inputIndex++) + { + const VertexFormat &vertexFormat = inputLayout[inputIndex]; + + if (vertexFormat.mType == GL_NONE) + { + signature[inputIndex] = GL_NONE; + } + else + { + bool gpuConverted = ((mRenderer->getVertexConversionType(vertexFormat) & rx::VERTEX_CONVERT_GPU) != 0); + signature[inputIndex] = (gpuConverted ? GL_TRUE : GL_FALSE); + } + } +} + +} diff --git a/src/libGLESv2/renderer/d3d/DynamicHLSL.h b/src/libGLESv2/renderer/d3d/DynamicHLSL.h new file mode 100644 index 00000000..494f6782 --- /dev/null +++ b/src/libGLESv2/renderer/d3d/DynamicHLSL.h @@ -0,0 +1,96 @@ +// +// Copyright (c) 2014 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. +// +// DynamicHLSL.h: Interface for link and run-time HLSL generation +// + +#ifndef LIBGLESV2_RENDERER_DYNAMIC_HLSL_H_ +#define LIBGLESV2_RENDERER_DYNAMIC_HLSL_H_ + +#include "common/angleutils.h" +#include "libGLESv2/constants.h" + +namespace sh +{ +struct Attribute; +struct ShaderVariable; +} + +namespace gl +{ +class InfoLog; +struct VariableLocation; +struct LinkedVarying; +struct VertexAttribute; +struct VertexFormat; +struct PackedVarying; +} + +namespace rx +{ +class Renderer; +class ShaderD3D; +class VertexShaderD3D; +class FragmentShaderD3D; + +typedef const gl::PackedVarying *VaryingPacking[gl::IMPLEMENTATION_MAX_VARYING_VECTORS][4]; + +struct PixelShaderOuputVariable +{ + GLenum type; + std::string name; + std::string source; + size_t outputIndex; +}; + +class DynamicHLSL +{ + public: + explicit DynamicHLSL(rx::Renderer *const renderer); + + int packVaryings(gl::InfoLog &infoLog, VaryingPacking packing, rx::FragmentShaderD3D *fragmentShader, + rx::VertexShaderD3D *vertexShader, const std::vector& transformFeedbackVaryings); + std::string generateVertexShaderForInputLayout(const std::string &sourceShader, const gl::VertexFormat inputLayout[], + const sh::Attribute shaderAttributes[]) const; + std::string generatePixelShaderForOutputSignature(const std::string &sourceShader, const std::vector &outputVariables, + bool usesFragDepth, const std::vector &outputLayout) const; + bool generateShaderLinkHLSL(gl::InfoLog &infoLog, int registers, const VaryingPacking packing, + std::string& pixelHLSL, std::string& vertexHLSL, + rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader, + const std::vector& transformFeedbackVaryings, + std::vector *linkedVaryings, + std::map *programOutputVars, + std::vector *outPixelShaderKey, + bool *outUsesFragDepth) const; + + std::string generateGeometryShaderHLSL(int registers, rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader) const; + void getInputLayoutSignature(const gl::VertexFormat inputLayout[], GLenum signature[]) const; + + private: + DISALLOW_COPY_AND_ASSIGN(DynamicHLSL); + + rx::Renderer *const mRenderer; + + struct SemanticInfo; + + std::string getVaryingSemantic(bool pointSize) const; + SemanticInfo getSemanticInfo(int startRegisters, bool fragCoord, bool pointCoord, bool pointSize, + bool pixelShader) const; + std::string generateVaryingLinkHLSL(const SemanticInfo &info, const std::string &varyingHLSL) const; + std::string generateVaryingHLSL(rx::VertexShaderD3D *shader) const; + void storeUserLinkedVaryings(const rx::VertexShaderD3D *vertexShader, std::vector *linkedVaryings) const; + void storeBuiltinLinkedVaryings(const SemanticInfo &info, std::vector *linkedVaryings) const; + void defineOutputVariables(rx::FragmentShaderD3D *fragmentShader, std::map *programOutputVars) const; + std::string generatePointSpriteHLSL(int registers, rx::FragmentShaderD3D *fragmentShader, rx::VertexShaderD3D *vertexShader) const; + + // Prepend an underscore + static std::string decorateVariable(const std::string &name); + + std::string generateAttributeConversionHLSL(const gl::VertexFormat &vertexFormat, const sh::ShaderVariable &shaderAttrib) const; +}; + +} + +#endif // LIBGLESV2_RENDERER_DYNAMIC_HLSL_H_ diff --git a/src/libGLESv2/renderer/d3d/ShaderD3D.h b/src/libGLESv2/renderer/d3d/ShaderD3D.h index ce856eac..cc0ef81a 100644 --- a/src/libGLESv2/renderer/d3d/ShaderD3D.h +++ b/src/libGLESv2/renderer/d3d/ShaderD3D.h @@ -12,18 +12,13 @@ #include "libGLESv2/Shader.h" #include "libGLESv2/renderer/ShaderImpl.h" -namespace gl -{ -class DynamicHLSL; -} - namespace rx { class Renderer; class ShaderD3D : public ShaderImpl { - friend class gl::DynamicHLSL; + friend class DynamicHLSL; public: ShaderD3D(rx::Renderer *renderer); @@ -95,7 +90,7 @@ class ShaderD3D : public ShaderImpl class VertexShaderD3D : public ShaderD3D { - friend class gl::DynamicHLSL; + friend class DynamicHLSL; public: VertexShaderD3D(rx::Renderer *renderer); @@ -120,7 +115,7 @@ class VertexShaderD3D : public ShaderD3D class FragmentShaderD3D : public ShaderD3D { - friend class gl::DynamicHLSL; + friend class DynamicHLSL; public: FragmentShaderD3D(rx::Renderer *renderer); -- cgit v1.2.3