// // Copyright (c) 2002-2014 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. // // // Build the intermediate representation. // #include #include #include #include #include #include #include "common/mathutil.h" #include "common/matrix_utils.h" #include "compiler/translator/Diagnostics.h" #include "compiler/translator/IntermNode.h" #include "compiler/translator/SymbolTable.h" #include "compiler/translator/util.h" namespace sh { namespace { const float kPi = 3.14159265358979323846f; const float kDegreesToRadiansMultiplier = kPi / 180.0f; const float kRadiansToDegreesMultiplier = 180.0f / kPi; TPrecision GetHigherPrecision(TPrecision left, TPrecision right) { return left > right ? left : right; } TConstantUnion *Vectorize(const TConstantUnion &constant, size_t size) { TConstantUnion *constUnion = new TConstantUnion[size]; for (unsigned int i = 0; i < size; ++i) constUnion[i] = constant; return constUnion; } void UndefinedConstantFoldingError(const TSourceLoc &loc, TOperator op, TBasicType basicType, TDiagnostics *diagnostics, TConstantUnion *result) { diagnostics->warning(loc, "operation result is undefined for the values passed in", GetOperatorString(op)); switch (basicType) { case EbtFloat: result->setFConst(0.0f); break; case EbtInt: result->setIConst(0); break; case EbtUInt: result->setUConst(0u); break; case EbtBool: result->setBConst(false); break; default: break; } } float VectorLength(const TConstantUnion *paramArray, size_t paramArraySize) { float result = 0.0f; for (size_t i = 0; i < paramArraySize; i++) { float f = paramArray[i].getFConst(); result += f * f; } return sqrtf(result); } float VectorDotProduct(const TConstantUnion *paramArray1, const TConstantUnion *paramArray2, size_t paramArraySize) { float result = 0.0f; for (size_t i = 0; i < paramArraySize; i++) result += paramArray1[i].getFConst() * paramArray2[i].getFConst(); return result; } TIntermTyped *CreateFoldedNode(const TConstantUnion *constArray, const TIntermTyped *originalNode, TQualifier qualifier) { if (constArray == nullptr) { return nullptr; } TIntermTyped *folded = new TIntermConstantUnion(constArray, originalNode->getType()); folded->getTypePointer()->setQualifier(qualifier); folded->setLine(originalNode->getLine()); return folded; } angle::Matrix GetMatrix(const TConstantUnion *paramArray, const unsigned int &rows, const unsigned int &cols) { std::vector elements; for (size_t i = 0; i < rows * cols; i++) elements.push_back(paramArray[i].getFConst()); // Transpose is used since the Matrix constructor expects arguments in row-major order, // whereas the paramArray is in column-major order. Rows/cols parameters are also flipped below // so that the created matrix will have the expected dimensions after the transpose. return angle::Matrix(elements, cols, rows).transpose(); } angle::Matrix GetMatrix(const TConstantUnion *paramArray, const unsigned int &size) { std::vector elements; for (size_t i = 0; i < size * size; i++) elements.push_back(paramArray[i].getFConst()); // Transpose is used since the Matrix constructor expects arguments in row-major order, // whereas the paramArray is in column-major order. return angle::Matrix(elements, size).transpose(); } void SetUnionArrayFromMatrix(const angle::Matrix &m, TConstantUnion *resultArray) { // Transpose is used since the input Matrix is in row-major order, // whereas the actual result should be in column-major order. angle::Matrix result = m.transpose(); std::vector resultElements = result.elements(); for (size_t i = 0; i < resultElements.size(); i++) resultArray[i].setFConst(resultElements[i]); } } // namespace anonymous //////////////////////////////////////////////////////////////// // // Member functions of the nodes used for building the tree. // //////////////////////////////////////////////////////////////// void TIntermTyped::setTypePreservePrecision(const TType &t) { TPrecision precision = getPrecision(); mType = t; ASSERT(mType.getBasicType() != EbtBool || precision == EbpUndefined); mType.setPrecision(precision); } #define REPLACE_IF_IS(node, type, original, replacement) \ if (node == original) \ { \ node = static_cast(replacement); \ return true; \ } bool TIntermLoop::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { ASSERT(original != nullptr); // This risks replacing multiple children. REPLACE_IF_IS(mInit, TIntermNode, original, replacement); REPLACE_IF_IS(mCond, TIntermTyped, original, replacement); REPLACE_IF_IS(mExpr, TIntermTyped, original, replacement); REPLACE_IF_IS(mBody, TIntermBlock, original, replacement); return false; } bool TIntermBranch::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mExpression, TIntermTyped, original, replacement); return false; } bool TIntermSwizzle::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { ASSERT(original->getAsTyped()->getType() == replacement->getAsTyped()->getType()); REPLACE_IF_IS(mOperand, TIntermTyped, original, replacement); return false; } bool TIntermBinary::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mLeft, TIntermTyped, original, replacement); REPLACE_IF_IS(mRight, TIntermTyped, original, replacement); return false; } bool TIntermUnary::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { ASSERT(original->getAsTyped()->getType() == replacement->getAsTyped()->getType()); REPLACE_IF_IS(mOperand, TIntermTyped, original, replacement); return false; } bool TIntermInvariantDeclaration::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mSymbol, TIntermSymbol, original, replacement); return false; } bool TIntermFunctionDefinition::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mPrototype, TIntermFunctionPrototype, original, replacement); REPLACE_IF_IS(mBody, TIntermBlock, original, replacement); return false; } bool TIntermAggregate::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { return replaceChildNodeInternal(original, replacement); } bool TIntermBlock::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { return replaceChildNodeInternal(original, replacement); } bool TIntermFunctionPrototype::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { return replaceChildNodeInternal(original, replacement); } bool TIntermDeclaration::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { return replaceChildNodeInternal(original, replacement); } bool TIntermAggregateBase::replaceChildNodeInternal(TIntermNode *original, TIntermNode *replacement) { for (size_t ii = 0; ii < getSequence()->size(); ++ii) { REPLACE_IF_IS((*getSequence())[ii], TIntermNode, original, replacement); } return false; } bool TIntermAggregateBase::replaceChildNodeWithMultiple(TIntermNode *original, const TIntermSequence &replacements) { for (auto it = getSequence()->begin(); it < getSequence()->end(); ++it) { if (*it == original) { it = getSequence()->erase(it); getSequence()->insert(it, replacements.begin(), replacements.end()); return true; } } return false; } bool TIntermAggregateBase::insertChildNodes(TIntermSequence::size_type position, const TIntermSequence &insertions) { if (position > getSequence()->size()) { return false; } auto it = getSequence()->begin() + position; getSequence()->insert(it, insertions.begin(), insertions.end()); return true; } TIntermAggregate *TIntermAggregate::CreateFunctionCall(const TFunction &func, TIntermSequence *arguments) { TIntermAggregate *callNode = new TIntermAggregate(func.getReturnType(), EOpCallFunctionInAST, arguments); callNode->getFunctionSymbolInfo()->setFromFunction(func); return callNode; } TIntermAggregate *TIntermAggregate::CreateFunctionCall(const TType &type, const TSymbolUniqueId &id, const TName &name, TIntermSequence *arguments) { TIntermAggregate *callNode = new TIntermAggregate(type, EOpCallFunctionInAST, arguments); callNode->getFunctionSymbolInfo()->setId(id); callNode->getFunctionSymbolInfo()->setNameObj(name); return callNode; } TIntermAggregate *TIntermAggregate::CreateBuiltInFunctionCall(const TFunction &func, TIntermSequence *arguments) { TIntermAggregate *callNode = new TIntermAggregate(func.getReturnType(), EOpCallBuiltInFunction, arguments); callNode->getFunctionSymbolInfo()->setFromFunction(func); // Note that name needs to be set before texture function type is determined. callNode->setBuiltInFunctionPrecision(); return callNode; } TIntermAggregate *TIntermAggregate::CreateConstructor(const TType &type, TIntermSequence *arguments) { return new TIntermAggregate(type, EOpConstruct, arguments); } TIntermAggregate *TIntermAggregate::Create(const TType &type, TOperator op, TIntermSequence *arguments) { TIntermAggregate *node = new TIntermAggregate(type, op, arguments); ASSERT(op != EOpCallFunctionInAST); // Should use CreateFunctionCall ASSERT(op != EOpCallBuiltInFunction); // Should use CreateBuiltInFunctionCall ASSERT(!node->isConstructor()); // Should use CreateConstructor return node; } TIntermAggregate::TIntermAggregate(const TType &type, TOperator op, TIntermSequence *arguments) : TIntermOperator(op), mUseEmulatedFunction(false), mGotPrecisionFromChildren(false) { if (arguments != nullptr) { mArguments.swap(*arguments); } setTypePrecisionAndQualifier(type); } void TIntermAggregate::setTypePrecisionAndQualifier(const TType &type) { setType(type); mType.setQualifier(EvqTemporary); if (!isFunctionCall()) { if (isConstructor()) { // Structs should not be precision qualified, the individual members may be. // Built-in types on the other hand should be precision qualified. if (getBasicType() != EbtStruct) { setPrecisionFromChildren(); } } else { setPrecisionForBuiltInOp(); } if (areChildrenConstQualified()) { mType.setQualifier(EvqConst); } } } bool TIntermAggregate::areChildrenConstQualified() { for (TIntermNode *&arg : mArguments) { TIntermTyped *typedArg = arg->getAsTyped(); if (typedArg && typedArg->getQualifier() != EvqConst) { return false; } } return true; } void TIntermAggregate::setPrecisionFromChildren() { mGotPrecisionFromChildren = true; if (getBasicType() == EbtBool) { mType.setPrecision(EbpUndefined); return; } TPrecision precision = EbpUndefined; TIntermSequence::iterator childIter = mArguments.begin(); while (childIter != mArguments.end()) { TIntermTyped *typed = (*childIter)->getAsTyped(); if (typed) precision = GetHigherPrecision(typed->getPrecision(), precision); ++childIter; } mType.setPrecision(precision); } void TIntermAggregate::setPrecisionForBuiltInOp() { ASSERT(!isConstructor()); ASSERT(!isFunctionCall()); if (!setPrecisionForSpecialBuiltInOp()) { setPrecisionFromChildren(); } } bool TIntermAggregate::setPrecisionForSpecialBuiltInOp() { switch (mOp) { case EOpBitfieldExtract: mType.setPrecision(mArguments[0]->getAsTyped()->getPrecision()); mGotPrecisionFromChildren = true; return true; case EOpBitfieldInsert: mType.setPrecision(GetHigherPrecision(mArguments[0]->getAsTyped()->getPrecision(), mArguments[1]->getAsTyped()->getPrecision())); mGotPrecisionFromChildren = true; return true; case EOpUaddCarry: case EOpUsubBorrow: mType.setPrecision(EbpHigh); return true; default: return false; } } void TIntermAggregate::setBuiltInFunctionPrecision() { // All built-ins returning bool should be handled as ops, not functions. ASSERT(getBasicType() != EbtBool); ASSERT(mOp == EOpCallBuiltInFunction); TPrecision precision = EbpUndefined; for (TIntermNode *arg : mArguments) { TIntermTyped *typed = arg->getAsTyped(); // ESSL spec section 8: texture functions get their precision from the sampler. if (typed && IsSampler(typed->getBasicType())) { precision = typed->getPrecision(); break; } } // ESSL 3.0 spec section 8: textureSize always gets highp precision. // All other functions that take a sampler are assumed to be texture functions. if (mFunctionInfo.getName().find("textureSize") == 0) mType.setPrecision(EbpHigh); else mType.setPrecision(precision); } TString TIntermAggregate::getSymbolTableMangledName() const { ASSERT(!isConstructor()); switch (mOp) { case EOpCallInternalRawFunction: case EOpCallBuiltInFunction: case EOpCallFunctionInAST: return TFunction::GetMangledNameFromCall(mFunctionInfo.getName(), mArguments); default: TString opString = GetOperatorString(mOp); return TFunction::GetMangledNameFromCall(opString, mArguments); } } bool TIntermAggregate::hasSideEffects() const { if (isFunctionCall() && mFunctionInfo.isKnownToNotHaveSideEffects()) { for (TIntermNode *arg : mArguments) { if (arg->getAsTyped()->hasSideEffects()) { return true; } } return false; } // Conservatively assume most aggregate operators have side-effects return true; } void TIntermBlock::appendStatement(TIntermNode *statement) { // Declaration nodes with no children can appear if it was an empty declaration or if all the // declarators just added constants to the symbol table instead of generating code. We still // need to add the declaration to the AST in that case because it might be relevant to the // validity of switch/case. if (statement != nullptr) { mStatements.push_back(statement); } } void TIntermFunctionPrototype::appendParameter(TIntermSymbol *parameter) { ASSERT(parameter != nullptr); mParameters.push_back(parameter); } void TIntermDeclaration::appendDeclarator(TIntermTyped *declarator) { ASSERT(declarator != nullptr); ASSERT(declarator->getAsSymbolNode() != nullptr || (declarator->getAsBinaryNode() != nullptr && declarator->getAsBinaryNode()->getOp() == EOpInitialize)); ASSERT(mDeclarators.empty() || declarator->getType().sameNonArrayType(mDeclarators.back()->getAsTyped()->getType())); mDeclarators.push_back(declarator); } bool TIntermTernary::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mCondition, TIntermTyped, original, replacement); REPLACE_IF_IS(mTrueExpression, TIntermTyped, original, replacement); REPLACE_IF_IS(mFalseExpression, TIntermTyped, original, replacement); return false; } bool TIntermIfElse::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mCondition, TIntermTyped, original, replacement); REPLACE_IF_IS(mTrueBlock, TIntermBlock, original, replacement); REPLACE_IF_IS(mFalseBlock, TIntermBlock, original, replacement); return false; } bool TIntermSwitch::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mInit, TIntermTyped, original, replacement); REPLACE_IF_IS(mStatementList, TIntermBlock, original, replacement); ASSERT(mStatementList); return false; } bool TIntermCase::replaceChildNode(TIntermNode *original, TIntermNode *replacement) { REPLACE_IF_IS(mCondition, TIntermTyped, original, replacement); return false; } TIntermTyped::TIntermTyped(const TIntermTyped &node) : TIntermNode(), mType(node.mType) { // Copy constructor is disallowed for TIntermNode in order to disallow it for subclasses that // don't explicitly allow it, so normal TIntermNode constructor is used to construct the copy. // We need to manually copy any fields of TIntermNode besides handling fields in TIntermTyped. mLine = node.mLine; } bool TIntermTyped::isConstructorWithOnlyConstantUnionParameters() { TIntermAggregate *constructor = getAsAggregate(); if (!constructor || !constructor->isConstructor()) { return false; } for (TIntermNode *&node : *constructor->getSequence()) { if (!node->getAsConstantUnion()) return false; } return true; } TIntermConstantUnion::TIntermConstantUnion(const TIntermConstantUnion &node) : TIntermTyped(node) { mUnionArrayPointer = node.mUnionArrayPointer; } void TFunctionSymbolInfo::setFromFunction(const TFunction &function) { setName(function.getName()); setId(TSymbolUniqueId(function)); } TFunctionSymbolInfo::TFunctionSymbolInfo(const TSymbolUniqueId &id) : mId(new TSymbolUniqueId(id)), mKnownToNotHaveSideEffects(false) { } TFunctionSymbolInfo::TFunctionSymbolInfo(const TFunctionSymbolInfo &info) : mName(info.mName), mId(nullptr), mKnownToNotHaveSideEffects(info.mKnownToNotHaveSideEffects) { if (info.mId) { mId = new TSymbolUniqueId(*info.mId); } } TFunctionSymbolInfo &TFunctionSymbolInfo::operator=(const TFunctionSymbolInfo &info) { mName = info.mName; if (info.mId) { mId = new TSymbolUniqueId(*info.mId); } else { mId = nullptr; } return *this; } void TFunctionSymbolInfo::setId(const TSymbolUniqueId &id) { mId = new TSymbolUniqueId(id); } const TSymbolUniqueId &TFunctionSymbolInfo::getId() const { ASSERT(mId); return *mId; } TIntermAggregate::TIntermAggregate(const TIntermAggregate &node) : TIntermOperator(node), mUseEmulatedFunction(node.mUseEmulatedFunction), mGotPrecisionFromChildren(node.mGotPrecisionFromChildren), mFunctionInfo(node.mFunctionInfo) { for (TIntermNode *arg : node.mArguments) { TIntermTyped *typedArg = arg->getAsTyped(); ASSERT(typedArg != nullptr); TIntermTyped *argCopy = typedArg->deepCopy(); mArguments.push_back(argCopy); } } TIntermAggregate *TIntermAggregate::shallowCopy() const { TIntermSequence *copySeq = new TIntermSequence(); copySeq->insert(copySeq->begin(), getSequence()->begin(), getSequence()->end()); TIntermAggregate *copyNode = new TIntermAggregate(mType, mOp, copySeq); *copyNode->getFunctionSymbolInfo() = mFunctionInfo; copyNode->setLine(mLine); return copyNode; } TIntermSwizzle::TIntermSwizzle(const TIntermSwizzle &node) : TIntermTyped(node) { TIntermTyped *operandCopy = node.mOperand->deepCopy(); ASSERT(operandCopy != nullptr); mOperand = operandCopy; mSwizzleOffsets = node.mSwizzleOffsets; } TIntermBinary::TIntermBinary(const TIntermBinary &node) : TIntermOperator(node), mAddIndexClamp(node.mAddIndexClamp) { TIntermTyped *leftCopy = node.mLeft->deepCopy(); TIntermTyped *rightCopy = node.mRight->deepCopy(); ASSERT(leftCopy != nullptr && rightCopy != nullptr); mLeft = leftCopy; mRight = rightCopy; } TIntermUnary::TIntermUnary(const TIntermUnary &node) : TIntermOperator(node), mUseEmulatedFunction(node.mUseEmulatedFunction) { TIntermTyped *operandCopy = node.mOperand->deepCopy(); ASSERT(operandCopy != nullptr); mOperand = operandCopy; } TIntermTernary::TIntermTernary(const TIntermTernary &node) : TIntermTyped(node) { TIntermTyped *conditionCopy = node.mCondition->deepCopy(); TIntermTyped *trueCopy = node.mTrueExpression->deepCopy(); TIntermTyped *falseCopy = node.mFalseExpression->deepCopy(); ASSERT(conditionCopy != nullptr && trueCopy != nullptr && falseCopy != nullptr); mCondition = conditionCopy; mTrueExpression = trueCopy; mFalseExpression = falseCopy; } bool TIntermOperator::isAssignment() const { return IsAssignment(mOp); } bool TIntermOperator::isMultiplication() const { switch (mOp) { case EOpMul: case EOpMatrixTimesMatrix: case EOpMatrixTimesVector: case EOpMatrixTimesScalar: case EOpVectorTimesMatrix: case EOpVectorTimesScalar: return true; default: return false; } } bool TIntermOperator::isConstructor() const { return (mOp == EOpConstruct); } bool TIntermOperator::isFunctionCall() const { switch (mOp) { case EOpCallFunctionInAST: case EOpCallBuiltInFunction: case EOpCallInternalRawFunction: return true; default: return false; } } TOperator TIntermBinary::GetMulOpBasedOnOperands(const TType &left, const TType &right) { if (left.isMatrix()) { if (right.isMatrix()) { return EOpMatrixTimesMatrix; } else { if (right.isVector()) { return EOpMatrixTimesVector; } else { return EOpMatrixTimesScalar; } } } else { if (right.isMatrix()) { if (left.isVector()) { return EOpVectorTimesMatrix; } else { return EOpMatrixTimesScalar; } } else { // Neither operand is a matrix. if (left.isVector() == right.isVector()) { // Leave as component product. return EOpMul; } else { return EOpVectorTimesScalar; } } } } TOperator TIntermBinary::GetMulAssignOpBasedOnOperands(const TType &left, const TType &right) { if (left.isMatrix()) { if (right.isMatrix()) { return EOpMatrixTimesMatrixAssign; } else { // right