// //Copyright (C) 2016 Google, Inc. //Copyright (C) 2016 LunarG, Inc. // //All rights reserved. // //Redistribution and use in source and binary forms, with or without //modification, are permitted provided that the following conditions //are met: // // Redistributions of source code must retain the above copyright // notice, this list of conditions and the following disclaimer. // // Redistributions in binary form must reproduce the above // copyright notice, this list of conditions and the following // disclaimer in the documentation and/or other materials provided // with the distribution. // // Neither the name of Google, Inc., nor the names of its // contributors may be used to endorse or promote products derived // from this software without specific prior written permission. // //THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS //"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT //LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS //FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE //COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, //INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, //BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; //LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER //CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT //LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN //ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE //POSSIBILITY OF SUCH DAMAGE. // // // This is a set of mutually recursive methods implementing the HLSL grammar. // Generally, each returns // - through an argument: a type specifically appropriate to which rule it // recognized // - through the return value: true/false to indicate whether or not it // recognized its rule // // As much as possible, only grammar recognition should happen in this file, // with all other work being farmed out to hlslParseHelper.cpp, which in turn // will build the AST. // // The next token, yet to be "accepted" is always sitting in 'token'. // When a method says it accepts a rule, that means all tokens involved // in the rule will have been consumed, and none left in 'token'. // #include "hlslTokens.h" #include "hlslGrammar.h" namespace glslang { // Root entry point to this recursive decent parser. // Return true if compilation unit was successfully accepted. bool HlslGrammar::parse() { advanceToken(); return acceptCompilationUnit(); } void HlslGrammar::expected(const char* syntax) { parseContext.error(token.loc, "Expected", syntax, ""); } void HlslGrammar::unimplemented(const char* error) { parseContext.error(token.loc, "Unimplemented", error, ""); } // Only process the next token if it is an identifier. // Return true if it was an identifier. bool HlslGrammar::acceptIdentifier(HlslToken& idToken) { if (peekTokenClass(EHTokIdentifier)) { idToken = token; advanceToken(); return true; } return false; } // compilationUnit // : list of externalDeclaration // bool HlslGrammar::acceptCompilationUnit() { TIntermNode* unitNode = nullptr; while (! peekTokenClass(EHTokNone)) { // externalDeclaration TIntermNode* declarationNode; if (! acceptDeclaration(declarationNode)) return false; // hook it up unitNode = intermediate.growAggregate(unitNode, declarationNode); } // set root of AST intermediate.setTreeRoot(unitNode); return true; } // sampler_state // : LEFT_BRACE [sampler_state_assignment ... ] RIGHT_BRACE // // sampler_state_assignment // : sampler_state_identifier EQUAL value SEMICOLON // // sampler_state_identifier // : ADDRESSU // | ADDRESSV // | ADDRESSW // | BORDERCOLOR // | FILTER // | MAXANISOTROPY // | MAXLOD // | MINLOD // | MIPLODBIAS // bool HlslGrammar::acceptSamplerState() { // TODO: this should be genericized to accept a list of valid tokens and // return token/value pairs. Presently it is specific to texture values. if (! acceptTokenClass(EHTokLeftBrace)) return true; parseContext.warn(token.loc, "unimplemented", "immediate sampler state", ""); do { // read state name HlslToken state; if (! acceptIdentifier(state)) break; // end of list // FXC accepts any case TString stateName = *state.string; std::transform(stateName.begin(), stateName.end(), stateName.begin(), ::tolower); if (! acceptTokenClass(EHTokAssign)) { expected("assign"); return false; } if (stateName == "minlod" || stateName == "maxlod") { if (! peekTokenClass(EHTokIntConstant)) { expected("integer"); return false; } TIntermTyped* lod = nullptr; if (! acceptLiteral(lod)) // should never fail, since we just looked for an integer return false; } else if (stateName == "maxanisotropy") { if (! peekTokenClass(EHTokIntConstant)) { expected("integer"); return false; } TIntermTyped* maxAnisotropy = nullptr; if (! acceptLiteral(maxAnisotropy)) // should never fail, since we just looked for an integer return false; } else if (stateName == "filter") { HlslToken filterMode; if (! acceptIdentifier(filterMode)) { expected("filter mode"); return false; } } else if (stateName == "addressu" || stateName == "addressv" || stateName == "addressw") { HlslToken addrMode; if (! acceptIdentifier(addrMode)) { expected("texture address mode"); return false; } } else if (stateName == "miplodbias") { TIntermTyped* lodBias = nullptr; if (! acceptLiteral(lodBias)) { expected("lod bias"); return false; } } else if (stateName == "bordercolor") { return false; } else { expected("texture state"); return false; } // SEMICOLON if (! acceptTokenClass(EHTokSemicolon)) { expected("semicolon"); return false; } } while (true); if (! acceptTokenClass(EHTokRightBrace)) return false; return true; } // sampler_declaration_dx9 // : SAMPLER identifier EQUAL sampler_type sampler_state // bool HlslGrammar::acceptSamplerDeclarationDX9(TType& /*type*/) { if (! acceptTokenClass(EHTokSampler)) return false; // TODO: remove this when DX9 style declarations are implemented. unimplemented("Direct3D 9 sampler declaration"); // read sampler name HlslToken name; if (! acceptIdentifier(name)) { expected("sampler name"); return false; } if (! acceptTokenClass(EHTokAssign)) { expected("="); return false; } return false; } // declaration // : sampler_declaration_dx9 post_decls SEMICOLON // | fully_specified_type declarator_list SEMICOLON // | fully_specified_type identifier function_parameters post_decls compound_statement // function definition // | fully_specified_type identifier sampler_state post_decls compound_statement // sampler definition // | typedef declaration // // declarator_list // : declarator COMMA declarator COMMA declarator... // zero or more declarators // // declarator // : identifier array_specifier post_decls // | identifier array_specifier post_decls EQUAL assignment_expression // | identifier function_parameters post_decls // function prototype // // Parsing has to go pretty far in to know whether it's a variable, prototype, or // function definition, so the implementation below doesn't perfectly divide up the grammar // as above. (The 'identifier' in the first item in init_declarator list is the // same as 'identifier' for function declarations.) // // 