diff options
Diffstat (limited to 'lib/Target/X86')
41 files changed, 3431 insertions, 1509 deletions
diff --git a/lib/Target/X86/AsmParser/X86AsmParser.cpp b/lib/Target/X86/AsmParser/X86AsmParser.cpp index 4ed5534a6..fef5cfe 100644 --- a/lib/Target/X86/AsmParser/X86AsmParser.cpp +++ b/lib/Target/X86/AsmParser/X86AsmParser.cpp @@ -13,6 +13,7 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringSwitch.h" #include "llvm/ADT/Twine.h" +#include "llvm/MC/MCContext.h" #include "llvm/MC/MCExpr.h" #include "llvm/MC/MCInst.h" #include "llvm/MC/MCParser/MCAsmLexer.h" @@ -32,11 +33,445 @@ using namespace llvm; namespace { struct X86Operand; +static const char OpPrecedence[] = { + 0, // IC_PLUS + 0, // IC_MINUS + 1, // IC_MULTIPLY + 1, // IC_DIVIDE + 2, // IC_RPAREN + 3, // IC_LPAREN + 0, // IC_IMM + 0 // IC_REGISTER +}; + class X86AsmParser : public MCTargetAsmParser { MCSubtargetInfo &STI; MCAsmParser &Parser; ParseInstructionInfo *InstInfo; private: + enum InfixCalculatorTok { + IC_PLUS = 0, + IC_MINUS, + IC_MULTIPLY, + IC_DIVIDE, + IC_RPAREN, + IC_LPAREN, + IC_IMM, + IC_REGISTER + }; + + class InfixCalculator { + typedef std::pair< InfixCalculatorTok, int64_t > ICToken; + SmallVector<InfixCalculatorTok, 4> InfixOperatorStack; + SmallVector<ICToken, 4> PostfixStack; + + public: + int64_t popOperand() { + assert (!PostfixStack.empty() && "Poped an empty stack!"); + ICToken Op = PostfixStack.pop_back_val(); + assert ((Op.first == IC_IMM || Op.first == IC_REGISTER) + && "Expected and immediate or register!"); + return Op.second; + } + void pushOperand(InfixCalculatorTok Op, int64_t Val = 0) { + assert ((Op == IC_IMM || Op == IC_REGISTER) && + "Unexpected operand!"); + PostfixStack.push_back(std::make_pair(Op, Val)); + } + + void popOperator() { InfixOperatorStack.pop_back_val(); } + void pushOperator(InfixCalculatorTok Op) { + // Push the new operator if the stack is empty. + if (InfixOperatorStack.empty()) { + InfixOperatorStack.push_back(Op); + return; + } + + // Push the new operator if it has a higher precedence than the operator + // on the top of the stack or the operator on the top of the stack is a + // left parentheses. + unsigned Idx = InfixOperatorStack.size() - 1; + InfixCalculatorTok StackOp = InfixOperatorStack[Idx]; + if (OpPrecedence[Op] > OpPrecedence[StackOp] || StackOp == IC_LPAREN) { + InfixOperatorStack.push_back(Op); + return; + } + + // The operator on the top of the stack has higher precedence than the + // new operator. + unsigned ParenCount = 0; + while (1) { + // Nothing to process. + if (InfixOperatorStack.empty()) + break; + + Idx = InfixOperatorStack.size() - 1; + StackOp = InfixOperatorStack[Idx]; + if (!(OpPrecedence[StackOp] >= OpPrecedence[Op] || ParenCount)) + break; + + // If we have an even parentheses count and we see a left parentheses, + // then stop processing. + if (!ParenCount && StackOp == IC_LPAREN) + break; + + if (StackOp == IC_RPAREN) { + ++ParenCount; + InfixOperatorStack.pop_back_val(); + } else if (StackOp == IC_LPAREN) { + --ParenCount; + InfixOperatorStack.pop_back_val(); + } else { + InfixOperatorStack.pop_back_val(); + PostfixStack.push_back(std::make_pair(StackOp, 0)); + } + } + // Push the new operator. + InfixOperatorStack.push_back(Op); + } + int64_t execute() { + // Push any remaining operators onto the postfix stack. + while (!InfixOperatorStack.empty()) { + InfixCalculatorTok StackOp = InfixOperatorStack.pop_back_val(); + if (StackOp != IC_LPAREN && StackOp != IC_RPAREN) + PostfixStack.push_back(std::make_pair(StackOp, 0)); + } + + if (PostfixStack.empty()) + return 0; + + SmallVector<ICToken, 16> OperandStack; + for (unsigned i = 0, e = PostfixStack.size(); i != e; ++i) { + ICToken Op = PostfixStack[i]; + if (Op.first == IC_IMM || Op.first == IC_REGISTER) { + OperandStack.push_back(Op); + } else { + assert (OperandStack.size() > 1 && "Too few operands."); + int64_t Val; + ICToken Op2 = OperandStack.pop_back_val(); + ICToken Op1 = OperandStack.pop_back_val(); + switch (Op.first) { + default: + report_fatal_error("Unexpected operator!"); + break; + case IC_PLUS: + Val = Op1.second + Op2.second; + OperandStack.push_back(std::make_pair(IC_IMM, Val)); + break; + case IC_MINUS: + Val = Op1.second - Op2.second; + OperandStack.push_back(std::make_pair(IC_IMM, Val)); + break; + case IC_MULTIPLY: + assert (Op1.first == IC_IMM && Op2.first == IC_IMM && + "Multiply operation with an immediate and a register!"); + Val = Op1.second * Op2.second; + OperandStack.push_back(std::make_pair(IC_IMM, Val)); + break; + case IC_DIVIDE: + assert (Op1.first == IC_IMM && Op2.first == IC_IMM && + "Divide operation with an immediate and a register!"); + assert (Op2.second != 0 && "Division by zero!"); + Val = Op1.second / Op2.second; + OperandStack.push_back(std::make_pair(IC_IMM, Val)); + break; + } + } + } + assert (OperandStack.size() == 1 && "Expected a single result."); + return OperandStack.pop_back_val().second; + } + }; + + enum IntelExprState { + IES_PLUS, + IES_MINUS, + IES_MULTIPLY, + IES_DIVIDE, + IES_LBRAC, + IES_RBRAC, + IES_LPAREN, + IES_RPAREN, + IES_REGISTER, + IES_INTEGER, + IES_IDENTIFIER, + IES_ERROR + }; + + class IntelExprStateMachine { + IntelExprState State, PrevState; + unsigned BaseReg, IndexReg, TmpReg, Scale; + int64_t Imm; + const MCExpr *Sym; + StringRef SymName; + bool StopOnLBrac, AddImmPrefix; + InfixCalculator IC; + InlineAsmIdentifierInfo Info; + public: + IntelExprStateMachine(int64_t imm, bool stoponlbrac, bool addimmprefix) : + State(IES_PLUS), PrevState(IES_ERROR), BaseReg(0), IndexReg(0), TmpReg(0), + Scale(1), Imm(imm), Sym(0), StopOnLBrac(stoponlbrac), + AddImmPrefix(addimmprefix) { Info.clear(); } + + unsigned getBaseReg() { return BaseReg; } + unsigned getIndexReg() { return IndexReg; } + unsigned getScale() { return Scale; } + const MCExpr *getSym() { return Sym; } + StringRef getSymName() { return SymName; } + int64_t getImm() { return Imm + IC.execute(); } + bool isValidEndState() { return State == IES_RBRAC; } + bool getStopOnLBrac() { return StopOnLBrac; } + bool getAddImmPrefix() { return AddImmPrefix; } + bool hadError() { return State == IES_ERROR; } + + InlineAsmIdentifierInfo &getIdentifierInfo() { + return Info; + } + + void onPlus() { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_INTEGER: + case IES_RPAREN: + case IES_REGISTER: + State = IES_PLUS; + IC.pushOperator(IC_PLUS); + if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { + // If we already have a BaseReg, then assume this is the IndexReg with + // a scale of 1. + if (!BaseReg) { + BaseReg = TmpReg; + } else { + assert (!IndexReg && "BaseReg/IndexReg already set!"); + IndexReg = TmpReg; + Scale = 1; + } + } + break; + } + PrevState = CurrState; + } + void onMinus() { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_PLUS: + case IES_MULTIPLY: + case IES_DIVIDE: + case IES_LPAREN: + case IES_RPAREN: + case IES_LBRAC: + case IES_RBRAC: + case IES_INTEGER: + case IES_REGISTER: + State = IES_MINUS; + // Only push the minus operator if it is not a unary operator. + if (!(CurrState == IES_PLUS || CurrState == IES_MINUS || + CurrState == IES_MULTIPLY || CurrState == IES_DIVIDE || + CurrState == IES_LPAREN || CurrState == IES_LBRAC)) + IC.pushOperator(IC_MINUS); + if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { + // If we already have a BaseReg, then assume this is the IndexReg with + // a scale of 1. + if (!BaseReg) { + BaseReg = TmpReg; + } else { + assert (!IndexReg && "BaseReg/IndexReg already set!"); + IndexReg = TmpReg; + Scale = 1; + } + } + break; + } + PrevState = CurrState; + } + void onRegister(unsigned Reg) { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_PLUS: + case IES_LPAREN: + State = IES_REGISTER; + TmpReg = Reg; + IC.pushOperand(IC_REGISTER); + break; + case IES_MULTIPLY: + // Index Register - Scale * Register + if (PrevState == IES_INTEGER) { + assert (!IndexReg && "IndexReg already set!"); + State = IES_REGISTER; + IndexReg = Reg; + // Get the scale and replace the 'Scale * Register' with '0'. + Scale = IC.popOperand(); + IC.pushOperand(IC_IMM); + IC.popOperator(); + } else { + State = IES_ERROR; + } + break; + } + PrevState = CurrState; + } + void onIdentifierExpr(const MCExpr *SymRef, StringRef SymRefName) { + PrevState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_PLUS: + case IES_MINUS: + State = IES_INTEGER; + Sym = SymRef; + SymName = SymRefName; + IC.pushOperand(IC_IMM); + break; + } + } + void onInteger(int64_t TmpInt) { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_PLUS: + case IES_MINUS: + case IES_DIVIDE: + case IES_MULTIPLY: + case IES_LPAREN: + State = IES_INTEGER; + if (PrevState == IES_REGISTER && CurrState == IES_MULTIPLY) { + // Index Register - Register * Scale + assert (!IndexReg && "IndexReg already set!"); + IndexReg = TmpReg; + Scale = TmpInt; + // Get the scale and replace the 'Register * Scale' with '0'. + IC.popOperator(); + } else if ((PrevState == IES_PLUS || PrevState == IES_MINUS || + PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE || + PrevState == IES_LPAREN || PrevState == IES_LBRAC) && + CurrState == IES_MINUS) { + // Unary minus. No need to pop the minus operand because it was never + // pushed. + IC.pushOperand(IC_IMM, -TmpInt); // Push -Imm. + } else { + IC.pushOperand(IC_IMM, TmpInt); + } + break; + } + PrevState = CurrState; + } + void onStar() { + PrevState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_INTEGER: + case IES_REGISTER: + case IES_RPAREN: + State = IES_MULTIPLY; + IC.pushOperator(IC_MULTIPLY); + break; + } + } + void onDivide() { + PrevState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_INTEGER: + case IES_RPAREN: + State = IES_DIVIDE; + IC.pushOperator(IC_DIVIDE); + break; + } + } + void onLBrac() { + PrevState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_RBRAC: + State = IES_PLUS; + IC.pushOperator(IC_PLUS); + break; + } + } + void onRBrac() { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_INTEGER: + case IES_REGISTER: + case IES_RPAREN: + State = IES_RBRAC; + if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) { + // If we already have a BaseReg, then assume this is the IndexReg with + // a scale of 1. + if (!BaseReg) { + BaseReg = TmpReg; + } else { + assert (!IndexReg && "BaseReg/IndexReg already set!"); + IndexReg = TmpReg; + Scale = 1; + } + } + break; + } + PrevState = CurrState; + } + void onLParen() { + IntelExprState CurrState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_PLUS: + case IES_MINUS: + case IES_MULTIPLY: + case IES_DIVIDE: + case IES_LPAREN: + // FIXME: We don't handle this type of unary minus, yet. + if ((PrevState == IES_PLUS || PrevState == IES_MINUS || + PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE || + PrevState == IES_LPAREN || PrevState == IES_LBRAC) && + CurrState == IES_MINUS) { + State = IES_ERROR; + break; + } + State = IES_LPAREN; + IC.pushOperator(IC_LPAREN); + break; + } + PrevState = CurrState; + } + void onRParen() { + PrevState = State; + switch (State) { + default: + State = IES_ERROR; + break; + case IES_INTEGER: + case IES_REGISTER: + case IES_RPAREN: + State = IES_RPAREN; + IC.pushOperator(IC_RPAREN); + break; + } + } + }; + MCAsmParser &getParser() const { return Parser; } MCAsmLexer &getLexer() const { return Parser.getLexer(); } @@ -56,14 +491,24 @@ private: X86Operand *ParseOperand(); X86Operand *ParseATTOperand(); X86Operand *ParseIntelOperand(); - X86Operand *ParseIntelOffsetOfOperator(SMLoc StartLoc); - X86Operand *ParseIntelOperator(SMLoc StartLoc, unsigned OpKind); - X86Operand *ParseIntelMemOperand(unsigned SegReg, SMLoc StartLoc); - X86Operand *ParseIntelBracExpression(unsigned SegReg, unsigned Size); + X86Operand *ParseIntelOffsetOfOperator(); + X86Operand *ParseIntelDotOperator(const MCExpr *Disp, const MCExpr *&NewDisp); + X86Operand *ParseIntelOperator(unsigned OpKind); + X86Operand *ParseIntelMemOperand(unsigned SegReg, int64_t ImmDisp, + SMLoc StartLoc); + X86Operand *ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End); + X86Operand *ParseIntelBracExpression(unsigned SegReg, SMLoc Start, + int64_t ImmDisp, unsigned Size); + X86Operand *ParseIntelIdentifier(const MCExpr *&Val, StringRef &Identifier, + InlineAsmIdentifierInfo &Info, SMLoc &End); + X86Operand *ParseMemOperand(unsigned SegReg, SMLoc StartLoc); - bool ParseIntelDotOperator(const MCExpr *Disp, const MCExpr **NewDisp, - SmallString<64> &Err); + X86Operand *CreateMemForInlineAsm(unsigned SegReg, const MCExpr *Disp, + unsigned BaseReg, unsigned IndexReg, + unsigned Scale, SMLoc Start, SMLoc End, + unsigned Size, StringRef Identifier, + InlineAsmIdentifierInfo &Info); bool ParseDirectiveWord(unsigned Size, SMLoc L); bool ParseDirectiveCode(StringRef IDVal, SMLoc L); @@ -93,6 +538,10 @@ private: setAvailableFeatures(FB); } + bool isParsingIntelSyntax() { + return getParser().getAssemblerDialect(); + } + /// @name Auto-generated Matcher Functions /// { @@ -115,10 +564,6 @@ public: SmallVectorImpl<MCParsedAsmOperand*> &Operands); virtual bool ParseDirective(AsmToken DirectiveID); - - bool isParsingIntelSyntax() { - return getParser().getAssemblerDialect(); - } }; } // end anonymous namespace @@ -168,6 +613,8 @@ struct X86Operand : public MCParsedAsmOperand { SMLoc StartLoc, EndLoc; SMLoc OffsetOfLoc; + StringRef SymName; + void *OpDecl; bool AddressOf; struct TokOp { @@ -181,7 +628,6 @@ struct X86Operand : public MCParsedAsmOperand { struct ImmOp { const MCExpr *Val; - bool NeedAsmRewrite; }; struct MemOp { @@ -191,7 +637,6 @@ struct X86Operand : public MCParsedAsmOperand { unsigned IndexReg; unsigned Scale; unsigned Size; - bool NeedSizeDir; }; union { @@ -204,6 +649,9 @@ struct X86Operand : public MCParsedAsmOperand { X86Operand(KindTy K, SMLoc Start, SMLoc End) : Kind(K), StartLoc(Start), EndLoc(End) {} + StringRef getSymName() { return SymName; } + void *getOpDecl() { return OpDecl; } + /// getStartLoc - Get the location of the first token of this operand. SMLoc getStartLoc() const { return StartLoc; } /// getEndLoc - Get the location of the last token of this operand. @@ -236,11 +684,6 @@ struct X86Operand : public MCParsedAsmOperand { return Imm.Val; } - bool needAsmRewrite() const { - assert(Kind == Immediate && "Invalid access!"); - return Imm.NeedAsmRewrite; - } - const MCExpr *getMemDisp() const { assert(Kind == Memory && "Invalid access!"); return Mem.Disp; @@ -337,11 +780,6 @@ struct X86Operand : public MCParsedAsmOperand { return isImmSExti64i32Value(CE->getValue()); } - unsigned getMemSize() const { - assert(Kind == Memory && "Invalid access!"); - return Mem.Size; - } - bool isOffsetOf() const { return OffsetOfLoc.getPointer(); } @@ -350,11 +788,6 @@ struct X86Operand : public MCParsedAsmOperand { return AddressOf; } - bool needSizeDirective() const { - assert(Kind == Memory && "Invalid access!"); - return Mem.NeedSizeDir; - } - bool isMem() const { return Kind == Memory; } bool isMem8() const { return Kind == Memory && (!Mem.Size || Mem.Size == 8); @@ -482,25 +915,28 @@ struct X86Operand : public MCParsedAsmOperand { static X86Operand *CreateReg(unsigned RegNo, SMLoc StartLoc, SMLoc EndLoc, bool AddressOf = false, - SMLoc OffsetOfLoc = SMLoc()) { + SMLoc OffsetOfLoc = SMLoc(), + StringRef SymName = StringRef(), + void *OpDecl = 0) { X86Operand *Res = new X86Operand(Register, StartLoc, EndLoc); Res->Reg.RegNo = RegNo; Res->AddressOf = AddressOf; Res->OffsetOfLoc = OffsetOfLoc; + Res->SymName = SymName; + Res->OpDecl = OpDecl; return Res; } - static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc, - bool NeedRewrite = true){ + static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc){ X86Operand *Res = new X86Operand(Immediate, StartLoc, EndLoc); Res->Imm.Val = Val; - Res->Imm.NeedAsmRewrite = NeedRewrite; return Res; } /// Create an absolute memory operand. static X86Operand *CreateMem(const MCExpr *Disp, SMLoc StartLoc, SMLoc EndLoc, - unsigned Size = 0, bool NeedSizeDir = false) { + unsigned Size = 0, StringRef SymName = StringRef(), + void *OpDecl = 0) { X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc); Res->Mem.SegReg = 0; Res->Mem.Disp = Disp; @@ -508,8 +944,9 @@ struct X86Operand : public MCParsedAsmOperand { Res->Mem.IndexReg = 0; Res->Mem.Scale = 1; Res->Mem.Size = Size; - Res->Mem.NeedSizeDir = NeedSizeDir; - Res->AddressOf = false; + Res->SymName = SymName; + Res->OpDecl = OpDecl; + Res->AddressOf = false; return Res; } @@ -517,7 +954,9 @@ struct X86Operand : public MCParsedAsmOperand { static X86Operand *CreateMem(unsigned SegReg, const MCExpr *Disp, unsigned BaseReg, unsigned IndexReg, unsigned Scale, SMLoc StartLoc, SMLoc EndLoc, - unsigned Size = 0, bool NeedSizeDir = false) { + unsigned Size = 0, + StringRef SymName = StringRef(), + void *OpDecl = 0) { // We should never just have a displacement, that should be parsed as an // absolute memory operand. assert((SegReg || BaseReg || IndexReg) && "Invalid memory operand!"); @@ -532,8 +971,9 @@ struct X86Operand : public MCParsedAsmOperand { Res->Mem.IndexReg = IndexReg; Res->Mem.Scale = Scale; Res->Mem.Size = Size; - Res->Mem.NeedSizeDir = NeedSizeDir; - Res->AddressOf = false; + Res->SymName = SymName; + Res->OpDecl = OpDecl; + Res->AddressOf = false; return Res; } }; @@ -689,251 +1129,104 @@ static unsigned getIntelMemOperandSize(StringRef OpStr) { return Size; } -enum IntelBracExprState { - IBES_START, - IBES_LBRAC, - IBES_RBRAC, - IBES_REGISTER, - IBES_REGISTER_STAR, - IBES_REGISTER_STAR_INTEGER, - IBES_INTEGER, - IBES_INTEGER_STAR, - IBES_INDEX_REGISTER, - IBES_IDENTIFIER, - IBES_DISP_EXPR, - IBES_MINUS, - IBES_ERROR -}; - -class IntelBracExprStateMachine { - IntelBracExprState State; - unsigned BaseReg, IndexReg, Scale; - int64_t Disp; - - unsigned TmpReg; - int64_t TmpInteger; - - bool isPlus; - -public: - IntelBracExprStateMachine(MCAsmParser &parser) : - State(IBES_START), BaseReg(0), IndexReg(0), Scale(1), Disp(0), - TmpReg(0), TmpInteger(0), isPlus(true) {} - - unsigned getBaseReg() { return BaseReg; } - unsigned getIndexReg() { return IndexReg; } - unsigned getScale() { return Scale; } - int64_t getDisp() { return Disp; } - bool isValidEndState() { return State == IBES_RBRAC; } - - void onPlus() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_INTEGER: - State = IBES_START; - if (isPlus) - Disp += TmpInteger; - else - Disp -= TmpInteger; - break; - case IBES_REGISTER: - State = IBES_START; - // If we already have a BaseReg, then assume this is the IndexReg with a - // scale of 1. - if (!BaseReg) { - BaseReg = TmpReg; - } else { - assert (!IndexReg && "BaseReg/IndexReg already set!"); - IndexReg = TmpReg; - Scale = 1; - } - break; - case IBES_INDEX_REGISTER: - State = IBES_START; - break; - } - isPlus = true; - } - void onMinus() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_START: - State = IBES_MINUS; - break; - case IBES_INTEGER: - State = IBES_START; - if (isPlus) - Disp += TmpInteger; - else - Disp -= TmpInteger; - break; - case IBES_REGISTER: - State = IBES_START; - // If we already have a BaseReg, then assume this is the IndexReg with a - // scale of 1. - if (!BaseReg) { - BaseReg = TmpReg; - } else { - assert (!IndexReg && "BaseReg/IndexReg already set!"); - IndexReg = TmpReg; - Scale = 1; - } - break; - case IBES_INDEX_REGISTER: - State = IBES_START; - break; - } - isPlus = false; - } - void onRegister(unsigned Reg) { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_START: - State = IBES_REGISTER; - TmpReg = Reg; - break; - case IBES_INTEGER_STAR: - assert (!IndexReg && "IndexReg already set!"); - State = IBES_INDEX_REGISTER; - IndexReg = Reg; - Scale = TmpInteger; - break; - } - } - void onDispExpr() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_START: - State = IBES_DISP_EXPR; - break; - } - } - void onInteger(int64_t TmpInt) { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_START: - State = IBES_INTEGER; - TmpInteger = TmpInt; - break; - case IBES_MINUS: - State = IBES_INTEGER; - TmpInteger = TmpInt; - break; - case IBES_REGISTER_STAR: - assert (!IndexReg && "IndexReg already set!"); - State = IBES_INDEX_REGISTER; - IndexReg = TmpReg; - Scale = TmpInt; - break; - } - } - void onStar() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_INTEGER: - State = IBES_INTEGER_STAR; - break; - case IBES_REGISTER: - State = IBES_REGISTER_STAR; - break; +X86Operand * +X86AsmParser::CreateMemForInlineAsm(unsigned SegReg, const MCExpr *Disp, + unsigned BaseReg, unsigned IndexReg, + unsigned Scale, SMLoc Start, SMLoc End, + unsigned Size, StringRef Identifier, + InlineAsmIdentifierInfo &Info){ + if (isa<MCSymbolRefExpr>(Disp)) { + // If this is not a VarDecl then assume it is a FuncDecl or some other label + // reference. We need an 'r' constraint here, so we need to create register + // operand to ensure proper matching. Just pick a GPR based on the size of + // a pointer. + if (!Info.IsVarDecl) { + unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX; + return X86Operand::CreateReg(RegNo, Start, End, /*AddressOf=*/true, + SMLoc(), Identifier, Info.OpDecl); } - } - void onLBrac() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_RBRAC: - State = IBES_START; - isPlus = true; - break; + if (!Size) { + Size = Info.Type * 8; // Size is in terms of bits in this context. + if (Size) + InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_SizeDirective, Start, + /*Len=*/0, Size)); } } - void onRBrac() { - switch (State) { - default: - State = IBES_ERROR; - break; - case IBES_DISP_EXPR: - State = IBES_RBRAC; - break; - case IBES_INTEGER: - State = IBES_RBRAC; - if (isPlus) - Disp += TmpInteger; - else - Disp -= TmpInteger; - break; - case IBES_REGISTER: - State = IBES_RBRAC; - // If we already have a BaseReg, then assume this is the IndexReg with a - // scale of 1. - if (!BaseReg) { - BaseReg = TmpReg; - } else { - assert (!IndexReg && "BaseReg/IndexReg already set!"); - IndexReg = TmpReg; - Scale = 1; + + // When parsing inline assembly we set the base register to a non-zero value + // if we don't know the actual value at this time. This is necessary to + // get the matching correct in some cases. + BaseReg = BaseReg ? BaseReg : 1; + return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, Start, + End, Size, Identifier, Info.OpDecl); +} + +static void +RewriteIntelBracExpression(SmallVectorImpl<AsmRewrite> *AsmRewrites, + StringRef SymName, int64_t ImmDisp, + int64_t FinalImmDisp, SMLoc &BracLoc, + SMLoc &StartInBrac, SMLoc &End) { + // Remove the '[' and ']' from the IR string. + AsmRewrites->push_back(AsmRewrite(AOK_Skip, BracLoc, 1)); + AsmRewrites->push_back(AsmRewrite(AOK_Skip, End, 1)); + + // If ImmDisp is non-zero, then we parsed a displacement before the + // bracketed expression (i.e., ImmDisp [ BaseReg + Scale*IndexReg + Disp]) + // If ImmDisp doesn't match the displacement computed by the state machine + // then we have an additional displacement in the bracketed expression. + if (ImmDisp != FinalImmDisp) { + if (ImmDisp) { + // We have an immediate displacement before the bracketed expression. + // Adjust this to match the final immediate displacement. + bool Found = false; + for (SmallVectorImpl<AsmRewrite>::iterator I = AsmRewrites->begin(), + E = AsmRewrites->end(); I != E; ++I) { + if ((*I).Loc.getPointer() > BracLoc.getPointer()) + continue; + if ((*I).Kind == AOK_ImmPrefix || (*I).Kind == AOK_Imm) { + assert (!Found && "ImmDisp already rewritten."); + (*I).Kind = AOK_Imm; + (*I).Len = BracLoc.getPointer() - (*I).Loc.getPointer(); + (*I).Val = FinalImmDisp; + Found = true; + break; + } } - break; - case IBES_INDEX_REGISTER: - State = IBES_RBRAC; - break; + assert (Found && "Unable to rewrite ImmDisp."); + } else { + // We have a symbolic and an immediate displacement, but no displacement + // before the bracketed expression. Put the immediate displacement + // before the bracketed expression. + AsmRewrites->push_back(AsmRewrite(AOK_Imm, BracLoc, 0, FinalImmDisp)); } } -}; + // Remove all the ImmPrefix rewrites within the brackets. + for (SmallVectorImpl<AsmRewrite>::iterator I = AsmRewrites->begin(), + E = AsmRewrites->end(); I != E; ++I) { + if ((*I).Loc.getPointer() < StartInBrac.getPointer()) + continue; + if ((*I).Kind == AOK_ImmPrefix) + (*I).Kind = AOK_Delete; + } + const char *SymLocPtr = SymName.data(); + // Skip everything before the symbol. + if (unsigned Len = SymLocPtr - StartInBrac.getPointer()) { + assert(Len > 0 && "Expected a non-negative length."); + AsmRewrites->push_back(AsmRewrite(AOK_Skip, StartInBrac, Len)); + } + // Skip everything after the symbol. + if (unsigned Len = End.getPointer() - (SymLocPtr + SymName.size())) { + SMLoc Loc = SMLoc::getFromPointer(SymLocPtr + SymName.size()); + assert(Len > 0 && "Expected a non-negative length."); + AsmRewrites->push_back(AsmRewrite(AOK_Skip, Loc, Len)); + } +} -X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, - unsigned Size) { +X86Operand * +X86AsmParser::ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End) { const AsmToken &Tok = Parser.getTok(); - SMLoc Start = Tok.getLoc(), End = Tok.getEndLoc(); - - // Eat '[' - if (getLexer().isNot(AsmToken::LBrac)) - return ErrorOperand(Start, "Expected '[' token!"); - Parser.Lex(); - unsigned TmpReg = 0; - - // Try to handle '[' 'symbol' ']' - if (getLexer().is(AsmToken::Identifier)) { - if (ParseRegister(TmpReg, Start, End)) { - const MCExpr *Disp; - if (getParser().parseExpression(Disp, End)) - return 0; - - if (getLexer().isNot(AsmToken::RBrac)) - return ErrorOperand(Parser.getTok().getLoc(), "Expected ']' token!"); - // Adjust the EndLoc due to the ']'. - End = SMLoc::getFromPointer(Parser.getTok().getEndLoc().getPointer()-1); - Parser.Lex(); - return X86Operand::CreateMem(Disp, Start, End, Size); - } - } - - // Parse [ BaseReg + Scale*IndexReg + Disp ]. bool Done = false; - IntelBracExprStateMachine SM(Parser); - - // If we parsed a register, then the end loc has already been set and - // the identifier has already been lexed. We also need to update the - // state. - if (TmpReg) - SM.onRegister(TmpReg); - - const MCExpr *Disp = 0; while (!Done) { bool UpdateLocLex = true; @@ -941,6 +1234,10 @@ X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, // identifier. Don't try an parse it as a register. if (Tok.getString().startswith(".")) break; + + // If we're parsing an immediate expression, we don't expect a '['. + if (SM.getStopOnLBrac() && getLexer().getKind() == AsmToken::LBrac) + break; switch (getLexer().getKind()) { default: { @@ -950,82 +1247,185 @@ X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, } return ErrorOperand(Tok.getLoc(), "Unexpected token!"); } + case AsmToken::EndOfStatement: { + Done = true; + break; + } case AsmToken::Identifier: { - // This could be a register or a displacement expression. - if(!ParseRegister(TmpReg, Start, End)) { + // This could be a register or a symbolic displacement. + unsigned TmpReg; + const MCExpr *Val; + SMLoc IdentLoc = Tok.getLoc(); + StringRef Identifier = Tok.getString(); + if(!ParseRegister(TmpReg, IdentLoc, End)) { SM.onRegister(TmpReg); UpdateLocLex = false; break; - } else if (!getParser().parseExpression(Disp, End)) { - SM.onDispExpr(); + } else { + if (!isParsingInlineAsm()) { + if (getParser().parsePrimaryExpr(Val, End)) + return ErrorOperand(Tok.getLoc(), "Unexpected identifier!"); + } else { + InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo(); + if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info, End)) + return Err; + } + SM.onIdentifierExpr(Val, Identifier); UpdateLocLex = false; break; } return ErrorOperand(Tok.getLoc(), "Unexpected identifier!"); } - case AsmToken::Integer: { - int64_t Val = Tok.getIntVal(); - SM.onInteger(Val); + case AsmToken::Integer: + if (isParsingInlineAsm() && SM.getAddImmPrefix()) + InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, + Tok.getLoc())); + SM.onInteger(Tok.getIntVal()); break; - } case AsmToken::Plus: SM.onPlus(); break; case AsmToken::Minus: SM.onMinus(); break; case AsmToken::Star: SM.onStar(); break; + case AsmToken::Slash: SM.onDivide(); break; case AsmToken::LBrac: SM.onLBrac(); break; case AsmToken::RBrac: SM.onRBrac(); break; + case AsmToken::LParen: SM.onLParen(); break; + case AsmToken::RParen: SM.onRParen(); break; } + if (SM.hadError()) + return ErrorOperand(Tok.getLoc(), "Unexpected token!"); + if (!Done && UpdateLocLex) { End = Tok.getLoc(); Parser.Lex(); // Consume the token. } } + return 0; +} - if (!Disp) - Disp = MCConstantExpr::Create(SM.getDisp(), getContext()); +X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, SMLoc Start, + int64_t ImmDisp, + unsigned Size) { + const AsmToken &Tok = Parser.getTok(); + SMLoc BracLoc = Tok.getLoc(), End = Tok.getEndLoc(); + if (getLexer().isNot(AsmToken::LBrac)) + return ErrorOperand(BracLoc, "Expected '[' token!"); + Parser.Lex(); // Eat '[' + + SMLoc StartInBrac = Tok.getLoc(); + // Parse [ Symbol + ImmDisp ] and [ BaseReg + Scale*IndexReg + ImmDisp ]. We + // may have already parsed an immediate displacement before the bracketed + // expression. + IntelExprStateMachine SM(ImmDisp, /*StopOnLBrac=*/false, /*AddImmPrefix=*/true); + if (X86Operand *Err = ParseIntelExpression(SM, End)) + return Err; + + const MCExpr *Disp; + if (const MCExpr *Sym = SM.getSym()) { + // A symbolic displacement. + Disp = Sym; + if (isParsingInlineAsm()) + RewriteIntelBracExpression(InstInfo->AsmRewrites, SM.getSymName(), + ImmDisp, SM.getImm(), BracLoc, StartInBrac, + End); + } else { + // An immediate displacement only. + Disp = MCConstantExpr::Create(SM.getImm(), getContext()); + } // Parse the dot operator (e.g., [ebx].foo.bar). if (Tok.getString().startswith(".")) { - SmallString<64> Err; const MCExpr *NewDisp; - if (ParseIntelDotOperator(Disp, &NewDisp, Err)) - return ErrorOperand(Tok.getLoc(), Err); + if (X86Operand *Err = ParseIntelDotOperator(Disp, NewDisp)) + return Err; - End = Parser.getTok().getEndLoc(); + End = Tok.getEndLoc(); Parser.Lex(); // Eat the field. Disp = NewDisp; } int BaseReg = SM.getBaseReg(); int IndexReg = SM.getIndexReg(); - - // handle [-42] - if (!BaseReg && !IndexReg) { - if (!SegReg) - return X86Operand::CreateMem(Disp, Start, End); - else - return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, Start, End, Size); + int Scale = SM.getScale(); + if (!isParsingInlineAsm()) { + // handle [-42] + if (!BaseReg && !IndexReg) { + if (!SegReg) + return X86Operand::CreateMem(Disp, Start, End, Size); + else + return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, Start, End, Size); + } + return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, Start, + End, Size); } - int Scale = SM.getScale(); - return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, - Start, End, Size); + InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo(); + return CreateMemForInlineAsm(SegReg, Disp, BaseReg, IndexReg, Scale, Start, + End, Size, SM.getSymName(), Info); +} + +// Inline assembly may use variable names with namespace alias qualifiers. +X86Operand *X86AsmParser::ParseIntelIdentifier(const MCExpr *&Val, + StringRef &Identifier, + InlineAsmIdentifierInfo &Info, + SMLoc &End) { + assert (isParsingInlineAsm() && "Expected to be parsing inline assembly."); + Val = 0; + + StringRef LineBuf(Identifier.data()); + SemaCallback->LookupInlineAsmIdentifier(LineBuf, Info); + unsigned BufLen = LineBuf.size(); + assert (BufLen && "Expected a non-zero length identifier."); + + // Advance the token stream based on what the frontend parsed. + const AsmToken &Tok = Parser.getTok(); + AsmToken IdentEnd = Tok; + while (BufLen > 0) { + IdentEnd = Tok; + BufLen -= Tok.getString().size(); + getLexer().Lex(); // Consume the token. + } + if (BufLen != 0) + return ErrorOperand(IdentEnd.getLoc(), + "Frontend parser mismatch with asm lexer!"); + End = IdentEnd.getEndLoc(); + + // Create the symbol reference. + Identifier = LineBuf; + MCSymbol *Sym = getContext().GetOrCreateSymbol(Identifier); + MCSymbolRefExpr::VariantKind Variant = MCSymbolRefExpr::VK_None; + Val = MCSymbolRefExpr::Create(Sym, Variant, getParser().getContext()); + return 0; } /// ParseIntelMemOperand - Parse intel style memory operand. -X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, SMLoc Start) { +X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, + int64_t ImmDisp, + SMLoc Start) { const AsmToken &Tok = Parser.getTok(); SMLoc End; unsigned Size = getIntelMemOperandSize(Tok.getString()); if (Size) { - Parser.Lex(); - assert ((Tok.getString() == "PTR" || Tok.getString() == "ptr") && - "Unexpected token!"); - Parser.Lex(); + Parser.Lex(); // Eat operand size (e.g., byte, word). + if (Tok.getString() != "PTR" && Tok.getString() != "ptr") + return ErrorOperand(Start, "Expected 'PTR' or 'ptr' token!"); + Parser.Lex(); // Eat ptr. + } + + // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. + if (getLexer().is(AsmToken::Integer)) { + if (isParsingInlineAsm()) + InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, + Tok.getLoc())); + int64_t ImmDisp = Tok.getIntVal(); + Parser.Lex(); // Eat the integer. + if (getLexer().isNot(AsmToken::LBrac)) + return ErrorOperand(Start, "Expected '[' token!"); + return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); } if (getLexer().is(AsmToken::LBrac)) - return ParseIntelBracExpression(SegReg, Size); + return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); if (!ParseRegister(SegReg, Start, End)) { // Handel SegReg : [ ... ] @@ -1034,63 +1434,36 @@ X86Operand *X86AsmParser::ParseIntelMemOperand(unsigned SegReg, SMLoc Start) { Parser.Lex(); // Eat : if (getLexer().isNot(AsmToken::LBrac)) return ErrorOperand(Start, "Expected '[' token!"); - return ParseIntelBracExpression(SegReg, Size); + return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size); } - const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext()); - if (getParser().parseExpression(Disp, End)) - return 0; + const MCExpr *Val; + if (!isParsingInlineAsm()) { + if (getParser().parsePrimaryExpr(Val, End)) + return ErrorOperand(Tok.getLoc(), "Unexpected token!"); - bool NeedSizeDir = false; - bool IsVarDecl = false; - if (isParsingInlineAsm()) { - if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Disp)) { - const MCSymbol &Sym = SymRef->getSymbol(); - // FIXME: The SemaLookup will fail if the name is anything other then an - // identifier. - // FIXME: Pass a valid SMLoc. - unsigned tLength, tSize, tType; - SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, tLength, - tSize, tType, IsVarDecl); - if (!Size) - Size = tType * 8; // Size is in terms of bits in this context. - NeedSizeDir = Size > 0; - } + return X86Operand::CreateMem(Val, Start, End, Size); } - if (!isParsingInlineAsm()) - return X86Operand::CreateMem(Disp, Start, End, Size); - else { - // If this is not a VarDecl then assume it is a FuncDecl or some other label - // reference. We need an 'r' constraint here, so we need to create register - // operand to ensure proper matching. Just pick a GPR based on the size of - // a pointer. - if (!IsVarDecl) { - unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX; - return X86Operand::CreateReg(RegNo, Start, End, /*AddressOf=*/true); - } - // When parsing inline assembly we set the base register to a non-zero value - // as we don't know the actual value at this time. This is necessary to - // get the matching correct in some cases. - return X86Operand::CreateMem(/*SegReg*/0, Disp, /*BaseReg*/1, /*IndexReg*/0, - /*Scale*/1, Start, End, Size, NeedSizeDir); - } + InlineAsmIdentifierInfo Info; + StringRef Identifier = Tok.getString(); + if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info, End)) + return Err; + return CreateMemForInlineAsm(/*SegReg=*/0, Val, /*BaseReg=*/0,/*IndexReg=*/0, + /*Scale=*/1, Start, End, Size, Identifier, Info); } /// Parse the '.' operator. -bool X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, - const MCExpr **NewDisp, - SmallString<64> &Err) { - AsmToken Tok = *&Parser.getTok(); - uint64_t OrigDispVal, DotDispVal; +X86Operand *X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, + const MCExpr *&NewDisp) { + const AsmToken &Tok = Parser.getTok(); + int64_t OrigDispVal, DotDispVal; // FIXME: Handle non-constant expressions. - if (const MCConstantExpr *OrigDisp = dyn_cast<MCConstantExpr>(Disp)) { + if (const MCConstantExpr *OrigDisp = dyn_cast<MCConstantExpr>(Disp)) OrigDispVal = OrigDisp->getValue(); - } else { - Err = "Non-constant offsets are not supported!"; - return true; - } + else + return ErrorOperand(Tok.getLoc(), "Non-constant offsets are not supported!"); // Drop the '.'. StringRef DotDispStr = Tok.getString().drop_front(1); @@ -1100,23 +1473,15 @@ bool X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, APInt DotDisp; DotDispStr.getAsInteger(10, DotDisp); DotDispVal = DotDisp.getZExtValue(); - } else if (Tok.is(AsmToken::Identifier)) { - // We should only see an identifier when parsing the original inline asm. - // The front-end should rewrite this in terms of immediates. - assert (isParsingInlineAsm() && "Unexpected field name!"); - + } else if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) { unsigned DotDisp; std::pair<StringRef, StringRef> BaseMember = DotDispStr.split('.'); if (SemaCallback->LookupInlineAsmField(BaseMember.first, BaseMember.second, - DotDisp)) { - Err = "Unable to lookup field reference!"; - return true; - } + DotDisp)) + return ErrorOperand(Tok.getLoc(), "Unable to lookup field reference!"); DotDispVal = DotDisp; - } else { - Err = "Unexpected token type!"; - return true; - } + } else + return ErrorOperand(Tok.getLoc(), "Unexpected token type!"); if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) { SMLoc Loc = SMLoc::getFromPointer(DotDispStr.data()); @@ -1126,22 +1491,23 @@ bool X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp, Val)); } - *NewDisp = MCConstantExpr::Create(OrigDispVal + DotDispVal, getContext()); - return false; + NewDisp = MCConstantExpr::Create(OrigDispVal + DotDispVal, getContext()); + return 0; } /// Parse the 'offset' operator. This operator is used to specify the /// location rather then the content of a variable. -X86Operand *X86AsmParser::ParseIntelOffsetOfOperator(SMLoc Start) { - SMLoc OffsetOfLoc = Start; +X86Operand *X86AsmParser::ParseIntelOffsetOfOperator() { + const AsmToken &Tok = Parser.getTok(); + SMLoc OffsetOfLoc = Tok.getLoc(); Parser.Lex(); // Eat offset. - Start = Parser.getTok().getLoc(); - assert (Parser.getTok().is(AsmToken::Identifier) && "Expected an identifier"); - SMLoc End; const MCExpr *Val; - if (getParser().parseExpression(Val, End)) - return ErrorOperand(Start, "Unable to parse expression!"); + InlineAsmIdentifierInfo Info; + SMLoc Start = Tok.getLoc(), End; + StringRef Identifier = Tok.getString(); + if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info, End)) + return Err; // Don't emit the offset operator. InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, OffsetOfLoc, 7)); @@ -1151,7 +1517,7 @@ X86Operand *X86AsmParser::ParseIntelOffsetOfOperator(SMLoc Start) { // the size of a pointer. unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX; return X86Operand::CreateReg(RegNo, Start, End, /*GetAddress=*/true, - OffsetOfLoc); + OffsetOfLoc, Identifier, Info.OpDecl); } enum IntelOperatorKind { @@ -1166,34 +1532,24 @@ enum IntelOperatorKind { /// variable. A variable's size is the product of its LENGTH and TYPE. The /// TYPE operator returns the size of a C or C++ type or variable. If the /// variable is an array, TYPE returns the size of a single element. -X86Operand *X86AsmParser::ParseIntelOperator(SMLoc Start, unsigned OpKind) { - SMLoc TypeLoc = Start; - Parser.Lex(); // Eat offset. - Start = Parser.getTok().getLoc(); - assert (Parser.getTok().is(AsmToken::Identifier) && "Expected an identifier"); +X86Operand *X86AsmParser::ParseIntelOperator(unsigned OpKind) { + const AsmToken &Tok = Parser.getTok(); + SMLoc TypeLoc = Tok.getLoc(); + Parser.Lex(); // Eat operator. - SMLoc End; - const MCExpr *Val; - if (getParser().parseExpression(Val, End)) - return 0; + const MCExpr *Val = 0; + InlineAsmIdentifierInfo Info; + SMLoc Start = Tok.getLoc(), End; + StringRef Identifier = Tok.getString(); + if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info, End)) + return Err; - unsigned Length = 0, Size = 0, Type = 0; - if (const MCSymbolRefExpr *SymRef = dyn_cast<MCSymbolRefExpr>(Val)) { - const MCSymbol &Sym = SymRef->getSymbol(); - // FIXME: The SemaLookup will fail if the name is anything other then an - // identifier. - // FIXME: Pass a valid SMLoc. - bool IsVarDecl; - if (!SemaCallback->LookupInlineAsmIdentifier(Sym.getName(), NULL, Length, - Size, Type, IsVarDecl)) - return ErrorOperand(Start, "Unable to lookup expr!"); - } - unsigned CVal; + unsigned CVal = 0; switch(OpKind) { default: llvm_unreachable("Unexpected operand kind!"); - case IOK_LENGTH: CVal = Length; break; - case IOK_SIZE: CVal = Size; break; - case IOK_TYPE: CVal = Type; break; + case IOK_LENGTH: CVal = Info.Length; break; + case IOK_SIZE: CVal = Info.Size; break; + case IOK_TYPE: CVal = Info.Type; break; } // Rewrite the type operator and the C or C++ type or variable in terms of an @@ -1202,32 +1558,58 @@ X86Operand *X86AsmParser::ParseIntelOperator(SMLoc Start, unsigned OpKind) { InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, TypeLoc, Len, CVal)); const MCExpr *Imm = MCConstantExpr::Create(CVal, getContext()); - return X86Operand::CreateImm(Imm, Start, End, /*NeedAsmRewrite*/false); + return X86Operand::CreateImm(Imm, Start, End); } X86Operand *X86AsmParser::ParseIntelOperand() { - SMLoc Start = Parser.getTok().getLoc(), End; - StringRef AsmTokStr = Parser.getTok().getString(); + const AsmToken &Tok = Parser.getTok(); + SMLoc Start = Tok.getLoc(), End; // Offset, length, type and size operators. if (isParsingInlineAsm()) { + StringRef AsmTokStr = Tok.getString(); if (AsmTokStr == "offset" || AsmTokStr == "OFFSET") - return ParseIntelOffsetOfOperator(Start); + return ParseIntelOffsetOfOperator(); if (AsmTokStr == "length" || AsmTokStr == "LENGTH") - return ParseIntelOperator(Start, IOK_LENGTH); + return ParseIntelOperator(IOK_LENGTH); if (AsmTokStr == "size" || AsmTokStr == "SIZE") - return ParseIntelOperator(Start, IOK_SIZE); + return ParseIntelOperator(IOK_SIZE); if (AsmTokStr == "type" || AsmTokStr == "TYPE") - return ParseIntelOperator(Start, IOK_TYPE); + return ParseIntelOperator(IOK_TYPE); } // Immediate. - if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Real) || - getLexer().is(AsmToken::Minus)) { - const MCExpr *Val; - if (!getParser().parseExpression(Val, End)) { - return X86Operand::CreateImm(Val, Start, End); + if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Minus) || + getLexer().is(AsmToken::LParen)) { + AsmToken StartTok = Tok; + IntelExprStateMachine SM(/*Imm=*/0, /*StopOnLBrac=*/true, + /*AddImmPrefix=*/false); + if (X86Operand *Err = ParseIntelExpression(SM, End)) + return Err; + + int64_t Imm = SM.getImm(); + if (isParsingInlineAsm()) { + unsigned Len = Tok.getLoc().getPointer() - Start.getPointer(); + if (StartTok.getString().size() == Len) + // Just add a prefix if this wasn't a complex immediate expression. + InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, Start)); + else + // Otherwise, rewrite the complex expression as a single immediate. + InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, Start, Len, Imm)); + } + + if (getLexer().isNot(AsmToken::LBrac)) { + const MCExpr *ImmExpr = MCConstantExpr::Create(Imm, getContext()); + return X86Operand::CreateImm(ImmExpr, Start, End); } + + // Only positive immediates are valid. + if (Imm < 0) + return ErrorOperand(Start, "expected a positive immediate displacement " + "before bracketed expr."); + + // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ]. + return ParseIntelMemOperand(/*SegReg=*/0, Imm, Start); } // Register. @@ -1239,11 +1621,11 @@ X86Operand *X86AsmParser::ParseIntelOperand() { return X86Operand::CreateReg(RegNo, Start, End); getParser().Lex(); // Eat the colon. - return ParseIntelMemOperand(RegNo, Start); + return ParseIntelMemOperand(/*SegReg=*/RegNo, /*Disp=*/0, Start); } // Memory operand. - return ParseIntelMemOperand(0, Start); + return ParseIntelMemOperand(/*SegReg=*/0, /*Disp=*/0, Start); } X86Operand *X86AsmParser::ParseATTOperand() { @@ -1267,7 +1649,6 @@ X86Operand *X86AsmParser::ParseATTOperand() { if (getLexer().isNot(AsmToken::Colon)) return X86Operand::CreateReg(RegNo, Start, End); - getParser().Lex(); // Eat the colon. return ParseMemOperand(RegNo, Start); } diff --git a/lib/Target/X86/CMakeLists.txt b/lib/Target/X86/CMakeLists.txt index d14899d..7cb71f0 100644 --- a/lib/Target/X86/CMakeLists.txt +++ b/lib/Target/X86/CMakeLists.txt @@ -33,6 +33,7 @@ set(sources X86TargetObjectFile.cpp X86TargetTransformInfo.cpp X86VZeroUpper.cpp + X86FixupLEAs.cpp ) if( CMAKE_CL_64 ) diff --git a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c index 85d8a99..e40edba 100644 --- a/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c +++ b/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c @@ -61,7 +61,7 @@ static int modRMRequired(OpcodeType type, InstructionContext insnContext, uint8_t opcode) { const struct ContextDecision* decision = 0; - + switch (type) { case ONEBYTE: decision = &ONEBYTE_SYM; @@ -102,7 +102,7 @@ static InstrUID decode(OpcodeType type, uint8_t opcode, uint8_t modRM) { const struct ModRMDecision* dec = 0; - + switch (type) { case ONEBYTE: dec = &ONEBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; @@ -123,7 +123,7 @@ static InstrUID decode(OpcodeType type, dec = &THREEBYTEA7_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; break; } - + switch (dec->modrm_type) { default: debug("Corrupt table! Unknown modrm_type"); @@ -171,10 +171,10 @@ static const struct InstructionSpecifier *specifierForUID(InstrUID uid) { */ static int consumeByte(struct InternalInstruction* insn, uint8_t* byte) { int ret = insn->reader(insn->readerArg, byte, insn->readerCursor); - + if (!ret) ++(insn->readerCursor); - + return ret; } @@ -238,19 +238,19 @@ CONSUME_FUNC(consumeUInt64, uint64_t) */ static void dbgprintf(struct InternalInstruction* insn, const char* format, - ...) { + ...) { char buffer[256]; va_list ap; - + if (!insn->dlog) return; - + va_start(ap, format); (void)vsnprintf(buffer, sizeof(buffer), format, ap); va_end(ap); - + insn->dlog(insn->dlogArg, buffer); - + return; } @@ -305,27 +305,40 @@ static int readPrefixes(struct InternalInstruction* insn) { BOOL prefixGroups[4] = { FALSE }; uint64_t prefixLocation; uint8_t byte = 0; - + BOOL hasAdSize = FALSE; BOOL hasOpSize = FALSE; - + dbgprintf(insn, "readPrefixes()"); - + while (isPrefix) { prefixLocation = insn->readerCursor; - + if (consumeByte(insn, &byte)) return -1; /* - * If the first byte is a LOCK prefix break and let it be disassembled - * as a lock "instruction", by creating an <MCInst #xxxx LOCK_PREFIX>. - * FIXME there is currently no way to get the disassembler to print the - * lock prefix if it is not the first byte. + * If the byte is a LOCK/REP/REPNE prefix and not a part of the opcode, then + * break and let it be disassembled as a normal "instruction". */ - if (insn->readerCursor - 1 == insn->startLocation && byte == 0xf0) - break; - + if (insn->readerCursor - 1 == insn->startLocation + && (byte == 0xf0 || byte == 0xf2 || byte == 0xf3)) { + uint8_t nextByte; + if (byte == 0xf0) + break; + if (lookAtByte(insn, &nextByte)) + return -1; + if (insn->mode == MODE_64BIT && (nextByte & 0xf0) == 0x40) { + if (consumeByte(insn, &nextByte)) + return -1; + if (lookAtByte(insn, &nextByte)) + return -1; + unconsumeByte(insn); + } + if (nextByte != 0x0f && nextByte != 0x90) + break; + } + switch (byte) { case 0xf0: /* LOCK */ case 0xf2: /* REPNE/REPNZ */ @@ -387,21 +400,21 @@ static int readPrefixes(struct InternalInstruction* insn) { isPrefix = FALSE; break; } - + if (isPrefix) dbgprintf(insn, "Found prefix 0x%hhx", byte); } - + insn->vexSize = 0; - + if (byte == 0xc4) { uint8_t byte1; - + if (lookAtByte(insn, &byte1)) { dbgprintf(insn, "Couldn't read second byte of VEX"); return -1; } - + if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) { insn->vexSize = 3; insn->necessaryPrefixLocation = insn->readerCursor - 1; @@ -410,67 +423,67 @@ static int readPrefixes(struct InternalInstruction* insn) { unconsumeByte(insn); insn->necessaryPrefixLocation = insn->readerCursor - 1; } - + if (insn->vexSize == 3) { insn->vexPrefix[0] = byte; consumeByte(insn, &insn->vexPrefix[1]); consumeByte(insn, &insn->vexPrefix[2]); /* We simulate the REX prefix for simplicity's sake */ - + if (insn->mode == MODE_64BIT) { - insn->rexPrefix = 0x40 + insn->rexPrefix = 0x40 | (wFromVEX3of3(insn->vexPrefix[2]) << 3) | (rFromVEX2of3(insn->vexPrefix[1]) << 2) | (xFromVEX2of3(insn->vexPrefix[1]) << 1) | (bFromVEX2of3(insn->vexPrefix[1]) << 0); } - + switch (ppFromVEX3of3(insn->vexPrefix[2])) { default: break; case VEX_PREFIX_66: - hasOpSize = TRUE; + hasOpSize = TRUE; break; } - + dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx 0x%hhx", insn->vexPrefix[0], insn->vexPrefix[1], insn->vexPrefix[2]); } } else if (byte == 0xc5) { uint8_t byte1; - + if (lookAtByte(insn, &byte1)) { dbgprintf(insn, "Couldn't read second byte of VEX"); return -1; } - + if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) { insn->vexSize = 2; } else { unconsumeByte(insn); } - + if (insn->vexSize == 2) { insn->vexPrefix[0] = byte; consumeByte(insn, &insn->vexPrefix[1]); - + if (insn->mode == MODE_64BIT) { - insn->rexPrefix = 0x40 + insn->rexPrefix = 0x40 | (rFromVEX2of2(insn->vexPrefix[1]) << 2); } - + switch (ppFromVEX2of2(insn->vexPrefix[1])) { default: break; case VEX_PREFIX_66: - hasOpSize = TRUE; + hasOpSize = TRUE; break; } - + dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx", insn->vexPrefix[0], insn->vexPrefix[1]); } } @@ -478,17 +491,17 @@ static int readPrefixes(struct InternalInstruction* insn) { if (insn->mode == MODE_64BIT) { if ((byte & 0xf0) == 0x40) { uint8_t opcodeByte; - + if (lookAtByte(insn, &opcodeByte) || ((opcodeByte & 0xf0) == 0x40)) { dbgprintf(insn, "Redundant REX prefix"); return -1; } - + insn->rexPrefix = byte; insn->necessaryPrefixLocation = insn->readerCursor - 2; - + dbgprintf(insn, "Found REX prefix 0x%hhx", byte); - } else { + } else { unconsumeByte(insn); insn->necessaryPrefixLocation = insn->readerCursor - 1; } @@ -526,7 +539,7 @@ static int readPrefixes(struct InternalInstruction* insn) { insn->immediateSize = (hasOpSize ? 