should be scalar, but this may not be validated yet. return EOpMatrixTimesScalarAssign; } } else { if (right.isMatrix()) { // Left should be a vector, but this may not be validated yet. return EOpVectorTimesMatrixAssign; } else { // Neither operand is a matrix. if (left.isVector() == right.isVector()) { // Leave as component product. return EOpMulAssign; } else { // left should be vector and right should be scalar, but this may not be validated // yet. return EOpVectorTimesScalarAssign; } } } } // // Make sure the type of a unary operator is appropriate for its // combination of operation and operand type. // void TIntermUnary::promote() { if (mOp == EOpArrayLength) { // Special case: the qualifier of .length() doesn't depend on the operand qualifier. setType(TType(EbtInt, EbpUndefined, EvqConst)); return; } TQualifier resultQualifier = EvqTemporary; if (mOperand->getQualifier() == EvqConst) resultQualifier = EvqConst; unsigned char operandPrimarySize = static_cast(mOperand->getType().getNominalSize()); switch (mOp) { case EOpFloatBitsToInt: setType(TType(EbtInt, EbpHigh, resultQualifier, operandPrimarySize)); break; case EOpFloatBitsToUint: setType(TType(EbtUInt, EbpHigh, resultQualifier, operandPrimarySize)); break; case EOpIntBitsToFloat: case EOpUintBitsToFloat: setType(TType(EbtFloat, EbpHigh, resultQualifier, operandPrimarySize)); break; case EOpPackSnorm2x16: case EOpPackUnorm2x16: case EOpPackHalf2x16: case EOpPackUnorm4x8: case EOpPackSnorm4x8: setType(TType(EbtUInt, EbpHigh, resultQualifier)); break; case EOpUnpackSnorm2x16: case EOpUnpackUnorm2x16: setType(TType(EbtFloat, EbpHigh, resultQualifier, 2)); break; case EOpUnpackHalf2x16: setType(TType(EbtFloat, EbpMedium, resultQualifier, 2)); break; case EOpUnpackUnorm4x8: case EOpUnpackSnorm4x8: setType(TType(EbtFloat, EbpMedium, resultQualifier, 4)); break; case EOpAny: case EOpAll: setType(TType(EbtBool, EbpUndefined, resultQualifier)); break; case EOpLength: case EOpDeterminant: setType(TType(EbtFloat, mOperand->getType().getPrecision(), resultQualifier)); break; case EOpTranspose: setType(TType(EbtFloat, mOperand->getType().getPrecision(), resultQualifier, static_cast(mOperand->getType().getRows()), static_cast(mOperand->getType().getCols()))); break; case EOpIsInf: case EOpIsNan: setType(TType(EbtBool, EbpUndefined, resultQualifier, operandPrimarySize)); break; case EOpBitfieldReverse: setType(TType(mOperand->getBasicType(), EbpHigh, resultQualifier, operandPrimarySize)); break; case EOpBitCount: setType(TType(EbtInt, EbpLow, resultQualifier, operandPrimarySize)); break; case EOpFindLSB: setType(TType(EbtInt, EbpLow, resultQualifier, operandPrimarySize)); break; case EOpFindMSB: setType(TType(EbtInt, EbpLow, resultQualifier, operandPrimarySize)); break; default: setType(mOperand->getType()); mType.setQualifier(resultQualifier); break; } } TIntermSwizzle::TIntermSwizzle(TIntermTyped *operand, const TVector &swizzleOffsets) : TIntermTyped(TType(EbtFloat, EbpUndefined)), mOperand(operand), mSwizzleOffsets(swizzleOffsets) { ASSERT(mSwizzleOffsets.size() <= 4); promote(); } TIntermUnary::TIntermUnary(TOperator op, TIntermTyped *operand) : TIntermOperator(op), mOperand(operand), mUseEmulatedFunction(false) { promote(); } TIntermBinary::TIntermBinary(TOperator op, TIntermTyped *left, TIntermTyped *right) : TIntermOperator(op), mLeft(left), mRight(right), mAddIndexClamp(false) { promote(); } TIntermInvariantDeclaration::TIntermInvariantDeclaration(TIntermSymbol *symbol, const TSourceLoc &line) : TIntermNode(), mSymbol(symbol) { ASSERT(symbol); setLine(line); } TIntermTernary::TIntermTernary(TIntermTyped *cond, TIntermTyped *trueExpression, TIntermTyped *falseExpression) : TIntermTyped(trueExpression->getType()), mCondition(cond), mTrueExpression(trueExpression), mFalseExpression(falseExpression) { getTypePointer()->setQualifier( TIntermTernary::DetermineQualifier(cond, trueExpression, falseExpression)); } TIntermLoop::TIntermLoop(TLoopType type, TIntermNode *init, TIntermTyped *cond, TIntermTyped *expr, TIntermBlock *body) : mType(type), mInit(init), mCond(cond), mExpr(expr), mBody(body) { // Declaration nodes with no children can appear if all the declarators just added constants to // the symbol table instead of generating code. They're no-ops so don't add them to the tree. if (mInit && mInit->getAsDeclarationNode() && mInit->getAsDeclarationNode()->getSequence()->empty()) { mInit = nullptr; } } TIntermIfElse::TIntermIfElse(TIntermTyped *cond, TIntermBlock *trueB, TIntermBlock *falseB) : TIntermNode(), mCondition(cond), mTrueBlock(trueB), mFalseBlock(falseB) { // Prune empty false blocks so that there won't be unnecessary operations done on it. if (mFalseBlock && mFalseBlock->getSequence()->empty()) { mFalseBlock = nullptr; } } TIntermSwitch::TIntermSwitch(TIntermTyped *init, TIntermBlock *statementList) : TIntermNode(), mInit(init), mStatementList(statementList) { ASSERT(mStatementList); } void TIntermSwitch::setStatementList(TIntermBlock *statementList) { ASSERT(statementList); mStatementList = statementList; } // static TQualifier TIntermTernary::DetermineQualifier(TIntermTyped *cond, TIntermTyped *trueExpression, TIntermTyped *falseExpression) { if (cond->getQualifier() == EvqConst && trueExpression->getQualifier() == EvqConst && falseExpression->getQualifier() == EvqConst) { return EvqConst; } return EvqTemporary; } TIntermTyped *TIntermTernary::fold() { if (mCondition->getAsConstantUnion()) { if (mCondition->getAsConstantUnion()->getBConst(0)) { mTrueExpression->getTypePointer()->setQualifier(mType.getQualifier()); return mTrueExpression; } else { mFalseExpression->getTypePointer()->setQualifier(mType.getQualifier()); return mFalseExpression; } } return this; } void TIntermSwizzle::promote() { TQualifier resultQualifier = EvqTemporary; if (mOperand->getQualifier() == EvqConst) resultQualifier = EvqConst; auto numFields = mSwizzleOffsets.size(); setType(TType(mOperand->getBasicType(), mOperand->getPrecision(), resultQualifier, static_cast(numFields))); } bool TIntermSwizzle::hasDuplicateOffsets() const { int offsetCount[4] = {0u, 0u, 0u, 0u}; for (const auto offset : mSwizzleOffsets) { offsetCount[offset]++; if (offsetCount[offset] > 1) { return true; } } return false; } bool TIntermSwizzle::offsetsMatch(int offset) const { return mSwizzleOffsets.size() == 1 && mSwizzleOffsets[0] == offset; } void TIntermSwizzle::writeOffsetsAsXYZW(TInfoSinkBase *out) const { for (const int offset : mSwizzleOffsets) { switch (offset) { case 0: *out << "x"; break; case 1: *out << "y"; break; case 2: *out << "z"; break; case 3: *out << "w"; break; default: UNREACHABLE(); } } } TQualifier TIntermBinary::GetCommaQualifier(int shaderVersion, const TIntermTyped *left, const TIntermTyped *right) { // ESSL3.00 section 12.43: The result of a sequence operator is not a constant-expression. if (shaderVersion >= 300 || left->getQualifier() != EvqConst || right->getQualifier() != EvqConst) { return EvqTemporary; } return EvqConst; } // Establishes the type of the result of the binary operation. void TIntermBinary::promote() { ASSERT(!isMultiplication() || mOp == GetMulOpBasedOnOperands(mLeft->getType(), mRight->getType())); // Comma is handled as a special case. Note that the comma node qualifier depends on the shader // version and so is not being set here. if (mOp == EOpComma) { setType(mRight->getType()); return; } // Base assumption: just make the type the same as the left // operand. Then only deviations from this need be coded. setType(mLeft->getType()); TQualifier resultQualifier = EvqConst; // Binary operations results in temporary variables unless both // operands are const. if (mLeft->getQualifier() != EvqConst || mRight->getQualifier() != EvqConst) { resultQualifier = EvqTemporary; getTypePointer()->setQualifier(EvqTemporary); } // Handle indexing ops. switch (mOp) { case EOpIndexDirect: case EOpIndexIndirect: if (mLeft->isArray()) { mType.toArrayElementType(); } else if (mLeft->isMatrix()) { setType(TType(mLeft->getBasicType(), mLeft->getPrecision(), resultQualifier, static_cast(mLeft->getRows()))); } else if (mLeft->isVector()) { setType(TType(mLeft->getBasicType(), mLeft->getPrecision(), resultQualifier)); } else { UNREACHABLE(); } return; case EOpIndexDirectStruct: { const TFieldList &fields = mLeft->getType().getStruct()->fields(); const int i = mRight->getAsConstantUnion()->getIConst(0); setType(*fields[i]->type()); getTypePointer()->setQualifier(resultQualifier); return; } case EOpIndexDirectInterfaceBlock: { const TFieldList &fields = mLeft->getType().getInterfaceBlock()->fields(); const int i = mRight->getAsConstantUnion()->getIConst(0); setType(*fields[i]->type()); getTypePointer()->setQualifier(resultQualifier); return; } default: break; } ASSERT(mLeft->isArray() == mRight->isArray()); // The result gets promoted to the highest precision. TPrecision higherPrecision = GetHigherPrecision(mLeft->getPrecision(), mRight->getPrecision()); getTypePointer()->setPrecision(higherPrecision); const int nominalSize = std::max(mLeft->getNominalSize(), mRight->getNominalSize()); // // All scalars or structs. Code after this test assumes this case is removed! // if (nominalSize == 1) { switch (mOp) { // // Promote to conditional // case EOpEqual: case EOpNotEqual: case EOpLessThan: case EOpGreaterThan: case EOpLessThanEqual: case EOpGreaterThanEqual: setType(TType(EbtBool, EbpUndefined, resultQualifier)); break; // // And and Or operate on conditionals // case EOpLogicalAnd: case EOpLogicalXor: case EOpLogicalOr: ASSERT(mLeft->getBasicType() == EbtBool && mRight->getBasicType() == EbtBool); setType(TType(EbtBool, EbpUndefined, resultQualifier)); break; default: break; } return; } // If we reach here, at least one of the operands