'node' could get populated if the declaration creates code, like an initializer // or a function body. // bool HlslGrammar::acceptDeclaration(TIntermNode*& node) { node = nullptr; bool list = false; // typedef bool typedefDecl = acceptTokenClass(EHTokTypedef); TType type; // DX9 sampler declaration use a different syntax if (acceptSamplerDeclarationDX9(type)) return true; // fully_specified_type if (! acceptFullySpecifiedType(type)) return false; if (type.getQualifier().storage == EvqTemporary && parseContext.symbolTable.atGlobalLevel()) { if (type.getBasicType() == EbtSampler) { // Sampler/textures are uniform by default (if no explicit qualifier is present) in // HLSL. This line silently converts samplers *explicitly* declared static to uniform, // which is incorrect but harmless. type.getQualifier().storage = EvqUniform; } else { type.getQualifier().storage = EvqGlobal; } } // identifier HlslToken idToken; while (acceptIdentifier(idToken)) { // function_parameters TFunction* function = new TFunction(idToken.string, type); if (acceptFunctionParameters(*function)) { // post_decls acceptPostDecls(type); // compound_statement (function body definition) or just a prototype? if (peekTokenClass(EHTokLeftBrace)) { if (list) parseContext.error(idToken.loc, "function body can't be in a declarator list", "{", ""); if (typedefDecl) parseContext.error(idToken.loc, "function body can't be in a typedef", "{", ""); return acceptFunctionDefinition(*function, node); } else { if (typedefDecl) parseContext.error(idToken.loc, "function typedefs not implemented", "{", ""); parseContext.handleFunctionDeclarator(idToken.loc, *function, true); } } else { // a variable declaration // array_specifier TArraySizes* arraySizes = nullptr; acceptArraySpecifier(arraySizes); // samplers accept immediate sampler state if (type.getBasicType() == EbtSampler) { if (! acceptSamplerState()) return false; } // post_decls acceptPostDecls(type); // EQUAL assignment_expression TIntermTyped* expressionNode = nullptr; if (acceptTokenClass(EHTokAssign)) { if (typedefDecl) parseContext.error(idToken.loc, "can't have an initializer", "typedef", ""); if (! acceptAssignmentExpression(expressionNode)) { expected("initializer"); return false; } } if (typedefDecl) parseContext.declareTypedef(idToken.loc, *idToken.string, type, arraySizes); else { // Declare the variable and add any initializer code to the AST. // The top-level node is always made into an aggregate, as that's // historically how the AST has been. node = intermediate.growAggregate(node, parseContext.declareVariable(idToken.loc, *idToken.string, type, arraySizes, expressionNode), idToken.loc); } } if (acceptTokenClass(EHTokComma)) { list = true; continue; } }; // The top-level node is a sequence. if (node != nullptr) node->getAsAggregate()->setOperator(EOpSequence); // SEMICOLON if (! acceptTokenClass(EHTokSemicolon)) { expected(";"); return false; } return true; } // control_declaration // : fully_specified_type identifier EQUAL expression // bool HlslGrammar::acceptControlDeclaration(TIntermNode*& node) { node = nullptr; // fully_specified_type TType type; if (! acceptFullySpecifiedType(type)) return false; // identifier HlslToken idToken; if (! acceptIdentifier(idToken)) { expected("identifier"); return false; } // EQUAL TIntermTyped* expressionNode = nullptr; if (! acceptTokenClass(EHTokAssign)) { expected("="); return false; } // expression if (! acceptExpression(expressionNode)) { expected("initializer"); return false; } node = parseContext.declareVariable(idToken.loc, *idToken.string, type, 0, expressionNode); return true; } // fully_specified_type // : type_specifier // | type_qualifier type_specifier // bool HlslGrammar::acceptFullySpecifiedType(TType& type) { // type_qualifier TQualifier qualifier; qualifier.clear(); acceptQualifier(qualifier); // type_specifier if (! acceptType(type)) return false; type.getQualifier() = qualifier; return true; } // type_qualifier // : qualifier qualifier ... // // Zero or more of these, so this can't return false. // void HlslGrammar::acceptQualifier(TQualifier& qualifier) { do { switch (peek()) { case EHTokStatic: // normal glslang default break; case EHTokExtern: // TODO: no meaning in glslang? break; case EHTokShared: // TODO: hint break; case EHTokGroupShared: qualifier.storage = EvqShared; break; case EHTokUniform: qualifier.storage = EvqUniform; break; case EHTokConst: qualifier.storage = EvqConst; break; case EHTokVolatile: qualifier.volatil = true; break; case EHTokLinear: qualifier.storage = EvqVaryingIn; qualifier.smooth = true; break; case EHTokCentroid: qualifier.centroid = true; break; case EHTokNointerpolation: qualifier.flat = true; break; case EHTokNoperspective: qualifier.nopersp = true; break; case EHTokSample: qualifier.sample = true; break; case EHTokRowMajor: qualifier.layoutMatrix = ElmRowMajor; break; case EHTokColumnMajor: qualifier.layoutMatrix = ElmColumnMajor; break; case EHTokPrecise: qualifier.noContraction = true; break; case EHTokIn: qualifier.storage = EvqIn; break; case EHTokOut: qualifier.storage = EvqOut; break; case EHTokInOut: qualifier.storage = EvqInOut; break; default: return; } advanceToken(); } while (true); } // template_type // : FLOAT // | DOUBLE // | INT // | DWORD // | UINT // | BOOL // bool HlslGrammar::acceptTemplateType(TBasicType& basicType) { switch (peek()) { case EHTokFloat: basicType = EbtFloat; break; case EHTokDouble: basicType = EbtDouble; break; case EHTokInt: case EHTokDword: basicType = EbtInt; break; case EHTokUint: basicType = EbtUint; break; case EHTokBool: basicType = EbtBool; break; default: return false; } advanceToken(); return true; } // vector_template_type // : VECTOR // | VECTOR LEFT_ANGLE template_type COMMA integer_literal RIGHT_ANGLE // bool HlslGrammar::acceptVectorTemplateType(TType& type) { if (! acceptTokenClass(EHTokVector)) return false; if (! acceptTokenClass(EHTokLeftAngle)) { // in HLSL, 'vector' alone means float4. new(&type) TType(EbtFloat, EvqTemporary, 4); return true; } TBasicType basicType; if (! acceptTemplateType(basicType)) { expected("scalar type"); return false; } // COMMA if (! acceptTokenClass(EHTokComma)) { expected(","); return false; } // integer if (! peekTokenClass(EHTokIntConstant)) { expected("literal integer"); return false; } TIntermTyped* vecSize; if (! acceptLiteral(vecSize)) return false; const int vecSizeI = vecSize->getAsConstantUnion()->getConstArray()[0].getIConst(); new(&type) TType(basicType, EvqTemporary, vecSizeI); if (vecSizeI == 1) type.makeVector(); if (!acceptTokenClass(EHTokRightAngle)) { expected("right angle bracket"); return false; } return true; } // matrix_template_type // : MATRIX // | MATRIX LEFT_ANGLE template_type COMMA integer_literal COMMA integer_literal RIGHT_ANGLE // bool HlslGrammar::acceptMatrixTemplateType(TType& type) { if (! acceptTokenClass(EHTokMatrix)) return false; if (! acceptTokenClass(EHTokLeftAngle)) { // in HLSL, 'matrix' alone means float4x4. new(&type) TType(EbtFloat, EvqTemporary, 0, 4, 4); return true; } TBasicType basicType; if (! acceptTemplateType(basicType)) { expected("scalar type"); return false; } // COMMA if (! acceptTokenClass(EHTokComma)) { expected(","); return false; } // integer rows if (! peekTokenClass(EHTokIntConstant)) { expected("literal integer"); return false; } TIntermTyped* rows; if (! acceptLiteral(rows)) return false; // COMMA if (! acceptTokenClass(EHTokComma)) { expected(","); return false; } // integer cols if (! peekTokenClass(EHTokIntConstant)) { expected("literal integer"); return false; } TIntermTyped* cols; if (! acceptLiteral(cols)) return false; new(&type) TType(basicType, EvqTemporary, 0, cols->getAsConstantUnion()->getConstArray()[0].getIConst(), rows->getAsConstantUnion()->getConstArray()[0].getIConst()); if (!acceptTokenClass(EHTokRightAngle)) { expected("right angle bracket"); return false; } return true; } // sampler_type // : SAMPLER // | SAMPLER1D // | SAMPLER2D // | SAMPLER3D // | SAMPLERCUBE // | SAMPLERSTATE // | SAMPLERCOMPARISONSTATE bool HlslGrammar::acceptSamplerType(TType& type) { // read sampler type const EHlslTokenClass samplerType = peek(); TSamplerDim dim = EsdNone; switch (samplerType) { case EHTokSampler: break; case EHTokSampler1d: dim = Esd1D; break; case EHTokSampler2d: dim = Esd2D; break; case EHTokSampler3d: dim = Esd3D; break; case EHTokSamplerCube: dim = EsdCube; break; case EHTokSamplerState: break; case EHTokSamplerComparisonState: break; default: return false; // not a sampler declaration } advanceToken(); // consume the sampler type keyword TArraySizes* arraySizes = nullptr; // TODO: array bool shadow = false; // TODO: shadow TSampler sampler; sampler.setPureSampler(shadow); type.shallowCopy(TType(sampler, EvqUniform, arraySizes)); return true; } // texture_type // | BUFFER // | TEXTURE1D // | TEXTURE1DARRAY // | TEXTURE2D // | TEXTURE2DARRAY // | TEXTURE3D // | TEXTURECUBE // | TEXTURECUBEARRAY // | TEXTURE2DMS // | TEXTURE2DMSARRAY bool HlslGrammar::acceptTextureType(TType& type) { const EHlslTokenClass textureType = peek(); TSamplerDim dim = EsdNone; bool array = false; bool ms = false; switch (textureType) { case EHTokBuffer: dim = EsdBuffer; break; case EHTokTexture1d: dim = Esd1D; break; case EHTokTexture1darray: dim = Esd1D; array = true; break; case EHTokTexture2d: dim = Esd2D; break; case EHTokTexture2darray: dim = Esd2D; array = true; break; case EHTokTexture3d: dim = Esd3D; break; case EHTokTextureCube: dim = EsdCube; break; case EHTokTextureCubearray: dim = EsdCube; array = true; break; case EHTokTexture2DMS: dim = Esd2D; ms = true; break; case EHTokTexture2DMSarray: dim = Esd2D; array = true; ms = true; break; default: return false; // not a texture declaration } advanceToken(); // consume the texture object keyword TType txType(EbtFloat, EvqUniform, 4); // default type is float4 TIntermTyped* msCount = nullptr; // texture type: required for multisample types! if (acceptTokenClass(EHTokLeftAngle)) { if (! acceptType(txType)) { expected("scalar or vector type"); return false; } const TBasicType basicRetType = txType.getBasicType() ; if (basicRetType != EbtFloat && basicRetType != EbtUint && basicRetType != EbtInt) { unimplemented("basic type in texture"); return false; } if (!txType.isScalar() && !txType.isVector()) { expected("scalar or vector type"); return false; } if (txType.getVectorSize() != 1 && txType.getVectorSize() != 4) { // TODO: handle vec2/3 types expected("vector size not yet supported in texture type"); return false; } if (ms && acceptTokenClass(EHTokComma)) { // read sample count for multisample types, if given if (! peekTokenClass(EHTokIntConstant)) { expected("multisample count"); return false; } if (! acceptLiteral(msCount)) // should never fail, since we just found an integer return false; } if (! acceptTokenClass(EHTokRightAngle)) { expected("right angle bracket"); return false; } } else if (ms) { expected("texture type for multisample"); return false; } TArraySizes* arraySizes = nullptr; const bool shadow = txType.isScalar() || (txType.isVector() && txType.getVectorSize() == 1); TSampler sampler; sampler.setTexture(txType.getBasicType(), dim, array, shadow, ms); type.shallowCopy(TType(sampler, EvqUniform, arraySizes)); return true; } // If token is for a type, update 'type' with the type information, // and return true and advance. // Otherwise, return false, and don't advance bool HlslGrammar::acceptType(TType& type) { switch (peek()) { case EHTokVector: return acceptVectorTemplateType(type); break; case EHTokMatrix: return acceptMatrixTemplateType(type); break; case EHTokSampler: // fall through case EHTokSampler1d: // ... case EHTokSampler2d: // ... case EHTokSampler3d: // ... case EHTokSamplerCube: // ... case EHTokSamplerState: // ... case EHTokSamplerComparisonState: // ... return acceptSamplerType(type); break; case EHTokBuffer: // fall through case EHTokTexture1d: // ... case EHTokTexture1darray: // ... case EHTokTexture2d: // ... case EHTokTexture2darray: // ... case EHTokTexture3d: // ... case EHTokTextureCube: // ... case EHTokTextureCubearray: // ... case EHTokTexture2DMS: // ... case EHTokTexture2DMSarray: // ... return acceptTextureType(type); break; case EHTokStruct: return acceptStruct(type); break; case EHTokIdentifier: // An identifier could be for a user-defined type. // Note we cache the symbol table lookup, to save for a later rule // when this is not a type. token.symbol = parseContext.symbolTable.find(*token.string); if (token.symbol && token.symbol->getAsVariable() && token.symbol->getAsVariable()->isUserType()) { type.shallowCopy(token.symbol->getType()); advanceToken(); return true; } else return false; case EHTokVoid: new(&type) TType(EbtVoid); break; case EHTokFloat: new(&type) TType(EbtFloat); break; case EHTokFloat1: new(&type) TType(EbtFloat); type.makeVector(); break; case EHTokFloat2: new(&type) TType(EbtFloat, EvqTemporary, 2); break; case EHTokFloat3: new(&type) TType(EbtFloat, EvqTemporary, 3); break; case EHTokFloat4: new(&type) TType(EbtFloat, EvqTemporary, 4); break; case EHTokDouble: new(&type) TType(EbtDouble); break; case EHTokDouble1: new(&type) TType(EbtDouble); type.makeVector(); break; case EHTokDouble2: new(&type) TType(EbtDouble, EvqTemporary, 2); break; case EHTokDouble3: new(&type) TType(EbtDouble, EvqTemporary, 3); break; case EHTokDouble4: new(&type) TType(EbtDouble, EvqTemporary, 4); break; case EHTokInt: case EHTokDword: new(&type) TType(EbtInt); break; case EHTokInt1: new(&type) TType(EbtInt); type.makeVector(); break; case EHTokInt2: new(&type) TType(EbtInt, EvqTemporary, 2); break; case EHTokInt3: new(&type) TType(EbtInt, EvqTemporary, 3); break; case EHTokInt4: new(&type) TType(EbtInt, EvqTemporary, 4); break; case EHTokUint: new(&type) TType(EbtUint); break; case EHTokUint1: new(&type) TType(EbtUint); type.makeVector(); break; case EHTokUint2: new(&type) TType(EbtUint, EvqTemporary, 2); break; case EHTokUint3: new(&type) TType(EbtUint, EvqTemporary, 3); break; case EHTokUint4: new(&type) TType(EbtUint, EvqTemporary, 4); break; case EHTokBool: new(&type) TType(EbtBool); break; case EHTokBool1: new(&type) TType(EbtBool); type.makeVector(); break; case EHTokBool2: new(&type) TType(EbtBool, EvqTemporary, 2); break; case EHTokBool3: new(&type) TType(EbtBool, EvqTemporary, 3); break; case EHTokBool4: new(&type) TType(EbtBool, EvqTemporary, 4); break; case EHTokInt1x1: new(&type) TType(EbtInt, EvqTemporary, 0, 1, 1); break; case EHTokInt1x2: new(&type) TType(EbtInt, EvqTemporary, 0, 2, 1); break; case EHTokInt1x3: new(&type) TType(EbtInt, EvqTemporary, 0, 3, 1); break; case EHTokInt1x4: new(&type) TType(EbtInt, EvqTemporary, 0, 4, 1); break; case EHTokInt2x1: new(&type) TType(EbtInt, EvqTemporary, 0, 1, 2); break; case EHTokInt2x2: new(&type) TType(EbtInt, EvqTemporary, 0, 2, 2); break; case EHTokInt2x3: new(&type) TType(EbtInt, EvqTemporary, 0, 3, 2); break; case EHTokInt2x4: new(&type) TType(EbtInt, EvqTemporary, 0, 4, 2); break; case EHTokInt3x1: new(&type) TType(EbtInt, EvqTemporary, 0, 1, 3); break; case EHTokInt3x2: new(&type) TType(EbtInt, EvqTemporary, 0, 2, 3); break; case EHTokInt3x3: new(&type) TType(EbtInt, EvqTemporary, 0, 3, 3); break; case EHTokInt3x4: new(&type) TType(EbtInt, EvqTemporary, 0, 4, 3); break; case EHTokInt4x1: new(&type) TType(EbtInt, EvqTemporary, 0, 1, 4); break; case EHTokInt4x2: new(&type) TType(EbtInt, EvqTemporary, 0, 2, 4); break; case EHTokInt4x3: new(&type) TType(EbtInt, EvqTemporary, 0, 3, 4); break; case EHTokInt4x4: new(&type) TType(EbtInt, EvqTemporary, 0, 4, 4); break; case EHTokUint1x1: new(&type) TType(EbtUint, EvqTemporary, 0, 1, 1); break; case EHTokUint1x2: new(&type) TType(EbtUint, EvqTemporary, 0, 2, 1); break; case EHTokUint1x3: new(&type) TType(EbtUint, EvqTemporary, 0, 3, 1); break; case EHTokUint1x4: new(&type) TType(EbtUint, EvqTemporary, 0, 4, 1); break; case EHTokUint2x1: new(&type) TType(EbtUint, EvqTemporary, 0, 1, 2); break; case EHTokUint2x2: new(&type) TType(EbtUint, EvqTemporary, 0, 2, 2); break; case EHTokUint2x3: new(&type) TType(EbtUint, EvqTemporary, 0, 3, 2); break; case EHTokUint2x4: new(&type) TType(EbtUint, EvqTemporary, 0, 4, 2); break; case EHTokUint3x1: new(&type) TType(EbtUint, EvqTemporary, 0, 1, 3); break; case EHTokUint3x2: new(&type) TType(EbtUint, EvqTemporary, 0, 2, 3); break; case EHTokUint3x3: new(&type) TType(EbtUint, EvqTemporary, 0, 3, 3); break; case EHTokUint3x4: new(&type) TType(EbtUint, EvqTemporary, 0, 4, 3); break; case EHTokUint4x1: new(&type) TType(EbtUint, EvqTemporary, 0, 1, 4); break; case EHTokUint4x2: new(&type) TType(EbtUint, EvqTemporary, 0, 2, 4); break; case EHTokUint4x3: new(&type) TType(EbtUint, EvqTemporary, 0, 3, 4); break; case EHTokUint4x4: new(&type) TType(EbtUint, EvqTemporary, 0, 4, 4); break; case EHTokBool1x1: new(&type) TType(EbtBool, EvqTemporary, 0, 1, 1); break; case EHTokBool1x2: new(&type) TType(EbtBool, EvqTemporary, 0, 2, 1); break; case EHTokBool1x3: new(&type) TType(EbtBool, EvqTemporary, 0, 3, 1); break; case EHTokBool1x4: new(&type) TType(EbtBool, EvqTemporary, 0, 4, 1); break; case EHTokBool2x1: new(&type) TType(EbtBool, EvqTemporary, 0, 1, 2); break; case EHTokBool2x2: new(&type) TType(EbtBool, EvqTemporary, 0, 2, 2); break; case EHTokBool2x3: new(&type) TType(EbtBool, EvqTemporary, 0, 3, 2); break; case EHTokBool2x4: new(&type) TType(EbtBool, EvqTemporary, 0, 4, 2); break; case EHTokBool3x1: new(&type) TType(EbtBool, EvqTemporary, 0, 1, 3); break; case EHTokBool3x2: new(&type) TType(EbtBool, EvqTemporary, 0, 2, 3); break; case EHTokBool3x3: new(&type) TType(EbtBool, EvqTemporary, 0, 3, 3); break; case EHTokBool3x4: new(&type) TType(EbtBool, EvqTemporary, 0, 4, 3); break; case EHTokBool4x1: new(&type) TType(EbtBool, EvqTemporary, 0, 1, 4); break; case EHTokBool4x2: new(&type) TType(EbtBool, EvqTemporary, 0, 2, 4); break; case EHTokBool4x3: new(&type) TType(EbtBool, EvqTemporary, 0, 3, 4); break; case EHTokBool4x4: new(&type) TType(EbtBool, EvqTemporary, 0, 4, 4); break; case EHTokFloat1x1: new(&type) TType(EbtFloat, EvqTemporary, 0, 1, 1); break; case EHTokFloat1x2: new(&type) TType(EbtFloat, EvqTemporary, 0, 2, 1); break; case EHTokFloat1x3: new(&type) TType(EbtFloat, EvqTemporary, 0, 3, 1); break; case EHTokFloat1x4: new(&type) TType(EbtFloat, EvqTemporary, 0, 4, 1); break; case EHTokFloat2x1: new(&type) TType(EbtFloat, EvqTemporary, 0, 1, 2); break; case EHTokFloat2x2: new(&type) TType(EbtFloat, EvqTemporary, 0, 2, 2); break; case EHTokFloat2x3: new(&type) TType(EbtFloat, EvqTemporary, 0, 3, 2); break; case EHTokFloat2x4: new(&type) TType(EbtFloat, EvqTemporary, 0, 4, 2); break; case EHTokFloat3x1: new(&type) TType(EbtFloat, EvqTemporary, 0, 1, 3); break; case EHTokFloat3x2: new(&type) TType(EbtFloat, EvqTemporary, 0, 2, 3); break; case EHTokFloat3x3: new(&type) TType(EbtFloat, EvqTemporary, 0, 3, 3); break; case EHTokFloat3x4: new(&type) TType(EbtFloat, EvqTemporary, 0, 4, 3); break; case EHTokFloat4x1: new(&type) TType(EbtFloat, EvqTemporary, 0, 1, 4); break; case EHTokFloat4x2: new(&type) TType(EbtFloat, EvqTemporary, 0, 2, 4); break; case EHTokFloat4x3: new(&type) TType(EbtFloat, EvqTemporary, 0, 3, 4); break; case EHTokFloat4x4: new(&type) TType(EbtFloat, EvqTemporary, 0, 4, 4); break; case EHTokDouble1x1: new(&type) TType(EbtDouble, EvqTemporary, 0, 1, 1); break; case EHTokDouble1x2: new(&type) TType(EbtDouble, EvqTemporary, 0, 2, 1); break; case EHTokDouble1x3: new(&type) TType(EbtDouble, EvqTemporary, 0, 3, 1); break; case EHTokDouble1x4: new(&type) TType(EbtDouble, EvqTemporary, 0, 4, 1); break; case EHTokDouble2x1: new(&type) TType(EbtDouble, EvqTemporary, 0, 1, 2); break; case EHTokDouble2x2: new(&type) TType(EbtDouble, EvqTemporary, 0, 2, 2); break; case EHTokDouble2x3: new(&type) TType(EbtDouble, EvqTemporary, 0, 3, 2); break; case EHTokDouble2x4: new(&type) TType(EbtDouble, EvqTemporary, 0, 4, 2); break; case EHTokDouble3x1: new(&type) TType(EbtDouble, EvqTemporary, 0, 1, 3); break; case EHTokDouble3x2: new(&type) TType(EbtDouble, EvqTemporary, 0, 2, 3); break; case EHTokDouble3x3: new(&type) TType(EbtDouble, EvqTemporary, 0, 3, 3); break; case EHTokDouble3x4: new(&type) TType(EbtDouble, EvqTemporary, 0, 4, 3); break; case EHTokDouble4x1: new(&type) TType(EbtDouble, EvqTemporary, 0, 1, 4); break; case EHTokDouble4x2: new(&type) TType(EbtDouble, EvqTemporary, 0, 2, 4); break; case EHTokDouble4x3: new(&type) TType(EbtDouble, EvqTemporary, 0, 3, 4); break; case EHTokDouble4x4: new(&type) TType(EbtDouble, EvqTemporary, 0, 4, 4); break; default: return false; } advanceToken(); return true; } // struct // : STRUCT IDENTIFIER LEFT_BRACE struct_declaration_list RIGHT_BRACE // | STRUCT LEFT_BRACE struct_declaration_list RIGHT_BRACE // bool HlslGrammar::acceptStruct(TType& type) { // STRUCT if (! acceptTokenClass(EHTokStruct)) return false; // IDENTIFIER TString structName = ""; if (peekTokenClass(EHTokIdentifier)) { structName = *token.string; advanceToken(); } // LEFT_BRACE if (! acceptTokenClass(EHTokLeftBrace)) { expected("{"); return false; } // struct_declaration_list TTypeList* typeList; if (! acceptStructDeclarationList(typeList)) { expected("struct member declarations"); return