2 : 4); } } - + return 0; } @@ -537,22 +550,22 @@ static int readPrefixes(struct InternalInstruction* insn) { * @param insn - The instruction whose opcode is to be read. * @return - 0 if the opcode could be read successfully; nonzero otherwise. */ -static int readOpcode(struct InternalInstruction* insn) { +static int readOpcode(struct InternalInstruction* insn) { /* Determine the length of the primary opcode */ - + uint8_t current; - + dbgprintf(insn, "readOpcode()"); - + insn->opcodeType = ONEBYTE; - + if (insn->vexSize == 3) { switch (mmmmmFromVEX2of3(insn->vexPrefix[1])) { default: dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)", mmmmmFromVEX2of3(insn->vexPrefix[1])); - return -1; + return -1; case 0: break; case VEX_LOB_0F: @@ -564,7 +577,7 @@ static int readOpcode(struct InternalInstruction* insn) { insn->threeByteEscape = 0x38; insn->opcodeType = THREEBYTE_38; return consumeByte(insn, &insn->opcode); - case VEX_LOB_0F3A: + case VEX_LOB_0F3A: insn->twoByteEscape = 0x0f; insn->threeByteEscape = 0x3a; insn->opcodeType = THREEBYTE_3A; @@ -577,68 +590,68 @@ static int readOpcode(struct InternalInstruction* insn) { insn->opcodeType = TWOBYTE; return consumeByte(insn, &insn->opcode); } - + if (consumeByte(insn, ¤t)) return -1; - + if (current == 0x0f) { dbgprintf(insn, "Found a two-byte escape prefix (0x%hhx)", current); - + insn->twoByteEscape = current; - + if (consumeByte(insn, ¤t)) return -1; - + if (current == 0x38) { dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); - + insn->threeByteEscape = current; - + if (consumeByte(insn, ¤t)) return -1; - + insn->opcodeType = THREEBYTE_38; } else if (current == 0x3a) { dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); - + insn->threeByteEscape = current; - + if (consumeByte(insn, ¤t)) return -1; - + insn->opcodeType = THREEBYTE_3A; } else if (current == 0xa6) { dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); - + insn->threeByteEscape = current; - + if (consumeByte(insn, ¤t)) return -1; - + insn->opcodeType = THREEBYTE_A6; } else if (current == 0xa7) { dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); - + insn->threeByteEscape = current; - + if (consumeByte(insn, ¤t)) return -1; - + insn->opcodeType = THREEBYTE_A7; } else { dbgprintf(insn, "Didn't find a three-byte escape prefix"); - + insn->opcodeType = TWOBYTE; } } - + /* * At this point we have consumed the full opcode. * Anything we consume from here on must be unconsumed. */ - + insn->opcode = current; - + return 0; } @@ -660,19 +673,19 @@ static int getIDWithAttrMask(uint16_t* instructionID, struct InternalInstruction* insn, uint8_t attrMask) { BOOL hasModRMExtension; - + uint8_t instructionClass; instructionClass = contextForAttrs(attrMask); - + hasModRMExtension = modRMRequired(insn->opcodeType, instructionClass, insn->opcode); - + if (hasModRMExtension) { if (readModRM(insn)) return -1; - + *instructionID = decode(insn->opcodeType, instructionClass, insn->opcode, @@ -683,7 +696,7 @@ static int getIDWithAttrMask(uint16_t* instructionID, insn->opcode, 0); } - + return 0; } @@ -696,7 +709,7 @@ static int getIDWithAttrMask(uint16_t* instructionID, */ static BOOL is16BitEquivalent(const char* orig, const char* equiv) { off_t i; - + for (i = 0;; i++) { if (orig[i] == '\0' && equiv[i] == '\0') return TRUE; @@ -715,8 +728,8 @@ static BOOL is16BitEquivalent(const char* orig, const char* equiv) { } /* - * getID - Determines the ID of an instruction, consuming the ModR/M byte as - * appropriate for extended and escape opcodes. Determines the attributes and + * getID - Determines the ID of an instruction, consuming the ModR/M byte as + * appropriate for extended and escape opcodes. Determines the attributes and * context for the instruction before doing so. * * @param insn - The instruction whose ID is to be determined. @@ -726,21 +739,21 @@ static BOOL is16BitEquivalent(const char* orig, const char* equiv) { static int getID(struct InternalInstruction* insn, const void *miiArg) { uint8_t attrMask; uint16_t instructionID; - + dbgprintf(insn, "getID()"); - + attrMask = ATTR_NONE; if (insn->mode == MODE_64BIT) attrMask |= ATTR_64BIT; - + if (insn->vexSize) { attrMask |= ATTR_VEX; if (insn->vexSize == 3) { switch (ppFromVEX3of3(insn->vexPrefix[2])) { case VEX_PREFIX_66: - attrMask |= ATTR_OPSIZE; + attrMask |= ATTR_OPSIZE; break; case VEX_PREFIX_F3: attrMask |= ATTR_XS; @@ -749,14 +762,14 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) { attrMask |= ATTR_XD; break; } - + if (lFromVEX3of3(insn->vexPrefix[2])) attrMask |= ATTR_VEXL; } else if (insn->vexSize == 2) { switch (ppFromVEX2of2(insn->vexPrefix[1])) { case VEX_PREFIX_66: - attrMask |= ATTR_OPSIZE; + attrMask |= ATTR_OPSIZE; break; case VEX_PREFIX_F3: attrMask |= ATTR_XS; @@ -765,7 +778,7 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) { attrMask |= ATTR_XD; break; } - + if (lFromVEX2of2(insn->vexPrefix[1])) attrMask |= ATTR_VEXL; } @@ -836,26 +849,26 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) { * conservative, but in the specific case where OpSize is present but not * in the right place we check if there's a 16-bit operation. */ - + const struct InstructionSpecifier *spec; uint16_t instructionIDWithOpsize; const char *specName, *specWithOpSizeName; - + spec = specifierForUID(instructionID); - + if (getIDWithAttrMask(&instructionIDWithOpsize, insn, attrMask | ATTR_OPSIZE)) { - /* + /* * ModRM required with OpSize but not present; give up and return version * without OpSize set */ - + insn->instructionID = instructionID; insn->spec = spec; return 0; } - + specName = x86DisassemblerGetInstrName(instructionID, miiArg); specWithOpSizeName = x86DisassemblerGetInstrName(instructionIDWithOpsize, miiArg); @@ -882,10 +895,10 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) { const struct InstructionSpecifier *specWithNewOpcode; spec = specifierForUID(instructionID); - + /* Borrow opcode from one of the other XCHGar opcodes */ insn->opcode = 0x91; - + if (getIDWithAttrMask(&instructionIDWithNewOpcode, insn, attrMask)) { @@ -906,10 +919,10 @@ static int getID(struct InternalInstruction* insn, const void *miiArg) { return 0; } - + insn->instructionID = instructionID; insn->spec = specifierForUID(insn->instructionID); - + return 0; } @@ -924,14 +937,14 @@ static int readSIB(struct InternalInstruction* insn) { SIBIndex sibIndexBase = 0; SIBBase sibBaseBase = 0; uint8_t index, base; - + dbgprintf(insn, "readSIB()"); - + if (insn->consumedSIB) return 0; - + insn->consumedSIB = TRUE; - + switch (insn->addressSize) { case 2: dbgprintf(insn, "SIB-based addressing doesn't work in 16-bit mode"); @@ -949,9 +962,9 @@ static int readSIB(struct InternalInstruction* insn) { if (consumeByte(insn, &insn->sib)) return -1; - + index = indexFromSIB(insn->sib) | (xFromREX(insn->rexPrefix) << 3); - + switch (index) { case 0x4: insn->sibIndex = SIB_INDEX_NONE; @@ -963,7 +976,7 @@ static int readSIB(struct InternalInstruction* insn) { insn->sibIndex = SIB_INDEX_NONE; break; } - + switch (scaleFromSIB(insn->sib)) { case 0: insn->sibScale = 1; @@ -978,9 +991,9 @@ static int readSIB(struct InternalInstruction* insn) { insn->sibScale = 8; break; } - + base = baseFromSIB(insn->sib) | (bFromREX(insn->rexPrefix) << 3); - + switch (base) { case 0x5: switch (modFromModRM(insn->modRM)) { @@ -990,12 +1003,12 @@ static int readSIB(struct InternalInstruction* insn) { break; case 0x1: insn->eaDisplacement = EA_DISP_8; - insn->sibBase = (insn->addressSize == 4 ? + insn->sibBase = (insn->addressSize == 4 ? SIB_BASE_EBP : SIB_BASE_RBP); break; case 0x2: insn->eaDisplacement = EA_DISP_32; - insn->sibBase = (insn->addressSize == 4 ? + insn->sibBase = (insn->addressSize == 4 ? SIB_BASE_EBP : SIB_BASE_RBP); break; case 0x3: @@ -1007,7 +1020,7 @@ static int readSIB(struct InternalInstruction* insn) { insn->sibBase = (SIBBase)(sibBaseBase + base); break; } - + return 0; } @@ -1015,22 +1028,22 @@ static int readSIB(struct InternalInstruction* insn) { * readDisplacement - Consumes the displacement of an instruction. * * @param insn - The instruction whose displacement is to be read. - * @return - 0 if the displacement byte was successfully read; nonzero + * @return - 0 if the displacement byte was successfully read; nonzero * otherwise. */ -static int readDisplacement(struct InternalInstruction* insn) { +static int readDisplacement(struct InternalInstruction* insn) { int8_t d8; int16_t d16; int32_t d32; - + dbgprintf(insn, "readDisplacement()"); - + if (insn->consumedDisplacement) return 0; - + insn->consumedDisplacement = TRUE; insn->displacementOffset = insn->readerCursor - insn->startLocation; - + switch (insn->eaDisplacement) { case EA_DISP_NONE: insn->consumedDisplacement = FALSE; @@ -1051,7 +1064,7 @@ static int readDisplacement(struct InternalInstruction* insn) { insn->displacement = d32; break; } - + insn->consumedDisplacement = TRUE; return 0; } @@ -1063,22 +1076,22 @@ static int readDisplacement(struct InternalInstruction* insn) { * @param insn - The instruction whose addressing information is to be read. * @return - 0 if the information was successfully read; nonzero otherwise. */ -static int readModRM(struct InternalInstruction* insn) { +static int readModRM(struct InternalInstruction* insn) { uint8_t mod, rm, reg; - + dbgprintf(insn, "readModRM()"); - + if (insn->consumedModRM) return 0; - + if (consumeByte(insn, &insn->modRM)) return -1; insn->consumedModRM = TRUE; - + mod = modFromModRM(insn->modRM); rm = rmFromModRM(insn->modRM); reg = regFromModRM(insn->modRM); - + /* * This goes by insn->registerSize to pick the correct register, which messes * up if we're using (say) XMM or 8-bit register operands. That gets fixed in @@ -1098,16 +1111,16 @@ static int readModRM(struct InternalInstruction* insn) { insn->eaRegBase = EA_REG_RAX; break; } - + reg |= rFromREX(insn->rexPrefix) << 3; rm |= bFromREX(insn->rexPrefix) << 3; - + insn->reg = (Reg)(insn->regBase + reg); - + switch (insn->addressSize) { case 2: insn->eaBaseBase = EA_BASE_BX_SI; - + switch (mod) { case 0x0: if (rm == 0x6) { @@ -1142,14 +1155,14 @@ static int readModRM(struct InternalInstruction* insn) { case 4: case 8: insn->eaBaseBase = (insn->addressSize == 4 ? EA_BASE_EAX : EA_BASE_RAX); - + switch (mod) { case 0x0: insn->eaDisplacement = EA_DISP_NONE; /* readSIB may override this */ switch (rm) { case 0x4: case 0xc: /* in case REXW.b is set */ - insn->eaBase = (insn->addressSize == 4 ? + insn->eaBase = (insn->addressSize == 4 ? EA_BASE_sib : EA_BASE_sib64); readSIB(insn); if (readDisplacement(insn)) @@ -1191,7 +1204,7 @@ static int readModRM(struct InternalInstruction* insn) { } break; } /* switch (insn->addressSize) */ - + return 0; } @@ -1274,12 +1287,12 @@ GENERIC_FIXUP_FUNC(fixupRMValue, insn->eaRegBase, EA_REG) * @return - 0 if fixup was successful; -1 if the register returned was * invalid for its class. */ -static int fixupReg(struct InternalInstruction *insn, +static int fixupReg(struct InternalInstruction *insn, const struct OperandSpecifier *op) { uint8_t valid; - + dbgprintf(insn, "fixupReg()"); - + switch ((OperandEncoding)op->encoding) { default: debug("Expected a REG or R/M encoding in fixupReg"); @@ -1311,12 +1324,12 @@ static int fixupReg(struct InternalInstruction *insn, } break; } - + return 0; } /* - * readOpcodeModifier - Reads an operand from the opcode field of an + * readOpcodeModifier - Reads an operand from the opcode field of an * instruction. Handles AddRegFrm instructions. * * @param insn - The instruction whose opcode field is to be read. @@ -1326,12 +1339,12 @@ static int fixupReg(struct InternalInstruction *insn, */ static int readOpcodeModifier(struct InternalInstruction* insn) { dbgprintf(insn, "readOpcodeModifier()"); - + if (insn->consumedOpcodeModifier) return 0; - + insn->consumedOpcodeModifier = TRUE; - + switch (insn->spec->modifierType) { default: debug("Unknown modifier type."); @@ -1345,11 +1358,11 @@ static int readOpcodeModifier(struct InternalInstruction* insn) { case MODIFIER_MODRM: insn->opcodeModifier = insn->modRM - insn->spec->modifierBase; return 0; - } + } } /* - * readOpcodeRegister - Reads an operand from the opcode field of an + * readOpcodeRegister - Reads an operand from the opcode field of an * instruction and interprets it appropriately given the operand width. * Handles AddRegFrm instructions. * @@ -1364,39 +1377,39 @@ static int readOpcodeRegister(struct InternalInstruction* insn, uint8_t size) { if (readOpcodeModifier(insn)) return -1; - + if (size == 0) size = insn->registerSize; - + switch (size) { case 1: - insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix) << 3) + insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix) << 3) | insn->opcodeModifier)); - if (insn->rexPrefix && + if (insn->rexPrefix && insn->opcodeRegister >= MODRM_REG_AL + 0x4 && insn->opcodeRegister < MODRM_REG_AL + 0x8) { insn->opcodeRegister = (Reg)(MODRM_REG_SPL + (insn->opcodeRegister - MODRM_REG_AL - 4)); } - + break; case 2: insn->opcodeRegister = (Reg)(MODRM_REG_AX - + ((bFromREX(insn->rexPrefix) << 3) + + ((bFromREX(insn->rexPrefix) << 3) | insn->opcodeModifier)); break; case 4: insn->opcodeRegister = (Reg)(MODRM_REG_EAX - + ((bFromREX(insn->rexPrefix) << 3) + + ((bFromREX(insn->rexPrefix) << 3) | insn->opcodeModifier)); break; case 8: - insn->opcodeRegister = (Reg)(MODRM_REG_RAX - + ((bFromREX(insn->rexPrefix) << 3) + insn->opcodeRegister = (Reg)(MODRM_REG_RAX + + ((bFromREX(insn->rexPrefix) << 3) | insn->opcodeModifier)); break; } - + return 0; } @@ -1414,20 +1427,20 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) { uint16_t imm16; uint32_t imm32; uint64_t imm64; - + dbgprintf(insn, "readImmediate()"); - + if (insn->numImmediatesConsumed == 2) { debug("Already consumed two immediates"); return -1; } - + if (size == 0) size = insn->immediateSize; else insn->immediateSize = size; insn->immediateOffset = insn->readerCursor - insn->startLocation; - + switch (size) { case 1: if (consumeByte(insn, &imm8)) @@ -1450,9 +1463,9 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) { insn->immediates[insn->numImmediatesConsumed] = imm64; break; } - + insn->numImmediatesConsumed++; - + return 0; } @@ -1465,7 +1478,7 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) { */ static int readVVVV(struct InternalInstruction* insn) { dbgprintf(insn, "readVVVV()"); - + if (insn->vexSize == 3) insn->vvvv = vvvvFromVEX3of3(insn->vexPrefix[2]); else if (insn->vexSize == 2) @@ -1490,14 +1503,14 @@ static int readOperands(struct InternalInstruction* insn) { int index; int hasVVVV, needVVVV; int sawRegImm = 0; - + dbgprintf(insn, "readOperands()"); /* If non-zero vvvv specified, need to make sure one of the operands uses it. */ hasVVVV = !readVVVV(insn); needVVVV = hasVVVV && (insn->vvvv != 0); - + for (index = 0; index < X86_MAX_OPERANDS; ++index) { switch (x86OperandSets[insn->spec->operands][index].encoding) { case ENCODING_NONE: @@ -1599,7 +1612,7 @@ static int readOperands(struct InternalInstruction* insn) { /* If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail */ if (needVVVV) return -1; - + return 0; } @@ -1607,7 +1620,7 @@ static int readOperands(struct InternalInstruction* insn) { * decodeInstruction - Reads and interprets a full instruction provided by the * user. * - * @param insn - A pointer to the instruction to be populated. Must be + * @param insn - A pointer to the instruction to be populated. Must be * pre-allocated. * @param reader - The function to be used to read the instruction's bytes. * @param readerArg - A generic argument to be passed to the reader to store @@ -1632,7 +1645,7 @@ int decodeInstruction(struct InternalInstruction* insn, uint64_t startLoc, DisassemblerMode mode) { memset(insn, 0, sizeof(struct InternalInstruction)); - + insn->reader = reader; insn->readerArg = readerArg; insn->dlog = logger; @@ -1641,7 +1654,7 @@ int decodeInstruction(struct InternalInstruction* insn, insn->readerCursor = startLoc; insn->mode = mode; insn->numImmediatesConsumed = 0; - + if (readPrefixes(insn) || readOpcode(insn) || getID(insn, miiArg) || @@ -1650,14 +1663,14 @@ int decodeInstruction(struct InternalInstruction* insn, return -1; insn->operands = &x86OperandSets[insn->spec->operands][0]; - + insn->length = insn->readerCursor - insn->startLocation; - + dbgprintf(insn, "Read from 0x%llx to 0x%llx: length %zu", startLoc, insn->readerCursor, insn->length); - + if (insn->length > 15) dbgprintf(insn, "Instruction exceeds 15-byte limit"); - + return 0; } diff --git a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h index 9e68388..d8f7278 100644 --- a/lib/Target/X86/MCTargetDesc/X86BaseInfo.h +++ b/lib/Target/X86/MCTargetDesc/X86BaseInfo.h @@ -20,6 +20,7 @@ #include "X86MCTargetDesc.h" #include "llvm/Support/DataTypes.h" #include "llvm/Support/ErrorHandling.h" +#include "llvm/MC/MCInstrInfo.h" namespace llvm { @@ -41,7 +42,6 @@ namespace X86 { AddrNumOperands = 5 }; } // end namespace X86; - /// X86II - This namespace holds all of the target specific flags that /// instruction info tracks. @@ -274,11 +274,12 @@ namespace X86II { //// MRM_XX - A mod/rm byte of exactly 0xXX. MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36, - MRM_C8 = 37, MRM_C9 = 38, MRM_E8 = 39, MRM_F0 = 40, - MRM_F8 = 41, MRM_F9 = 42, MRM_D0 = 45, MRM_D1 = 46, - MRM_D4 = 47, MRM_D5 = 48, MRM_D8 = 49, MRM_D9 = 50, - MRM_DA = 51, MRM_DB = 52, MRM_DC = 53, MRM_DD = 54, - MRM_DE = 55, MRM_DF = 56, + MRM_C8 = 37, MRM_C9 = 38, MRM_CA = 39, MRM_CB = 40, + MRM_E8 = 41, MRM_F0 = 42, MRM_F8 = 45, MRM_F9 = 46, + MRM_D0 = 47, MRM_D1 = 48, MRM_D4 = 49, MRM_D5 = 50, + MRM_D6 = 51, MRM_D8 = 52, MRM_D9 = 53, MRM_DA = 54, + MRM_DB = 55, MRM_DC = 56, MRM_DD = 57, MRM_DE = 58, + MRM_DF = 59, /// RawFrmImm8 - This is used for the ENTER instruction, which has two /// immediates, the first of which is a 16-bit immediate (specified by @@ -521,6 +522,26 @@ namespace X86II { } } + /// getOperandBias - compute any additional adjustment needed to + /// the offset to the start of the memory operand + /// in this instruction. + /// If this is a two-address instruction,skip one of the register operands. + /// FIXME: This should be handled during MCInst lowering. + inline int getOperandBias(const MCInstrDesc& Desc) + { + unsigned NumOps = Desc.getNumOperands(); + unsigned CurOp = 0; + if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0) + ++CurOp; + else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0) { + assert(Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1); + // Special case for GATHER with 2 TIED_TO operands + // Skip the first 2 operands: dst, mask_wb + CurOp += 2; + } + return CurOp; + } + /// getMemoryOperandNo - The function returns the MCInst operand # for the /// first field of the memory operand. If the instruction doesn't have a /// memory operand, this returns -1. @@ -574,17 +595,15 @@ namespace X86II { ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV). return FirstMemOp; } - case X86II::MRM_C1: case X86II::MRM_C2: - case X86II::MRM_C3: case X86II::MRM_C4: - case X86II::MRM_C8: case X86II::MRM_C9: - case X86II::MRM_E8: case X86II::MRM_F0: - case X86II::MRM_F8: case X86II::MRM_F9: - case X86II::MRM_D0: case X86II::MRM_D1: - case X86II::MRM_D4: case X86II::MRM_D5: - case X86II::MRM_D8: case X86II::MRM_D9: - case X86II::MRM_DA: case X86II::MRM_DB: - case X86II::MRM_DC: case X86II::MRM_DD: - case X86II::MRM_DE: case X86II::MRM_DF: + case X86II::MRM_C1: case X86II::MRM_C2: case X86II::MRM_C3: + case X86II::MRM_C4: case X86II::MRM_C8: case X86II::MRM_C9: + case X86II::MRM_CA: case X86II::MRM_CB: case X86II::MRM_E8: + case X86II::MRM_F0: case X86II::MRM_F8: case X86II::MRM_F9: + case X86II::MRM_D0: case X86II::MRM_D1: case X86II::MRM_D4: + case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D8: + case X86II::MRM_D9: case X86II::MRM_DA: case X86II::MRM_DB: + case X86II::MRM_DC: case X86II::MRM_DD: case X86II::MRM_DE: + case X86II::MRM_DF: return -1; } } diff --git a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp index 5fbefae..016af71 100644 --- a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp +++ b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp @@ -237,6 +237,14 @@ StartsWithGlobalOffsetTable(const MCExpr *Expr) { return GOT_Normal; } +static bool HasSecRelSymbolRef(const MCExpr *Expr) { + if (Expr->getKind() == MCExpr::SymbolRef) { + const MCSymbolRefExpr *Ref = static_cast<const MCSymbolRefExpr*>(Expr); + return Ref->getKind() == MCSymbolRefExpr::VK_SECREL; + } + return false; +} + void X86MCCodeEmitter:: EmitImmediate(const MCOperand &DispOp, SMLoc Loc, unsigned Size, MCFixupKind FixupKind, unsigned &CurByte, raw_ostream &OS, @@ -268,8 +276,13 @@ EmitImmediate(const MCOperand &DispOp, SMLoc Loc, unsigned Size, if (Kind == GOT_Normal) ImmOffset = CurByte; } else if (Expr->getKind() == MCExpr::SymbolRef) { - const MCSymbolRefExpr *Ref = static_cast<const MCSymbolRefExpr*>(Expr); - if (Ref->getKind() == MCSymbolRefExpr::VK_SECREL) { + if (HasSecRelSymbolRef(Expr)) { + FixupKind = MCFixupKind(FK_SecRel_4); + } + } else if (Expr->getKind() == MCExpr::Binary) { + const MCBinaryExpr *Bin = static_cast<const MCBinaryExpr*>(Expr); + if (HasSecRelSymbolRef(Bin->getLHS()) + || HasSecRelSymbolRef(Bin->getRHS())) { FixupKind = MCFixupKind(FK_SecRel_4); } } @@ -979,18 +992,8 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS, if ((TSFlags & X86II::FormMask) == X86II::Pseudo) return; - // If this is a two-address instruction, skip one of the register operands. - // FIXME: This should be handled during MCInst lowering. unsigned NumOps = Desc.getNumOperands(); - unsigned CurOp = 0; - if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0) - ++CurOp; - else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0) { - assert(Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1); - // Special case for GATHER with 2 TIED_TO operands - // Skip the first 2 operands: dst, mask_wb - CurOp += 2; - } + unsigned CurOp = X86II::getOperandBias(Desc); // Keep track of the current byte being emitted. unsigned CurByte = 0; @@ -1136,17 +1139,15 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS, TSFlags, CurByte, OS, Fixups); CurOp += X86::AddrNumOperands; break; - case X86II::MRM_C1: case X86II::MRM_C2: - case X86II::MRM_C3: case X86II::MRM_C4: - case X86II::MRM_C8: case X86II::MRM_C9: - case X86II::MRM_D0: case X86II::MRM_D1: - case X86II::MRM_D4: case X86II::MRM_D5: - case X86II::MRM_D8: case X86II::MRM_D9: - case X86II::MRM_DA: case X86II::MRM_DB: - case X86II::MRM_DC: case X86II::MRM_DD: - case X86II::MRM_DE: case X86II::MRM_DF: - case X86II::MRM_E8: case X86II::MRM_F0: - case X86II::MRM_F8: case X86II::MRM_F9: + case X86II::MRM_C1: case X86II::MRM_C2: case X86II::MRM_C3: + case X86II::MRM_C4: case X86II::MRM_C8: case X86II::MRM_C9: + case X86II::MRM_CA: case X86II::MRM_CB: case X86II::MRM_D0: + case X86II::MRM_D1: case X86II::MRM_D4: case X86II::MRM_D5: + case X86II::MRM_D6: case X86II::MRM_D8: case X86II::MRM_D9: + case X86II::MRM_DA: case X86II::MRM_DB: case X86II::MRM_DC: + case X86II::MRM_DD: case X86II::MRM_DE: case X86II::MRM_DF: + case X86II::MRM_E8: case X86II::MRM_F0: case X86II::MRM_F8: + case X86II::MRM_F9: EmitByte(BaseOpcode, CurByte, OS); unsigned char MRM; @@ -1158,10 +1159,13 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS, case X86II::MRM_C4: MRM = 0xC4; break; case X86II::MRM_C8: MRM = 0xC8; break; case X86II::MRM_C9: MRM = 0xC9; break; + case X86II::MRM_CA: MRM = 0xCA; break; + case X86II::MRM_CB: MRM = 0xCB; break; case X86II::MRM_D0: MRM = 0xD0; break; case X86II::MRM_D1: MRM = 0xD1; break; case X86II::MRM_D4: MRM = 0xD4; break; case X86II::MRM_D5: MRM = 0xD5; break; + case X86II::MRM_D6: MRM = 0xD6; break; case X86II::MRM_D8: MRM = 0xD8; break; case X86II::MRM_D9: MRM = 0xD9; break; case X86II::MRM_DA: MRM = 0xDA; break; diff --git a/lib/Target/X86/MCTargetDesc/X86WinCOFFObjectWriter.cpp b/lib/Target/X86/MCTargetDesc/X86WinCOFFObjectWriter.cpp index bc272ef..ed64a32 100644 --- a/lib/Target/X86/MCTargetDesc/X86WinCOFFObjectWriter.cpp +++ b/lib/Target/X86/MCTargetDesc/X86WinCOFFObjectWriter.cpp @@ -9,6 +9,8 @@ #include "MCTargetDesc/X86FixupKinds.h" #include "MCTargetDesc/X86MCTargetDesc.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCValue.h" #include "llvm/MC/MCWinCOFFObjectWriter.h" #include "llvm/Support/COFF.h" #include "llvm/Support/ErrorHandling.h" @@ -27,7 +29,9 @@ namespace { X86WinCOFFObjectWriter(bool Is64Bit_); ~X86WinCOFFObjectWriter(); - virtual unsigned getRelocType(unsigned FixupKind) const; + virtual unsigned getRelocType(const MCValue &Target, + const MCFixup &Fixup, + bool IsCrossSection) const LLVM_OVERRIDE; }; } @@ -38,7 +42,14 @@ X86WinCOFFObjectWriter::X86WinCOFFObjectWriter(bool Is64Bit_) X86WinCOFFObjectWriter::~X86WinCOFFObjectWriter() {} -unsigned X86WinCOFFObjectWriter::getRelocType(unsigned FixupKind) const { +unsigned X86WinCOFFObjectWriter::getRelocType(const MCValue &Target, + const MCFixup &Fixup, + bool IsCrossSection) const { + unsigned FixupKind = IsCrossSection ? FK_PCRel_4 : Fixup.getKind(); + + MCSymbolRefExpr::VariantKind Modifier = Target.isAbsolute() ? + MCSymbolRefExpr::VK_None : Target.getSymA()->getKind(); + switch (FixupKind) { case FK_PCRel_4: case X86::reloc_riprel_4byte: @@ -46,6 +57,9 @@ unsigned X86WinCOFFObjectWriter::getRelocType(unsigned FixupKind) const { return Is64Bit ? COFF::IMAGE_REL_AMD64_REL32 : COFF::IMAGE_REL_I386_REL32; case FK_Data_4: case X86::reloc_signed_4byte: + if (Modifier == MCSymbolRefExpr::VK_COFF_IMGREL32) + return Is64Bit ? COFF::IMAGE_REL_AMD64_ADDR32NB : + COFF::IMAGE_REL_I386_DIR32NB; return Is64Bit ? COFF::IMAGE_REL_AMD64_ADDR32 : COFF::IMAGE_REL_I386_DIR32; case FK_Data_8: if (Is64Bit) diff --git a/lib/Target/X86/X86.h b/lib/Target/X86/X86.h index 1f9919f..947002f 100644 --- a/lib/Target/X86/X86.h +++ b/lib/Target/X86/X86.h @@ -69,6 +69,11 @@ ImmutablePass *createX86TargetTransformInfoPass(const X86TargetMachine *TM); /// createX86PadShortFunctions - Return a pass that pads short functions /// with NOOPs. This will prevent a stall when returning on the Atom. FunctionPass *createX86PadShortFunctions(); +/// createX86FixupLEAs - Return a a pass that selectively replaces +/// certain instructions (like add, sub, inc, dec, some shifts, +/// and some multiplies) by equivalent LEA instructions, in order +/// to eliminate execution delays in some Atom processors. +FunctionPass *createX86FixupLEAs(); } // End llvm namespace diff --git a/lib/Target/X86/X86.td b/lib/Target/X86/X86.td index 0216252..87bb68d 100644 --- a/lib/Target/X86/X86.td +++ b/lib/Target/X86/X86.td @@ -120,8 +120,14 @@ def FeatureBMI2 : SubtargetFeature<"bmi2", "HasBMI2", "true", "Support BMI2 instructions">; def FeatureRTM : SubtargetFeature<"rtm", "HasRTM", "true", "Support RTM instructions">; +def FeatureHLE : SubtargetFeature<"hle", "HasHLE", "true", + "Support HLE">; def FeatureADX : SubtargetFeature<"adx", "HasADX", "true", "Support ADX instructions">; +def FeaturePRFCHW : SubtargetFeature<"prfchw", "HasPRFCHW", "true", + "Support PRFCHW instructions">; +def FeatureRDSEED : SubtargetFeature<"rdseed", "HasRDSEED", "true", + "Support RDSEED instruction">; def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true", "Use LEA for adjusting the stack pointer">; def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb", @@ -130,6 +136,11 @@ def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb", def FeaturePadShortFunctions : SubtargetFeature<"pad-short-functions", "PadShortFunctions", "true", "Pad short functions">; +def FeatureCallRegIndirect : SubtargetFeature<"call-reg-indirect", + "CallRegIndirect", "true", + "Call register indirect">; +def FeatureLEAUsesAG : SubtargetFeature<"lea-uses-ag", "LEAUsesAG", "true", + "LEA instruction needs inputs at AG stage">; //===----------------------------------------------------------------------===// // X86 processors supported. @@ -143,9 +154,6 @@ def ProcIntelAtom : SubtargetFeature<"atom", "X86ProcFamily", "IntelAtom", class Proc<string Name, list<SubtargetFeature> Features> : ProcessorModel<Name, GenericModel, Features>; -class AtomProc<string Name, list<SubtargetFeature> Features> - : ProcessorModel<Name, AtomModel, Features>; - def : Proc<"generic", []>; def : Proc<"i386", []>; def : Proc<"i486", []>; @@ -162,46 +170,62 @@ def : Proc<"pentium4", [FeatureSSE2]>; def : Proc<"pentium4m", [FeatureSSE2, FeatureSlowBTMem]>; def : Proc<"x86-64", [FeatureSSE2, Feature64Bit, FeatureSlowBTMem, FeatureFastUAMem]>; -def : Proc<"yonah", [FeatureSSE3, FeatureSlowBTMem]>; -def : Proc<"prescott", [FeatureSSE3, FeatureSlowBTMem]>; -def : Proc<"nocona", [FeatureSSE3, FeatureCMPXCHG16B, - FeatureSlowBTMem]>; -def : Proc<"core2", [FeatureSSSE3, FeatureCMPXCHG16B, - FeatureSlowBTMem]>; -def : Proc<"penryn", [FeatureSSE41, FeatureCMPXCHG16B, - FeatureSlowBTMem]>; -def : AtomProc<"atom", [ProcIntelAtom, FeatureSSSE3, FeatureCMPXCHG16B, - FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP, - FeatureSlowDivide, FeaturePadShortFunctions]>; +// Intel Core Duo. +def : ProcessorModel<"yonah", SandyBridgeModel, + [FeatureSSE3, FeatureSlowBTMem]>; + +// NetBurst. +def : Proc<"prescott", [FeatureSSE3, FeatureSlowBTMem]>; +def : Proc<"nocona", [FeatureSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>; + +// Intel Core 2 Solo/Duo. +def : ProcessorModel<"core2", SandyBridgeModel, + [FeatureSSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>; +def : ProcessorModel<"penryn", SandyBridgeModel, + [FeatureSSE41, FeatureCMPXCHG16B, FeatureSlowBTMem]>; + +// Atom. +def : ProcessorModel<"atom", AtomModel, + [ProcIntelAtom, FeatureSSSE3, FeatureCMPXCHG16B, + FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP, + FeatureSlowDivide, + FeatureCallRegIndirect, + FeatureLEAUsesAG, + FeaturePadShortFunctions]>; + // "Arrandale" along with corei3 and corei5 -def : Proc<"corei7", [FeatureSSE42, FeatureCMPXCHG16B, - FeatureSlowBTMem, FeatureFastUAMem, - FeaturePOPCNT, FeatureAES]>; -def : Proc<"nehalem", [FeatureSSE42, FeatureCMPXCHG16B, - FeatureSlowBTMem, FeatureFastUAMem, - FeaturePOPCNT]>; +def : ProcessorModel<"corei7", SandyBridgeModel, + [FeatureSSE42, FeatureCMPXCHG16B, FeatureSlowBTMem, + FeatureFastUAMem, FeaturePOPCNT, FeatureAES]>; + +def : ProcessorModel<"nehalem", SandyBridgeModel, + [FeatureSSE42, FeatureCMPXCHG16B, FeatureSlowBTMem, + FeatureFastUAMem, FeaturePOPCNT]>; // Westmere is a similar machine to nehalem with some additional features. // Westmere is the corei3/i5/i7 path from nehalem to sandybridge -def : Proc<"westmere", [FeatureSSE42, FeatureCMPXCHG16B, - FeatureSlowBTMem, FeatureFastUAMem, - FeaturePOPCNT, FeatureAES, FeaturePCLMUL]>; +def : ProcessorModel<"westmere", SandyBridgeModel, + [FeatureSSE42, FeatureCMPXCHG16B, FeatureSlowBTMem, + FeatureFastUAMem, FeaturePOPCNT, FeatureAES, + FeaturePCLMUL]>; // Sandy Bridge // SSE is not listed here since llvm treats AVX as a reimplementation of SSE, // rather than a superset. -def : Proc<"corei7-avx", [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem, - FeaturePOPCNT, FeatureAES, FeaturePCLMUL]>; +def : ProcessorModel<"corei7-avx", SandyBridgeModel, + [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem, + FeaturePOPCNT, FeatureAES, FeaturePCLMUL]>; // Ivy Bridge -def : Proc<"core-avx-i", [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem, - FeaturePOPCNT, FeatureAES, FeaturePCLMUL, - FeatureRDRAND, FeatureF16C, FeatureFSGSBase]>; +def : ProcessorModel<"core-avx-i", SandyBridgeModel, + [FeatureAVX, FeatureCMPXCHG16B, FeatureFastUAMem, + FeaturePOPCNT, FeatureAES, FeaturePCLMUL, FeatureRDRAND, + FeatureF16C, FeatureFSGSBase]>; // Haswell -def : Proc<"core-avx2", [FeatureAVX2, FeatureCMPXCHG16B, FeatureFastUAMem, - FeaturePOPCNT, FeatureAES, FeaturePCLMUL, - FeatureRDRAND, FeatureF16C, FeatureFSGSBase, - FeatureMOVBE, FeatureLZCNT, FeatureBMI, - FeatureBMI2, FeatureFMA, - FeatureRTM]>; +def : ProcessorModel<"core-avx2", HaswellModel, + [FeatureAVX2, FeatureCMPXCHG16B, FeatureFastUAMem, + FeaturePOPCNT, FeatureAES, FeaturePCLMUL, FeatureRDRAND, + FeatureF16C, FeatureFSGSBase, FeatureMOVBE, FeatureLZCNT, + FeatureBMI, FeatureBMI2, FeatureFMA, FeatureRTM, + FeatureHLE]>; def : Proc<"k6", [FeatureMMX]>; def : Proc<"k6-2", [Feature3DNow]>; @@ -279,6 +303,9 @@ def ATTAsmParser : AsmParser { def ATTAsmParserVariant : AsmParserVariant { int Variant = 0; + // Variant name. + string Name = "att"; + // Discard comments in assembly strings. string CommentDelimiter = "#"; @@ -289,6 +316,9 @@ def ATTAsmParserVariant : AsmParserVariant { def IntelAsmParserVariant : AsmParserVariant { int Variant = 1; + // Variant name. + string Name = "intel"; + // Discard comments in assembly strings. string CommentDelimiter = ";"; diff --git a/lib/Target/X86/X86AsmPrinter.cpp b/lib/Target/X86/X86AsmPrinter.cpp index ac5daec..6b228b0 100644 --- a/lib/Target/X86/X86AsmPrinter.cpp +++ b/lib/Target/X86/X86AsmPrinter.cpp @@ -201,7 +201,7 @@ void X86AsmPrinter::printSymbolOperand(const MachineOperand &MO, case X86II::MO_TLVP_PIC_BASE: O << "@TLVP" << '-' << *MF->getPICBaseSymbol(); break; - case X86II::MO_SECREL: O << "@SECREL"; break; + case X86II::MO_SECREL: O << "@SECREL32"; break; } } diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td index b516be0..9eafbd5 100644 --- a/lib/Target/X86/X86CallingConv.td +++ b/lib/Target/X86/X86CallingConv.td @@ -387,8 +387,8 @@ def CC_X86_32_ThisCall : CallingConv<[ // Promote i8/i16 arguments to i32. CCIfType<[i8, i16], CCPromoteToType<i32>>, - // Pass sret arguments indirectly through EAX - CCIfSRet<CCAssignToReg<[EAX]>>, + // Pass sret arguments indirectly through stack. + CCIfSRet<CCAssignToStack<4, 4>>, // The first integer argument is passed in ECX CCIfType<[i32], CCAssignToReg<[ECX]>>, diff --git a/lib/Target/X86/X86CodeEmitter.cpp b/lib/Target/X86/X86CodeEmitter.cpp index 2518e02..8fea6ed 100644 --- a/lib/Target/X86/X86CodeEmitter.cpp +++ b/lib/Target/X86/X86CodeEmitter.cpp @@ -1451,6 +1451,14 @@ void Emitter<CodeEmitter>::emitInstruction(MachineInstr &MI, MCE.emitByte(BaseOpcode); MCE.emitByte(0xC9); break; + case X86II::MRM_CA: + MCE.emitByte(BaseOpcode); + MCE.emitByte(0xCA); + break; + case X86II::MRM_CB: + MCE.emitByte(BaseOpcode); + MCE.emitByte(0xCB); + break; case X86II::MRM_E8: MCE.emitByte(BaseOpcode); MCE.emitByte(0xE8); diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp index 85155f5..cf44bd0 100644 --- a/lib/Target/X86/X86FastISel.cpp +++ b/lib/Target/X86/X86FastISel.cpp @@ -68,12 +68,12 @@ public: virtual bool TargetSelectInstruction(const Instruction *I); - /// TryToFoldLoad - The specified machine instr operand is a vreg, and that + /// \brief The specified machine instr operand is a vreg, and that /// vreg is being provided by the specified load instruction. If possible, /// try to fold the load as an operand to the instruction, returning true if /// possible. - virtual bool TryToFoldLoad(MachineInstr *MI, unsigned OpNo, - const LoadInst *LI); + virtual bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo, + const LoadInst *LI); virtual bool FastLowerArguments(); @@ -107,6 +107,8 @@ private: bool X86SelectShift(const Instruction *I); + bool X86SelectDivRem(const Instruction *I); + bool X86SelectSelect(const Instruction *I); bool X86SelectTrunc(const Instruction *I); @@ -691,11 +693,6 @@ bool X86FastISel::X86SelectStore(const Instruction *I) { if (S->isAtomic()) return false; - unsigned SABIAlignment = - TD.getABITypeAlignment(S->getValueOperand()->getType()); - if (S->getAlignment() != 0 && S->getAlignment() < SABIAlignment) - return false; - MVT VT; if (!isTypeLegal(I->getOperand(0)->getType(), VT, /*AllowI1=*/true)) return false; @@ -816,14 +813,16 @@ bool X86FastISel::X86SelectRet(const Instruction *I) { // The x86-64 ABI for returning structs by value requires that we copy // the sret argument into %rax for the return. We saved the argument into // a virtual register in the entry block, so now we copy the value out - // and into %rax. - if (Subtarget->is64Bit() && F.hasStructRetAttr()) { + // and into %rax. We also do the same with %eax for Win32. + if (F.hasStructRetAttr() && + (Subtarget->is64Bit() || Subtarget->isTargetWindows())) { unsigned Reg = X86MFInfo->getSRetReturnReg(); assert(Reg && "SRetReturnReg should have been set in LowerFormalArguments()!"); + unsigned RetReg = Subtarget->is64Bit() ? X86::RAX : X86::EAX; BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY), - X86::RAX).addReg(Reg); - RetRegs.push_back(X86::RAX); + RetReg).addReg(Reg); + RetRegs.push_back(RetReg); } // Now emit the RET. @@ -1233,6 +1232,124 @@ bool X86FastISel::X86SelectShift(const Instruction *I) { return true; } +bool X86FastISel::X86SelectDivRem(const Instruction *I) { + const static unsigned NumTypes = 4; // i8, i16, i32, i64 + const static unsigned NumOps = 4; // SDiv, SRem, UDiv, URem + const static bool S = true; // IsSigned + const static bool U = false; // !IsSigned + const static unsigned Copy = TargetOpcode::COPY; + // For the X86 DIV/IDIV instruction, in most cases the dividend + // (numerator) must be in a specific register pair highreg:lowreg, + // producing the quotient in lowreg and the remainder in highreg. + // For most data types, to set up the instruction, the dividend is + // copied into lowreg, and lowreg is sign-extended or zero-extended + // into highreg. The exception is i8, where the dividend is defined + // as a single register rather than a register pair, and we + // therefore directly sign-extend or zero-extend the dividend into + // lowreg, instead of copying, and ignore the highreg. + const static struct DivRemEntry { + // The following portion depends only on the data type. + const TargetRegisterClass *RC; + unsigned LowInReg; // low part of the register pair + unsigned HighInReg; // high part of the register pair + // The following portion depends on both the data type and the operation. + struct DivRemResult { + unsigned OpDivRem; // The specific DIV/IDIV opcode to use. + unsigned OpSignExtend; // Opcode for sign-extending lowreg into + // highreg, or copying a zero into highreg. + unsigned OpCopy; // Opcode for copying dividend into lowreg, or + // zero/sign-extending into lowreg for i8. + unsigned DivRemResultReg; // Register containing the desired result. + bool IsOpSigned; // Whether to use signed or unsigned form. + } ResultTable[NumOps]; + } OpTable[NumTypes] = { + { &X86::GR8RegClass, X86::AX, 0, { + { X86::IDIV8r, 0, X86::MOVSX16rr8, X86::AL, S }, // SDiv + { X86::IDIV8r, 0, X86::MOVSX16rr8, X86::AH, S }, // SRem + { X86::DIV8r, 0, X86::MOVZX16rr8, X86::AL, U }, // UDiv + { X86::DIV8r, 0, X86::MOVZX16rr8, X86::AH, U }, // URem + } + }, // i8 + { &X86::GR16RegClass, X86::AX, X86::DX, { + { X86::IDIV16r, X86::CWD, Copy, X86::AX, S }, // SDiv + { X86::IDIV16r, X86::CWD, Copy, X86::DX, S }, // SRem + { X86::DIV16r, X86::MOV16r0, Copy, X86::AX, U }, // UDiv + { X86::DIV16r, X86::MOV16r0, Copy, X86::DX, U }, // URem + } + }, // i16 + { &X86::GR32RegClass, X86::EAX, X86::EDX, { + { X86::IDIV32r, X86::CDQ, Copy, X86::EAX, S }, // SDiv + { X86::IDIV32r, X86::CDQ, Copy, X86::EDX, S }, // SRem + { X86::DIV32r, X86::MOV32r0, Copy, X86::EAX, U }, // UDiv + { X86::DIV32r, X86::MOV32r0, Copy, X86::EDX, U }, // URem + } + }, // i32 + { &X86::GR64RegClass, X86::RAX, X86::RDX, { + { X86::IDIV64r, X86::CQO, Copy, X86::RAX, S }, // SDiv + { X86::IDIV64r, X86::CQO, Copy, X86::RDX, S }, // SRem + { X86::DIV64r, X86::MOV64r0, Copy, X86::RAX, U }, // UDiv + { X86::DIV64r, X86::MOV64r0, Copy, X86::RDX, U }, // URem + } + }, // i64 + }; + + MVT VT; + if (!isTypeLegal(I->getType(), VT)) + return false; + + unsigned TypeIndex, OpIndex; + switch (VT.SimpleTy) { + default: return false; + case MVT::i8: TypeIndex = 0; break; + case MVT::i16: TypeIndex = 1; break; + case MVT::i32: TypeIndex = 2; break; + case MVT::i64: TypeIndex = 3; + if (!Subtarget->is64Bit()) + return false; + break; + } + + switch (I->getOpcode()) { + default: llvm_unreachable("Unexpected div/rem opcode"); + case Instruction::SDiv: OpIndex = 0; break; + case Instruction::SRem: OpIndex = 1; break; + case Instruction::UDiv: OpIndex = 2; break; + case Instruction::URem: OpIndex = 3; break; + } + + const DivRemEntry &TypeEntry = OpTable[TypeIndex]; + const DivRemEntry::DivRemResult &OpEntry = TypeEntry.ResultTable[OpIndex]; + unsigned Op0Reg = getRegForValue(I->getOperand(0)); + if (Op0Reg == 0) + return false; + unsigned Op1Reg = getRegForValue(I->getOperand(1)); + if (Op1Reg == 0) + return false; + + // Move op0 into low-order input register. + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(OpEntry.OpCopy), TypeEntry.LowInReg).addReg(Op0Reg); + // Zero-extend or sign-extend into high-order input register. + if (OpEntry.OpSignExtend) { + if (OpEntry.IsOpSigned) + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(OpEntry.OpSignExtend)); + else + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(OpEntry.OpSignExtend), TypeEntry.HighInReg); + } + // Generate the DIV/IDIV instruction. + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(OpEntry.OpDivRem)).addReg(Op1Reg); + // Copy output register into result register. + unsigned ResultReg = createResultReg(TypeEntry.RC); + BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, + TII.get(Copy), ResultReg).addReg(OpEntry.DivRemResultReg); + UpdateValueMap(I, ResultReg); + + return true; +} + bool X86FastISel::X86SelectSelect(const Instruction *I) { MVT VT; if (!isTypeLegal(I->getType(), VT)) @@ -1526,7 +1643,7 @@ bool X86FastISel::FastLowerArguments() { if (!FuncInfo.CanLowerReturn) return false; - if (Subtarget->isTargetWindows()) + if (Subtarget->isTargetWin64()) return false; const Function *F = FuncInfo.Fn; @@ -2082,6 +2199,11 @@ X86FastISel::TargetSelectInstruction(const Instruction *I) { case Instruction::AShr: case Instruction::Shl: return X86SelectShift(I); + case Instruction::SDiv: + case Instruction::UDiv: + case Instruction::SRem: + case Instruction::URem: + return X86SelectDivRem(I); case Instruction::Select: return X86SelectSelect(I); case Instruction::Trunc: @@ -2273,12 +2395,8 @@ unsigned X86FastISel::TargetMaterializeFloatZero(const ConstantFP *CF) { } -/// TryToFoldLoad - The specified machine instr operand is a vreg, and that -/// vreg is being provided by the specified load instruction. If possible, -/// try to fold the load as an operand to the instruction, returning true if -/// possible. -bool X86FastISel::TryToFoldLoad(MachineInstr *MI, unsigned OpNo, - const LoadInst *LI) { +bool X86FastISel::tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo, + const LoadInst *LI) { X86AddressMode AM; if (!X86SelectAddress(LI->getOperand(0), AM)) return false; diff --git a/lib/Target/X86/X86FixupLEAs.cpp b/lib/Target/X86/X86FixupLEAs.cpp new file mode 100644 index 0000000..0dd034c --- /dev/null +++ b/lib/Target/X86/X86FixupLEAs.cpp @@ -0,0 +1,253 @@ +//===-- X86FixupLEAs.cpp - use or replace LEA instructions -----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines the pass which will find instructions which +// can be re-written as LEA instructions in order to reduce pipeline +// delays for some models of the Intel Atom family. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "x86-fixup-LEAs" +#include "X86.h" +#include "X86InstrInfo.h" +#include "X86Subtarget.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/CodeGen/LiveVariables.h" +#include "llvm/CodeGen/MachineFunctionPass.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/Passes.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetInstrInfo.h" +using namespace llvm; + +STATISTIC(NumLEAs, "Number of LEA instructions created"); + +namespace { + class FixupLEAPass : public MachineFunctionPass { + enum RegUsageState { RU_NotUsed, RU_Write, RU_Read }; + static char ID; + /// \brief Loop over all of the instructions in the basic block + /// replacing applicable instructions with LEA instructions, + /// where appropriate. + bool processBasicBlock(MachineFunction &MF, MachineFunction::iterator MFI); + + virtual const char *getPassName() const { return "X86 Atom LEA Fixup";} + + /// \brief Given a machine register, look for the instruction + /// which writes it in the current basic block. If found, + /// try to replace it with an equivalent LEA instruction. + /// If replacement succeeds, then also process the the newly created + /// instruction. + void seekLEAFixup(MachineOperand& p, MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI); + + /// \brief Given a memory access or LEA instruction + /// whose address mode uses a base and/or index register, look for + /// an opportunity to replace the instruction which sets the base or index + /// register with an equivalent LEA instruction. + void processInstruction(MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI); + + /// \brief Determine if an instruction references a machine register + /// and, if so, whether it reads or writes the register. + RegUsageState usesRegister(MachineOperand& p, + MachineBasicBlock::iterator I); + + /// \brief Step backwards through a basic block, looking + /// for an instruction which writes a register within + /// a maximum of INSTR_DISTANCE_THRESHOLD instruction latency cycles. + MachineBasicBlock::iterator searchBackwards(MachineOperand& p, + MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI); + + /// \brief if an instruction can be converted to an + /// equivalent LEA, insert the new instruction into the basic block + /// and return a pointer to it. Otherwise, return zero. + MachineInstr* postRAConvertToLEA(MachineFunction::iterator &MFI, + MachineBasicBlock::iterator &MBBI) const; + + public: + FixupLEAPass() : MachineFunctionPass(ID) {} + + /// \brief Loop over all of the basic blocks, + /// replacing instructions by equivalent LEA instructions + /// if needed and when possible. + virtual bool runOnMachineFunction(MachineFunction &MF); + + private: + MachineFunction *MF; + const TargetMachine *TM; + const TargetInstrInfo *TII; // Machine instruction info. + + }; + char FixupLEAPass::ID = 0; +} + +MachineInstr * +FixupLEAPass::postRAConvertToLEA(MachineFunction::iterator &MFI, + MachineBasicBlock::iterator &MBBI) const { + MachineInstr* MI = MBBI; + MachineInstr* NewMI; + switch (MI->getOpcode()) { + case X86::MOV32rr: + case X86::MOV64rr: { + const MachineOperand& Src = MI->getOperand(1); + const MachineOperand& Dest = MI->getOperand(0); + NewMI = BuildMI(*MF, MI->getDebugLoc(), + TII->get( MI->getOpcode() == X86::MOV32rr ? X86::LEA32r : X86::LEA64r)) + .addOperand(Dest) + .addOperand(Src).addImm(1).addReg(0).addImm(0).addReg(0); + MFI->insert(MBBI, NewMI); // Insert the new inst + return NewMI; + } + case X86::ADD64ri32: + case X86::ADD64ri8: + case X86::ADD64ri32_DB: + case X86::ADD64ri8_DB: + case X86::ADD32ri: + case X86::ADD32ri8: + case X86::ADD32ri_DB: + case X86::ADD32ri8_DB: + case X86::ADD16ri: + case X86::ADD16ri8: + case X86::ADD16ri_DB: + case X86::ADD16ri8_DB: + if (!MI->getOperand(2).isImm()) { + // convertToThreeAddress will call getImm() + // which requires isImm() to be true + return 0; + } + } + return TII->convertToThreeAddress(MFI, MBBI, 0); +} + +FunctionPass *llvm::createX86FixupLEAs() { + return new FixupLEAPass(); +} + +bool FixupLEAPass::runOnMachineFunction(MachineFunction &Func) { + MF = &Func; + TII = Func.getTarget().getInstrInfo(); + TM = &MF->getTarget(); + + DEBUG(dbgs() << "Start X86FixupLEAs\n";); + // Process all basic blocks. + for (MachineFunction::iterator I = Func.begin(), E = Func.end(); I != E; ++I) + processBasicBlock(Func, I); + DEBUG(dbgs() << "End X86FixupLEAs\n";); + + return true; +} + +FixupLEAPass::RegUsageState FixupLEAPass::usesRegister(MachineOperand& p, + MachineBasicBlock::iterator I) { + RegUsageState RegUsage = RU_NotUsed; + MachineInstr* MI = I; + + for (unsigned int i = 0; i < MI->getNumOperands(); ++i) { + MachineOperand& opnd = MI->getOperand(i); + if (opnd.isReg() && opnd.getReg() == p.getReg()){ + if (opnd.isDef()) + return RU_Write; + RegUsage = RU_Read; + } + } + return RegUsage; +} + +/// getPreviousInstr - Given a reference to an instruction in a basic +/// block, return a reference to the previous instruction in the block, +/// wrapping around to the last instruction of the block if the block +/// branches to itself. +static inline bool getPreviousInstr(MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI) { + if (I == MFI->begin()) { + if (MFI->isPredecessor(MFI)) { + I = --MFI->end(); + return true; + } + else + return false; + } + --I; + return true; +} + +MachineBasicBlock::iterator FixupLEAPass::searchBackwards(MachineOperand& p, + MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI) { + int InstrDistance = 1; + MachineBasicBlock::iterator CurInst; + static const int INSTR_DISTANCE_THRESHOLD = 5; + + CurInst = I; + bool Found; + Found = getPreviousInstr(CurInst, MFI); + while( Found && I != CurInst) { + if (CurInst->isCall() || CurInst->isInlineAsm()) + break; + if (InstrDistance > INSTR_DISTANCE_THRESHOLD) + break; // too far back to make a difference + if (usesRegister(p, CurInst) == RU_Write){ + return CurInst; + } + InstrDistance += TII->getInstrLatency(TM->getInstrItineraryData(), CurInst); + Found = getPreviousInstr(CurInst, MFI); + } + return 0; +} + +void FixupLEAPass::processInstruction(MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI) { + // Process a load, store, or LEA instruction. + MachineInstr *MI = I; + int opcode = MI->getOpcode(); + const MCInstrDesc& Desc = MI->getDesc(); + int AddrOffset = X86II::getMemoryOperandNo(Desc.TSFlags, opcode); + if (AddrOffset >= 0) { + AddrOffset += X86II::getOperandBias(Desc); + MachineOperand& p = MI->getOperand(AddrOffset + X86::AddrBaseReg); + if (p.isReg() && p.getReg() != X86::ESP) { + seekLEAFixup(p, I, MFI); + } + MachineOperand& q = MI->getOperand(AddrOffset + X86::AddrIndexReg); + if (q.isReg() && q.getReg() != X86::ESP) { + seekLEAFixup(q, I, MFI); + } + } +} + +void FixupLEAPass::seekLEAFixup(MachineOperand& p, + MachineBasicBlock::iterator& I, + MachineFunction::iterator MFI) { + MachineBasicBlock::iterator MBI = searchBackwards(p, I, MFI); + if (MBI) { + MachineInstr* NewMI = postRAConvertToLEA(MFI, MBI); + if (NewMI) { + ++NumLEAs; + DEBUG(dbgs() << "Candidate to replace:"; MBI->dump();); + // now to replace with an equivalent LEA... + DEBUG(dbgs() << "Replaced by: "; NewMI->dump();); + MFI->erase(MBI); + MachineBasicBlock::iterator J = + static_cast<MachineBasicBlock::iterator> (NewMI); + processInstruction(J, MFI); + } + } +} + +bool FixupLEAPass::processBasicBlock(MachineFunction &MF, + MachineFunction::iterator MFI) { + + for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I) + processInstruction(I, MFI); + return false; +} diff --git a/lib/Target/X86/X86FrameLowering.cpp b/lib/Target/X86/X86FrameLowering.cpp index 54cbd40..16e1e42 100644 --- a/lib/Target/X86/X86FrameLowering.cpp +++ b/lib/Target/X86/X86FrameLowering.cpp @@ -528,11 +528,11 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { if (!MI.getFlag(MachineInstr::FrameSetup)) break; // We don't exect any more prolog instructions. - if (ExpectEnd) return 0; + if (ExpectEnd) return CU::UNWIND_MODE_DWARF; if (Opc == PushInstr) { // If there are too many saved registers, we cannot use compact encoding. - if (SavedRegIdx >= CU_NUM_SAVED_REGS) return 0; + if (SavedRegIdx >= CU_NUM_SAVED_REGS) return CU::UNWIND_MODE_DWARF; SavedRegs[SavedRegIdx++] = MI.getOperand(0).getReg(); StackAdjust += OffsetSize; @@ -542,7 +542,7 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { unsigned DstReg = MI.getOperand(0).getReg(); if (DstReg != FramePtr || SrcReg != StackPtr) - return 0; + return CU::UNWIND_MODE_DWARF; StackAdjust = 0; memset(SavedRegs, 0, sizeof(SavedRegs)); @@ -552,7 +552,7 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { Opc == X86::SUB32ri || Opc == X86::SUB32ri8) { if (StackSize) // We already have a stack size. - return 0; + return CU::UNWIND_MODE_DWARF; if (!MI.getOperand(0).isReg() || MI.getOperand(0).getReg() != MI.getOperand(1).getReg() || @@ -560,7 +560,7 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { // We need this to be a stack adjustment pointer. Something like: // // %RSP<def> = SUB64ri8 %RSP, 48 - return 0; + return CU::UNWIND_MODE_DWARF; StackSize = MI.getOperand(2).getImm() / StackDivide; SubtractInstrIdx += InstrOffset; @@ -574,31 +574,31 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { if (HasFP) { if ((StackAdjust & 0xFF) != StackAdjust) // Offset was too big for compact encoding. - return 0; + return CU::UNWIND_MODE_DWARF; // Get the encoding of the saved registers when we have a frame pointer. uint32_t RegEnc = encodeCompactUnwindRegistersWithFrame(SavedRegs, Is64Bit); - if (RegEnc == ~0U) return 0; + if (RegEnc == ~0U) return CU::UNWIND_MODE_DWARF; - CompactUnwindEncoding |= 0x01000000; + CompactUnwindEncoding |= CU::UNWIND_MODE_BP_FRAME; CompactUnwindEncoding |= (StackAdjust & 0xFF) << 16; - CompactUnwindEncoding |= RegEnc & 0x7FFF; + CompactUnwindEncoding |= RegEnc & CU::UNWIND_BP_FRAME_REGISTERS; } else { ++StackAdjust; uint32_t TotalStackSize = StackAdjust + StackSize; if ((TotalStackSize & 0xFF) == TotalStackSize) { // Frameless stack with a small stack size. - CompactUnwindEncoding |= 0x02000000; + CompactUnwindEncoding |= CU::UNWIND_MODE_STACK_IMMD; // Encode the stack size. CompactUnwindEncoding |= (TotalStackSize & 0xFF) << 16; } else { if ((StackAdjust & 0x7) != StackAdjust) // The extra stack adjustments are too big for us to handle. - return 0; + return CU::UNWIND_MODE_DWARF; // Frameless stack with an offset too large for us to encode compactly. - CompactUnwindEncoding |= 0x03000000; + CompactUnwindEncoding |= CU::UNWIND_MODE_STACK_IND; // Encode the offset to the nnnnnn value in the 'subl $nnnnnn, ESP' // instruction. @@ -616,10 +616,11 @@ uint32_t X86FrameLowering::getCompactUnwindEncoding(MachineFunction &MF) const { uint32_t RegEnc = encodeCompactUnwindRegistersWithoutFrame(SavedRegs, SavedRegIdx, Is64Bit); - if (RegEnc == ~0U) return 0; + if (RegEnc == ~0U) return CU::UNWIND_MODE_DWARF; // Encode the register encoding. - CompactUnwindEncoding |= RegEnc & 0x3FF; + CompactUnwindEncoding |= + RegEnc & CU::UNWIND_FRAMELESS_STACK_REG_PERMUTATION; } return CompactUnwindEncoding; diff --git a/lib/Target/X86/X86FrameLowering.h b/lib/Target/X86/X86FrameLowering.h index 3f08b9a..6e309d8 100644 --- a/lib/Target/X86/X86FrameLowering.h +++ b/lib/Target/X86/X86FrameLowering.h @@ -19,8 +19,35 @@ #include "llvm/Target/TargetFrameLowering.h" namespace llvm { - class