is vector or matrix. // The other operand could be a scalar, vector, or matrix. TBasicType basicType = mLeft->getBasicType(); switch (mOp) { case EOpMul: break; case EOpMatrixTimesScalar: if (mRight->isMatrix()) { setType(TType(basicType, higherPrecision, resultQualifier, static_cast(mRight->getCols()), static_cast(mRight->getRows()))); } break; case EOpMatrixTimesVector: setType(TType(basicType, higherPrecision, resultQualifier, static_cast(mLeft->getRows()), 1)); break; case EOpMatrixTimesMatrix: setType(TType(basicType, higherPrecision, resultQualifier, static_cast(mRight->getCols()), static_cast(mLeft->getRows()))); break; case EOpVectorTimesScalar: setType(TType(basicType, higherPrecision, resultQualifier, static_cast(nominalSize), 1)); break; case EOpVectorTimesMatrix: setType(TType(basicType, higherPrecision, resultQualifier, static_cast(mRight->getCols()), 1)); break; case EOpMulAssign: case EOpVectorTimesScalarAssign: case EOpVectorTimesMatrixAssign: case EOpMatrixTimesScalarAssign: case EOpMatrixTimesMatrixAssign: ASSERT(mOp == GetMulAssignOpBasedOnOperands(mLeft->getType(), mRight->getType())); break; case EOpAssign: case EOpInitialize: ASSERT((mLeft->getNominalSize() == mRight->getNominalSize()) && (mLeft->getSecondarySize() == mRight->getSecondarySize())); break; case EOpAdd: case EOpSub: case EOpDiv: case EOpIMod: case EOpBitShiftLeft: case EOpBitShiftRight: case EOpBitwiseAnd: case EOpBitwiseXor: case EOpBitwiseOr: case EOpAddAssign: case EOpSubAssign: case EOpDivAssign: case EOpIModAssign: case EOpBitShiftLeftAssign: case EOpBitShiftRightAssign: case EOpBitwiseAndAssign: case EOpBitwiseXorAssign: case EOpBitwiseOrAssign: { const int secondarySize = std::max(mLeft->getSecondarySize(), mRight->getSecondarySize()); setType(TType(basicType, higherPrecision, resultQualifier, static_cast(nominalSize), static_cast(secondarySize))); ASSERT(!mLeft->isArray() && !mRight->isArray()); break; } case EOpEqual: case EOpNotEqual: case EOpLessThan: case EOpGreaterThan: case EOpLessThanEqual: case EOpGreaterThanEqual: ASSERT((mLeft->getNominalSize() == mRight->getNominalSize()) && (mLeft->getSecondarySize() == mRight->getSecondarySize())); setType(TType(EbtBool, EbpUndefined, resultQualifier)); break; case EOpIndexDirect: case EOpIndexIndirect: case EOpIndexDirectInterfaceBlock: case EOpIndexDirectStruct: // These ops should be already fully handled. UNREACHABLE(); break; default: UNREACHABLE(); break; } } const TConstantUnion *TIntermConstantUnion::foldIndexing(int index) { if (isArray()) { ASSERT(index < static_cast(getType().getOutermostArraySize())); TType arrayElementType = getType(); arrayElementType.toArrayElementType(); size_t arrayElementSize = arrayElementType.getObjectSize(); return &mUnionArrayPointer[arrayElementSize * index]; } else if (isMatrix()) { ASSERT(index < getType().getCols()); int size = getType().getRows(); return &mUnionArrayPointer[size * index]; } else if (isVector()) { ASSERT(index < getType().getNominalSize()); return &mUnionArrayPointer[index]; } else { UNREACHABLE(); return nullptr; } } TIntermTyped *TIntermSwizzle::fold() { TIntermConstantUnion *operandConstant = mOperand->getAsConstantUnion(); if (operandConstant == nullptr) { return this; } TConstantUnion *constArray = new TConstantUnion[mSwizzleOffsets.size()]; for (size_t i = 0; i < mSwizzleOffsets.size(); ++i) { constArray[i] = *operandConstant->foldIndexing(mSwizzleOffsets.at(i)); } return CreateFoldedNode(constArray, this, mType.getQualifier()); } TIntermTyped *TIntermBinary::fold(TDiagnostics *diagnostics) { TIntermConstantUnion *leftConstant = mLeft->getAsConstantUnion(); TIntermConstantUnion *rightConstant = mRight->getAsConstantUnion(); switch (mOp) { case EOpComma: { if (mLeft->hasSideEffects()) { return this; } mRight->getTypePointer()->setQualifier(mType.getQualifier()); return mRight; } case EOpIndexDirect: { if (leftConstant == nullptr || rightConstant == nullptr) { return this; } int index = rightConstant->getIConst(0); const TConstantUnion *constArray = leftConstant->foldIndexing(index); if (!constArray) { return this; } return CreateFoldedNode(constArray, this, mType.getQualifier()); } case EOpIndexDirectStruct: { if (leftConstant == nullptr || rightConstant == nullptr) { return this; } const TFieldList &fields = mLeft->getType().getStruct()->fields(); size_t index = static_cast(rightConstant->getIConst(0)); size_t previousFieldsSize = 0; for (size_t i = 0; i < index; ++i) { previousFieldsSize += fields[i]->type()->getObjectSize(); } const TConstantUnion *constArray = leftConstant->getUnionArrayPointer(); return CreateFoldedNode(constArray + previousFieldsSize, this, mType.getQualifier()); } case EOpIndexIndirect: case EOpIndexDirectInterfaceBlock: // Can never be constant folded. return this; default: { if (leftConstant == nullptr || rightConstant == nullptr) { return this; } TConstantUnion *constArray = leftConstant->foldBinary(mOp, rightConstant, diagnostics, mLeft->getLine()); if (!constArray) { return this; } // Nodes may be constant folded without being qualified as constant. return CreateFoldedNode(constArray, this, mType.getQualifier()); } } } TIntermTyped *TIntermUnary::fold(TDiagnostics *diagnostics) { TConstantUnion *constArray = nullptr; if (mOp == EOpArrayLength) { // The size of runtime-sized arrays may only be determined at runtime. if (mOperand->hasSideEffects() || mOperand->getType().isUnsizedArray()) { return this; } constArray = new TConstantUnion[1]; constArray->setIConst(mOperand->getOutermostArraySize()); } else { TIntermConstantUnion *operandConstant = mOperand->getAsConstantUnion(); if (operandConstant == nullptr) { return this; } switch (mOp) { case EOpAny: case EOpAll: case EOpLength: case EOpTranspose: case EOpDeterminant: case EOpInverse: case EOpPackSnorm2x16: case EOpUnpackSnorm2x16: case EOpPackUnorm2x16: case EOpUnpackUnorm2x16: case EOpPackHalf2x16: case EOpUnpackHalf2x16: case EOpPackUnorm4x8: case EOpPackSnorm4x8: case EOpUnpackUnorm4x8: case EOpUnpackSnorm4x8: constArray = operandConstant->foldUnaryNonComponentWise(mOp); break; default: constArray = operandConstant->foldUnaryComponentWise(mOp, diagnostics); break; } } if (constArray == nullptr) { return this; } // Nodes may be constant folded without being qualified as constant. return CreateFoldedNode(constArray, this, mType.getQualifier()); } TIntermTyped *TIntermAggregate::fold(TDiagnostics *diagnostics) { // Make sure that all params are constant before actual constant folding. for (auto *param : *getSequence()) { if (param->getAsConstantUnion() == nullptr) { return this; } } TConstantUnion *constArray = nullptr; if (isConstructor()) constArray = TIntermConstantUnion::FoldAggregateConstructor(this); else constArray = TIntermConstantUnion::FoldAggregateBuiltIn(this, diagnostics); // Nodes may be constant folded without being qualified as constant. return CreateFoldedNode(constArray, this, getQualifier()); } // // The fold functions see if an operation on a constant can be done in place, // without generating run-time code. // // Returns the constant value to keep using or nullptr. // TConstantUnion *TIntermConstantUnion::foldBinary(TOperator op, TIntermConstantUnion *rightNode, TDiagnostics *diagnostics, const TSourceLoc &line) { const TConstantUnion *leftArray = getUnionArrayPointer(); const TConstantUnion *rightArray = rightNode->getUnionArrayPointer(); ASSERT(leftArray && rightArray); size_t objectSize = getType().getObjectSize(); // for a case like float f = vec4(2, 3, 4, 5) + 1.2; if (rightNode->getType().getObjectSize() == 1 && objectSize > 1) { rightArray = Vectorize(*rightNode->getUnionArrayPointer(), objectSize); } else if (rightNode->getType().getObjectSize() > 1 && objectSize == 1) { // for a case like float f = 1.2 + vec4(2, 3, 4, 5); leftArray = Vectorize(*getUnionArrayPointer(), rightNode->getType().getObjectSize()); objectSize = rightNode->getType().getObjectSize(); } TConstantUnion *resultArray = nullptr; switch (op) { case EOpAdd: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = TConstantUnion::add(leftArray[i], rightArray[i], diagnostics, line); break; case EOpSub: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = TConstantUnion::sub(leftArray[i], rightArray[i], diagnostics, line); break; case EOpMul: case EOpVectorTimesScalar: case EOpMatrixTimesScalar: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = TConstantUnion::mul(leftArray[i], rightArray[i], diagnostics, line); break; case EOpMatrixTimesMatrix: { // TODO(jmadll): This code should check for overflows. ASSERT(getType().getBasicType() == EbtFloat && rightNode->getBasicType() == EbtFloat); const int leftCols = getCols(); const int leftRows = getRows(); const int rightCols = rightNode->getType().getCols(); const int rightRows = rightNode->getType().getRows(); const int resultCols = rightCols; const int resultRows = leftRows; resultArray = new TConstantUnion[resultCols * resultRows]; for (int row = 0; row < resultRows; row++) { for (int column = 0; column < resultCols; column++) { resultArray[resultRows * column + row].setFConst(0.0f); for (int i = 0; i < leftCols; i++) { resultArray[resultRows * column + row].setFConst( resultArray[resultRows * column + row].getFConst() + leftArray[i * leftRows + row].getFConst() * rightArray[column * rightRows + i].getFConst()); } } } } break; case EOpDiv: case EOpIMod: { resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) { switch (getType().getBasicType()) { case EbtFloat: { ASSERT(op == EOpDiv); float dividend = leftArray[i].getFConst(); float divisor = rightArray[i].getFConst(); if (divisor == 0.0f) { if (dividend == 0.0f) { diagnostics->warning( getLine(), "Zero divided by zero during constant folding generated