false; } // RIGHT_BRACE if (! acceptTokenClass(EHTokRightBrace)) { expected("}"); return false; } // create the user-defined type new(&type) TType(typeList, structName); // If it was named, which means it can be reused later, add // it to the symbol table. if (structName.size() > 0) { TVariable* userTypeDef = new TVariable(&structName, type, true); if (! parseContext.symbolTable.insert(*userTypeDef)) parseContext.error(token.loc, "redefinition", structName.c_str(), "struct"); } return true; } // struct_declaration_list // : struct_declaration SEMI_COLON struct_declaration SEMI_COLON ... // // struct_declaration // : fully_specified_type struct_declarator COMMA struct_declarator ... // // struct_declarator // : IDENTIFIER post_decls // | IDENTIFIER array_specifier post_decls // bool HlslGrammar::acceptStructDeclarationList(TTypeList*& typeList) { typeList = new TTypeList(); do { // success on seeing the RIGHT_BRACE coming up if (peekTokenClass(EHTokRightBrace)) return true; // struct_declaration // fully_specified_type TType memberType; if (! acceptFullySpecifiedType(memberType)) { expected("member type"); return false; } // struct_declarator COMMA struct_declarator ... do { // peek IDENTIFIER if (! peekTokenClass(EHTokIdentifier)) { expected("member name"); return false; } // add it to the list of members TTypeLoc member = { new TType(EbtVoid), token.loc }; member.type->shallowCopy(memberType); member.type->setFieldName(*token.string); typeList->push_back(member); // accept IDENTIFIER advanceToken(); // array_specifier TArraySizes* arraySizes = nullptr; acceptArraySpecifier(arraySizes); if (arraySizes) typeList->back().type->newArraySizes(*arraySizes); acceptPostDecls(*member.type); // success on seeing the SEMICOLON coming up if (peekTokenClass(EHTokSemicolon)) break; // COMMA if (! acceptTokenClass(EHTokComma)) { expected(","); return false; } } while (true); // SEMI_COLON if (! acceptTokenClass(EHTokSemicolon)) { expected(";"); return false; } } while (true); } // function_parameters // : LEFT_PAREN parameter_declaration COMMA parameter_declaration ... RIGHT_PAREN // | LEFT_PAREN VOID RIGHT_PAREN // bool HlslGrammar::acceptFunctionParameters(TFunction& function) { // LEFT_PAREN if (! acceptTokenClass(EHTokLeftParen)) return false; // VOID RIGHT_PAREN if (! acceptTokenClass(EHTokVoid)) { do { // parameter_declaration if (! acceptParameterDeclaration(function)) break; // COMMA if (! acceptTokenClass(EHTokComma)) break; } while (true); } // RIGHT_PAREN if (! acceptTokenClass(EHTokRightParen)) { expected(")"); return false; } return true; } // parameter_declaration // : fully_specified_type post_decls // | fully_specified_type identifier array_specifier post_decls // bool HlslGrammar::acceptParameterDeclaration(TFunction& function) { // fully_specified_type TType* type = new TType; if (! acceptFullySpecifiedType(*type)) return false; // identifier HlslToken idToken; acceptIdentifier(idToken); // array_specifier TArraySizes* arraySizes = nullptr; acceptArraySpecifier(arraySizes); if (arraySizes) type->newArraySizes(*arraySizes); // post_decls acceptPostDecls(*type); parseContext.paramFix(*type); TParameter param = { idToken.string, type }; function.addParameter(param); return true; } // Do the work to create the function definition in addition to // parsing the body (compound_statement). bool HlslGrammar::acceptFunctionDefinition(TFunction& function, TIntermNode*& node) { TFunction* functionDeclarator = parseContext.handleFunctionDeclarator(token.loc, function, false /* not prototype */); // This does a pushScope() node = parseContext.handleFunctionDefinition(token.loc, *functionDeclarator); // compound_statement TIntermNode* functionBody = nullptr; if (acceptCompoundStatement(functionBody)) { node = intermediate.growAggregate(node, functionBody); intermediate.setAggregateOperator(node, EOpFunction, functionDeclarator->getType(), token.loc); node->getAsAggregate()->setName(functionDeclarator->getMangledName().c_str()); parseContext.popScope(); return true; } return false; } // Accept an expression with parenthesis around it, where // the parenthesis ARE NOT expression parenthesis, but the // syntactically required ones like in "if ( expression )". // // Also accepts a declaration expression; "if (int a = expression)". // // Note this one is not set up to be speculative; as it gives // errors if not found. // bool HlslGrammar::acceptParenExpression(TIntermTyped*& expression) { // LEFT_PAREN if (! acceptTokenClass(EHTokLeftParen)) expected("("); bool decl = false; TIntermNode* declNode = nullptr; decl = acceptControlDeclaration(declNode); if (decl) { if (declNode == nullptr || declNode->getAsTyped() == nullptr) { expected("initialized declaration"); return false; } else expression = declNode->getAsTyped(); } else { // no declaration if (! acceptExpression(expression)) { expected("expression"); return false; } } // RIGHT_PAREN if (! acceptTokenClass(EHTokRightParen)) expected(")"); return true; } // The top-level full expression recognizer. // // expression // : assignment_expression COMMA assignment_expression COMMA assignment_expression ... // bool HlslGrammar::acceptExpression(TIntermTyped*& node) { node = nullptr; // assignment_expression if (! acceptAssignmentExpression(node)) return false; if (! peekTokenClass(EHTokComma)) return true; do { // ... COMMA TSourceLoc loc = token.loc; advanceToken(); // ... assignment_expression TIntermTyped* rightNode = nullptr; if (! acceptAssignmentExpression(rightNode)) { expected("assignment expression"); return false; } node = intermediate.addComma(node, rightNode, loc); if (! peekTokenClass(EHTokComma)) return true; } while (true); } // initializer // : LEFT_BRACE initializer_list RIGHT_BRACE // // initializer_list // : assignment_expression COMMA assignment_expression COMMA ... // bool HlslGrammar::acceptInitializer(TIntermTyped*& node) { // LEFT_BRACE if (! acceptTokenClass(EHTokLeftBrace)) return false; // initializer_list TSourceLoc loc = token.loc; node = nullptr; do { // assignment_expression TIntermTyped* expr; if (! acceptAssignmentExpression(expr)) { expected("assignment expression in initializer list"); return false; } node = intermediate.growAggregate(node, expr, loc); // COMMA if (acceptTokenClass(EHTokComma)) continue; // RIGHT_BRACE if (acceptTokenClass(EHTokRightBrace)) return true; expected(", or }"); return false; } while (true); } // Accept an assignment expression, where assignment operations // associate right-to-left. That is, it is implicit, for example // // a op (b op (c op d)) // // assigment_expression // : binary_expression op binary_expression op binary_expression ... // | initializer // bool HlslGrammar::acceptAssignmentExpression(TIntermTyped*& node) { // initializer if (peekTokenClass(EHTokLeftBrace)) { if (acceptInitializer(node)) return true; expected("initializer"); return false; } // binary_expression if (! acceptBinaryExpression(node, PlLogicalOr)) return false; // assignment operation? TOperator assignOp = HlslOpMap::assignment(peek()); if (assignOp == EOpNull) return