MCSymbol; - class X86TargetMachine; + +namespace CU { + + /// Compact unwind encoding values. + enum CompactUnwindEncodings { + /// [RE]BP based frame where [RE]BP is pused on the stack immediately after + /// the return address, then [RE]SP is moved to [RE]BP. + UNWIND_MODE_BP_FRAME = 0x01000000, + + /// A frameless function with a small constant stack size. + UNWIND_MODE_STACK_IMMD = 0x02000000, + + /// A frameless function with a large constant stack size. + UNWIND_MODE_STACK_IND = 0x03000000, + + /// No compact unwind encoding is available. + UNWIND_MODE_DWARF = 0x04000000, + + /// Mask for encoding the frame registers. + UNWIND_BP_FRAME_REGISTERS = 0x00007FFF, + + /// Mask for encoding the frameless registers. + UNWIND_FRAMELESS_STACK_REG_PERMUTATION = 0x000003FF + }; + +} // end CU namespace + +class MCSymbol; +class X86TargetMachine; class X86FrameLowering : public TargetFrameLowering { const X86TargetMachine &TM; diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp index 00fbe69..968b358 100644 --- a/lib/Target/X86/X86ISelDAGToDAG.cpp +++ b/lib/Target/X86/X86ISelDAGToDAG.cpp @@ -444,7 +444,9 @@ void X86DAGToDAGISel::PreprocessISelDAG() { SDNode *N = I++; // Preincrement iterator to avoid invalidation issues. if (OptLevel != CodeGenOpt::None && - (N->getOpcode() == X86ISD::CALL || + // Only does this when target favors doesn't favor register indirect + // call. + ((N->getOpcode() == X86ISD::CALL && !Subtarget->callRegIndirect()) || (N->getOpcode() == X86ISD::TC_RETURN && // Only does this if load can be folded into TC_RETURN. (Subtarget->is64Bit() || @@ -1501,8 +1503,7 @@ SDNode *X86DAGToDAGISel::SelectAtomic64(SDNode *Node, unsigned Opc) { MemOp[0] = cast<MemSDNode>(Node)->getMemOperand(); const SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, In2L, In2H, Chain}; SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(), - MVT::i32, MVT::i32, MVT::Other, Ops, - array_lengthof(Ops)); + MVT::i32, MVT::i32, MVT::Other, Ops); cast<MachineSDNode>(ResNode)->setMemRefs(MemOp, MemOp + 1); return ResNode; } @@ -1718,7 +1719,7 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadArith(SDNode *Node, EVT NVT) { Op = ADD; break; } - + Val = getAtomicLoadArithTargetConstant(CurDAG, dl, Op, NVT, Val); bool isUnOp = !Val.getNode(); bool isCN = Val.getNode() && (Val.getOpcode() == ISD::TargetConstant); @@ -1770,12 +1771,10 @@ SDNode *X86DAGToDAGISel::SelectAtomicLoadArith(SDNode *Node, EVT NVT) { MemOp[0] = cast<MemSDNode>(Node)->getMemOperand(); if (isUnOp) { SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Chain }; - Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, - array_lengthof(Ops)), 0); + Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops), 0); } else { SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Val, Chain }; - Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, - array_lengthof(Ops)), 0); + Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops), 0); } cast<MachineSDNode>(Ret)->setMemRefs(MemOp, MemOp + 1); SDValue RetVals[] = { Undef, Ret }; @@ -1969,8 +1968,7 @@ SDNode *X86DAGToDAGISel::SelectGather(SDNode *Node, unsigned Opc) { SDValue Segment = CurDAG->getRegister(0, MVT::i32); const SDValue Ops[] = { VSrc, Base, getI8Imm(Scale->getSExtValue()), VIdx, Disp, Segment, VMask, Chain}; - SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(), - VTs, Ops, array_lengthof(Ops)); + SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(), VTs, Ops); // Node has 2 outputs: VDst and MVT::Other. // ResNode has 3 outputs: VDst, VMask_wb, and MVT::Other. // We replace VDst of Node with VDst of ResNode, and Other of Node with Other @@ -2184,7 +2182,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { SDVTList VTs = CurDAG->getVTList(NVT, NVT, MVT::i32); SDValue Ops[] = {N1, InFlag}; - SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops, 2); + SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops); ReplaceUses(SDValue(Node, 0), SDValue(CNode, 0)); ReplaceUses(SDValue(Node, 1), SDValue(CNode, 1)); @@ -2265,16 +2263,14 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { InFlag }; if (MOpc == X86::MULX32rm || MOpc == X86::MULX64rm) { SDVTList VTs = CurDAG->getVTList(NVT, NVT, MVT::Other, MVT::Glue); - SDNode *CNode = CurDAG->getMachineNode(MOpc, dl, VTs, Ops, - array_lengthof(Ops)); + SDNode *CNode = CurDAG->getMachineNode(MOpc, dl, VTs, Ops); ResHi = SDValue(CNode, 0); ResLo = SDValue(CNode, 1); Chain = SDValue(CNode, 2); InFlag = SDValue(CNode, 3); } else { SDVTList VTs = CurDAG->getVTList(MVT::Other, MVT::Glue); - SDNode *CNode = CurDAG->getMachineNode(MOpc, dl, VTs, Ops, - array_lengthof(Ops)); + SDNode *CNode = CurDAG->getMachineNode(MOpc, dl, VTs, Ops); Chain = SDValue(CNode, 0); InFlag = SDValue(CNode, 1); } @@ -2285,15 +2281,13 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { SDValue Ops[] = { N1, InFlag }; if (Opc == X86::MULX32rr || Opc == X86::MULX64rr) { SDVTList VTs = CurDAG->getVTList(NVT, NVT, MVT::Glue); - SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops, - array_lengthof(Ops)); + SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops); ResHi = SDValue(CNode, 0); ResLo = SDValue(CNode, 1); InFlag = SDValue(CNode, 2); } else { SDVTList VTs = CurDAG->getVTList(MVT::Glue); - SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops, - array_lengthof(Ops)); + SDNode *CNode = CurDAG->getMachineNode(Opc, dl, VTs, Ops); InFlag = SDValue(CNode, 0); } } @@ -2341,6 +2335,9 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { DEBUG(dbgs() << "=> "; ResHi.getNode()->dump(CurDAG); dbgs() << '\n'); } + // Propagate ordering to the last node, for now. + CurDAG->AssignOrdering(InFlag.getNode(), CurDAG->GetOrdering(Node)); + return NULL; } @@ -2407,8 +2404,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, N0.getOperand(0) }; Move = SDValue(CurDAG->getMachineNode(X86::MOVZX32rm8, dl, MVT::i32, - MVT::Other, Ops, - array_lengthof(Ops)), 0); + MVT::Other, Ops), 0); Chain = Move.getValue(1); ReplaceUses(N0.getValue(1), Chain); } else { @@ -2439,8 +2435,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, N1.getOperand(0), InFlag }; SDNode *CNode = - CurDAG->getMachineNode(MOpc, dl, MVT::Other, MVT::Glue, Ops, - array_lengthof(Ops)); + CurDAG->getMachineNode(MOpc, dl, MVT::Other, MVT::Glue, Ops); InFlag = SDValue(CNode, 1); // Update the chain. ReplaceUses(N1.getValue(1), SDValue(CNode, 0)); @@ -2672,8 +2667,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) { unsigned newOpc = getFusedLdStOpcode(LdVT, Opc); MachineSDNode *Result = CurDAG->getMachineNode(newOpc, Node->getDebugLoc(), - MVT::i32, MVT::Other, Ops, - array_lengthof(Ops)); + MVT::i32, MVT::Other, Ops); Result->setMemRefs(MemOp, MemOp + 2); ReplaceUses(SDValue(StoreNode, 0), SDValue(Result, 1)); diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index e6858bc..b587336 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -163,10 +163,28 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) Subtarget = &TM.getSubtarget<X86Subtarget>(); X86ScalarSSEf64 = Subtarget->hasSSE2(); X86ScalarSSEf32 = Subtarget->hasSSE1(); - RegInfo = TM.getRegisterInfo(); TD = getDataLayout(); + resetOperationActions(); +} + +void X86TargetLowering::resetOperationActions() { + const TargetMachine &TM = getTargetMachine(); + static bool FirstTimeThrough = true; + + // If none of the target options have changed, then we don't need to reset the + // operation actions. + if (!FirstTimeThrough && TO == TM.Options) return; + + if (!FirstTimeThrough) { + // Reinitialize the actions. + initActions(); + FirstTimeThrough = false; + } + + TO = TM.Options; + // Set up the TargetLowering object. static const MVT IntVTs[] = { MVT::i8, MVT::i16, MVT::i32, MVT::i64 }; @@ -470,7 +488,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::SETCC , MVT::i64 , Custom); } setOperationAction(ISD::EH_RETURN , MVT::Other, Custom); - // NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intened to support + // NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intended to support // SjLj exception handling but a light-weight setjmp/longjmp replacement to // support continuation, user-level threading, and etc.. As a result, no // other SjLj exception interfaces are implemented and please don't build @@ -508,16 +526,8 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) if (Subtarget->hasSSE1()) setOperationAction(ISD::PREFETCH , MVT::Other, Legal); - setOperationAction(ISD::MEMBARRIER , MVT::Other, Custom); setOperationAction(ISD::ATOMIC_FENCE , MVT::Other, Custom); - // On X86 and X86-64, atomic operations are lowered to locked instructions. - // Locked instructions, in turn, have implicit fence semantics (all memory - // operations are flushed before issuing the locked instruction, and they - // are not buffered), so we can fold away the common pattern of - // fence-atomic-fence. - setShouldFoldAtomicFences(true); - // Expand certain atomics for (unsigned i = 0; i != array_lengthof(IntVTs); ++i) { MVT VT = IntVTs[i]; @@ -1053,23 +1063,16 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::SRA, MVT::v8i16, Custom); setOperationAction(ISD::SRA, MVT::v16i8, Custom); - if (Subtarget->hasInt256()) { - setOperationAction(ISD::SRL, MVT::v2i64, Legal); - setOperationAction(ISD::SRL, MVT::v4i32, Legal); - - setOperationAction(ISD::SHL, MVT::v2i64, Legal); - setOperationAction(ISD::SHL, MVT::v4i32, Legal); + // In the customized shift lowering, the legal cases in AVX2 will be + // recognized. + setOperationAction(ISD::SRL, MVT::v2i64, Custom); + setOperationAction(ISD::SRL, MVT::v4i32, Custom); - setOperationAction(ISD::SRA, MVT::v4i32, Legal); - } else { - setOperationAction(ISD::SRL, MVT::v2i64, Custom); - setOperationAction(ISD::SRL, MVT::v4i32, Custom); + setOperationAction(ISD::SHL, MVT::v2i64, Custom); + setOperationAction(ISD::SHL, MVT::v4i32, Custom); - setOperationAction(ISD::SHL, MVT::v2i64, Custom); - setOperationAction(ISD::SHL, MVT::v4i32, Custom); + setOperationAction(ISD::SRA, MVT::v4i32, Custom); - setOperationAction(ISD::SRA, MVT::v4i32, Custom); - } setOperationAction(ISD::SDIV, MVT::v8i16, Custom); setOperationAction(ISD::SDIV, MVT::v4i32, Custom); } @@ -1118,6 +1121,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::FP_TO_SINT, MVT::v8i16, Custom); setOperationAction(ISD::FP_TO_SINT, MVT::v8i32, Legal); + setOperationAction(ISD::SINT_TO_FP, MVT::v8i16, Promote); setOperationAction(ISD::SINT_TO_FP, MVT::v8i32, Legal); setOperationAction(ISD::FP_ROUND, MVT::v4f32, Legal); @@ -1186,14 +1190,6 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::VSELECT, MVT::v32i8, Legal); - setOperationAction(ISD::SRL, MVT::v4i64, Legal); - setOperationAction(ISD::SRL, MVT::v8i32, Legal); - - setOperationAction(ISD::SHL, MVT::v4i64, Legal); - setOperationAction(ISD::SHL, MVT::v8i32, Legal); - - setOperationAction(ISD::SRA, MVT::v8i32, Legal); - setOperationAction(ISD::SDIV, MVT::v8i32, Custom); } else { setOperationAction(ISD::ADD, MVT::v4i64, Custom); @@ -1210,15 +1206,17 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) setOperationAction(ISD::MUL, MVT::v8i32, Custom); setOperationAction(ISD::MUL, MVT::v16i16, Custom); // Don't lower v32i8 because there is no 128-bit byte mul + } - setOperationAction(ISD::SRL, MVT::v4i64, Custom); - setOperationAction(ISD::SRL, MVT::v8i32, Custom); + // In the customized shift lowering, the legal cases in AVX2 will be + // recognized. + setOperationAction(ISD::SRL, MVT::v4i64, Custom); + setOperationAction(ISD::SRL, MVT::v8i32, Custom); - setOperationAction(ISD::SHL, MVT::v4i64, Custom); - setOperationAction(ISD::SHL, MVT::v8i32, Custom); + setOperationAction(ISD::SHL, MVT::v4i64, Custom); + setOperationAction(ISD::SHL, MVT::v8i32, Custom); - setOperationAction(ISD::SRA, MVT::v8i32, Custom); - } + setOperationAction(ISD::SRA, MVT::v8i32, Custom); // Custom lower several nodes for 256-bit types. for (int i = MVT::FIRST_VECTOR_VALUETYPE; @@ -1356,7 +1354,6 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM) MaxStoresPerMemmove = 8; // For @llvm.memmove -> sequence of stores MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 8 : 4; setPrefLoopAlignment(4); // 2^4 bytes. - BenefitFromCodePlacementOpt = true; // Predictable cmov don't hurt on atom because it's in-order. PredictableSelectIsExpensive = !Subtarget->isAtom(); @@ -1679,10 +1676,11 @@ X86TargetLowering::LowerReturn(SDValue Chain, // The x86-64 ABIs require that for returning structs by value we copy // the sret argument into %rax/%eax (depending on ABI) for the return. + // Win32 requires us to put the sret argument to %eax as well. // We saved the argument into a virtual register in the entry block, // so now we copy the value out and into %rax/%eax. - if (Subtarget->is64Bit() && - DAG.getMachineFunction().getFunction()->hasStructRetAttr()) { + if (DAG.getMachineFunction().getFunction()->hasStructRetAttr() && + (Subtarget->is64Bit() || Subtarget->isTargetWindows())) { MachineFunction &MF = DAG.getMachineFunction(); X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>(); unsigned Reg = FuncInfo->getSRetReturnReg(); @@ -1690,12 +1688,14 @@ X86TargetLowering::LowerReturn(SDValue Chain, "SRetReturnReg should have been set in LowerFormalArguments()."); SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy()); - unsigned RetValReg = Subtarget->isTarget64BitILP32() ? X86::EAX : X86::RAX; + unsigned RetValReg + = (Subtarget->is64Bit() && !Subtarget->isTarget64BitILP32()) ? + X86::RAX : X86::EAX; Chain = DAG.getCopyToReg(Chain, dl, RetValReg, Val, Flag); Flag = Chain.getValue(1); // RAX/EAX now acts like a return value. - RetOps.push_back(DAG.getRegister(RetValReg, MVT::i64)); + RetOps.push_back(DAG.getRegister(RetValReg, getPointerTy())); } RetOps[0] = Chain; // Update chain. @@ -1795,7 +1795,7 @@ X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag, if (isScalarFPTypeInSSEReg(VA.getValVT())) CopyVT = MVT::f80; SDValue Ops[] = { Chain, InFlag }; Chain = SDValue(DAG.getMachineNode(X86::FpPOP_RETVAL, dl, CopyVT, - MVT::Other, MVT::Glue, Ops, 2), 1); + MVT::Other, MVT::Glue, Ops), 1); Val = Chain.getValue(0); // Round the f80 to the right size, which also moves it to the appropriate @@ -2049,9 +2049,11 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain, // The x86-64 ABIs require that for returning structs by value we copy // the sret argument into %rax/%eax (depending on ABI) for the return. + // Win32 requires us to put the sret argument to %eax as well. // Save the argument into a virtual register so that we can access it // from the return points. - if (Is64Bit && MF.getFunction()->hasStructRetAttr()) { + if (MF.getFunction()->hasStructRetAttr() && + (Subtarget->is64Bit() || Subtarget->isTargetWindows())) { X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>(); unsigned Reg = FuncInfo->getSRetReturnReg(); if (!Reg) { @@ -4412,13 +4414,15 @@ static SDValue getZeroVector(EVT VT, const X86Subtarget *Subtarget, if (Subtarget->hasInt256()) { // AVX2 SDValue Cst = DAG.getTargetConstant(0, MVT::i32); SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst }; - Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8); + Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, + array_lengthof(Ops)); } else { // 256-bit logic and arithmetic instructions in AVX are all // floating-point, no support for integer ops. Emit fp zeroed vectors. SDValue Cst = DAG.getTargetConstantFP(+0.0, MVT::f32); SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst }; - Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8f32, Ops, 8); + Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8f32, Ops, + array_lengthof(Ops)); } } else llvm_unreachable("Unexpected vector type"); @@ -4439,7 +4443,8 @@ static SDValue getOnesVector(MVT VT, bool HasInt256, SelectionDAG &DAG, if (VT.is256BitVector()) { if (HasInt256) { // AVX2 SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst }; - Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, 8); + Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops, + array_lengthof(Ops)); } else { // AVX Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst); Vec = Concat128BitVectors(Vec, Vec, MVT::v8i32, 8, DAG, dl); @@ -5109,7 +5114,8 @@ static SDValue EltsFromConsecutiveLoads(EVT VT, SmallVectorImpl<SDValue> &Elts, SDVTList Tys = DAG.getVTList(MVT::v2i64, MVT::Other); SDValue Ops[] = { LDBase->getChain(), LDBase->getBasePtr() }; SDValue ResNode = - DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, DL, Tys, Ops, 2, MVT::i64, + DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, DL, Tys, Ops, + array_lengthof(Ops), MVT::i64, LDBase->getPointerInfo(), LDBase->getAlignment(), false/*isVolatile*/, true/*ReadMem*/, @@ -7632,10 +7638,10 @@ GetTLSADDR(SelectionDAG &DAG, SDValue Chain, GlobalAddressSDNode *GA, if (InFlag) { SDValue Ops[] = { Chain, TGA, *InFlag }; - Chain = DAG.getNode(CallType, dl, NodeTys, Ops, 3); + Chain = DAG.getNode(CallType, dl, NodeTys, Ops, array_lengthof(Ops)); } else { SDValue Ops[] = { Chain, TGA }; - Chain = DAG.getNode(CallType, dl, NodeTys, Ops, 2); + Chain = DAG.getNode(CallType, dl, NodeTys, Ops, array_lengthof(Ops)); } // TLSADDR will be codegen'ed as call. Inform MFI that function has calls. @@ -7945,7 +7951,7 @@ SDValue X86TargetLowering::LowerShiftParts(SDValue Op, SelectionDAG &DAG) const{ } SDValue Ops[2] = { Lo, Hi }; - return DAG.getMergeValues(Ops, 2, dl); + return DAG.getMergeValues(Ops, array_lengthof(Ops), dl); } SDValue X86TargetLowering::LowerSINT_TO_FP(SDValue Op, @@ -8228,8 +8234,8 @@ SDValue X86TargetLowering::LowerUINT_TO_FP(SDValue Op, SDVTList Tys = DAG.getVTList(MVT::f80, MVT::Other); SDValue Ops[] = { Store, StackSlot, DAG.getValueType(MVT::i64) }; - SDValue Fild = DAG.getMemIntrinsicNode(X86ISD::FILD, dl, Tys, Ops, 3, - MVT::i64, MMO); + SDValue Fild = DAG.getMemIntrinsicNode(X86ISD::FILD, dl, Tys, Ops, + array_lengthof(Ops), MVT::i64, MMO); APInt FF(32, 0x5F800000ULL); @@ -8321,8 +8327,8 @@ X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, MachineMemOperand *MMO = MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI), MachineMemOperand::MOLoad, MemSize, MemSize); - Value = DAG.getMemIntrinsicNode(X86ISD::FLD, DL, Tys, Ops, 3, - DstTy, MMO); + Value = DAG.getMemIntrinsicNode(X86ISD::FLD, DL, Tys, Ops, + array_lengthof(Ops), DstTy, MMO); Chain = Value.getValue(1); SSFI = MF.getFrameInfo()->CreateStackObject(MemSize, MemSize, false); StackSlot = DAG.getFrameIndex(SSFI, getPointerTy()); @@ -8336,7 +8342,8 @@ X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, // Build the FP_TO_INT*_IN_MEM SDValue Ops[] = { Chain, Value, StackSlot }; SDValue FIST = DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::Other), - Ops, 3, DstTy, MMO); + Ops, array_lengthof(Ops), DstTy, + MMO); return std::make_pair(FIST, StackSlot); } else { SDValue ftol = DAG.getNode(X86ISD::WIN_FTOL, DL, @@ -8348,8 +8355,8 @@ X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG, MVT::i32, eax.getValue(2)); SDValue Ops[] = { eax, edx }; SDValue pair = IsReplace - ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops, 2) - : DAG.getMergeValues(Ops, 2, DL); + ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops, array_lengthof(Ops)) + : DAG.getMergeValues(Ops, array_lengthof(Ops), DL); return std::make_pair(pair, SDValue()); } } @@ -9340,11 +9347,49 @@ static SDValue LowerVSETCC(SDValue Op, const X86Subtarget *Subtarget, if (Swap) std::swap(Op0, Op1); + // Since SSE has no unsigned integer comparisons, we need to flip the sign + // bits of the inputs before performing those operations. + if (FlipSigns) { + EVT EltVT = VT.getVectorElementType(); + SDValue SignBit = DAG.getConstant(APInt::getSignBit(EltVT.getSizeInBits()), + EltVT); + std::vector<SDValue> SignBits(VT.getVectorNumElements(), SignBit); + SDValue SignVec = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &SignBits[0], + SignBits.size()); + Op0 = DAG.getNode(ISD::XOR, dl, VT, Op0, SignVec); + Op1 = DAG.getNode(ISD::XOR, dl, VT, Op1, SignVec); + } + // Check that the operation in question is available (most are plain SSE2, // but PCMPGTQ and PCMPEQQ have different requirements). if (VT == MVT::v2i64) { - if (Opc == X86ISD::PCMPGT && !Subtarget->hasSSE42()) - return SDValue(); + if (Opc == X86ISD::PCMPGT && !Subtarget->hasSSE42()) { + assert(Subtarget->hasSSE2() && "Don't know how to lower!"); + + // First cast everything to the right type, + Op0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op0); + Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op1); + + // Emulate PCMPGTQ with (hi1 > hi2) | ((hi1 == hi2) & (lo1 > lo2)) + SDValue GT = DAG.getNode(X86ISD::PCMPGT, dl, MVT::v4i32, Op0, Op1); + SDValue EQ = DAG.getNode(X86ISD::PCMPEQ, dl, MVT::v4i32, Op0, Op1); + + // Create masks for only the low parts/high parts of the 64 bit integers. + const int MaskHi[] = { 1, 1, 3, 3 }; + const int MaskLo[] = { 0, 0, 2, 2 }; + SDValue EQHi = DAG.getVectorShuffle(MVT::v4i32, dl, EQ, EQ, MaskHi); + SDValue GTLo = DAG.getVectorShuffle(MVT::v4i32, dl, GT, GT, MaskLo); + SDValue GTHi = DAG.getVectorShuffle(MVT::v4i32, dl, GT, GT, MaskHi); + + SDValue Result = DAG.getNode(ISD::AND, dl, MVT::v4i32, EQHi, GTLo); + Result = DAG.getNode(ISD::OR, dl, MVT::v4i32, Result, GTHi); + + if (Invert) + Result = DAG.getNOT(dl, Result, MVT::v4i32); + + return DAG.getNode(ISD::BITCAST, dl, VT, Result); + } + if (Opc == X86ISD::PCMPEQ && !Subtarget->hasSSE41()) { // If pcmpeqq is missing but pcmpeqd is available synthesize pcmpeqq with // pcmpeqd + pshufd + pand. @@ -9369,19 +9414,6 @@ static SDValue LowerVSETCC(SDValue Op, const X86Subtarget *Subtarget, } } - // Since SSE has no unsigned integer comparisons, we need to flip the sign - // bits of the inputs before performing those operations. - if (FlipSigns) { - EVT EltVT = VT.getVectorElementType(); - SDValue SignBit = DAG.getConstant(APInt::getSignBit(EltVT.getSizeInBits()), - EltVT); - std::vector<SDValue> SignBits(VT.getVectorNumElements(), SignBit); - SDValue SignVec = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &SignBits[0], - SignBits.size()); - Op0 = DAG.getNode(ISD::XOR, dl, VT, Op0, SignVec); - Op1 = DAG.getNode(ISD::XOR, dl, VT, Op1, SignVec); - } - SDValue Result = DAG.getNode(Opc, dl, VT, Op0, Op1); // If the logical-not of the result is required, perform that now. @@ -10922,28 +10954,47 @@ static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) { switch (IntNo) { default: return SDValue(); // Don't custom lower most intrinsics. - // RDRAND intrinsics. + // RDRAND/RDSEED intrinsics. case Intrinsic::x86_rdrand_16: case Intrinsic::x86_rdrand_32: - case Intrinsic::x86_rdrand_64: { + case Intrinsic::x86_rdrand_64: + case Intrinsic::x86_rdseed_16: + case Intrinsic::x86_rdseed_32: + case Intrinsic::x86_rdseed_64: { + unsigned Opcode = (IntNo == Intrinsic::x86_rdseed_16 || + IntNo == Intrinsic::x86_rdseed_32 || + IntNo == Intrinsic::x86_rdseed_64) ? X86ISD::RDSEED : + X86ISD::RDRAND; // Emit the node with the right value type. SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::Glue, MVT::Other); - SDValue Result = DAG.getNode(X86ISD::RDRAND, dl, VTs, Op.getOperand(0)); + SDValue Result = DAG.getNode(Opcode, dl, VTs, Op.getOperand(0)); - // If the value returned by RDRAND was valid (CF=1), return 1. Otherwise - // return the value from Rand, which is always 0, casted to i32. + // If the value returned by RDRAND/RDSEED was valid (CF=1), return 1. + // Otherwise return the value from Rand, which is always 0, casted to i32. SDValue Ops[] = { DAG.getZExtOrTrunc(Result, dl, Op->getValueType(1)), DAG.getConstant(1, Op->getValueType(1)), DAG.getConstant(X86::COND_B, MVT::i32), SDValue(Result.getNode(), 1) }; SDValue isValid = DAG.getNode(X86ISD::CMOV, dl, DAG.getVTList(Op->getValueType(1), MVT::Glue), - Ops, 4); + Ops, array_lengthof(Ops)); // Return { result, isValid, chain }. return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), Result, isValid, SDValue(Result.getNode(), 2)); } + + // XTEST intrinsics. + case Intrinsic::x86_xtest: { + SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::Other); + SDValue InTrans = DAG.getNode(X86ISD::XTEST, dl, VTs, Op.getOperand(0)); + SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8, + DAG.getConstant(X86::COND_NE, MVT::i8), + InTrans); + SDValue Ret = DAG.getNode(ISD::ZERO_EXTEND, dl, Op->getValueType(0), SetCC); + return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), + Ret, SDValue(InTrans.getNode(), 1)); + } } } @@ -10979,7 +11030,10 @@ SDValue X86TargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const { EVT VT = Op.getValueType(); DebugLoc dl = Op.getDebugLoc(); // FIXME probably not meaningful unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue(); - unsigned FrameReg = Subtarget->is64Bit() ? X86::RBP : X86::EBP; + unsigned FrameReg = RegInfo->getFrameRegister(DAG.getMachineFunction()); + assert(((FrameReg == X86::RBP && VT == MVT::i64) || + (FrameReg == X86::EBP && VT == MVT::i32)) && + "Invalid Frame Register!"); SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, VT); while (Depth--) FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr, @@ -10999,21 +11053,23 @@ SDValue X86TargetLowering::LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const { SDValue Handler = Op.getOperand(2); DebugLoc dl = Op.getDebugLoc(); - SDValue Frame = DAG.getCopyFromReg(DAG.getEntryNode(), dl, - Subtarget->is64Bit() ? X86::RBP : X86::EBP, - getPointerTy()); - unsigned StoreAddrReg = (Subtarget->is64Bit() ? X86::RCX : X86::ECX); - - SDValue StoreAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), Frame, - DAG.getIntPtrConstant(RegInfo->getSlotSize())); - StoreAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), StoreAddr, Offset); + EVT PtrVT = getPointerTy(); + unsigned FrameReg = RegInfo->getFrameRegister(DAG.getMachineFunction()); + assert(((FrameReg == X86::RBP && PtrVT == MVT::i64) || + (FrameReg == X86::EBP && PtrVT == MVT::i32)) && + "Invalid Frame Register!"); + SDValue Frame = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, PtrVT); + unsigned StoreAddrReg = (PtrVT == MVT::i64) ? X86::RCX : X86::ECX; + + SDValue StoreAddr = DAG.getNode(ISD::ADD, dl, PtrVT, Frame, + DAG.getIntPtrConstant(RegInfo->getSlotSize())); + StoreAddr = DAG.getNode(ISD::ADD, dl, PtrVT, StoreAddr, Offset); Chain = DAG.getStore(Chain, dl, Handler, StoreAddr, MachinePointerInfo(), false, false, 0); Chain = DAG.getCopyToReg(Chain, dl, StoreAddrReg, StoreAddr); - return DAG.getNode(X86ISD::EH_RETURN, dl, - MVT::Other, - Chain, DAG.getRegister(StoreAddrReg, getPointerTy())); + return DAG.getNode(X86ISD::EH_RETURN, dl, MVT::Other, Chain, + DAG.getRegister(StoreAddrReg, PtrVT)); } SDValue X86TargetLowering::lowerEH_SJLJ_SETJMP(SDValue Op, @@ -11224,7 +11280,8 @@ SDValue X86TargetLowering::LowerFLT_ROUNDS_(SDValue Op, SDValue Ops[] = { DAG.getEntryNode(), StackSlot }; SDValue Chain = DAG.getMemIntrinsicNode(X86ISD::FNSTCW16m, DL, DAG.getVTList(MVT::Other), - Ops, 2, MVT::i16, MMO); + Ops, array_lengthof(Ops), MVT::i16, + MMO); // Load FP Control Word from stack slot SDValue CWD = DAG.getLoad(MVT::i16, DL, Chain, StackSlot, @@ -11491,16 +11548,13 @@ SDValue X86TargetLowering::LowerSDIV(SDValue Op, SelectionDAG &DAG) const { return SDValue(); } -SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const { - +static SDValue LowerScalarImmediateShift(SDValue Op, SelectionDAG &DAG, + const X86Subtarget *Subtarget) { EVT VT = Op.getValueType(); DebugLoc dl = Op.getDebugLoc(); SDValue R = Op.getOperand(0); SDValue Amt = Op.getOperand(1); - if (!Subtarget->hasSSE2()) - return SDValue(); - // Optimize shl/srl/sra with constant shift amount. if (isSplatVector(Amt.getNode())) { SDValue SclrAmt = Amt->getOperand(0); @@ -11611,6 +11665,224 @@ SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const { } } + // Special case in 32-bit mode, where i64 is expanded into high and low parts. + if (!Subtarget->is64Bit() && + (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64)) && + Amt.getOpcode() == ISD::BITCAST && + Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) { + Amt = Amt.getOperand(0); + unsigned Ratio = Amt.getValueType().getVectorNumElements() / + VT.getVectorNumElements(); + unsigned RatioInLog2 = Log2_32_Ceil(Ratio); + uint64_t ShiftAmt = 0; + for (unsigned i = 0; i != Ratio; ++i) { + ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i)); + if (C == 0) + return SDValue(); + // 6 == Log2(64) + ShiftAmt |= C->getZExtValue() << (i * (1 << (6 - RatioInLog2))); + } + // Check remaining shift amounts. + for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) { + uint64_t ShAmt = 0; + for (unsigned j = 0; j != Ratio; ++j) { + ConstantSDNode *C = + dyn_cast<ConstantSDNode>(Amt.getOperand(i + j)); + if (C == 0) + return SDValue(); + // 6 == Log2(64) + ShAmt |= C->getZExtValue() << (j * (1 << (6 - RatioInLog2))); + } + if (ShAmt != ShiftAmt) + return SDValue(); + } + switch (Op.getOpcode()) { + default: + llvm_unreachable("Unknown shift opcode!"); + case ISD::SHL: + return DAG.getNode(X86ISD::VSHLI, dl, VT, R, + DAG.getConstant(ShiftAmt, MVT::i32)); + case ISD::SRL: + return DAG.getNode(X86ISD::VSRLI, dl, VT, R, + DAG.getConstant(ShiftAmt, MVT::i32)); + case ISD::SRA: + return DAG.getNode(X86ISD::VSRAI, dl, VT, R, + DAG.getConstant(ShiftAmt, MVT::i32)); + } + } + + return SDValue(); +} + +static SDValue LowerScalarVariableShift(SDValue Op, SelectionDAG &DAG, + const X86Subtarget* Subtarget) { + EVT VT = Op.getValueType(); + DebugLoc dl = Op.getDebugLoc(); + SDValue R = Op.getOperand(0); + SDValue Amt = Op.getOperand(1); + + if ((VT == MVT::v2i64 && Op.getOpcode() != ISD::SRA) || + VT == MVT::v4i32 || VT == MVT::v8i16 || + (Subtarget->hasInt256() && + ((VT == MVT::v4i64 && Op.getOpcode() != ISD::SRA) || + VT == MVT::v8i32 || VT == MVT::v16i16))) { + SDValue BaseShAmt; + EVT EltVT = VT.getVectorElementType(); + + if (Amt.getOpcode() == ISD::BUILD_VECTOR) { + unsigned NumElts = VT.getVectorNumElements(); + unsigned i, j; + for (i = 0; i != NumElts; ++i) { + if (Amt.getOperand(i).getOpcode() == ISD::UNDEF) + continue; + break; + } + for (j = i; j != NumElts; ++j) { + SDValue Arg = Amt.getOperand(j); + if (Arg.getOpcode() == ISD::UNDEF) continue; + if (Arg != Amt.getOperand(i)) + break; + } + if (i != NumElts && j == NumElts) + BaseShAmt = Amt.getOperand(i); + } else { + if (Amt.getOpcode() == ISD::EXTRACT_SUBVECTOR) + Amt = Amt.getOperand(0); + if (Amt.getOpcode() == ISD::VECTOR_SHUFFLE && + cast<ShuffleVectorSDNode>(Amt)->isSplat()) { + SDValue InVec = Amt.getOperand(0); + if (InVec.getOpcode() == ISD::BUILD_VECTOR) { + unsigned NumElts = InVec.getValueType().getVectorNumElements(); + unsigned i = 0; + for (; i != NumElts; ++i) { + SDValue Arg = InVec.getOperand(i); + if (Arg.getOpcode() == ISD::UNDEF) continue; + BaseShAmt = Arg; + break; + } + } else if (InVec.getOpcode() == ISD::INSERT_VECTOR_ELT) { + if (ConstantSDNode *C = + dyn_cast<ConstantSDNode>(InVec.getOperand(2))) { + unsigned SplatIdx = + cast<ShuffleVectorSDNode>(Amt)->getSplatIndex(); + if (C->getZExtValue() == SplatIdx) + BaseShAmt = InVec.getOperand(1); + } + } + if (BaseShAmt.getNode() == 0) + BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, Amt, + DAG.getIntPtrConstant(0)); + } + } + + if (BaseShAmt.getNode()) { + if (EltVT.bitsGT(MVT::i32)) + BaseShAmt = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, BaseShAmt); + else if (EltVT.bitsLT(MVT::i32)) + BaseShAmt = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, BaseShAmt); + + switch (Op.getOpcode()) { + default: + llvm_unreachable("Unknown shift opcode!"); + case ISD::SHL: + switch (VT.getSimpleVT().SimpleTy) { + default: return SDValue(); + case MVT::v2i64: + case MVT::v4i32: + case MVT::v8i16: + case MVT::v4i64: + case MVT::v8i32: + case MVT::v16i16: + return getTargetVShiftNode(X86ISD::VSHLI, dl, VT, R, BaseShAmt, DAG); + } + case ISD::SRA: + switch (VT.getSimpleVT().SimpleTy) { + default: return SDValue(); + case MVT::v4i32: + case MVT::v8i16: + case MVT::v8i32: + case MVT::v16i16: + return getTargetVShiftNode(X86ISD::VSRAI, dl, VT, R, BaseShAmt, DAG); + } + case ISD::SRL: + switch (VT.getSimpleVT().SimpleTy) { + default: return SDValue(); + case MVT::v2i64: + case MVT::v4i32: + case MVT::v8i16: + case MVT::v4i64: + case MVT::v8i32: + case MVT::v16i16: + return getTargetVShiftNode(X86ISD::VSRLI, dl, VT, R, BaseShAmt, DAG); + } + } + } + } + + // Special case in 32-bit mode, where i64 is expanded into high and low parts. + if (!Subtarget->is64Bit() && + (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64)) && + Amt.getOpcode() == ISD::BITCAST && + Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) { + Amt = Amt.getOperand(0); + unsigned Ratio = Amt.getValueType().getVectorNumElements() / + VT.getVectorNumElements(); + std::vector<SDValue> Vals(Ratio); + for (unsigned i = 0; i != Ratio; ++i) + Vals[i] = Amt.getOperand(i); + for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) { + for (unsigned j = 0; j != Ratio; ++j) + if (Vals[j] != Amt.getOperand(i + j)) + return SDValue(); + } + switch (Op.getOpcode()) { + default: + llvm_unreachable("Unknown shift opcode!"); + case ISD::SHL: + return DAG.getNode(X86ISD::VSHL, dl, VT, R, Op.getOperand(1)); + case ISD::SRL: + return DAG.getNode(X86ISD::VSRL, dl, VT, R, Op.getOperand(1)); + case ISD::SRA: + return DAG.getNode(X86ISD::VSRA, dl, VT, R, Op.getOperand(1)); + } + } + + return SDValue(); +} + +SDValue X86TargetLowering::LowerShift(SDValue Op, SelectionDAG &DAG) const { + + EVT VT = Op.getValueType(); + DebugLoc dl = Op.getDebugLoc(); + SDValue R = Op.getOperand(0); + SDValue Amt = Op.getOperand(1); + SDValue V; + + if (!Subtarget->hasSSE2()) + return SDValue(); + + V = LowerScalarImmediateShift(Op, DAG, Subtarget); + if (V.getNode()) + return V; + + V = LowerScalarVariableShift(Op, DAG, Subtarget); + if (V.getNode()) + return V; + + // AVX2 has VPSLLV/VPSRAV/VPSRLV. + if (Subtarget->hasInt256()) { + if (Op.getOpcode() == ISD::SRL && + (VT == MVT::v2i64 || VT == MVT::v4i32 || + VT == MVT::v4i64 || VT == MVT::v8i32)) + return Op; + if (Op.getOpcode() == ISD::SHL && + (VT == MVT::v2i64 || VT == MVT::v4i32 || + VT == MVT::v4i64 || VT == MVT::v8i32)) + return Op; + if (Op.getOpcode() == ISD::SRA && (VT == MVT::v4i32 || VT == MVT::v8i32)) + return Op; + } + // Lower SHL with variable shift amount. if (VT == MVT::v4i32 && Op->getOpcode() == ISD::SHL) { Op = DAG.getNode(ISD::SHL, dl, VT, Amt, DAG.getConstant(23, VT)); @@ -11827,59 +12099,28 @@ SDValue X86TargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op, // fall through case MVT::v4i32: case MVT::v8i16: { - SDValue Tmp1 = getTargetVShiftNode(X86ISD::VSHLI, dl, VT, - Op.getOperand(0), ShAmt, DAG); + // (sext (vzext x)) -> (vsext x) + SDValue Op0 = Op.getOperand(0); + SDValue Op00 = Op0.getOperand(0); + SDValue Tmp1; + // Hopefully, this VECTOR_SHUFFLE is just a VZEXT. + if (Op0.getOpcode() == ISD::BITCAST && + Op00.getOpcode() == ISD::VECTOR_SHUFFLE) + Tmp1 = LowerVectorIntExtend(Op00, DAG); + if (Tmp1.getNode()) { + SDValue Tmp1Op0 = Tmp1.getOperand(0); + assert(Tmp1Op0.getOpcode() == X86ISD::VZEXT && + "This optimization is invalid without a VZEXT."); + return DAG.getNode(X86ISD::VSEXT, dl, VT, Tmp1Op0.getOperand(0)); + } + + // If the above didn't work, then just use Shift-Left + Shift-Right. + Tmp1 = getTargetVShiftNode(X86ISD::VSHLI, dl, VT, Op0, ShAmt, DAG); return getTargetVShiftNode(X86ISD::VSRAI, dl, VT, Tmp1, ShAmt, DAG); } } } -static SDValue LowerMEMBARRIER(SDValue Op, const X86Subtarget *Subtarget, - SelectionDAG &DAG) { - DebugLoc dl = Op.getDebugLoc(); - - // Go ahead and emit the fence on x86-64 even if we asked for no-sse2. - // There isn't any reason to disable it if the target processor supports it. - if (!Subtarget->hasSSE2() && !Subtarget->is64Bit()) { - SDValue Chain = Op.getOperand(0); - SDValue Zero = DAG.getConstant(0, MVT::i32); - SDValue Ops[] = { - DAG.getRegister(X86::ESP, MVT::i32), // Base - DAG.getTargetConstant(1, MVT::i8), // Scale - DAG.getRegister(0, MVT::i32), // Index - DAG.getTargetConstant(0, MVT::i32), // Disp - DAG.getRegister(0, MVT::i32), // Segment. - Zero, - Chain - }; - SDNode *Res = - DAG.getMachineNode(X86::OR32mrLocked, dl, MVT::Other, Ops, - array_lengthof(Ops)); - return SDValue(Res, 0); - } - - unsigned isDev = cast<ConstantSDNode>(Op.getOperand(5))->getZExtValue(); - if (!isDev) - return DAG.getNode(X86ISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0)); - - unsigned Op1 = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue(); - unsigned Op2 = cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue(); - unsigned Op3 = cast<ConstantSDNode>(Op.getOperand(3))->getZExtValue(); - unsigned Op4 = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue(); - - // def : Pat<(membarrier (i8 0), (i8 0), (i8 0), (i8 1), (i8 1)), (SFENCE)>; - if (!Op1 && !Op2 && !Op3 && Op4) - return DAG.getNode(X86ISD::SFENCE, dl, MVT::Other, Op.getOperand(0)); - - // def : Pat<(membarrier (i8 1), (i8 0), (i8 0), (i8 0), (i8 1)), (LFENCE)>; - if (Op1 && !Op2 && !Op3 && !Op4) - return DAG.getNode(X86ISD::LFENCE, dl, MVT::Other, Op.getOperand(0)); - - // def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm), (i8 1)), - // (MFENCE)>; - return DAG.getNode(X86ISD::MFENCE, dl, MVT::Other, Op.getOperand(0)); -} - static SDValue LowerATOMIC_FENCE(SDValue Op, const X86Subtarget *Subtarget, SelectionDAG &DAG) { DebugLoc dl = Op.getDebugLoc(); @@ -11908,9 +12149,7 @@ static SDValue LowerATOMIC_FENCE(SDValue Op, const X86Subtarget *Subtarget, Zero, Chain }; - SDNode *Res = - DAG.getMachineNode(X86::OR32mrLocked, dl, MVT::Other, Ops, - array_lengthof(Ops)); + SDNode *Res = DAG.getMachineNode(X86::OR32mrLocked, dl, MVT::Other, Ops); return SDValue(Res, 0); } @@ -11944,7 +12183,7 @@ static SDValue LowerCMP_SWAP(SDValue Op, const X86Subtarget *Subtarget, SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue); MachineMemOperand *MMO = cast<AtomicSDNode>(Op)->getMemOperand(); SDValue Result = DAG.getMemIntrinsicNode(X86ISD::LCMPXCHG_DAG, DL, Tys, - Ops, 5, T, MMO); + Ops, array_lengthof(Ops), T, MMO); SDValue cpOut = DAG.getCopyFromReg(Result.getValue(0), DL, Reg, T, Result.getValue(1)); return cpOut; @@ -11966,7 +12205,7 @@ static SDValue LowerREADCYCLECOUNTER(SDValue Op, const X86Subtarget *Subtarget, DAG.getNode(ISD::OR, dl, MVT::i64, rax, Tmp), rdx.getValue(1) }; - return DAG.getMergeValues(Ops, 2, dl); + return DAG.getMergeValues(Ops, array_lengthof(Ops), dl); } SDValue X86TargetLowering::LowerBITCAST(SDValue Op, SelectionDAG &DAG) const { @@ -12060,7 +12299,8 @@ SDValue X86TargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const { assert(Subtarget->isTargetDarwin() && Subtarget->is64Bit()); // For MacOSX, we want to call an alternative entry point: __sincos_stret, - // which returns the values in two XMM registers. + // which returns the values as { float, float } (in XMM0) or + // { double, double } (which is returned in XMM0, XMM1). DebugLoc dl = Op.getDebugLoc(); SDValue Arg = Op.getOperand(0); EVT ArgVT = Arg.getValueType(); @@ -12075,14 +12315,16 @@ SDValue X86TargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const { Entry.isZExt = false; Args.push_back(Entry); + bool isF64 = ArgVT == MVT::f64; // Only optimize x86_64 for now. i386 is a bit messy. For f32, // the small struct {f32, f32} is returned in (eax, edx). For f64, // the results are returned via SRet in memory. - const char *LibcallName = (ArgVT == MVT::f64) - ? "__sincos_stret" : "__sincosf_stret"; + const char *LibcallName = isF64 ? "__sincos_stret" : "__sincosf_stret"; SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy()); - StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL); + Type *RetTy = isF64 + ? (Type*)StructType::get(ArgTy, ArgTy, NULL) + : (Type*)VectorType::get(ArgTy, 4); TargetLowering:: CallLoweringInfo CLI(DAG.getEntryNode(), RetTy, false, false, false, false, 0, @@ -12090,7 +12332,18 @@ SDValue X86TargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const { /*doesNotRet=*/false, /*isReturnValueUsed*/true, Callee, Args, DAG, dl); std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI); - return CallResult.first; + + if (isF64) + // Returned in xmm0 and xmm1. + return CallResult.first; + + // Returned in bits 0:31 and 32:64 xmm0. + SDValue SinVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ArgVT, + CallResult.first, DAG.getIntPtrConstant(0)); + SDValue CosVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, ArgVT, + CallResult.first, DAG.getIntPtrConstant(1)); + SDVTList Tys = DAG.getVTList(ArgVT, ArgVT); + return DAG.getNode(ISD::MERGE_VALUES, dl, Tys, SinVal, CosVal); } /// LowerOperation - Provide custom lowering hooks for some operations. @@ -12099,7 +12352,6 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const { switch (Op.getOpcode()) { default: llvm_unreachable("Should not custom lower this!"); case ISD::SIGN_EXTEND_INREG: return LowerSIGN_EXTEND_INREG(Op,DAG); - case ISD::MEMBARRIER: return LowerMEMBARRIER(Op, Subtarget, DAG); case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, Subtarget, DAG); case ISD::ATOMIC_CMP_SWAP: return LowerCMP_SWAP(Op, Subtarget, DAG); case ISD::ATOMIC_LOAD_SUB: return LowerLOAD_SUB(Op,DAG); @@ -12216,7 +12468,7 @@ ReplaceATOMIC_BINARY_64(SDNode *Node, SmallVectorImpl<SDValue>&Results, SDValue Ops[] = { Chain, In1, In2L, In2H }; SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other); SDValue Result = - DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, 4, MVT::i64, + DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, array_lengthof(Ops), MVT::i64, cast<MemSDNode>(Node)->getMemOperand()); SDValue OpsF[] = { Result.getValue(0), Result.getValue(1)}; Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF, 2)); @@ -12296,7 +12548,8 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N, eax.getValue(2)); // Use a buildpair to merge the two 32-bit values into a 64-bit one. SDValue Ops[] = { eax, edx }; - Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Ops, 2)); + Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Ops, + array_lengthof(Ops))); Results.push_back(edx.getValue(1)); return; } @@ -12335,7 +12588,7 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N, unsigned Opcode = Regs64bit ? X86ISD::LCMPXCHG16_DAG : X86ISD::LCMPXCHG8_DAG; SDValue Result = DAG.getMemIntrinsicNode(Opcode, dl, Tys, - Ops, 3, T, MMO); + Ops, array_lengthof(Ops), T, MMO); SDValue cpOutL = DAG.getCopyFromReg(Result.getValue(0), dl, Regs64bit ? X86::RAX : X86::EAX, HalfT, Result.getValue(1)); @@ -12547,6 +12800,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const { case X86ISD::WIN_FTOL: return "X86ISD::WIN_FTOL"; case X86ISD::SAHF: return "X86ISD::SAHF"; case X86ISD::RDRAND: return "X86ISD::RDRAND"; + case X86ISD::RDSEED: return "X86ISD::RDSEED"; case X86ISD::FMADD: return "X86ISD::FMADD"; case X86ISD::FMSUB: return "X86ISD::FMSUB"; case X86ISD::FNMADD: return "X86ISD::FNMADD"; @@ -12555,6 +12809,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const { case X86ISD::FMSUBADD: return "X86ISD::FMSUBADD"; case X86ISD::PCMPESTRI: return "X86ISD::PCMPESTRI"; case X86ISD::PCMPISTRI: return "X86ISD::PCMPISTRI"; + case X86ISD::XTEST: return "X86ISD::XTEST"; } } @@ -14820,7 +15075,8 @@ static SDValue PerformShuffleCombine256(SDNode *N, SelectionDAG &DAG, SDVTList Tys = DAG.getVTList(MVT::v4i64, MVT::Other); SDValue Ops[] = { Ld->getChain(), Ld->getBasePtr() }; SDValue ResNode = - DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops, 2, + DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops, + array_lengthof(Ops), Ld->getMemoryVT(), Ld->getPointerInfo(), Ld->getAlignment(), @@ -15512,6 +15768,51 @@ static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG, if (unsigned Op = matchIntegerMINMAX(Cond, VT, LHS, RHS, DAG, Subtarget)) return DAG.getNode(Op, DL, N->getValueType(0), LHS, RHS); + // Simplify vector selection if the selector will be produced by CMPP*/PCMP*. + if (!DCI.isBeforeLegalize() && N->getOpcode() == ISD::VSELECT && + Cond.getOpcode() == ISD::SETCC) { + + assert(Cond.getValueType().isVector() && + "vector select expects a vector selector!"); + + EVT IntVT = Cond.getValueType(); + bool TValIsAllOnes = ISD::isBuildVectorAllOnes(LHS.getNode()); + bool FValIsAllZeros = ISD::isBuildVectorAllZeros(RHS.getNode()); + + if (!TValIsAllOnes && !FValIsAllZeros) { + // Try invert the condition if true value is not all 1s and false value + // is not all 0s. + bool TValIsAllZeros = ISD::isBuildVectorAllZeros(LHS.getNode()); + bool FValIsAllOnes = ISD::isBuildVectorAllOnes(RHS.getNode()); + + if (TValIsAllZeros || FValIsAllOnes) { + SDValue CC = Cond.getOperand(2); + ISD::CondCode NewCC = + ISD::getSetCCInverse(cast<CondCodeSDNode>(CC)->get(), + Cond.getOperand(0).getValueType().isInteger()); + Cond = DAG.getSetCC(DL, IntVT, Cond.getOperand(0), Cond.getOperand(1), NewCC); + std::swap(LHS, RHS); + TValIsAllOnes = FValIsAllOnes; + FValIsAllZeros = TValIsAllZeros; + } + } + + if (TValIsAllOnes || FValIsAllZeros) { + SDValue Ret; + + if (TValIsAllOnes && FValIsAllZeros) + Ret = Cond; + else if (TValIsAllOnes) + Ret = DAG.getNode(ISD::OR, DL, IntVT, Cond, + DAG.getNode(ISD::BITCAST, DL, IntVT, RHS)); + else if (FValIsAllZeros) + Ret = DAG.getNode(ISD::AND, DL, IntVT, Cond, + DAG.getNode(ISD::BITCAST, DL, IntVT, LHS)); + + return DAG.getNode(ISD::BITCAST, DL, VT, Ret); + } + } + // If we know that this node is legal then we know that it is going to be // matched by one of the SSE/AVX BLEND instructions. These instructions only // depend on the highest bit in each word. Try to use SimplifyDemandedBits @@ -15572,6 +15873,7 @@ static SDValue checkBoolTestSetCCCombine(SDValue Cmp, X86::CondCode &CC) { SDValue SetCC; const ConstantSDNode* C = 0; bool needOppositeCond = (CC == X86::COND_E); + bool checkAgainstTrue = false; // Is it a comparison against 1? if ((C = dyn_cast<ConstantSDNode>(Op1))) SetCC = Op2; @@ -15580,17 +15882,46 @@ static SDValue checkBoolTestSetCCCombine(SDValue Cmp, X86::CondCode &CC) { else // Quit if all operands are not constants. return SDValue(); - if (C->getZExtValue() == 1) + if (C->getZExtValue() == 1) { needOppositeCond = !needOppositeCond; - else if (C->getZExtValue() != 0) + checkAgainstTrue = true; + } else if (C->getZExtValue() != 0) // Quit if the constant is neither 0 or 1. return SDValue(); - // Skip 'zext' node. - if (SetCC.getOpcode() == ISD::ZERO_EXTEND) - SetCC = SetCC.getOperand(0); + bool truncatedToBoolWithAnd = false; + // Skip (zext $x), (trunc $x), or (and $x, 1) node. + while (SetCC.getOpcode() == ISD::ZERO_EXTEND || + SetCC.getOpcode() == ISD::TRUNCATE || + SetCC.getOpcode() == ISD::AND) { + if (SetCC.getOpcode() == ISD::AND) { + int OpIdx = -1; + ConstantSDNode *CS; + if ((CS = dyn_cast<ConstantSDNode>(SetCC.getOperand(0))) && + CS->getZExtValue() == 1) + OpIdx = 1; + if ((CS = dyn_cast<ConstantSDNode>(SetCC.getOperand(1))) && + CS->getZExtValue() == 1) + OpIdx = 0; + if (OpIdx == -1) + break; + SetCC = SetCC.getOperand(OpIdx); + truncatedToBoolWithAnd = true; + } else + SetCC = SetCC.getOperand(0); + } switch (SetCC.getOpcode()) { + case X86ISD::SETCC_CARRY: + // Since SETCC_CARRY gives output based on R = CF ? ~0 : 0, it's unsafe to + // simplify it if the result of SETCC_CARRY is not canonicalized to 0 or 1, + // i.e. it's a comparison against true but the result of SETCC_CARRY is not + // truncated to i1 using 'and'. + if (checkAgainstTrue && !truncatedToBoolWithAnd) + break; + assert(X86::CondCode(SetCC.getConstantOperandVal(0)) == X86::COND_B && + "Invalid use of SETCC_CARRY!"); + // FALL THROUGH case X86ISD::SETCC: // Set the condition code or opposite one if necessary. CC = X86::CondCode(SetCC.getConstantOperandVal(0)); @@ -15606,9 +15937,15 @@ static SDValue checkBoolTestSetCCCombine(SDValue Cmp, X86::CondCode &CC) { return SDValue(); // Quit if false value is not a constant. if (!FVal) { - // A special case for rdrand, where 0 is set if false cond is found. SDValue Op = SetCC.getOperand(0); - if (Op.getOpcode() != X86ISD::RDRAND) + // Skip 'zext' or 'trunc' node. + if (Op.getOpcode() == ISD::ZERO_EXTEND || + Op.getOpcode() == ISD::TRUNCATE) + Op = Op.getOperand(0); + // A special case for rdrand/rdseed, where 0 is set if false cond is + // found. + if ((Op.getOpcode() != X86ISD::RDRAND && + Op.getOpcode() != X86ISD::RDSEED) || Op.getResNo() != 0) return SDValue(); } // Quit if false value is not the constant 0 or 1. @@ -15920,124 +16257,12 @@ static SDValue PerformSHLCombine(SDNode *N, SelectionDAG &DAG) { static SDValue PerformShiftCombine(SDNode* N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const X86Subtarget *Subtarget) { - EVT VT = N->getValueType(0); if (N->getOpcode() == ISD::SHL) { SDValue V = PerformSHLCombine(N, DAG); if (V.getNode()) return V; } - // On X86 with SSE2 support, we can transform this to a vector shift if - // all elements are shifted by the same amount. We can't do this in legalize - // because the a constant vector is typically transformed to a constant pool - // so we have no knowledge of the shift amount. - if (!Subtarget->hasSSE2()) - return SDValue(); - - if (VT != MVT::v2i64 && VT != MVT::v4i32 && VT != MVT::v8i16 && - (!Subtarget->hasInt256() || - (VT != MVT::v4i64 && VT != MVT::v8i32 && VT != MVT::v16i16))) - return SDValue(); - - SDValue ShAmtOp = N->getOperand(1); - EVT EltVT = VT.getVectorElementType(); - DebugLoc DL = N->getDebugLoc(); - SDValue BaseShAmt = SDValue(); - if (ShAmtOp.getOpcode() == ISD::BUILD_VECTOR) { - unsigned NumElts = VT.getVectorNumElements(); - unsigned i = 0; - for (; i != NumElts; ++i) { - SDValue Arg = ShAmtOp.getOperand(i); - if (Arg.getOpcode() == ISD::UNDEF) continue; - BaseShAmt = Arg; - break; - } - // Handle the case where the build_vector is all undef - // FIXME: Should DAG allow this? - if (i == NumElts) - return SDValue(); - - for (; i != NumElts; ++i) { - SDValue Arg = ShAmtOp.getOperand(i); - if (Arg.getOpcode() == ISD::UNDEF) continue; - if (Arg != BaseShAmt) { - return SDValue(); - } - } - } else if (ShAmtOp.getOpcode() == ISD::VECTOR_SHUFFLE && - cast<ShuffleVectorSDNode>(ShAmtOp)->isSplat()) { - SDValue InVec = ShAmtOp.getOperand(0); - if (InVec.getOpcode() == ISD::BUILD_VECTOR) { - unsigned NumElts = InVec.getValueType().getVectorNumElements(); - unsigned i = 0; - for (; i != NumElts; ++i) { - SDValue Arg = InVec.getOperand(i); - if (Arg.getOpcode() == ISD::UNDEF) continue; - BaseShAmt = Arg; - break; - } - } else if (InVec.getOpcode() == ISD::INSERT_VECTOR_ELT) { - if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(InVec.getOperand(2))) { - unsigned SplatIdx= cast<ShuffleVectorSDNode>(ShAmtOp)->getSplatIndex(); - if (C->getZExtValue() == SplatIdx) - BaseShAmt = InVec.getOperand(1); - } - } - if (BaseShAmt.getNode() == 0) { - // Don't create instructions with illegal types after legalize - // types has run. - if (!DAG.getTargetLoweringInfo().isTypeLegal(EltVT) && - !DCI.isBeforeLegalize()) - return SDValue(); - - BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, ShAmtOp, - DAG.getIntPtrConstant(0)); - } - } else - return SDValue(); - - // The shift amount is an i32. - if (EltVT.bitsGT(MVT::i32)) - BaseShAmt = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, BaseShAmt); - else if (EltVT.bitsLT(MVT::i32)) - BaseShAmt = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, BaseShAmt); - - // The shift amount is identical so we can do a vector shift. - SDValue ValOp = N->getOperand(0); - switch (N->getOpcode()) { - default: - llvm_unreachable("Unknown shift opcode!"); - case ISD::SHL: - switch (VT.getSimpleVT().SimpleTy) { - default: return SDValue(); - case MVT::v2i64: - case MVT::v4i32: - case MVT::v8i16: - case MVT::v4i64: - case MVT::v8i32: - case MVT::v16i16: - return getTargetVShiftNode(X86ISD::VSHLI, DL, VT, ValOp, BaseShAmt, DAG); - } - case ISD::SRA: - switch (VT.getSimpleVT().SimpleTy) { - default: return SDValue(); - case MVT::v4i32: - case MVT::v8i16: - case MVT::v8i32: - case MVT::v16i16: - return getTargetVShiftNode(X86ISD::VSRAI, DL, VT, ValOp, BaseShAmt, DAG); - } - case ISD::SRL: - switch (VT.getSimpleVT().SimpleTy) { - default: return SDValue(); - case MVT::v2i64: - case MVT::v4i32: - case MVT::v8i16: - case MVT::v4i64: - case MVT::v8i32: - case MVT::v16i16: - return getTargetVShiftNode(X86ISD::VSRLI, DL, VT, ValOp, BaseShAmt, DAG); - } - } + return SDValue(); } // CMPEQCombine - Recognize the distinctive (AND (setcc ...) (setcc ..)) @@ -16348,13 +16573,19 @@ static SDValue PerformOrCombine(SDNode *N, SelectionDAG &DAG, // Validate