NaN", "/"); resultArray[i].setFConst(std::numeric_limits::quiet_NaN()); } else { diagnostics->warning(getLine(), "Divide by zero during constant folding", "/"); bool negativeResult = std::signbit(dividend) != std::signbit(divisor); resultArray[i].setFConst( negativeResult ? -std::numeric_limits::infinity() : std::numeric_limits::infinity()); } } else if (gl::isInf(dividend) && gl::isInf(divisor)) { diagnostics->warning(getLine(), "Infinity divided by infinity during constant " "folding generated NaN", "/"); resultArray[i].setFConst(std::numeric_limits::quiet_NaN()); } else { float result = dividend / divisor; if (!gl::isInf(dividend) && gl::isInf(result)) { diagnostics->warning( getLine(), "Constant folded division overflowed to infinity", "/"); } resultArray[i].setFConst(result); } break; } case EbtInt: if (rightArray[i] == 0) { diagnostics->warning( getLine(), "Divide by zero error during constant folding", "/"); resultArray[i].setIConst(INT_MAX); } else { int lhs = leftArray[i].getIConst(); int divisor = rightArray[i].getIConst(); if (op == EOpDiv) { // Check for the special case where the minimum representable number // is // divided by -1. If left alone this leads to integer overflow in // C++. // ESSL 3.00.6 section 4.1.3 Integers: // "However, for the case where the minimum representable value is // divided by -1, it is allowed to return either the minimum // representable value or the maximum representable value." if (lhs == -0x7fffffff - 1 && divisor == -1) { resultArray[i].setIConst(0x7fffffff); } else { resultArray[i].setIConst(lhs / divisor); } } else { ASSERT(op == EOpIMod); if (lhs < 0 || divisor < 0) { // ESSL 3.00.6 section 5.9: Results of modulus are undefined // when // either one of the operands is negative. diagnostics->warning(getLine(), "Negative modulus operator operand " "encountered during constant folding", "%"); resultArray[i].setIConst(0); } else { resultArray[i].setIConst(lhs % divisor); } } } break; case EbtUInt: if (rightArray[i] == 0) { diagnostics->warning( getLine(), "Divide by zero error during constant folding", "/"); resultArray[i].setUConst(UINT_MAX); } else { if (op == EOpDiv) { resultArray[i].setUConst(leftArray[i].getUConst() / rightArray[i].getUConst()); } else { ASSERT(op == EOpIMod); resultArray[i].setUConst(leftArray[i].getUConst() % rightArray[i].getUConst()); } } break; default: UNREACHABLE(); return nullptr; } } } break; case EOpMatrixTimesVector: { // TODO(jmadll): This code should check for overflows. ASSERT(rightNode->getBasicType() == EbtFloat); const int matrixCols = getCols(); const int matrixRows = getRows(); resultArray = new TConstantUnion[matrixRows]; for (int matrixRow = 0; matrixRow < matrixRows; matrixRow++) { resultArray[matrixRow].setFConst(0.0f); for (int col = 0; col < matrixCols; col++) { resultArray[matrixRow].setFConst( resultArray[matrixRow].getFConst() + leftArray[col * matrixRows + matrixRow].getFConst() * rightArray[col].getFConst()); } } } break; case EOpVectorTimesMatrix: { // TODO(jmadll): This code should check for overflows. ASSERT(getType().getBasicType() == EbtFloat); const int matrixCols = rightNode->getType().getCols(); const int matrixRows = rightNode->getType().getRows(); resultArray = new TConstantUnion[matrixCols]; for (int matrixCol = 0; matrixCol < matrixCols; matrixCol++) { resultArray[matrixCol].setFConst(0.0f); for (int matrixRow = 0; matrixRow < matrixRows; matrixRow++) { resultArray[matrixCol].setFConst( resultArray[matrixCol].getFConst() + leftArray[matrixRow].getFConst() * rightArray[matrixCol * matrixRows + matrixRow].getFConst()); } } } break; case EOpLogicalAnd: { resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) { resultArray[i] = leftArray[i] && rightArray[i]; } } break; case EOpLogicalOr: { resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) { resultArray[i] = leftArray[i] || rightArray[i]; } } break; case EOpLogicalXor: { ASSERT(getType().getBasicType() == EbtBool); resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) { resultArray[i].setBConst(leftArray[i] != rightArray[i]); } } break; case EOpBitwiseAnd: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = leftArray[i] & rightArray[i]; break; case EOpBitwiseXor: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = leftArray[i] ^ rightArray[i]; break; case EOpBitwiseOr: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = leftArray[i] | rightArray[i]; break; case EOpBitShiftLeft: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = TConstantUnion::lshift(leftArray[i], rightArray[i], diagnostics, line); break; case EOpBitShiftRight: resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) resultArray[i] = TConstantUnion::rshift(leftArray[i], rightArray[i], diagnostics, line); break; case EOpLessThan: ASSERT(objectSize == 1); resultArray = new TConstantUnion[1]; resultArray->setBConst(*leftArray < *rightArray); break; case EOpGreaterThan: ASSERT(objectSize == 1); resultArray = new TConstantUnion[1]; resultArray->setBConst(*leftArray > *rightArray); break; case EOpLessThanEqual: ASSERT(objectSize == 1); resultArray = new TConstantUnion[1]; resultArray->setBConst(!(*leftArray > *rightArray)); break; case EOpGreaterThanEqual: ASSERT(objectSize == 1); resultArray = new TConstantUnion[1]; resultArray->setBConst(!(*leftArray < *rightArray)); break; case EOpEqual: case EOpNotEqual: { resultArray = new TConstantUnion[1]; bool equal = true; for (size_t i = 0; i < objectSize; i++) { if (leftArray[i] != rightArray[i]) { equal = false; break; // break out of for loop } } if (op == EOpEqual) { resultArray->setBConst(equal); } else { resultArray->setBConst(!equal); } } break; default: UNREACHABLE(); return nullptr; } return resultArray; } // The fold functions do operations on a constant at GLSL compile time, without generating run-time // code. Returns the constant value to keep using. Nullptr should not be returned. TConstantUnion *TIntermConstantUnion::foldUnaryNonComponentWise(TOperator op) { // Do operations where the return type may have a different number of components compared to the // operand type. const TConstantUnion *operandArray = getUnionArrayPointer(); ASSERT(operandArray); size_t objectSize = getType().getObjectSize(); TConstantUnion *resultArray = nullptr; switch (op) { case EOpAny: ASSERT(getType().getBasicType() == EbtBool); resultArray = new TConstantUnion(); resultArray->setBConst(false); for (size_t i = 0; i < objectSize; i++) { if (operandArray[i].getBConst()) { resultArray->setBConst(true); break; } } break; case EOpAll: ASSERT(getType().getBasicType() == EbtBool); resultArray = new TConstantUnion(); resultArray->setBConst(true); for (size_t i = 0; i < objectSize; i++) { if (!operandArray[i].getBConst()) { resultArray->setBConst(false); break; } } break; case EOpLength: ASSERT(getType().getBasicType() == EbtFloat); resultArray = new TConstantUnion(); resultArray->setFConst(VectorLength(operandArray, objectSize)); break; case EOpTranspose: { ASSERT(getType().getBasicType() == EbtFloat); resultArray = new TConstantUnion[objectSize]; angle::Matrix result = GetMatrix(operandArray, getType().getRows(), getType().getCols()).transpose(); SetUnionArrayFromMatrix(result, resultArray); break; } case EOpDeterminant: { ASSERT(getType().getBasicType() == EbtFloat); unsigned int size = getType().getNominalSize(); ASSERT(size >= 2 && size <= 4); resultArray = new TConstantUnion(); resultArray->setFConst(GetMatrix(operandArray, size).determinant()); break; } case EOpInverse: { ASSERT(getType().getBasicType() == EbtFloat); unsigned int size = getType().getNominalSize(); ASSERT(size >= 2 && size <= 4); resultArray = new TConstantUnion[objectSize]; angle::Matrix result = GetMatrix(operandArray, size).inverse(); SetUnionArrayFromMatrix(result, resultArray); break; } case EOpPackSnorm2x16: ASSERT(getType().getBasicType() == EbtFloat); ASSERT(getType().getNominalSize() == 2); resultArray = new TConstantUnion(); resultArray->setUConst( gl::packSnorm2x16(operandArray[0].getFConst(), operandArray[1].getFConst())); break; case EOpUnpackSnorm2x16: { ASSERT(getType().getBasicType() == EbtUInt); resultArray = new TConstantUnion[2]; float f1, f2; gl::unpackSnorm2x16(operandArray[0].getUConst(), &f1, &f2); resultArray[0].setFConst(f1); resultArray[1].setFConst(f2); break; } case EOpPackUnorm2x16: ASSERT(getType().getBasicType() == EbtFloat); ASSERT(getType().getNominalSize() == 2); resultArray = new TConstantUnion(); resultArray->setUConst( gl::packUnorm2x16(operandArray[0].getFConst(), operandArray[1].getFConst())); break; case EOpUnpackUnorm2x16: { ASSERT(getType().getBasicType() == EbtUInt); resultArray = new TConstantUnion[2]; float f1, f2; gl::unpackUnorm2x16(operandArray[0].getUConst(), &f1, &f2); resultArray[0].setFConst(f1); resultArray[1].setFConst(f2); break; } case EOpPackHalf2x16: ASSERT(getType().getBasicType() == EbtFloat); ASSERT(getType().getNominalSize() == 2); resultArray = new TConstantUnion(); resultArray->setUConst( gl::packHalf2x16(operandArray[0].getFConst(), operandArray[1].getFConst())); break; case EOpUnpackHalf2x16: { ASSERT(getType().getBasicType() == EbtUInt); resultArray = new TConstantUnion[2]; float f1, f2; gl::unpackHalf2x16(operandArray[0].getUConst(), &f1, &f2); resultArray[0].setFConst(f1); resultArray[1].setFConst(f2); break; } case EOpPackUnorm4x8: { ASSERT(getType().getBasicType() == EbtFloat); resultArray = new TConstantUnion(); resultArray->setUConst( gl::PackUnorm4x8(operandArray[0].getFConst(), operandArray[1].getFConst(), operandArray[2].getFConst(), operandArray[3].getFConst())); break; } case EOpPackSnorm4x8: { ASSERT(getType().getBasicType() == EbtFloat); resultArray = new TConstantUnion(); resultArray->setUConst( gl::PackSnorm4x8(operandArray[0].getFConst(), operandArray[1].getFConst(), operandArray[2].getFConst(), operandArray[3].getFConst())); break; } case