true; // assignment op TSourceLoc loc = token.loc; advanceToken(); // binary_expression // But, done by recursing this function, which automatically // gets the right-to-left associativity. TIntermTyped* rightNode = nullptr; if (! acceptAssignmentExpression(rightNode)) { expected("assignment expression"); return false; } node = intermediate.addAssign(assignOp, node, rightNode, loc); if (! peekTokenClass(EHTokComma)) return true; return true; } // Accept a binary expression, for binary operations that // associate left-to-right. This is, it is implicit, for example // // ((a op b) op c) op d // // binary_expression // : expression op expression op expression ... // // where 'expression' is the next higher level in precedence. // bool HlslGrammar::acceptBinaryExpression(TIntermTyped*& node, PrecedenceLevel precedenceLevel) { if (precedenceLevel > PlMul) return acceptUnaryExpression(node); // assignment_expression if (! acceptBinaryExpression(node, (PrecedenceLevel)(precedenceLevel + 1))) return false; TOperator op = HlslOpMap::binary(peek()); PrecedenceLevel tokenLevel = HlslOpMap::precedenceLevel(op); if (tokenLevel < precedenceLevel) return true; do { // ... op TSourceLoc loc = token.loc; advanceToken(); // ... expression TIntermTyped* rightNode = nullptr; if (! acceptBinaryExpression(rightNode, (PrecedenceLevel)(precedenceLevel + 1))) { expected("expression"); return false; } node = intermediate.addBinaryMath(op, node, rightNode, loc); if (! peekTokenClass(EHTokComma)) return true; } while (true); } // unary_expression // : (type) unary_expression // | + unary_expression // | - unary_expression // | ! unary_expression // | ~ unary_expression // | ++ unary_expression // | -- unary_expression // | postfix_expression // bool HlslGrammar::acceptUnaryExpression(TIntermTyped*& node) { // (type) unary_expression // Have to look two steps ahead, because this could be, e.g., a // postfix_expression instead, since that also starts with at "(". if (acceptTokenClass(EHTokLeftParen)) { TType castType; if (acceptType(castType)) { if (! acceptTokenClass(EHTokRightParen)) { expected(")"); return false; } // We've matched "(type)" now, get the expression to cast TSourceLoc loc = token.loc; if (! acceptUnaryExpression(node)) return false; // Hook it up like a constructor TFunction* constructorFunction = parseContext.handleConstructorCall(loc, castType); if (constructorFunction == nullptr) { expected("type that can be constructed"); return false; } TIntermTyped* arguments = nullptr; parseContext.handleFunctionArgument(constructorFunction, arguments, node); node = parseContext.handleFunctionCall(loc, constructorFunction, arguments); return true; } else { // This isn't a type cast, but it still started "(", so if it is a // unary expression, it can only be a postfix_expression, so try that. // Back it up first. recedeToken(); return acceptPostfixExpression(node); } } // peek for "op unary_expression" TOperator unaryOp = HlslOpMap::preUnary(peek()); // postfix_expression (if no unary operator) if (unaryOp == EOpNull) return acceptPostfixExpression(node); // op unary_expression TSourceLoc loc = token.loc; advanceToken(); if (! acceptUnaryExpression(node)) return false; // + is a no-op if (unaryOp == EOpAdd) return true; node = intermediate.addUnaryMath(unaryOp, node, loc); return node != nullptr; } // postfix_expression // : LEFT_PAREN expression RIGHT_PAREN // | literal // | constructor // | identifier // | function_call // | postfix_expression LEFT_BRACKET integer_expression RIGHT_BRACKET // | postfix_expression DOT IDENTIFIER // | postfix_expression INC_OP // | postfix_expression DEC_OP // bool HlslGrammar::acceptPostfixExpression(TIntermTyped*& node) { // Not implemented as self-recursive: // The logical "right recursion" is done with an loop at the end // idToken will pick up either a variable or a function name in a function call HlslToken idToken; // Find something before the postfix operations, as they can't operate // on nothing. So, no "return true", they fall through, only "return false". if (acceptTokenClass(EHTokLeftParen)) { // LEFT_PAREN expression RIGHT_PAREN if (! acceptExpression(node)) { expected("expression"); return false; } if (! acceptTokenClass(EHTokRightParen)) { expected(")"); return false; } } else if (acceptLiteral(node)) { // literal (nothing else to do yet), go on to the } else if (acceptConstructor(node)) { // constructor (nothing else to do yet) } else if (acceptIdentifier(idToken)) { // identifier or function_call name if (! peekTokenClass(EHTokLeftParen)) { node = parseContext.handleVariable(idToken.loc, idToken.symbol, token.string); } else if (acceptFunctionCall(idToken, node)) { // function_call (nothing else to do yet) } else { expected("function call arguments"); return false; } } else { // nothing found, can't post operate return false; } // Something was found, chain as many postfix operations as exist. do { TSourceLoc loc = token.loc; TOperator postOp = HlslOpMap::postUnary(peek()); // Consume only a valid post-unary operator, otherwise we are done. switch (postOp) { case EOpIndexDirectStruct: case EOpIndexIndirect: case EOpPostIncrement: case EOpPostDecrement: advanceToken(); break; default: return true; } // We have a valid post-unary operator, process it. switch (postOp) { case EOpIndexDirectStruct: { // DOT IDENTIFIER // includes swizzles and struct members HlslToken field; if (! acceptIdentifier(field)) { expected("swizzle or member"); return false; } TIntermTyped* base = node; // preserve for method function calls node = parseContext.handleDotDereference(field.loc, node, *field.string); // In the event of a method node, we look for an open paren and accept the function call. if (node->getAsMethodNode() != nullptr && peekTokenClass(EHTokLeftParen)) { if (! acceptFunctionCall(field, node, base)) { expected("function parameters"); return false; } } break; } case EOpIndexIndirect: { // LEFT_BRACKET integer_expression RIGHT_BRACKET TIntermTyped* indexNode = nullptr; if (! acceptExpression(indexNode) || ! peekTokenClass(EHTokRightBracket)) { expected("expression followed by ']'"); return false; } advanceToken(); node = parseContext.handleBracketDereference(indexNode->getLoc(), node, indexNode); break; } case EOpPostIncrement: // INC_OP // fall through case EOpPostDecrement: // DEC_OP node = intermediate.addUnaryMath(postOp, node, loc); break; default: assert(0); break; } } while (true); } // constructor // : type argument_list // bool HlslGrammar::acceptConstructor(TIntermTyped*& node) { // type TType type; if (acceptType(type)) { TFunction* constructorFunction = parseContext.handleConstructorCall(token.loc, type); if (constructorFunction == nullptr) return false; // arguments TIntermTyped* arguments = nullptr; if (! acceptArguments(constructorFunction, arguments)) { expected("constructor arguments"); return false; } // hook it up node = parseContext.handleFunctionCall(arguments->getLoc(), constructorFunction, arguments); return true; } return false; } // The function_call identifier was already recognized, and