that the Mask operand is a vector sra node. // FIXME: what to do for bytes, since there is a psignb/pblendvb, but // there is no psrai.b - if (Mask.getOpcode() != X86ISD::VSRAI) - return SDValue(); - - // Check that the SRA is all signbits. - SDValue SraC = Mask.getOperand(1); - unsigned SraAmt = cast<ConstantSDNode>(SraC)->getZExtValue(); unsigned EltBits = MaskVT.getVectorElementType().getSizeInBits(); + unsigned SraAmt = ~0; + if (Mask.getOpcode() == ISD::SRA) { + SDValue Amt = Mask.getOperand(1); + if (isSplatVector(Amt.getNode())) { + SDValue SclrAmt = Amt->getOperand(0); + if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(SclrAmt)) + SraAmt = C->getZExtValue(); + } + } else if (Mask.getOpcode() == X86ISD::VSRAI) { + SDValue SraC = Mask.getOperand(1); + SraAmt = cast<ConstantSDNode>(SraC)->getZExtValue(); + } if ((SraAmt + 1) != EltBits) return SDValue(); @@ -16528,11 +16759,10 @@ static SDValue PerformLOADCombine(SDNode *N, SelectionDAG &DAG, const TargetLowering &TLI = DAG.getTargetLoweringInfo(); unsigned RegSz = RegVT.getSizeInBits(); + // On Sandybridge unaligned 256bit loads are inefficient. ISD::LoadExtType Ext = Ld->getExtensionType(); unsigned Alignment = Ld->getAlignment(); - bool IsAligned = Alignment == 0 || Alignment == MemVT.getSizeInBits()/8; - - // On Sandybridge unaligned 256bit loads are inefficient. + bool IsAligned = Alignment == 0 || Alignment >= MemVT.getSizeInBits()/8; if (RegVT.is256BitVector() && !Subtarget->hasInt256() && !DCI.isBeforeLegalizeOps() && !IsAligned && Ext == ISD::NON_EXTLOAD) { unsigned NumElems = RegVT.getVectorNumElements(); @@ -16552,7 +16782,7 @@ static SDValue PerformLOADCombine(SDNode *N, SelectionDAG &DAG, SDValue Load2 = DAG.getLoad(HalfVT, dl, Ld->getChain(), Ptr, Ld->getPointerInfo(), Ld->isVolatile(), Ld->isNonTemporal(), Ld->isInvariant(), - std::max(Alignment/2U, 1U)); + std::min(16U, Alignment)); SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Load1.getValue(1), Load2.getValue(1)); @@ -16723,13 +16953,13 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG, DebugLoc dl = St->getDebugLoc(); SDValue StoredVal = St->getOperand(1); const TargetLowering &TLI = DAG.getTargetLoweringInfo(); - unsigned Alignment = St->getAlignment(); - bool IsAligned = Alignment == 0 || Alignment == VT.getSizeInBits()/8; // If we are saving a concatenation of two XMM registers, perform two stores. // On Sandy Bridge, 256-bit memory operations are executed by two // 128-bit ports. However, on Haswell it is better to issue a single 256-bit // memory operation. + unsigned Alignment = St->getAlignment(); + bool IsAligned = Alignment == 0 || Alignment >= VT.getSizeInBits()/8; if (VT.is256BitVector() && !Subtarget->hasInt256() && StVT == VT && !IsAligned) { unsigned NumElems = VT.getVectorNumElements(); @@ -16749,7 +16979,7 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG, SDValue Ch1 = DAG.getStore(St->getChain(), dl, Value1, Ptr1, St->getPointerInfo(), St->isVolatile(), St->isNonTemporal(), - std::max(Alignment/2U, 1U)); + std::min(16U, Alignment)); return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Ch0, Ch1); } diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h index da1dad0..2727e22 100644 --- a/lib/Target/X86/X86ISelLowering.h +++ b/lib/Target/X86/X86ISelLowering.h @@ -356,10 +356,17 @@ namespace llvm { // RDRAND - Get a random integer and indicate whether it is valid in CF. RDRAND, + // RDSEED - Get a NIST SP800-90B & C compliant random integer and + // indicate whether it is valid in CF. + RDSEED, + // PCMP*STRI PCMPISTRI, PCMPESTRI, + // XTEST - Test if in transactional execution. + XTEST, + // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG, // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG - // Atomic 64-bit binary operations. @@ -716,6 +723,9 @@ namespace llvm { SDValue BuildFILD(SDValue Op, EVT SrcVT, SDValue Chain, SDValue StackSlot, SelectionDAG &DAG) const; + /// \brief Reset the operation actions based on target options. + virtual void resetOperationActions(); + protected: std::pair<const TargetRegisterClass*, uint8_t> findRepresentativeClass(MVT VT) const; @@ -727,6 +737,10 @@ namespace llvm { const X86RegisterInfo *RegInfo; const DataLayout *TD; + /// Used to store the TargetOptions so that we don't waste time resetting + /// the operation actions unless we have to. + TargetOptions TO; + /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87 /// floating point ops. /// When SSE is available, use it for f32 operations. diff --git a/lib/Target/X86/X86Instr3DNow.td b/lib/Target/X86/X86Instr3DNow.td index bb362f5..ba1aede 100644 --- a/lib/Target/X86/X86Instr3DNow.td +++ b/lib/Target/X86/X86Instr3DNow.td @@ -84,13 +84,16 @@ defm PI2FD : I3DNow_conv_rm_int<0x0D, "pi2fd">; defm PMULHRW : I3DNow_binop_rm_int<0xB7, "pmulhrw">; -def FEMMS : I3DNow<0x0E, RawFrm, (outs), (ins), "femms", [(int_x86_mmx_femms)]>; +def FEMMS : I3DNow<0x0E, RawFrm, (outs), (ins), "femms", + [(int_x86_mmx_femms)]>; -def PREFETCH : I3DNow<0x0D, MRM0m, (outs), (ins i32mem:$addr), - "prefetch\t$addr", []>; +def PREFETCH : I3DNow<0x0D, MRM0m, (outs), (ins i8mem:$addr), + "prefetch\t$addr", + [(prefetch addr:$addr, (i32 0), imm, (i32 1))]>; -def PREFETCHW : I3DNow<0x0D, MRM1m, (outs), (ins i16mem:$addr), - "prefetchw\t$addr", []>; +def PREFETCHW : I<0x0D, MRM1m, (outs), (ins i8mem:$addr), "prefetchw\t$addr", + [(prefetch addr:$addr, (i32 1), (i32 3), (i32 1))]>, TB, + Requires<[HasPrefetchW]>; // "3DNowA" instructions defm PF2IW : I3DNow_conv_rm_int<0x1C, "pf2iw", "a">; diff --git a/lib/Target/X86/X86InstrArithmetic.td b/lib/Target/X86/X86InstrArithmetic.td index f406416..225e972 100644 --- a/lib/Target/X86/X86InstrArithmetic.td +++ b/lib/Target/X86/X86InstrArithmetic.td @@ -932,7 +932,8 @@ class BinOpMI8<string mnemonic, X86TypeInfo typeinfo, Format f, list<dag> pattern> : ITy<0x82, f, typeinfo, (outs), (ins typeinfo.MemOperand:$dst, typeinfo.Imm8Operand:$src), - mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM> { + mnemonic, "{$src, $dst|$dst, $src}", pattern, IIC_BIN_MEM>, + Sched<[WriteALULd, WriteRMW]> { let ImmT = Imm8; // Always 8-bit immediate. } @@ -964,7 +965,7 @@ class BinOpAI<bits<8> opcode, string mnemonic, X86TypeInfo typeinfo, Register areg, string operands> : ITy<opcode, RawFrm, typeinfo, (outs), (ins typeinfo.ImmOperand:$src), - mnemonic, operands, []> { + mnemonic, operands, []>, Sched<[WriteALU]> { let ImmT = typeinfo.ImmEncoding; let Uses = [areg]; let Defs = [areg]; @@ -1250,7 +1251,7 @@ let isCompare = 1, Defs = [EFLAGS] in { // register class is constrained to GR8_NOREX. let isPseudo = 1 in def TEST8ri_NOREX : I<0, Pseudo, (outs), (ins GR8_NOREX:$src, i8imm:$mask), - "", [], IIC_BIN_NONMEM>; + "", [], IIC_BIN_NONMEM>, Sched<[WriteALU]>; } //===----------------------------------------------------------------------===// @@ -1293,12 +1294,12 @@ let neverHasSideEffects = 1 in { let isCommutable = 1 in def rr : I<0xF6, MRMSrcReg, (outs RC:$dst1, RC:$dst2), (ins RC:$src), !strconcat(mnemonic, "\t{$src, $dst2, $dst1|$dst1, $dst2, $src}"), - [], IIC_MUL8>, T8XD, VEX_4V; + [], IIC_MUL8>, T8XD, VEX_4V, Sched<[WriteIMul]>; let mayLoad = 1 in def rm : I<0xF6, MRMSrcMem, (outs RC:$dst1, RC:$dst2), (ins x86memop:$src), !strconcat(mnemonic, "\t{$src, $dst2, $dst1|$dst1, $dst2, $src}"), - [], IIC_MUL8>, T8XD, VEX_4V; + [], IIC_MUL8>, T8XD, VEX_4V, Sched<[WriteIMulLd]>; } } @@ -1313,6 +1314,7 @@ let Predicates = [HasBMI2] in { // ADCX Instruction // let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in { + let SchedRW = [WriteALU] in { def ADCX32rr : I<0xF6, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src), "adcx{l}\t{$src, $dst|$dst, $src}", [], IIC_BIN_NONMEM>, T8, OpSize; @@ -1320,8 +1322,9 @@ let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in { def ADCX64rr : I<0xF6, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), "adcx{q}\t{$src, $dst|$dst, $src}", [], IIC_BIN_NONMEM>, T8, OpSize, REX_W, Requires<[In64BitMode]>; + } // SchedRW - let mayLoad = 1 in { + let mayLoad = 1, SchedRW = [WriteALULd] in { def ADCX32rm : I<0xF6, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src), "adcx{l}\t{$src, $dst|$dst, $src}", [], IIC_BIN_MEM>, T8, OpSize; @@ -1336,6 +1339,7 @@ let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in { // ADOX Instruction // let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in { + let SchedRW = [WriteALU] in { def ADOX32rr : I<0xF6, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src), "adox{l}\t{$src, $dst|$dst, $src}", [], IIC_BIN_NONMEM>, T8XS; @@ -1343,8 +1347,9 @@ let hasSideEffects = 0, Predicates = [HasADX], Defs = [EFLAGS] in { def ADOX64rr : I<0xF6, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), "adox{q}\t{$src, $dst|$dst, $src}", [], IIC_BIN_NONMEM>, T8XS, REX_W, Requires<[In64BitMode]>; + } // SchedRW - let mayLoad = 1 in { + let mayLoad = 1, SchedRW = [WriteALULd] in { def ADOX32rm : I<0xF6, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src), "adox{l}\t{$src, $dst|$dst, $src}", [], IIC_BIN_MEM>, T8XS; diff --git a/lib/Target/X86/X86InstrCMovSetCC.td b/lib/Target/X86/X86InstrCMovSetCC.td index 8f2d0a1..a967a4d 100644 --- a/lib/Target/X86/X86InstrCMovSetCC.td +++ b/lib/Target/X86/X86InstrCMovSetCC.td @@ -16,7 +16,7 @@ // SetCC instructions. multiclass CMOV<bits<8> opc, string Mnemonic, PatLeaf CondNode> { let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst", - isCommutable = 1 in { + isCommutable = 1, SchedRW = [WriteALU] in { def NAME#16rr : I<opc, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src1, GR16:$src2), !strconcat(Mnemonic, "{w}\t{$src2, $dst|$dst, $src2}"), @@ -37,7 +37,8 @@ multiclass CMOV<bits<8> opc, string Mnemonic, PatLeaf CondNode> { IIC_CMOV32_RR>, TB; } - let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst" in { + let Uses = [EFLAGS], Predicates = [HasCMov], Constraints = "$src1 = $dst", + SchedRW = [WriteALULd, ReadAfterLd] in { def NAME#16rm : I<opc, MRMSrcMem, (outs GR16:$dst), (ins GR16:$src1, i16mem:$src2), !strconcat(Mnemonic, "{w}\t{$src2, $dst|$dst, $src2}"), @@ -83,11 +84,11 @@ multiclass SETCC<bits<8> opc, string Mnemonic, PatLeaf OpNode> { def r : I<opc, MRM0r, (outs GR8:$dst), (ins), !strconcat(Mnemonic, "\t$dst"), [(set GR8:$dst, (X86setcc OpNode, EFLAGS))], - IIC_SET_R>, TB; + IIC_SET_R>, TB, Sched<[WriteALU]>; def m : I<opc, MRM0m, (outs), (ins i8mem:$dst), !strconcat(Mnemonic, "\t$dst"), [(store (X86setcc OpNode, EFLAGS), addr:$dst)], - IIC_SET_M>, TB; + IIC_SET_M>, TB, Sched<[WriteALU, WriteStore]>; } // Uses = [EFLAGS] } diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td index 734e598..d9ff0c6 100644 --- a/lib/Target/X86/X86InstrCompiler.td +++ b/lib/Target/X86/X86InstrCompiler.td @@ -149,11 +149,12 @@ let Defs = [EAX, EDX, EFLAGS], FPForm = SpecialFP in { //===----------------------------------------------------------------------===// // EH Pseudo Instructions // +let SchedRW = [WriteSystem] in { let isTerminator = 1, isReturn = 1, isBarrier = 1, hasCtrlDep = 1, isCodeGenOnly = 1 in { def EH_RETURN : I<0xC3, RawFrm, (outs), (ins GR32:$addr), "ret\t#eh_return, addr: $addr", - [(X86ehret GR32:$addr)], IIC_RET>; + [(X86ehret GR32:$addr)], IIC_RET>, Sched<[WriteJumpLd]>; } @@ -161,7 +162,7 @@ let isTerminator = 1, isReturn = 1, isBarrier = 1, hasCtrlDep = 1, isCodeGenOnly = 1 in { def EH_RETURN64 : I<0xC3, RawFrm, (outs), (ins GR64:$addr), "ret\t#eh_return, addr: $addr", - [(X86ehret GR64:$addr)], IIC_RET>; + [(X86ehret GR64:$addr)], IIC_RET>, Sched<[WriteJumpLd]>; } @@ -186,6 +187,7 @@ let hasSideEffects = 1, isBarrier = 1, isCodeGenOnly = 1, Requires<[In64BitMode]>; } } +} // SchedRW let isBranch = 1, isTerminator = 1, isCodeGenOnly = 1 in { def EH_SjLj_Setup : I<0, Pseudo, (outs), (ins brtarget:$dst), @@ -220,7 +222,7 @@ def MORESTACK_RET_RESTORE_R10 : I<0, Pseudo, (outs), (ins), let Defs = [EFLAGS], isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1 in { def MOV8r0 : I<0x30, MRMInitReg, (outs GR8 :$dst), (ins), "", - [(set GR8:$dst, 0)], IIC_ALU_NONMEM>; + [(set GR8:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>; // We want to rewrite MOV16r0 in terms of MOV32r0, because it's a smaller // encoding and avoids a partial-register update sometimes, but doing so @@ -229,11 +231,12 @@ def MOV8r0 : I<0x30, MRMInitReg, (outs GR8 :$dst), (ins), "", // to an MCInst. def MOV16r0 : I<0x31, MRMInitReg, (outs GR16:$dst), (ins), "", - [(set GR16:$dst, 0)], IIC_ALU_NONMEM>, OpSize; + [(set GR16:$dst, 0)], IIC_ALU_NONMEM>, OpSize, + Sched<[WriteZero]>; // FIXME: Set encoding to pseudo. def MOV32r0 : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "", - [(set GR32:$dst, 0)], IIC_ALU_NONMEM>; + [(set GR32:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>; } // We want to rewrite MOV64r0 in terms of MOV32r0, because it's sometimes a @@ -245,7 +248,7 @@ def MOV32r0 : I<0x31, MRMInitReg, (outs GR32:$dst), (ins), "", let Defs = [EFLAGS], isCodeGenOnly=1, AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in def MOV64r0 : I<0x31, MRMInitReg, (outs GR64:$dst), (ins), "", - [(set GR64:$dst, 0)], IIC_ALU_NONMEM>; + [(set GR64:$dst, 0)], IIC_ALU_NONMEM>, Sched<[WriteZero]>; // Materialize i64 constant where top 32-bits are zero. This could theoretically // use MOV32ri with a SUBREG_TO_REG to represent the zero-extension, however @@ -254,10 +257,10 @@ let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1 in def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src), "", [(set GR64:$dst, i64immZExt32:$src)], - IIC_ALU_NONMEM>; + IIC_ALU_NONMEM>, Sched<[WriteALU]>; // Use sbb to materialize carry bit. -let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1 in { +let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1, SchedRW = [WriteALU] in { // FIXME: These are pseudo ops that should be replaced with Pat<> patterns. // However, Pat<> can't replicate the destination reg into the inputs of the // result. @@ -320,6 +323,7 @@ def : Pat<(sub GR64:$op, (i64 (X86setcc_c X86_COND_B, EFLAGS))), //===----------------------------------------------------------------------===// // String Pseudo Instructions // +let SchedRW = [WriteMicrocoded] in { let Defs = [ECX,EDI,ESI], Uses = [ECX,EDI,ESI], isCodeGenOnly = 1 in { def REP_MOVSB_32 : I<0xA4, RawFrm, (outs), (ins), "{rep;movsb|rep movsb}", [(X86rep_movs i8)], IIC_REP_MOVS>, REP, @@ -382,6 +386,7 @@ let Defs = [RCX,RDI], isCodeGenOnly = 1 in { [(X86rep_stos i64)], IIC_REP_STOS>, REP, Requires<[In64BitMode]>; } +} // SchedRW //===----------------------------------------------------------------------===// // Thread Local Storage Instructions @@ -594,12 +599,13 @@ defm ATOMSWAP : PSEUDO_ATOMIC_LOAD_BINOP6432<"#ATOMSWAP">; let isCodeGenOnly = 1, Defs = [EFLAGS] in def OR32mrLocked : I<0x09, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$zero), "or{l}\t{$zero, $dst|$dst, $zero}", - [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK; + [], IIC_ALU_MEM>, Requires<[In32BitMode]>, LOCK, + Sched<[WriteALULd, WriteRMW]>; let hasSideEffects = 1 in def Int_MemBarrier : I<0, Pseudo, (outs), (ins), "#MEMBARRIER", - [(X86MemBarrier)]>; + [(X86MemBarrier)]>, Sched<[WriteLoad]>; // RegOpc corresponds to the mr version of the instruction // ImmOpc corresponds to the mi version of the instruction @@ -607,7 +613,8 @@ def Int_MemBarrier : I<0, Pseudo, (outs), (ins), // ImmMod corresponds to the instruction format of the mi and mi8 versions multiclass LOCK_ArithBinOp<bits<8> RegOpc, bits<8> ImmOpc, bits<8> ImmOpc8, Format ImmMod, string mnemonic> { -let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in { +let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1, + SchedRW = [WriteALULd, WriteRMW] in { def NAME#8mr : I<{RegOpc{7}, RegOpc{6}, RegOpc{5}, RegOpc{4}, RegOpc{3}, RegOpc{2}, RegOpc{1}, 0 }, @@ -694,7 +701,8 @@ defm LOCK_XOR : LOCK_ArithBinOp<0x30, 0x80, 0x83, MRM6m, "xor">; // Optimized codegen when the non-memory output is not used. multiclass LOCK_ArithUnOp<bits<8> Opc8, bits<8> Opc, Format Form, string mnemonic> { -let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1 in { +let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1, + SchedRW = [WriteALULd, WriteRMW] in { def NAME#8m : I<Opc8, Form, (outs), (ins i8mem :$dst), !strconcat(mnemonic, "{b}\t$dst"), @@ -728,7 +736,7 @@ let isCodeGenOnly = 1 in { multiclass LCMPXCHG_BinOp<bits<8> Opc8, bits<8> Opc, Format Form, string mnemonic, SDPatternOperator frag, InstrItinClass itin8, InstrItinClass itin> { -let isCodeGenOnly = 1 in { +let isCodeGenOnly = 1, SchedRW = [WriteALULd, WriteRMW] in { let Defs = [AL, EFLAGS], Uses = [AL] in def NAME#8 : I<Opc8, Form, (outs), (ins i8mem:$ptr, GR8:$swap), !strconcat(mnemonic, "{b}\t{$swap, $ptr|$ptr, $swap}"), @@ -748,14 +756,15 @@ let isCodeGenOnly = 1 in { } } -let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX] in { +let Defs = [EAX, EDX, EFLAGS], Uses = [EAX, EBX, ECX, EDX], + SchedRW = [WriteALULd, WriteRMW] in { defm LCMPXCHG8B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg8b", X86cas8, i64mem, IIC_CMPX_LOCK_8B>; } let Defs = [RAX, RDX, EFLAGS], Uses = [RAX, RBX, RCX, RDX], - Predicates = [HasCmpxchg16b] in { + Predicates = [HasCmpxchg16b], SchedRW = [WriteALULd, WriteRMW] in { defm LCMPXCHG16B : LCMPXCHG_UnOp<0xC7, MRM1m, "cmpxchg16b", X86cas16, i128mem, IIC_CMPX_LOCK_16B>, REX_W; @@ -768,7 +777,8 @@ defm LCMPXCHG : LCMPXCHG_BinOp<0xB0, 0xB1, MRMDestMem, "cmpxchg", multiclass ATOMIC_LOAD_BINOP<bits<8> opc8, bits<8> opc, string mnemonic, string frag, InstrItinClass itin8, InstrItinClass itin> { - let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1 in { + let Constraints = "$val = $dst", Defs = [EFLAGS], isCodeGenOnly = 1, + SchedRW = [WriteALULd, WriteRMW] in { def NAME#8 : I<opc8, MRMSrcMem, (outs GR8:$dst), (ins GR8:$val, i8mem:$ptr), !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"), @@ -990,9 +1000,6 @@ def : Pat<(i64 (X86Wrapper tglobaltlsaddr :$dst)), // This corresponds to add $foo@tpoff, %rax def : Pat<(add GR64:$src1, (X86Wrapper tglobaltlsaddr :$dst)), (ADD64ri32 GR64:$src1, tglobaltlsaddr :$dst)>; -// This corresponds to mov foo@tpoff(%rbx), %eax -def : Pat<(load (i64 (X86Wrapper tglobaltlsaddr :$dst))), - (MOV64rm tglobaltlsaddr :$dst)>; // Direct PC relative function call for small code model. 32-bit displacement @@ -1192,7 +1199,8 @@ def or_is_add : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),[{ // (or x1, x2) -> (add x1, x2) if two operands are known not to share bits. -let AddedComplexity = 5 in { // Try this before the selecting to OR +// Try this before the selecting to OR. +let AddedComplexity = 5, SchedRW = [WriteALU] in { let isConvertibleToThreeAddress = 1, Constraints = "$src1 = $dst", Defs = [EFLAGS] in { @@ -1239,7 +1247,7 @@ def ADD64ri32_DB : I<0, Pseudo, [(set GR64:$dst, (or_is_add GR64:$src1, i64immSExt32:$src2))]>; } -} // AddedComplexity +} // AddedComplexity, SchedRW //===----------------------------------------------------------------------===// diff --git a/lib/Target/X86/X86InstrControl.td b/lib/Target/X86/X86InstrControl.td index bfe9541..0e69651 100644 --- a/lib/Target/X86/X86InstrControl.td +++ b/lib/Target/X86/X86InstrControl.td @@ -20,7 +20,7 @@ // The X86retflag return instructions are variadic because we may add ST0 and // ST1 arguments when returning values on the x87 stack. let isTerminator = 1, isReturn = 1, isBarrier = 1, - hasCtrlDep = 1, FPForm = SpecialFP in { + hasCtrlDep = 1, FPForm = SpecialFP, SchedRW = [WriteJumpLd] in { def RET : I <0xC3, RawFrm, (outs), (ins variable_ops), "ret", [(X86retflag 0)], IIC_RET>; @@ -46,7 +46,7 @@ let isTerminator = 1, isReturn = 1, isBarrier = 1, } // Unconditional branches. -let isBarrier = 1, isBranch = 1, isTerminator = 1 in { +let isBarrier = 1, isBranch = 1, isTerminator = 1, SchedRW = [WriteJump] in { def JMP_4 : Ii32PCRel<0xE9, RawFrm, (outs), (ins brtarget:$dst), "jmp\t$dst", [(br bb:$dst)], IIC_JMP_REL>; def JMP_1 : Ii8PCRel<0xEB, RawFrm, (outs), (ins brtarget8:$dst), @@ -58,7 +58,7 @@ let isBarrier = 1, isBranch = 1, isTerminator = 1 in { } // Conditional Branches. -let isBranch = 1, isTerminator = 1, Uses = [EFLAGS] in { +let isBranch = 1, isTerminator = 1, Uses = [EFLAGS], SchedRW = [WriteJump] in { multiclass ICBr<bits<8> opc1, bits<8> opc4, string asm, PatFrag Cond> { def _1 : Ii8PCRel <opc1, RawFrm, (outs), (ins brtarget8:$dst), asm, [], IIC_Jcc>; @@ -85,7 +85,7 @@ defm JLE : ICBr<0x7E, 0x8E, "jle\t$dst", X86_COND_LE>; defm JG : ICBr<0x7F, 0x8F, "jg\t$dst" , X86_COND_G>; // jcx/jecx/jrcx instructions. -let isBranch = 1, isTerminator = 1 in { +let isBranch = 1, isTerminator = 1, SchedRW = [WriteJump] in { // These are the 32-bit versions of this instruction for the asmparser. In // 32-bit mode, the address size prefix is jcxz and the unprefixed version is // jecxz. @@ -110,36 +110,46 @@ let isBranch = 1, isTerminator = 1 in { // Indirect branches let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in { def JMP32r : I<0xFF, MRM4r, (outs), (ins GR32:$dst), "jmp{l}\t{*}$dst", - [(brind GR32:$dst)], IIC_JMP_REG>, Requires<[In32BitMode]>; + [(brind GR32:$dst)], IIC_JMP_REG>, Requires<[In32BitMode]>, + Sched<[WriteJump]>; def JMP32m : I<0xFF, MRM4m, (outs), (ins i32mem:$dst), "jmp{l}\t{*}$dst", - [(brind (loadi32 addr:$dst))], IIC_JMP_MEM>, Requires<[In32BitMode]>; + [(brind (loadi32 addr:$dst))], IIC_JMP_MEM>, + Requires<[In32BitMode]>, Sched<[WriteJumpLd]>; def JMP64r : I<0xFF, MRM4r, (outs), (ins GR64:$dst), "jmp{q}\t{*}$dst", - [(brind GR64:$dst)], IIC_JMP_REG>, Requires<[In64BitMode]>; + [(brind GR64:$dst)], IIC_JMP_REG>, Requires<[In64BitMode]>, + Sched<[WriteJump]>; def JMP64m : I<0xFF, MRM4m, (outs), (ins i64mem:$dst), "jmp{q}\t{*}$dst", - [(brind (loadi64 addr:$dst))], IIC_JMP_MEM>, Requires<[In64BitMode]>; + [(brind (loadi64 addr:$dst))], IIC_JMP_MEM>, + Requires<[In64BitMode]>, Sched<[WriteJumpLd]>; def FARJMP16i : Iseg16<0xEA, RawFrmImm16, (outs), (ins i16imm:$off, i16imm:$seg), - "ljmp{w}\t{$seg, $off|$off, $seg}", [], IIC_JMP_FAR_PTR>, OpSize; + "ljmp{w}\t{$seg, $off|$off, $seg}", [], + IIC_JMP_FAR_PTR>, OpSize, Sched<[WriteJump]>; def FARJMP32i : Iseg32<0xEA, RawFrmImm16, (outs), (ins i32imm:$off, i16imm:$seg), - "ljmp{l}\t{$seg, $off|$off, $seg}", [], IIC_JMP_FAR_PTR>; + "ljmp{l}\t{$seg, $off|$off, $seg}", [], + IIC_JMP_FAR_PTR>, Sched<[WriteJump]>; def FARJMP64 : RI<0xFF, MRM5m, (outs), (ins opaque80mem:$dst), - "ljmp{q}\t{*}$dst", [], IIC_JMP_FAR_MEM>; + "ljmp{q}\t{*}$dst", [], IIC_JMP_FAR_MEM>, + Sched<[WriteJump]>; def FARJMP16m : I<0xFF, MRM5m, (outs), (ins opaque32mem:$dst), - "ljmp{w}\t{*}$dst", [], IIC_JMP_FAR_MEM>, OpSize; + "ljmp{w}\t{*}$dst", [], IIC_JMP_FAR_MEM>, OpSize, + Sched<[WriteJumpLd]>; def FARJMP32m : I<0xFF, MRM5m, (outs), (ins opaque48mem:$dst), - "ljmp{l}\t{*}$dst", [], IIC_JMP_FAR_MEM>; + "ljmp{l}\t{*}$dst", [], IIC_JMP_FAR_MEM>, + Sched<[WriteJumpLd]>; } // Loop instructions - +let SchedRW = [WriteJump] in { def LOOP : Ii8PCRel<0xE2, RawFrm, (outs), (ins brtarget8:$dst), "loop\t$dst", [], IIC_LOOP>; def LOOPE : Ii8PCRel<0xE1, RawFrm, (outs), (ins brtarget8:$dst), "loope\t$dst", [], IIC_LOOPE>; def LOOPNE : Ii8PCRel<0xE0, RawFrm, (outs), (ins brtarget8:$dst), "loopne\t$dst", [], IIC_LOOPNE>; +} //===----------------------------------------------------------------------===// // Call Instructions... @@ -152,27 +162,32 @@ let isCall = 1 in let Uses = [ESP] in { def CALLpcrel32 : Ii32PCRel<0xE8, RawFrm, (outs), (ins i32imm_pcrel:$dst), - "call{l}\t$dst", [], IIC_CALL_RI>, Requires<[In32BitMode]>; + "call{l}\t$dst", [], IIC_CALL_RI>, + Requires<[In32BitMode]>, Sched<[WriteJump]>; def CALL32r : I<0xFF, MRM2r, (outs), (ins GR32:$dst), "call{l}\t{*}$dst", [(X86call GR32:$dst)], IIC_CALL_RI>, - Requires<[In32BitMode]>; + Requires<[In32BitMode]>, Sched<[WriteJump]>; def CALL32m : I<0xFF, MRM2m, (outs), (ins i32mem:$dst), - "call{l}\t{*}$dst", [(X86call (loadi32 addr:$dst))], IIC_CALL_MEM>, - Requires<[In32BitMode]>; + "call{l}\t{*}$dst", [(X86call (loadi32 addr:$dst))], + IIC_CALL_MEM>, + Requires<[In32BitMode,FavorMemIndirectCall]>, + Sched<[WriteJumpLd]>; def FARCALL16i : Iseg16<0x9A, RawFrmImm16, (outs), (ins i16imm:$off, i16imm:$seg), "lcall{w}\t{$seg, $off|$off, $seg}", [], - IIC_CALL_FAR_PTR>, OpSize; + IIC_CALL_FAR_PTR>, OpSize, Sched<[WriteJump]>; def FARCALL32i : Iseg32<0x9A, RawFrmImm16, (outs), (ins i32imm:$off, i16imm:$seg), "lcall{l}\t{$seg, $off|$off, $seg}", [], - IIC_CALL_FAR_PTR>; + IIC_CALL_FAR_PTR>, Sched<[WriteJump]>; def FARCALL16m : I<0xFF, MRM3m, (outs), (ins opaque32mem:$dst), - "lcall{w}\t{*}$dst", [], IIC_CALL_FAR_MEM>, OpSize; + "lcall{w}\t{*}$dst", [], IIC_CALL_FAR_MEM>, OpSize, + Sched<[WriteJumpLd]>; def FARCALL32m : I<0xFF, MRM3m, (outs), (ins opaque48mem:$dst), - "lcall{l}\t{*}$dst", [], IIC_CALL_FAR_MEM>; + "lcall{l}\t{*}$dst", [], IIC_CALL_FAR_MEM>, + Sched<[WriteJumpLd]>; // callw for 16 bit code for the assembler. let isAsmParserOnly = 1 in @@ -185,7 +200,7 @@ let isCall = 1 in // Tail call stuff. let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, - isCodeGenOnly = 1 in + isCodeGenOnly = 1, SchedRW = [WriteJumpLd] in let Uses = [ESP] in { def TCRETURNdi : PseudoI<(outs), (ins i32imm_pcrel:$dst, i32imm:$offset), []>; @@ -216,7 +231,7 @@ let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, // RSP is marked as a use to prevent stack-pointer assignments that appear // immediately before calls from potentially appearing dead. Uses for argument // registers are added manually. -let isCall = 1, Uses = [RSP] in { +let isCall = 1, Uses = [RSP], SchedRW = [WriteJump] in { // NOTE: this pattern doesn't match "X86call imm", because we do not know // that the offset between an arbitrary immediate and the call will fit in // the 32-bit pcrel field that we have. @@ -231,7 +246,7 @@ let isCall = 1, Uses = [RSP] in { def CALL64m : I<0xFF, MRM2m, (outs), (ins i64mem:$dst), "call{q}\t{*}$dst", [(X86call (loadi64 addr:$dst))], IIC_CALL_MEM>, - Requires<[In64BitMode]>; + Requires<[In64BitMode,FavorMemIndirectCall]>; def FARCALL64 : RI<0xFF, MRM3m, (outs), (ins opaque80mem:$dst), "lcall{q}\t{*}$dst", [], IIC_CALL_FAR_MEM>; @@ -245,13 +260,12 @@ let isCall = 1, isCodeGenOnly = 1 in def W64ALLOCA : Ii32PCRel<0xE8, RawFrm, (outs), (ins