EOpUnpackUnorm4x8: { ASSERT(getType().getBasicType() == EbtUInt); resultArray = new TConstantUnion[4]; float f[4]; gl::UnpackUnorm4x8(operandArray[0].getUConst(), f); for (size_t i = 0; i < 4; ++i) { resultArray[i].setFConst(f[i]); } break; } case EOpUnpackSnorm4x8: { ASSERT(getType().getBasicType() == EbtUInt); resultArray = new TConstantUnion[4]; float f[4]; gl::UnpackSnorm4x8(operandArray[0].getUConst(), f); for (size_t i = 0; i < 4; ++i) { resultArray[i].setFConst(f[i]); } break; } default: UNREACHABLE(); break; } return resultArray; } TConstantUnion *TIntermConstantUnion::foldUnaryComponentWise(TOperator op, TDiagnostics *diagnostics) { // Do unary operations where each component of the result is computed based on the corresponding // component of the operand. Also folds normalize, though the divisor in that case takes all // components into account. const TConstantUnion *operandArray = getUnionArrayPointer(); ASSERT(operandArray); size_t objectSize = getType().getObjectSize(); TConstantUnion *resultArray = new TConstantUnion[objectSize]; for (size_t i = 0; i < objectSize; i++) { switch (op) { case EOpNegative: switch (getType().getBasicType()) { case EbtFloat: resultArray[i].setFConst(-operandArray[i].getFConst()); break; case EbtInt: if (operandArray[i] == std::numeric_limits::min()) { // The minimum representable integer doesn't have a positive // counterpart, rather the negation overflows and in ESSL is supposed to // wrap back to the minimum representable integer. Make sure that we // don't actually let the negation overflow, which has undefined // behavior in C++. resultArray[i].setIConst(std::numeric_limits::min()); } else { resultArray[i].setIConst(-operandArray[i].getIConst()); } break; case EbtUInt: if (operandArray[i] == 0x80000000u) { resultArray[i].setUConst(0x80000000u); } else { resultArray[i].setUConst(static_cast( -static_cast(operandArray[i].getUConst()))); } break; default: UNREACHABLE(); return nullptr; } break; case EOpPositive: switch (getType().getBasicType()) { case EbtFloat: resultArray[i].setFConst(operandArray[i].getFConst()); break; case EbtInt: resultArray[i].setIConst(operandArray[i].getIConst()); break; case EbtUInt: resultArray[i].setUConst(static_cast( static_cast(operandArray[i].getUConst()))); break; default: UNREACHABLE(); return nullptr; } break; case EOpLogicalNot: switch (getType().getBasicType()) { case EbtBool: resultArray[i].setBConst(!operandArray[i].getBConst()); break; default: UNREACHABLE(); return nullptr; } break; case EOpBitwiseNot: switch (getType().getBasicType()) { case EbtInt: resultArray[i].setIConst(~operandArray[i].getIConst()); break; case EbtUInt: resultArray[i].setUConst(~operandArray[i].getUConst()); break; default: UNREACHABLE(); return nullptr; } break; case EOpRadians: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setFConst(kDegreesToRadiansMultiplier * operandArray[i].getFConst()); break; case EOpDegrees: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setFConst(kRadiansToDegreesMultiplier * operandArray[i].getFConst()); break; case EOpSin: foldFloatTypeUnary(operandArray[i], &sinf, &resultArray[i]); break; case EOpCos: foldFloatTypeUnary(operandArray[i], &cosf, &resultArray[i]); break; case EOpTan: foldFloatTypeUnary(operandArray[i], &tanf, &resultArray[i]); break; case EOpAsin: // For asin(x), results are undefined if |x| > 1, we are choosing to set result to // 0. if (fabsf(operandArray[i].getFConst()) > 1.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &asinf, &resultArray[i]); break; case EOpAcos: // For acos(x), results are undefined if |x| > 1, we are choosing to set result to // 0. if (fabsf(operandArray[i].getFConst()) > 1.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &acosf, &resultArray[i]); break; case EOpAtan: foldFloatTypeUnary(operandArray[i], &atanf, &resultArray[i]); break; case EOpSinh: foldFloatTypeUnary(operandArray[i], &sinhf, &resultArray[i]); break; case EOpCosh: foldFloatTypeUnary(operandArray[i], &coshf, &resultArray[i]); break; case EOpTanh: foldFloatTypeUnary(operandArray[i], &tanhf, &resultArray[i]); break; case EOpAsinh: foldFloatTypeUnary(operandArray[i], &asinhf, &resultArray[i]); break; case EOpAcosh: // For acosh(x), results are undefined if x < 1, we are choosing to set result to 0. if (operandArray[i].getFConst() < 1.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &acoshf, &resultArray[i]); break; case EOpAtanh: // For atanh(x), results are undefined if |x| >= 1, we are choosing to set result to // 0. if (fabsf(operandArray[i].getFConst()) >= 1.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &atanhf, &resultArray[i]); break; case EOpAbs: switch (getType().getBasicType()) { case EbtFloat: resultArray[i].setFConst(fabsf(operandArray[i].getFConst())); break; case EbtInt: resultArray[i].setIConst(abs(operandArray[i].getIConst())); break; default: UNREACHABLE(); return nullptr; } break; case EOpSign: switch (getType().getBasicType()) { case EbtFloat: { float fConst = operandArray[i].getFConst(); float fResult = 0.0f; if (fConst > 0.0f) fResult = 1.0f; else if (fConst < 0.0f) fResult = -1.0f; resultArray[i].setFConst(fResult); break; } case EbtInt: { int iConst = operandArray[i].getIConst(); int iResult = 0; if (iConst > 0) iResult = 1; else if (iConst < 0) iResult = -1; resultArray[i].setIConst(iResult); break; } default: UNREACHABLE(); return nullptr; } break; case EOpFloor: foldFloatTypeUnary(operandArray[i], &floorf, &resultArray[i]); break; case EOpTrunc: foldFloatTypeUnary(operandArray[i], &truncf, &resultArray[i]); break; case EOpRound: foldFloatTypeUnary(operandArray[i], &roundf, &resultArray[i]); break; case EOpRoundEven: { ASSERT(getType().getBasicType() == EbtFloat); float x = operandArray[i].getFConst(); float result; float fractPart = modff(x, &result); if (fabsf(fractPart) == 0.5f) result = 2.0f * roundf(x / 2.0f); else result = roundf(x); resultArray[i].setFConst(result); break; } case EOpCeil: foldFloatTypeUnary(operandArray[i], &ceilf, &resultArray[i]); break; case EOpFract: { ASSERT(getType().getBasicType() == EbtFloat); float x = operandArray[i].getFConst(); resultArray[i].setFConst(x - floorf(x)); break; } case EOpIsNan: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setBConst(gl::isNaN(operandArray[0].getFConst())); break; case EOpIsInf: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setBConst(gl::isInf(operandArray[0].getFConst())); break; case EOpFloatBitsToInt: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setIConst(gl::bitCast(operandArray[0].getFConst())); break; case EOpFloatBitsToUint: ASSERT(getType().getBasicType() == EbtFloat); resultArray[i].setUConst(gl::bitCast(operandArray[0].getFConst())); break; case EOpIntBitsToFloat: ASSERT(getType().getBasicType() == EbtInt); resultArray[i].setFConst(gl::bitCast(operandArray[0].getIConst())); break; case EOpUintBitsToFloat: ASSERT(getType().getBasicType() == EbtUInt); resultArray[i].setFConst(gl::bitCast(operandArray[0].getUConst())); break; case EOpExp: foldFloatTypeUnary(operandArray[i], &expf, &resultArray[i]); break; case EOpLog: // For log(x), results are undefined if x <= 0, we are choosing to set result to 0. if (operandArray[i].getFConst() <= 0.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &logf, &resultArray[i]); break; case EOpExp2: foldFloatTypeUnary(operandArray[i], &exp2f, &resultArray[i]); break; case EOpLog2: // For log2(x), results are undefined if x <= 0, we are choosing to set result to 0. // And log2f is not available on some plarforms like old android, so just using // log(x)/log(2) here. if (operandArray[i].getFConst() <= 0.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else { foldFloatTypeUnary(operandArray[i], &logf, &resultArray[i]); resultArray[i].setFConst(resultArray[i].getFConst() / logf(2.0f)); } break; case EOpSqrt: // For sqrt(x), results are undefined if x < 0, we are choosing to set result to 0. if (operandArray[i].getFConst() < 0.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else foldFloatTypeUnary(operandArray[i], &sqrtf, &resultArray[i]); break; case EOpInverseSqrt: // There is no stdlib built-in function equavalent for GLES built-in inversesqrt(), // so getting the square root first using builtin function sqrt() and then taking // its inverse. // Also, for inversesqrt(x), results are undefined if x <= 0, we are choosing to set // result to 0. if (operandArray[i].getFConst() <= 0.0f) UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); else { foldFloatTypeUnary(operandArray[i], &sqrtf, &resultArray[i]); resultArray[i].setFConst(1.0f / resultArray[i].getFConst()); } break; case EOpLogicalNotComponentWise: ASSERT(getType().getBasicType() == EbtBool); resultArray[i].setBConst(!operandArray[i].getBConst()); break; case EOpNormalize: { ASSERT(getType().getBasicType() == EbtFloat); float x = operandArray[i].getFConst(); float length = VectorLength(operandArray, objectSize); if (length) resultArray[i].setFConst(x / length); else UndefinedConstantFoldingError(getLine(), op, getType().getBasicType(), diagnostics, &resultArray[i]); break; } case EOpBitfieldReverse: { uint32_t value; if (getType().getBasicType() == EbtInt) { value = static_cast(operandArray[i].getIConst()); } else { ASSERT(getType().getBasicType() == EbtUInt); value = operandArray[i].getUConst(); } uint32_t result = gl::BitfieldReverse(value); if (getType().getBasicType() == EbtInt) { resultArray[i].setIConst(static_cast(result)); } else { resultArray[i].setUConst(result); } break; } case EOpBitCount: { uint32_t value; if (getType().getBasicType() == EbtInt) { value = static_cast(operandArray[i].getIConst()); } else { ASSERT(getType().getBasicType() == EbtUInt); value = operandArray[i].getUConst(); } int result = gl::BitCount(value); resultArray[i].setIConst(result); break; } case EOpFindLSB: { uint32_t value; if (getType().getBasicType() == EbtInt) { value = static_cast(operandArray[i].getIConst()); } else { ASSERT(getType().getBasicType() == EbtUInt); value = operandArray[i].getUConst(); } resultArray[i].setIConst(gl::FindLSB(value)); break; } case EOpFindMSB: { uint32_t value; if (getType().getBasicType() == EbtInt) { int intValue = operandArray[i].getIConst(); value = static_cast(intValue); if (intValue < 0) { // Look for zero instead of one in value. This also handles the intValue == // -1 special case, where the return value needs to be -1. value = ~value; } } else { ASSERT(getType().getBasicType() == EbtUInt); value = operandArray[i].getUConst(); } resultArray[i].setIConst(gl::FindMSB(value)); break; } case EOpDFdx: case EOpDFdy: case EOpFwidth: ASSERT(getType().getBasicType() == EbtFloat); // Derivatives of constant arguments should be 0. resultArray[i].setFConst(0.0f); break; default: return nullptr; } } return resultArray; } void TIntermConstantUnion::foldFloatTypeUnary(const TConstantUnion ¶meter, FloatTypeUnaryFunc builtinFunc, TConstantUnion *result) const { ASSERT(builtinFunc); ASSERT(getType().getBasicType() == EbtFloat); result->setFConst(builtinFunc(parameter.getFConst())); } // static TConstantUnion *TIntermConstantUnion::FoldAggregateConstructor(TIntermAggregate *aggregate) { ASSERT(aggregate->getSequence()->size() > 0u); size_t resultSize = aggregate->getType().getObjectSize(); TConstantUnion *resultArray = new TConstantUnion[resultSize]; TBasicType basicType = aggregate->getBasicType(); size_t resultIndex = 0u; if (aggregate->getSequence()->size() == 1u) { TIntermNode *argument = aggregate->getSequence()->front(); TIntermConstantUnion *argumentConstant = argument->getAsConstantUnion(); const TConstantUnion *argumentUnionArray = argumentConstant->getUnionArrayPointer(); // Check the special case of constructing a matrix diagonal from a single scalar, // or a vector from a single scalar. if (argumentConstant->getType().getObjectSize() == 1u) { if (aggregate->isMatrix()) { int resultCols = aggregate->getType().getCols(); int resultRows = aggregate->getType().getRows(); for (int col = 0; col < resultCols; ++col) { for (int row = 0; row < resultRows; ++row) { if (col == row) { resultArray[resultIndex].cast(basicType, argumentUnionArray[0]); } else { resultArray[resultIndex].setFConst(0.0f); } ++resultIndex; } } } else { while (resultIndex < resultSize) { resultArray[resultIndex].cast(basicType, argumentUnionArray[0]); ++resultIndex; } } ASSERT(resultIndex == resultSize); return resultArray; } else if (aggregate->isMatrix() && argumentConstant->isMatrix()) { // The special case of constructing a matrix from a matrix. int argumentCols = argumentConstant->getType().getCols(); int argumentRows = argumentConstant->getType().getRows(); int resultCols = aggregate->getType().getCols(); int resultRows = aggregate->getType().getRows(); for (int col = 0; col < resultCols; ++col) { for (int row = 0; row < resultRows; ++row) { if (col < argumentCols && row < argumentRows) { resultArray[resultIndex].cast(basicType, argumentUnionArray[col * argumentRows + row]); } else if (col == row) { resultArray[resultIndex].setFConst(1.0f); } else { resultArray[resultIndex].setFConst(0.0f); } ++resultIndex; } } ASSERT(resultIndex == resultSize); return resultArray; } } for (TIntermNode *&argument : *aggregate->getSequence()) { TIntermConstantUnion *argumentConstant = argument->getAsConstantUnion(); size_t argumentSize = argumentConstant->getType().getObjectSize(); const TConstantUnion *argumentUnionArray = argumentConstant->getUnionArrayPointer(); for (size_t i = 0u; i < argumentSize; ++i) { if (resultIndex >= resultSize) break; resultArray[resultIndex].cast(basicType, argumentUnionArray[i]); ++resultIndex; } } ASSERT(resultIndex == resultSize); return resultArray; } bool TIntermAggregate::CanFoldAggregateBuiltInOp(TOperator op) { switch (op) { case EOpAtan: case EOpPow: case EOpMod: case EOpMin: case EOpMax: case EOpClamp: case EOpMix: case EOpStep: case EOpSmoothStep: case EOpLdexp: case EOpMulMatrixComponentWise: case EOpOuterProduct: case EOpEqualComponentWise: case EOpNotEqualComponentWise: case EOpLessThanComponentWise: case EOpLessThanEqualComponentWise: case EOpGreaterThanComponentWise: case EOpGreaterThanEqualComponentWise: case EOpDistance: case EOpDot: case EOpCross: case EOpFaceforward: case EOpReflect: case EOpRefract: case EOpBitfieldExtract: case EOpBitfieldInsert: return true; default: return false; } } // static TConstantUnion *TIntermConstantUnion::FoldAggregateBuiltIn(TIntermAggregate *aggregate, TDiagnostics *diagnostics) { TOperator op = aggregate->getOp(); TIntermSequence *arguments = aggregate->getSequence(); unsigned int argsCount = static_cast(arguments->size()); std::vector unionArrays(argsCount); std::vector objectSizes(argsCount); size_t maxObjectSize = 0; TBasicType basicType = EbtVoid; TSourceLoc loc; for (unsigned int i = 0; i < argsCount; i++) { TIntermConstantUnion *argConstant = (*arguments)[i]->getAsConstantUnion(); ASSERT(argConstant != nullptr); // Should be checked already. if (i == 0) { basicType = argConstant->getType().getBasicType(); loc = argConstant->getLine(); } unionArrays[i] = argConstant->getUnionArrayPointer(); objectSizes[i] = argConstant->getType().getObjectSize(); if (objectSizes[i] > maxObjectSize) maxObjectSize = objectSizes[i]; } if (!(*arguments)[0]->getAsTyped()->isMatrix() && aggregate->getOp() != EOpOuterProduct) { for (unsigned int i = 0; i < argsCount; i++) if (objectSizes[i] != maxObjectSize) unionArrays[i] = Vectorize(*unionArrays[i], maxObjectSize); } TConstantUnion *resultArray = nullptr; switch (op) { case EOpAtan: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float y = unionArrays[0][i].getFConst(); float x = unionArrays[1][i].getFConst(); // Results are undefined if x and y are both 0. if (x == 0.0f && y == 0.0f) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else resultArray[i].setFConst(atan2f(y, x)); } break; } case EOpPow: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float x = unionArrays[0][i].getFConst(); float y = unionArrays[1][i].getFConst(); // Results are undefined if x < 0. // Results are undefined if x = 0 and y <= 0. if (x < 0.0f) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else if (x == 0.0f && y <= 0.0f) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else resultArray[i].setFConst(powf(x, y)); } break; } case EOpMod: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float x = unionArrays[0][i].getFConst(); float y = unionArrays[1][i].getFConst(); resultArray[i].setFConst(x - y * floorf(x / y)); } break; } case EOpMin: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setFConst( std::min(unionArrays[0][i].getFConst(), unionArrays[1][i].getFConst())); break; case EbtInt: resultArray[i].setIConst( std::min(unionArrays[0][i].getIConst(), unionArrays[1][i].getIConst())); break; case EbtUInt: resultArray[i].setUConst( std::min(unionArrays[0][i].getUConst(), unionArrays[1][i].getUConst())); break; default: UNREACHABLE(); break; } } break; } case EOpMax: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setFConst( std::max(unionArrays[0][i].getFConst(), unionArrays[1][i].getFConst())); break; case EbtInt: resultArray[i].setIConst( std::max(unionArrays[0][i].getIConst(), unionArrays[1][i].getIConst())); break; case EbtUInt: resultArray[i].setUConst( std::max(unionArrays[0][i].getUConst(), unionArrays[1][i].getUConst())); break; default: UNREACHABLE(); break; } } break; } case EOpStep: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) resultArray[i].setFConst( unionArrays[1][i].getFConst() < unionArrays[0][i].getFConst() ? 0.0f : 1.0f); break; } case EOpLessThanComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() < unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() < unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() < unionArrays[1][i].getUConst()); break; default: UNREACHABLE(); break; } } break; } case EOpLessThanEqualComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() <= unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() <= unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() <= unionArrays[1][i].getUConst()); break; default: UNREACHABLE(); break; } } break; } case EOpGreaterThanComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() > unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() > unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() > unionArrays[1][i].getUConst()); break; default: UNREACHABLE(); break; } } break; } case EOpGreaterThanEqualComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() >= unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() >= unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() >= unionArrays[1][i].getUConst()); break; default: UNREACHABLE(); break; } } } break; case EOpEqualComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() == unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() == unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() == unionArrays[1][i].getUConst()); break; case EbtBool: resultArray[i].setBConst(unionArrays[0][i].getBConst() == unionArrays[1][i].getBConst()); break; default: UNREACHABLE(); break; } } break; } case EOpNotEqualComponentWise: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: resultArray[i].setBConst(unionArrays[0][i].getFConst() != unionArrays[1][i].getFConst()); break; case EbtInt: resultArray[i].setBConst(unionArrays[0][i].getIConst() != unionArrays[1][i].getIConst()); break; case EbtUInt: resultArray[i].setBConst(unionArrays[0][i].getUConst() != unionArrays[1][i].getUConst()); break; case