passed in as idToken. // // function_call // : [idToken] arguments // bool HlslGrammar::acceptFunctionCall(HlslToken idToken, TIntermTyped*& node, TIntermTyped* base) { // arguments TFunction* function = new TFunction(idToken.string, TType(EbtVoid)); TIntermTyped* arguments = nullptr; // methods have an implicit first argument of the calling object. if (base != nullptr) parseContext.handleFunctionArgument(function, arguments, base); if (! acceptArguments(function, arguments)) return false; node = parseContext.handleFunctionCall(idToken.loc, function, arguments); return true; } // arguments // : LEFT_PAREN expression COMMA expression COMMA ... RIGHT_PAREN // // The arguments are pushed onto the 'function' argument list and // onto the 'arguments' aggregate. // bool HlslGrammar::acceptArguments(TFunction* function, TIntermTyped*& arguments) { // LEFT_PAREN if (! acceptTokenClass(EHTokLeftParen)) return false; do { // expression TIntermTyped* arg; if (! acceptAssignmentExpression(arg)) break; // hook it up parseContext.handleFunctionArgument(function, arguments, arg); // COMMA if (! acceptTokenClass(EHTokComma)) break; } while (true); // RIGHT_PAREN if (! acceptTokenClass(EHTokRightParen)) { expected(")"); return false; } return true; } bool HlslGrammar::acceptLiteral(TIntermTyped*& node) { switch (token.tokenClass) { case EHTokIntConstant: node = intermediate.addConstantUnion(token.i, token.loc, true); break; case EHTokFloatConstant: node = intermediate.addConstantUnion(token.d, EbtFloat, token.loc, true); break; case EHTokDoubleConstant: node = intermediate.addConstantUnion(token.d, EbtDouble, token.loc, true); break; case EHTokBoolConstant: node = intermediate.addConstantUnion(token.b, token.loc, true); break; default: return false; } advanceToken(); return true; } // compound_statement // : LEFT_CURLY statement statement ... RIGHT_CURLY // bool HlslGrammar::acceptCompoundStatement(TIntermNode*& retStatement) { TIntermAggregate* compoundStatement = nullptr; // LEFT_CURLY if (! acceptTokenClass(EHTokLeftBrace)) return false; // statement statement ... TIntermNode* statement = nullptr; while (acceptStatement(statement)) { TIntermBranch* branch = statement ? statement->getAsBranchNode() : nullptr; if (branch != nullptr && (branch->getFlowOp() == EOpCase || branch->getFlowOp() == EOpDefault)) { // hook up individual subsequences within a switch statement parseContext.wrapupSwitchSubsequence(compoundStatement, statement); compoundStatement = nullptr; } else { // hook it up to the growing compound statement compoundStatement = intermediate.growAggregate(compoundStatement, statement); } } if (compoundStatement) compoundStatement->setOperator(EOpSequence); retStatement = compoundStatement; // RIGHT_CURLY return acceptTokenClass(EHTokRightBrace); } bool HlslGrammar::acceptScopedStatement(TIntermNode*& statement) { parseContext.pushScope(); bool result = acceptStatement(statement); parseContext.popScope(); return result; } bool HlslGrammar::acceptScopedCompoundStatement(TIntermNode*& statement) { parseContext.pushScope(); bool result = acceptCompoundStatement(statement); parseContext.popScope(); return result; } // statement // : attributes attributed_statement // // attributed_statement // : compound_statement // | SEMICOLON // | expression SEMICOLON // | declaration_statement // | selection_statement // | switch_statement // | case_label // | iteration_statement // | jump_statement // bool HlslGrammar::acceptStatement(TIntermNode*& statement) { statement = nullptr; // attributes acceptAttributes(); // attributed_statement switch (peek()) { case EHTokLeftBrace: return acceptScopedCompoundStatement(statement); case EHTokIf: return acceptSelectionStatement(statement); case EHTokSwitch: return acceptSwitchStatement(statement); case EHTokFor: case EHTokDo: case EHTokWhile: return acceptIterationStatement(statement); case EHTokContinue: case EHTokBreak: case EHTokDiscard: case EHTokReturn: return acceptJumpStatement(statement); case EHTokCase: return acceptCaseLabel(statement); case EHTokDefault: return acceptDefaultLabel(statement); case EHTokSemicolon: return acceptTokenClass(EHTokSemicolon); case EHTokRightBrace: // Performance: not strictly necessary, but stops a bunch of hunting early, // and is how sequences of statements end. return false; default: { // declaration if (acceptDeclaration(statement)) return true; // expression TIntermTyped* node; if (acceptExpression(node)) statement = node; else return false; // SEMICOLON (following an expression) if (! acceptTokenClass(EHTokSemicolon)) { expected(";"); return false; } } } return true; } // attributes // : list of zero or more of: LEFT_BRACKET attribute RIGHT_BRACKET // // attribute: // : UNROLL // | UNROLL LEFT_PAREN literal RIGHT_PAREN // | FASTOPT // | ALLOW_UAV_CONDITION // | BRANCH // | FLATTEN // | FORCECASE // | CALL // void HlslGrammar::acceptAttributes() { // For now, accept the [ XXX(X) ] syntax, but drop. // TODO: subset to correct set? Pass on? do { // LEFT_BRACKET? if (! acceptTokenClass(EHTokLeftBracket)) return; // attribute if (peekTokenClass(EHTokIdentifier)) { // 'token.string' is the attribute advanceToken(); } else if (! peekTokenClass(EHTokRightBracket)) { expected("identifier"); advanceToken(); } // (x) if (acceptTokenClass(EHTokLeftParen)) { TIntermTyped* node; if (! acceptLiteral(node)) expected("literal"); // 'node' has the literal in it if (! acceptTokenClass(EHTokRightParen)) expected(")"); } // RIGHT_BRACKET if (acceptTokenClass(EHTokRightBracket)) continue; expected("]"); return; } while (true); } // selection_statement // : IF LEFT_PAREN expression RIGHT_PAREN statement // : IF LEFT_PAREN expression RIGHT_PAREN statement ELSE statement // bool HlslGrammar::acceptSelectionStatement(TIntermNode*& statement) { TSourceLoc loc = token.loc; // IF if (! acceptTokenClass(EHTokIf)) return false; // so that something declared in the condition is scoped to the lifetimes // of the then-else statements parseContext.pushScope(); // LEFT_PAREN expression RIGHT_PAREN TIntermTyped* condition; if (! acceptParenExpression(condition)) return false; // create the child statements TIntermNodePair thenElse = { nullptr, nullptr }; // then statement if (! acceptScopedStatement(thenElse.node1)) { expected("then statement"); return false; } // ELSE if (acceptTokenClass(EHTokElse)) { // else statement if (! acceptScopedStatement(thenElse.node2)) { expected("else statement"); return false; } } // Put the pieces together statement = intermediate.addSelection(condition, thenElse, loc); parseContext.popScope(); return true; } // switch_statement // : SWITCH LEFT_PAREN expression RIGHT_PAREN compound_statement // bool HlslGrammar::acceptSwitchStatement(TIntermNode*& statement) { // SWITCH TSourceLoc loc = token.loc; if (! acceptTokenClass(EHTokSwitch)) return false; // LEFT_PAREN expression RIGHT_PAREN parseContext.pushScope(); TIntermTyped* switchExpression; if (! acceptParenExpression(switchExpression)) { parseContext.popScope(); return false; } // compound_statement parseContext.pushSwitchSequence(new TIntermSequence); bool statementOkay = acceptCompoundStatement(statement); if (statementOkay) statement = parseContext.addSwitch(loc, switchExpression, statement ? statement->getAsAggregate() : nullptr); parseContext.popSwitchSequence(); parseContext.popScope(); return statementOkay; } // iteration_statement // : WHILE LEFT_PAREN condition RIGHT_PAREN statement // | DO LEFT_BRACE statement RIGHT_BRACE WHILE LEFT_PAREN expression RIGHT_PAREN SEMICOLON // | FOR LEFT_PAREN for_init_statement for_rest_statement RIGHT_PAREN statement // // Non-speculative, only call if it needs to be found; WHILE or DO or FOR already seen. bool HlslGrammar::acceptIterationStatement(TIntermNode*& statement) { TSourceLoc loc = token.loc; TIntermTyped* condition = nullptr; EHlslTokenClass loop = peek(); assert(loop == EHTokDo || loop == EHTokFor || loop == EHTokWhile); // WHILE or DO or FOR advanceToken(); switch (loop) { case EHTokWhile: // so that something declared in the condition is scoped to the lifetime // of the while sub-statement parseContext.pushScope(); parseContext.nestLooping(); // LEFT_PAREN condition RIGHT_PAREN if (! acceptParenExpression(condition)) return false; // statement if (! acceptScopedStatement(statement)) { expected("while sub-statement"); return false; } parseContext.unnestLooping(); parseContext.popScope(); statement = intermediate.addLoop(statement, condition, nullptr, true, loc); return true; case EHTokDo: parseContext.nestLooping(); if (! acceptTokenClass(EHTokLeftBrace)) expected("{"); // statement if (! peekTokenClass(EHTokRightBrace) && ! acceptScopedStatement(statement)) { expected("do sub-statement"); return false; } if (! acceptTokenClass(EHTokRightBrace)) expected("}"); // WHILE if (! acceptTokenClass(EHTokWhile)) { expected("while"); return false; } // LEFT_PAREN condition RIGHT_PAREN TIntermTyped* condition; if (! acceptParenExpression(condition)) return false; if (! acceptTokenClass(EHTokSemicolon)) expected(";"); parseContext.unnestLooping(); statement = intermediate.addLoop(statement, condition, 0, false, loc); return true; case EHTokFor: { // LEFT_PAREN if (! acceptTokenClass(EHTokLeftParen)) expected("("); // so that something declared in the condition is scoped to the lifetime // of the for sub-statement parseContext.pushScope(); // initializer TIntermNode* initNode = nullptr; if (! acceptControlDeclaration(initNode)) { TIntermTyped* initExpr = nullptr; acceptExpression(initExpr); initNode = initExpr; } // SEMI_COLON if (! acceptTokenClass(EHTokSemicolon)) expected(";"); parseContext.nestLooping(); // condition SEMI_COLON acceptExpression(condition); if (! acceptTokenClass(EHTokSemicolon)) expected(";"); // iterator SEMI_COLON TIntermTyped* iterator = nullptr; acceptExpression(iterator); if (! acceptTokenClass(EHTokRightParen)) expected(")"); // statement if (! acceptScopedStatement(statement)) { expected("for sub-statement"); return false; } statement = intermediate.addForLoop(statement, initNode, condition, iterator, true, loc); parseContext.popScope(); parseContext.unnestLooping(); return true; } default: return false; } } // jump_statement // : CONTINUE SEMICOLON // | BREAK SEMICOLON // | DISCARD SEMICOLON // | RETURN SEMICOLON // | RETURN expression SEMICOLON // bool HlslGrammar::acceptJumpStatement(TIntermNode*& statement) { EHlslTokenClass jump = peek(); switch (jump) { case EHTokContinue: case EHTokBreak: case EHTokDiscard: case EHTokReturn: advanceToken(); break; default: // not something we handle in this function return false; } switch (jump) { case EHTokContinue: statement = intermediate.addBranch(EOpContinue, token.loc); break; case EHTokBreak: statement = intermediate.addBranch(EOpBreak, token.loc); break; case EHTokDiscard: statement = intermediate.addBranch(EOpKill, token.loc); break; case EHTokReturn: { // expression TIntermTyped* node; if (acceptExpression(node)) { // hook it up statement = intermediate.addBranch(EOpReturn, node, token.loc); } else statement = intermediate.addBranch(EOpReturn, token.loc); break; } default: assert(0); return false; } // SEMICOLON if (! acceptTokenClass(EHTokSemicolon)) expected(";"); return true; } // case_label // : CASE expression COLON // bool HlslGrammar::acceptCaseLabel(TIntermNode*& statement) { TSourceLoc loc = token.loc; if (! acceptTokenClass(EHTokCase)) return false; TIntermTyped* expression; if (! acceptExpression(expression)) { expected("case expression"); return false; } if (! acceptTokenClass(EHTokColon)) { expected(":"); return false; } statement = parseContext.intermediate.addBranch(EOpCase, expression, loc); return true; } // default_label // : DEFAULT COLON // bool HlslGrammar::acceptDefaultLabel(TIntermNode*& statement) { TSourceLoc loc = token.loc; if (! acceptTokenClass(EHTokDefault)) return false; if (! acceptTokenClass(EHTokColon)) { expected(":"); return false; } statement = parseContext.intermediate.addBranch(EOpDefault, loc); return true; } // array_specifier // : LEFT_BRACKET integer_expression RGHT_BRACKET post_decls // optional // void HlslGrammar::acceptArraySpecifier(TArraySizes*& arraySizes) { arraySizes = nullptr; if (! acceptTokenClass(EHTokLeftBracket)) return; TSourceLoc loc = token.loc; TIntermTyped* sizeExpr; if (! acceptAssignmentExpression(sizeExpr)) { expected("array-sizing expression"); return; } if (! acceptTokenClass(EHTokRightBracket)) { expected("]"); return; } TArraySize arraySize; parseContext.arraySizeCheck(loc, sizeExpr, arraySize); arraySizes = new TArraySizes; arraySizes->addInnerSize(arraySize); } // post_decls // : COLON semantic // optional // COLON PACKOFFSET LEFT_PAREN ... RIGHT_PAREN // optional // COLON REGISTER // optional // annotations // optional // void HlslGrammar::acceptPostDecls(TType& type) { do { // COLON if (acceptTokenClass(EHTokColon)) { HlslToken idToken; if (acceptTokenClass(EHTokPackOffset)) { if (! acceptTokenClass(EHTokLeftParen)) { expected("("); return; } acceptTokenClass(EHTokIdentifier); acceptTokenClass(EHTokDot); acceptTokenClass(EHTokIdentifier); if (! acceptTokenClass(EHTokRightParen)) { expected(")"); break; } // TODO: process the packoffset information } else if (! acceptIdentifier(idToken)) { expected("semantic or packoffset or register"); return; } else if (*idToken.string == "register") { if (! acceptTokenClass(EHTokLeftParen)) { expected("("); return; } acceptTokenClass(EHTokIdentifier); acceptTokenClass(EHTokComma); acceptTokenClass(EHTokIdentifier); acceptTokenClass(EHTokLeftBracket); if (peekTokenClass(EHTokIntConstant)) advanceToken(); acceptTokenClass(EHTokRightBracket); if (! acceptTokenClass(EHTokRightParen)) { expected(")"); break; } // TODO: process the register information } else { // semantic, in idToken.string parseContext.handleSemantic(type, *idToken.string); } } else if (acceptTokenClass(EHTokLeftAngle)) { // TODO: process annotations, just accepting them for now do { if (peekTokenClass(EHTokNone)) return; if (acceptTokenClass(EHTokRightAngle)) break; advanceToken(); } while (true); } else break; } while (true); } } // end namespace glslang