EbtBool: resultArray[i].setBConst(unionArrays[0][i].getBConst() != unionArrays[1][i].getBConst()); break; default: UNREACHABLE(); break; } } break; } case EOpDistance: { ASSERT(basicType == EbtFloat); TConstantUnion *distanceArray = new TConstantUnion[maxObjectSize]; resultArray = new TConstantUnion(); for (size_t i = 0; i < maxObjectSize; i++) { float x = unionArrays[0][i].getFConst(); float y = unionArrays[1][i].getFConst(); distanceArray[i].setFConst(x - y); } resultArray->setFConst(VectorLength(distanceArray, maxObjectSize)); break; } case EOpDot: ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion(); resultArray->setFConst(VectorDotProduct(unionArrays[0], unionArrays[1], maxObjectSize)); break; case EOpCross: { ASSERT(basicType == EbtFloat && maxObjectSize == 3); resultArray = new TConstantUnion[maxObjectSize]; float x0 = unionArrays[0][0].getFConst(); float x1 = unionArrays[0][1].getFConst(); float x2 = unionArrays[0][2].getFConst(); float y0 = unionArrays[1][0].getFConst(); float y1 = unionArrays[1][1].getFConst(); float y2 = unionArrays[1][2].getFConst(); resultArray[0].setFConst(x1 * y2 - y1 * x2); resultArray[1].setFConst(x2 * y0 - y2 * x0); resultArray[2].setFConst(x0 * y1 - y0 * x1); break; } case EOpReflect: { ASSERT(basicType == EbtFloat); // genType reflect (genType I, genType N) : // For the incident vector I and surface orientation N, returns the reflection // direction: // I - 2 * dot(N, I) * N. resultArray = new TConstantUnion[maxObjectSize]; float dotProduct = VectorDotProduct(unionArrays[1], unionArrays[0], maxObjectSize); for (size_t i = 0; i < maxObjectSize; i++) { float result = unionArrays[0][i].getFConst() - 2.0f * dotProduct * unionArrays[1][i].getFConst(); resultArray[i].setFConst(result); } break; } case EOpMulMatrixComponentWise: { ASSERT(basicType == EbtFloat && (*arguments)[0]->getAsTyped()->isMatrix() && (*arguments)[1]->getAsTyped()->isMatrix()); // Perform component-wise matrix multiplication. resultArray = new TConstantUnion[maxObjectSize]; int size = (*arguments)[0]->getAsTyped()->getNominalSize(); angle::Matrix result = GetMatrix(unionArrays[0], size).compMult(GetMatrix(unionArrays[1], size)); SetUnionArrayFromMatrix(result, resultArray); break; } case EOpOuterProduct: { ASSERT(basicType == EbtFloat); size_t numRows = (*arguments)[0]->getAsTyped()->getType().getObjectSize(); size_t numCols = (*arguments)[1]->getAsTyped()->getType().getObjectSize(); resultArray = new TConstantUnion[numRows * numCols]; angle::Matrix result = GetMatrix(unionArrays[0], static_cast(numRows), 1) .outerProduct(GetMatrix(unionArrays[1], 1, static_cast(numCols))); SetUnionArrayFromMatrix(result, resultArray); break; } case EOpClamp: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { switch (basicType) { case EbtFloat: { float x = unionArrays[0][i].getFConst(); float min = unionArrays[1][i].getFConst(); float max = unionArrays[2][i].getFConst(); // Results are undefined if min > max. if (min > max) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else resultArray[i].setFConst(gl::clamp(x, min, max)); break; } case EbtInt: { int x = unionArrays[0][i].getIConst(); int min = unionArrays[1][i].getIConst(); int max = unionArrays[2][i].getIConst(); // Results are undefined if min > max. if (min > max) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else resultArray[i].setIConst(gl::clamp(x, min, max)); break; } case EbtUInt: { unsigned int x = unionArrays[0][i].getUConst(); unsigned int min = unionArrays[1][i].getUConst(); unsigned int max = unionArrays[2][i].getUConst(); // Results are undefined if min > max. if (min > max) UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); else resultArray[i].setUConst(gl::clamp(x, min, max)); break; } default: UNREACHABLE(); break; } } break; } case EOpMix: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float x = unionArrays[0][i].getFConst(); float y = unionArrays[1][i].getFConst(); TBasicType type = (*arguments)[2]->getAsTyped()->getType().getBasicType(); if (type == EbtFloat) { // Returns the linear blend of x and y, i.e., x * (1 - a) + y * a. float a = unionArrays[2][i].getFConst(); resultArray[i].setFConst(x * (1.0f - a) + y * a); } else // 3rd parameter is EbtBool { ASSERT(type == EbtBool); // Selects which vector each returned component comes from. // For a component of a that is false, the corresponding component of x is // returned. // For a component of a that is true, the corresponding component of y is // returned. bool a = unionArrays[2][i].getBConst(); resultArray[i].setFConst(a ? y : x); } } break; } case EOpSmoothStep: { ASSERT(basicType == EbtFloat); resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float edge0 = unionArrays[0][i].getFConst(); float edge1 = unionArrays[1][i].getFConst(); float x = unionArrays[2][i].getFConst(); // Results are undefined if edge0 >= edge1. if (edge0 >= edge1) { UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); } else { // Returns 0.0 if x <= edge0 and 1.0 if x >= edge1 and performs smooth // Hermite interpolation between 0 and 1 when edge0 < x < edge1. float t = gl::clamp((x - edge0) / (edge1 - edge0), 0.0f, 1.0f); resultArray[i].setFConst(t * t * (3.0f - 2.0f * t)); } } break; } case EOpLdexp: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; i++) { float x = unionArrays[0][i].getFConst(); int exp = unionArrays[1][i].getIConst(); if (exp > 128) { UndefinedConstantFoldingError(loc, op, basicType, diagnostics, &resultArray[i]); } else { resultArray[i].setFConst(gl::Ldexp(x, exp)); } } break; } case EOpFaceforward: { ASSERT(basicType == EbtFloat); // genType faceforward(genType N, genType I, genType Nref) : // If dot(Nref, I) < 0 return N, otherwise return -N. resultArray = new TConstantUnion[maxObjectSize]; float dotProduct = VectorDotProduct(unionArrays[2], unionArrays[1], maxObjectSize); for (size_t i = 0; i < maxObjectSize; i++) { if (dotProduct < 0) resultArray[i].setFConst(unionArrays[0][i].getFConst()); else resultArray[i].setFConst(-unionArrays[0][i].getFConst()); } break; } case EOpRefract: { ASSERT(basicType == EbtFloat); // genType refract(genType I, genType N, float eta) : // For the incident vector I and surface normal N, and the ratio of indices of // refraction eta, // return the refraction vector. The result is computed by // k = 1.0 - eta * eta * (1.0 - dot(N, I) * dot(N, I)) // if (k < 0.0) // return genType(0.0) // else // return eta * I - (eta * dot(N, I) + sqrt(k)) * N resultArray = new TConstantUnion[maxObjectSize]; float dotProduct = VectorDotProduct(unionArrays[1], unionArrays[0], maxObjectSize); for (size_t i = 0; i < maxObjectSize; i++) { float eta = unionArrays[2][i].getFConst(); float k = 1.0f - eta * eta * (1.0f - dotProduct * dotProduct); if (k < 0.0f) resultArray[i].setFConst(0.0f); else resultArray[i].setFConst(eta * unionArrays[0][i].getFConst() - (eta * dotProduct + sqrtf(k)) * unionArrays[1][i].getFConst()); } break; } case EOpBitfieldExtract: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; ++i) { int offset = unionArrays[1][0].getIConst(); int bits = unionArrays[2][0].getIConst(); if (bits == 0) { if (aggregate->getBasicType() == EbtInt) { resultArray[i].setIConst(0); } else { ASSERT(aggregate->getBasicType() == EbtUInt); resultArray[i].setUConst(0); } } else if (offset < 0 || bits < 0 || offset >= 32 || bits > 32 || offset + bits > 32) { UndefinedConstantFoldingError(loc, op, aggregate->getBasicType(), diagnostics, &resultArray[i]); } else { // bits can be 32 here, so we need to avoid bit shift overflow. uint32_t maskMsb = 1u << (bits - 1); uint32_t mask = ((maskMsb - 1u) | maskMsb) << offset; if (aggregate->getBasicType() == EbtInt) { uint32_t value = static_cast(unionArrays[0][i].getIConst()); uint32_t resultUnsigned = (value & mask) >> offset; if ((resultUnsigned & maskMsb) != 0) { // The most significant bits (from bits+1 to the most significant bit) // should be set to 1. uint32_t higherBitsMask = ((1u << (32 - bits)) - 1u) << bits; resultUnsigned |= higherBitsMask; } resultArray[i].setIConst(static_cast(resultUnsigned)); } else { ASSERT(aggregate->getBasicType() == EbtUInt); uint32_t value = unionArrays[0][i].getUConst(); resultArray[i].setUConst((value & mask) >> offset); } } } break; } case EOpBitfieldInsert: { resultArray = new TConstantUnion[maxObjectSize]; for (size_t i = 0; i < maxObjectSize; ++i) { int offset = unionArrays[2][0].getIConst(); int bits = unionArrays[3][0].getIConst(); if (bits == 0) { if (aggregate->getBasicType() == EbtInt) { int32_t base = unionArrays[0][i].getIConst(); resultArray[i].setIConst(base); } else { ASSERT(aggregate->getBasicType() == EbtUInt); uint32_t base = unionArrays[0][i].getUConst(); resultArray[i].setUConst(base); } } else if (offset < 0 || bits < 0 || offset >= 32 || bits > 32 || offset + bits > 32) { UndefinedConstantFoldingError(loc, op, aggregate->getBasicType(), diagnostics, &resultArray[i]); } else { // bits can be 32 here, so we need to avoid bit shift overflow. uint32_t maskMsb = 1u << (bits - 1); uint32_t insertMask = ((maskMsb - 1u) | maskMsb) << offset; uint32_t baseMask = ~insertMask; if (aggregate->getBasicType() == EbtInt) { uint32_t base = static_cast(unionArrays[0][i].getIConst()); uint32_t insert = static_cast(unionArrays[1][i].getIConst()); uint32_t resultUnsigned = (base & baseMask) | ((insert << offset) & insertMask); resultArray[i].setIConst(static_cast(resultUnsigned)); } else { ASSERT(aggregate->getBasicType() == EbtUInt); uint32_t base = unionArrays[0][i].getUConst(); uint32_t insert = unionArrays[1][i].getUConst(); resultArray[i].setUConst((base & baseMask) | ((insert << offset) & insertMask)); } } } break; } default: UNREACHABLE(); return nullptr; } return resultArray; } } // namespace sh