i64i32imm_pcrel:$dst), "call{q}\t$dst", [], IIC_CALL_RI>, - Requires<[IsWin64]>; + Requires<[IsWin64]>, Sched<[WriteJump]>; } let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1, - isCodeGenOnly = 1 in - let Uses = [RSP], - usesCustomInserter = 1 in { + isCodeGenOnly = 1, Uses = [RSP], usesCustomInserter = 1, + SchedRW = [WriteJump] in { def TCRETURNdi64 : PseudoI<(outs), (ins i64i32imm_pcrel:$dst, i32imm:$offset), []>; diff --git a/lib/Target/X86/X86InstrExtension.td b/lib/Target/X86/X86InstrExtension.td index 2eb454d..6dc7175 100644 --- a/lib/Target/X86/X86InstrExtension.td +++ b/lib/Target/X86/X86InstrExtension.td @@ -42,48 +42,54 @@ let neverHasSideEffects = 1 in { let neverHasSideEffects = 1 in { def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src), "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_R8>, - TB, OpSize; + TB, OpSize, Sched<[WriteALU]>; let mayLoad = 1 in def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src), "movs{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVSX_R16_M8>, - TB, OpSize; + TB, OpSize, Sched<[WriteALULd]>; } // neverHasSideEffects = 1 def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8:$src), "movs{bl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB; + [(set GR32:$dst, (sext GR8:$src))], IIC_MOVSX>, TB, + Sched<[WriteALU]>; def MOVSX32rm8 : I<0xBE, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src), "movs{bl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB; + [(set GR32:$dst, (sextloadi32i8 addr:$src))], IIC_MOVSX>, TB, + Sched<[WriteALULd]>; def MOVSX32rr16: I<0xBF, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src), "movs{wl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB; + [(set GR32:$dst, (sext GR16:$src))], IIC_MOVSX>, TB, + Sched<[WriteALU]>; def MOVSX32rm16: I<0xBF, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src), "movs{wl|x}\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (sextloadi32i16 addr:$src))], IIC_MOVSX>, - TB; + TB, Sched<[WriteALULd]>; let neverHasSideEffects = 1 in { def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8:$src), "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_R8>, - TB, OpSize; + TB, OpSize, Sched<[WriteALU]>; let mayLoad = 1 in def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem:$src), "movz{bw|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX_R16_M8>, - TB, OpSize; + TB, OpSize, Sched<[WriteALULd]>; } // neverHasSideEffects = 1 def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src), "movz{bl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB; + [(set GR32:$dst, (zext GR8:$src))], IIC_MOVZX>, TB, + Sched<[WriteALU]>; def MOVZX32rm8 : I<0xB6, MRMSrcMem, (outs GR32:$dst), (ins i8mem :$src), "movz{bl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB; + [(set GR32:$dst, (zextloadi32i8 addr:$src))], IIC_MOVZX>, TB, + Sched<[WriteALULd]>; def MOVZX32rr16: I<0xB7, MRMSrcReg, (outs GR32:$dst), (ins GR16:$src), "movz{wl|x}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB; + [(set GR32:$dst, (zext GR16:$src))], IIC_MOVZX>, TB, + Sched<[WriteALU]>; def MOVZX32rm16: I<0xB7, MRMSrcMem, (outs GR32:$dst), (ins i16mem:$src), "movz{wl|x}\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (zextloadi32i16 addr:$src))], IIC_MOVZX>, - TB; + TB, Sched<[WriteALULd]>; // These are the same as the regular MOVZX32rr8 and MOVZX32rm8 // except that they use GR32_NOREX for the output operand register class @@ -92,12 +98,12 @@ let neverHasSideEffects = 1, isCodeGenOnly = 1 in { def MOVZX32_NOREXrr8 : I<0xB6, MRMSrcReg, (outs GR32_NOREX:$dst), (ins GR8_NOREX:$src), "movz{bl|x}\t{$src, $dst|$dst, $src}", - [], IIC_MOVZX>, TB; + [], IIC_MOVZX>, TB, Sched<[WriteALU]>; let mayLoad = 1 in def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem, (outs GR32_NOREX:$dst), (ins i8mem_NOREX:$src), "movz{bl|x}\t{$src, $dst|$dst, $src}", - [], IIC_MOVZX>, TB; + [], IIC_MOVZX>, TB, Sched<[WriteALULd]>; } // MOVSX64rr8 always has a REX prefix and it has an 8-bit register @@ -106,38 +112,42 @@ def MOVZX32_NOREXrm8 : I<0xB6, MRMSrcMem, // were generalized, this would require a special register class. def MOVSX64rr8 : RI<0xBE, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src), "movs{bq|x}\t{$src, $dst|$dst, $src}", - [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB; + [(set GR64:$dst, (sext GR8:$src))], IIC_MOVSX>, TB, + Sched<[WriteALU]>; def MOVSX64rm8 : RI<0xBE, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src), "movs{bq|x}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (sextloadi64i8 addr:$src))], IIC_MOVSX>, - TB; + TB, Sched<[WriteALULd]>; def MOVSX64rr16: RI<0xBF, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src), "movs{wq|x}\t{$src, $dst|$dst, $src}", - [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB; + [(set GR64:$dst, (sext GR16:$src))], IIC_MOVSX>, TB, + Sched<[WriteALU]>; def MOVSX64rm16: RI<0xBF, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src), "movs{wq|x}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (sextloadi64i16 addr:$src))], IIC_MOVSX>, - TB; + TB, Sched<[WriteALULd]>; def MOVSX64rr32: RI<0x63, MRMSrcReg, (outs GR64:$dst), (ins GR32:$src), "movs{lq|xd}\t{$src, $dst|$dst, $src}", - [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>; + [(set GR64:$dst, (sext GR32:$src))], IIC_MOVSX>, + Sched<[WriteALU]>; def MOVSX64rm32: RI<0x63, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src), "movs{lq|xd}\t{$src, $dst|$dst, $src}", - [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>; + [(set GR64:$dst, (sextloadi64i32 addr:$src))], IIC_MOVSX>, + Sched<[WriteALULd]>; // movzbq and movzwq encodings for the disassembler def MOVZX64rr8_Q : RI<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8:$src), "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>, - TB; + TB, Sched<[WriteALU]>; def MOVZX64rm8_Q : RI<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem:$src), "movz{bq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>, - TB; + TB, Sched<[WriteALULd]>; def MOVZX64rr16_Q : RI<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src), "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>, - TB; + TB, Sched<[WriteALU]>; def MOVZX64rm16_Q : RI<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src), "movz{wq|x}\t{$src, $dst|$dst, $src}", [], IIC_MOVZX>, - TB; + TB, Sched<[WriteALULd]>; // FIXME: These should be Pat patterns. let isCodeGenOnly = 1 in { @@ -145,17 +155,19 @@ let isCodeGenOnly = 1 in { // Use movzbl instead of movzbq when the destination is a register; it's // equivalent due to implicit zero-extending, and it has a smaller encoding. def MOVZX64rr8 : I<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src), - "", [(set GR64:$dst, (zext GR8:$src))], IIC_MOVZX>, TB; + "", [(set GR64:$dst, (zext GR8:$src))], IIC_MOVZX>, TB, + Sched<[WriteALU]>; def MOVZX64rm8 : I<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src), "", [(set GR64:$dst, (zextloadi64i8 addr:$src))], IIC_MOVZX>, - TB; + TB, Sched<[WriteALULd]>; // Use movzwl instead of movzwq when the destination is a register; it's // equivalent due to implicit zero-extending, and it has a smaller encoding. def MOVZX64rr16: I<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src), - "", [(set GR64:$dst, (zext GR16:$src))], IIC_MOVZX>, TB; + "", [(set GR64:$dst, (zext GR16:$src))], IIC_MOVZX>, TB, + Sched<[WriteALU]>; def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src), "", [(set GR64:$dst, (zextloadi64i16 addr:$src))], - IIC_MOVZX>, TB; + IIC_MOVZX>, TB, Sched<[WriteALULd]>; // There's no movzlq instruction, but movl can be used for this purpose, using // implicit zero-extension. The preferred way to do 32-bit-to-64-bit zero @@ -165,9 +177,10 @@ def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src), // necessarily all zero. In such cases, we fall back to these explicit zext // instructions. def MOVZX64rr32 : I<0x89, MRMDestReg, (outs GR64:$dst), (ins GR32:$src), - "", [(set GR64:$dst, (zext GR32:$src))], IIC_MOVZX>; + "", [(set GR64:$dst, (zext GR32:$src))], IIC_MOVZX>, + Sched<[WriteALU]>; def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src), "", [(set GR64:$dst, (zextloadi64i32 addr:$src))], - IIC_MOVZX>; + IIC_MOVZX>, Sched<[WriteALULd]>; } diff --git a/lib/Target/X86/X86InstrFPStack.td b/lib/Target/X86/X86InstrFPStack.td index 568726e..2224a08 100644 --- a/lib/Target/X86/X86InstrFPStack.td +++ b/lib/Target/X86/X86InstrFPStack.td @@ -422,7 +422,7 @@ def IST_Fp32m80 : FpI_<(outs), (ins i32mem:$op, RFP80:$src), OneArgFP, []>; def IST_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP, []>; } -let mayLoad = 1 in { +let mayLoad = 1, SchedRW = [WriteLoad] in { def LD_F32m : FPI<0xD9, MRM0m, (outs), (ins f32mem:$src), "fld{s}\t$src", IIC_FLD>; def LD_F64m : FPI<0xDD, MRM0m, (outs), (ins f64mem:$src), "fld{l}\t$src", @@ -436,7 +436,7 @@ def ILD_F32m : FPI<0xDB, MRM0m, (outs), (ins i32mem:$src), "fild{l}\t$src", def ILD_F64m : FPI<0xDF, MRM5m, (outs), (ins i64mem:$src), "fild{ll}\t$src", IIC_FILD>; } -let mayStore = 1 in { +let mayStore = 1, SchedRW = [WriteStore] in { def ST_F32m : FPI<0xD9, MRM2m, (outs), (ins f32mem:$dst), "fst{s}\t$dst", IIC_FST>; def ST_F64m : FPI<0xDD, MRM2m, (outs), (ins f64mem:$dst), "fst{l}\t$dst", @@ -481,7 +481,7 @@ def ISTT_Fp64m80 : FpI_<(outs), (ins i64mem:$op, RFP80:$src), OneArgFP, [(X86fp_to_i64mem RFP80:$src, addr:$op)]>; } // Predicates = [HasSSE3] -let mayStore = 1 in { +let mayStore = 1, SchedRW = [WriteStore] in { def ISTT_FP16m : FPI<0xDF, MRM1m, (outs), (ins i16mem:$dst), "fisttp{s}\t$dst", IIC_FST>; def ISTT_FP32m : FPI<0xDB, MRM1m, (outs), (ins i32mem:$dst), "fisttp{l}\t$dst", @@ -491,6 +491,7 @@ def ISTT_FP64m : FPI<0xDD, MRM1m, (outs), (ins i64mem:$dst), } // FP Stack manipulation instructions. +let SchedRW = [WriteMove] in { def LD_Frr : FPI<0xC0, AddRegFrm, (outs), (ins RST:$op), "fld\t$op", IIC_FLD>, D9; def ST_Frr : FPI<0xD0, AddRegFrm, (outs), (ins RST:$op), "fst\t$op", @@ -499,6 +500,7 @@ def ST_FPrr : FPI<0xD8, AddRegFrm, (outs), (ins RST:$op), "fstp\t$op", IIC_FST>, DD; def XCH_F : FPI<0xC8, AddRegFrm, (outs), (ins RST:$op), "fxch\t$op", IIC_FXCH>, D9; +} // Floating point constant loads. let isReMaterializable = 1 in { @@ -516,19 +518,23 @@ def LD_Fp180 : FpI_<(outs RFP80:$dst), (ins), ZeroArgFP, [(set RFP80:$dst, fpimm1)]>; } +let SchedRW = [WriteZero] in { def LD_F0 : FPI<0xEE, RawFrm, (outs), (ins), "fldz", IIC_FLDZ>, D9; def LD_F1 : FPI<0xE8, RawFrm, (outs), (ins), "fld1", IIC_FIST>, D9; - +} // Floating point compares. +let SchedRW = [WriteFAdd] in { def UCOM_Fpr32 : FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP, [(set FPSW, (trunc (X86cmp RFP32:$lhs, RFP32:$rhs)))]>; def UCOM_Fpr64 : FpIf64<(outs), (ins RFP64:$lhs, RFP64:$rhs), CompareFP, [(set FPSW, (trunc (X86cmp RFP64:$lhs, RFP64:$rhs)))]>; def UCOM_Fpr80 : FpI_ <(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP, [(set FPSW, (trunc (X86cmp RFP80:$lhs, RFP80:$rhs)))]>; +} // SchedRW } // Defs = [FPSW] +let SchedRW = [WriteFAdd] in { // CC = ST(0) cmp ST(i) let Defs = [EFLAGS, FPSW] in { def UCOM_FpIr32: FpIf32<(outs), (ins RFP32:$lhs, RFP32:$rhs), CompareFP, @@ -566,8 +572,10 @@ def COM_FIr : FPI<0xF0, AddRegFrm, (outs), (ins RST:$reg), def COM_FIPr : FPI<0xF0, AddRegFrm, (outs), (ins RST:$reg), "fcompi\t$reg", IIC_FCOMI>, DF; } +} // SchedRW // Floating point flag ops. +let SchedRW = [WriteALU] in { let Defs = [AX], Uses = [FPSW] in def FNSTSW16r : I<0xE0, RawFrm, // AX = fp flags (outs), (ins), "fnstsw %ax", @@ -576,23 +584,26 @@ def FNSTSW16r : I<0xE0, RawFrm, // AX = fp flags def FNSTCW16m : I<0xD9, MRM7m, // [mem16] = X87 control world (outs), (ins i16mem:$dst), "fnstcw\t$dst", [(X86fp_cwd_get16 addr:$dst)], IIC_FNSTCW>; - +} // SchedRW let mayLoad = 1 in def FLDCW16m : I<0xD9, MRM5m, // X87 control world = [mem16] - (outs), (ins i16mem:$dst), "fldcw\t$dst", [], IIC_FLDCW>; + (outs), (ins i16mem:$dst), "fldcw\t$dst", [], IIC_FLDCW>, + Sched<[WriteLoad]>; // FPU control instructions +let SchedRW = [WriteMicrocoded] in { let Defs = [FPSW] in def FNINIT : I<0xE3, RawFrm, (outs), (ins), "fninit", [], IIC_FNINIT>, DB; def FFREE : FPI<0xC0, AddRegFrm, (outs), (ins RST:$reg), "ffree\t$reg", IIC_FFREE>, DD; - // Clear exceptions let Defs = [FPSW] in def FNCLEX : I<0xE2, RawFrm, (outs), (ins), "fnclex", [], IIC_FNCLEX>, DB; +} // SchedRW // Operandless floating-point instructions for the disassembler. +let SchedRW = [WriteMicrocoded] in { def WAIT : I<0x9B, RawFrm, (outs), (ins), "wait", [], IIC_WAIT>; def FNOP : I<0xD0, RawFrm, (outs), (ins), "fnop", [], IIC_FNOP>, D9; @@ -627,6 +638,7 @@ def FXRSTOR : I<0xAE, MRM1m, (outs), (ins opaque512mem:$src), def FXRSTOR64 : I<0xAE, MRM1m, (outs), (ins opaque512mem:$src), "fxrstorq\t$src", [], IIC_FXRSTOR>, TB, REX_W, Requires<[In64BitMode]>; +} // SchedRW //===----------------------------------------------------------------------===// // Non-Instruction Patterns diff --git a/lib/Target/X86/X86InstrFormats.td b/lib/Target/X86/X86InstrFormats.td index 44e574d..a71e024 100644 --- a/lib/Target/X86/X86InstrFormats.td +++ b/lib/Target/X86/X86InstrFormats.td @@ -35,24 +35,27 @@ def MRM_C3 : Format<35>; def MRM_C4 : Format<36>; def MRM_C8 : Format<37>; def MRM_C9 : Format<38>; -def MRM_E8 : Format<39>; -def MRM_F0 : Format<40>; -def MRM_F8 : Format<41>; -def MRM_F9 : Format<42>; +def MRM_CA : Format<39>; +def MRM_CB : Format<40>; +def MRM_E8 : Format<41>; +def MRM_F0 : Format<42>; def RawFrmImm8 : Format<43>; def RawFrmImm16 : Format<44>; -def MRM_D0 : Format<45>; -def MRM_D1 : Format<46>; -def MRM_D4 : Format<47>; -def MRM_D5 : Format<48>; -def MRM_D8 : Format<49>; -def MRM_D9 : Format<50>; -def MRM_DA : Format<51>; -def MRM_DB : Format<52>; -def MRM_DC : Format<53>; -def MRM_DD : Format<54>; -def MRM_DE : Format<55>; -def MRM_DF : Format<56>; +def MRM_F8 : Format<45>; +def MRM_F9 : Format<46>; +def MRM_D0 : Format<47>; +def MRM_D1 : Format<48>; +def MRM_D4 : Format<49>; +def MRM_D5 : Format<50>; +def MRM_D6 : Format<51>; +def MRM_D8 : Format<52>; +def MRM_D9 : Format<53>; +def MRM_DA : Format<54>; +def MRM_DB : Format<55>; +def MRM_DC : Format<56>; +def MRM_DD : Format<57>; +def MRM_DE : Format<58>; +def MRM_DF : Format<59>; // ImmType - This specifies the immediate type used by an instruction. This is // part of the ad-hoc solution used to emit machine instruction encodings by our @@ -208,47 +211,47 @@ class PseudoI<dag oops, dag iops, list<dag> pattern> } class I<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT, + list<dag> pattern, InstrItinClass itin = NoItinerary, Domain d = GenericDomain> : X86Inst<o, f, NoImm, outs, ins, asm, itin, d> { let Pattern = pattern; let CodeSize = 3; } class Ii8 <bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT, + list<dag> pattern, InstrItinClass itin = NoItinerary, Domain d = GenericDomain> : X86Inst<o, f, Imm8, outs, ins, asm, itin, d> { let Pattern = pattern; let CodeSize = 3; } class Ii8PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm8PCRel, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; } class Ii16<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm16, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; } class Ii32<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm32, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; } class Ii16PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm16PCRel, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; } class Ii32PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm32PCRel, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; @@ -257,12 +260,12 @@ class Ii32PCRel<bits<8> o, Format f, dag outs, dag ins, string asm, // FPStack Instruction Templates: // FPI - Floating Point Instruction template. class FPI<bits<8> o, Format F, dag outs, dag ins, string asm, - InstrItinClass itin = IIC_DEFAULT> + InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, [], itin> {} // FpI_ - Floating Point Pseudo Instruction template. Not Predicated. class FpI_<dag outs, dag ins, FPFormat fp, list<dag> pattern, - InstrItinClass itin = IIC_DEFAULT> + InstrItinClass itin = NoItinerary> : X86Inst<0, Pseudo, NoImm, outs, ins, "", itin> { let FPForm = fp; let Pattern = pattern; @@ -275,14 +278,14 @@ class FpI_<dag outs, dag ins, FPFormat fp, list<dag> pattern, // Iseg32 - 16-bit segment selector, 32-bit offset class Iseg16 <bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm16, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; } class Iseg32 <bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm32, outs, ins, asm, itin> { let Pattern = pattern; let CodeSize = 3; @@ -292,7 +295,7 @@ def __xs : XS; // SI - SSE 1 & 2 scalar instructions class SI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin> { let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX], !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2])); @@ -303,7 +306,7 @@ class SI<bits<8> o, Format F, dag outs, dag ins, string asm, // SIi8 - SSE 1 & 2 scalar instructions class SIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin> { let Predicates = !if(hasVEXPrefix /* VEX */, [HasAVX], !if(!eq(Prefix, __xs.Prefix), [UseSSE1], [UseSSE2])); @@ -350,25 +353,25 @@ class PIi8<bits<8> o, Format F, dag outs, dag ins, string asm, // VPSI - SSE1 instructions with TB prefix in AVX form. class SSI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE1]>; class SSIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE1]>; class PSI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, TB, Requires<[UseSSE1]>; class PSIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, TB, Requires<[UseSSE1]>; class VSSI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XS, Requires<[HasAVX]>; class VPSI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin, SSEPackedSingle>, TB, Requires<[HasAVX]>; @@ -388,42 +391,42 @@ class VPSI<bits<8> o, Format F, dag outs, dag ins, string asm, // MMX operands. class SDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[UseSSE2]>; class SDIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[UseSSE2]>; class S2SI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[UseSSE2]>; class S2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern>, XS, Requires<[UseSSE2]>; class PDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize, Requires<[UseSSE2]>; class PDIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize, Requires<[UseSSE2]>; class VSDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XD, Requires<[HasAVX]>; class VS2SI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin>, XS, Requires<[HasAVX]>; class VPDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, !strconcat("v", asm), pattern, itin, SSEPackedDouble>, TB, OpSize, Requires<[HasAVX]>; class MMXSDIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[HasSSE2]>; class MMXS2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern>, XS, Requires<[HasSSE2]>; // SSE3 Instruction Templates: @@ -433,15 +436,15 @@ class MMXS2SIi8<bits<8> o, Format F, dag outs, dag ins, string asm, // S3DI - SSE3 instructions with XD prefix. class S3SI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedSingle>, XS, Requires<[UseSSE3]>; class S3DI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, XD, Requires<[UseSSE3]>; class S3I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, TB, OpSize, Requires<[UseSSE3]>; @@ -458,19 +461,19 @@ class S3I<bits<8> o, Format F, dag outs, dag ins, string asm, // classes. They need to be enabled even if AVX is enabled. class SS38I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, Requires<[UseSSSE3]>; class SS3AI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, Requires<[UseSSSE3]>; class MMXSS38I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, Requires<[HasSSSE3]>; class MMXSS3AI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, Requires<[HasSSSE3]>; @@ -480,11 +483,11 @@ class MMXSS3AI<bits<8> o, Format F, dag outs, dag ins, string asm, // SS41AIi8 - SSE 4.1 instructions with TA prefix and ImmT == Imm8. // class SS48I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, Requires<[UseSSE41]>; class SS4AIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, Requires<[UseSSE41]>; @@ -492,19 +495,19 @@ class SS4AIi8<bits<8> o, Format F, dag outs, dag ins, string asm, // // SS428I - SSE 4.2 instructions with T8 prefix. class SS428I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, Requires<[UseSSE42]>; // SS42FI - SSE 4.2 instructions with T8XD prefix. // NOTE: 'HasSSE42' is used as SS42FI is only used for CRC32 insns. class SS42FI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, T8XD, Requires<[HasSSE42]>; // SS42AI = SSE 4.2 instructions with TA prefix class SS42AI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, Requires<[UseSSE42]>; @@ -514,11 +517,11 @@ class SS42AI<bits<8> o, Format F, dag outs, dag ins, string asm, // AVX8I - AVX instructions with T8 and OpSize prefix. // AVXAIi8 - AVX instructions with TA, OpSize prefix and ImmT = Imm8. class AVX8I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, OpSize, Requires<[HasAVX]>; class AVXAIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize, Requires<[HasAVX]>; @@ -528,11 +531,11 @@ class AVXAIi8<bits<8> o, Format F, dag outs, dag ins, string asm, // AVX28I - AVX2 instructions with T8 and OpSize prefix. // AVX2AIi8 - AVX2 instructions with TA, OpSize prefix and ImmT = Imm8. class AVX28I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, OpSize, Requires<[HasAVX2]>; class AVX2AIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize, Requires<[HasAVX2]>; @@ -541,53 +544,53 @@ class AVX2AIi8<bits<8> o, Format F, dag outs, dag ins, string asm, // AES8I // These use the same encoding as the SSE4.2 T8 and TA encodings. class AES8I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, T8, Requires<[HasAES]>; class AESAI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, Requires<[HasAES]>; // PCLMUL Instruction Templates class PCLMULIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize, Requires<[HasPCLMUL]>; class AVXPCLMULIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize, VEX_4V, Requires<[HasAVX, HasPCLMUL]>; // FMA3 Instruction Templates class FMA3<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, T8, - OpSize, VEX_4V, Requires<[HasFMA]>; + OpSize, VEX_4V, FMASC, Requires<[HasFMA]>; // FMA4 Instruction Templates class FMA4<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, TA, - OpSize, VEX_4V, VEX_I8IMM, Requires<[HasFMA4]>; + OpSize, VEX_4V, VEX_I8IMM, FMASC, Requires<[HasFMA4]>; // XOP 2, 3 and 4 Operand Instruction Template class IXOP<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, XOP, XOP9, Requires<[HasXOP]>; // XOP 2, 3 and 4 Operand Instruction Templates with imm byte class IXOPi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedDouble>, XOP, XOP8, Requires<[HasXOP]>; // XOP 5 operand instruction (VEX encoding!) class IXOP5<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag>pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag>pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin, SSEPackedInt>, TA, OpSize, VEX_4V, VEX_I8IMM, Requires<[HasXOP]>; @@ -595,33 +598,33 @@ class IXOP5<bits<8> o, Format F, dag outs, dag ins, string asm, // class RI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, REX_W; class RIi8 <bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, REX_W; class RIi32 <bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii32<o, F, outs, ins, asm, pattern, itin>, REX_W; class RIi64<bits<8> o, Format f, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : X86Inst<o, f, Imm64, outs, ins, asm, itin>, REX_W { let Pattern = pattern; let CodeSize = 3; } class RSSI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : SSI<o, F, outs, ins, asm, pattern, itin>, REX_W; class RSDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : SDI<o, F, outs, ins, asm, pattern, itin>, REX_W; class RPDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : PDI<o, F, outs, ins, asm, pattern, itin>, REX_W; class VRPDI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : VPDI<o, F, outs, ins, asm, pattern, itin>, VEX_W; // MMX Instruction templates @@ -635,23 +638,23 @@ class VRPDI<bits<8> o, Format F, dag outs, dag ins, string asm, // MMXID - MMX instructions with XD prefix. // MMXIS - MMX instructions with XS prefix. class MMXI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX]>; class MMXI64<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX,In64BitMode]>; class MMXRI<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, TB, REX_W, Requires<[HasMMX]>; class MMX2I<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : I<o, F, outs, ins, asm, pattern, itin>, TB, OpSize, Requires<[HasMMX]>; class MMXIi8<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, TB, Requires<[HasMMX]>; class MMXID<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, XD, Requires<[HasMMX]>; class MMXIS<bits<8> o, Format F, dag outs, dag ins, string asm, - list<dag> pattern, InstrItinClass itin = IIC_DEFAULT> + list<dag> pattern, InstrItinClass itin = NoItinerary> : Ii8<o, F, outs, ins, asm, pattern, itin>, XS, Requires<[HasMMX]>; diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp index 17714ac..7c0423f 100644 --- a/lib/Target/X86/X86InstrInfo.cpp +++ b/lib/Target/X86/X86InstrInfo.cpp @@ -3655,7 +3655,16 @@ X86InstrInfo::foldMemoryOperandImpl(MachineFunction &MF, const SmallVectorImpl<MachineOperand> &MOs, unsigned Size, unsigned Align) const { const DenseMap<unsigned, std::pair<unsigned,unsigned> > *OpcodeTablePtr = 0; + bool isCallRegIndirect = TM.getSubtarget<X86Subtarget>().callRegIndirect(); bool isTwoAddrFold = false; + + // Atom favors register form of call. So, we do not fold loads into calls + // when X86Subtarget is Atom. + if (isCallRegIndirect && + (MI->getOpcode() == X86::CALL32r || MI->getOpcode() == X86::CALL64r)) { + return NULL; + } + unsigned NumOps = MI->getDesc().getNumOperands(); bool isTwoAddr = NumOps > 1 && MI->getDesc().getOperandConstraint(1, MCOI::TIED_TO) != -1; @@ -4272,7 +4281,7 @@ X86InstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N, bool isAligned = (*MMOs.first) && (*MMOs.first)->getAlignment() >= Alignment; Load = DAG.getMachineNode(getLoadRegOpcode(0, RC, isAligned, TM), dl, - VT, MVT::Other, &AddrOps[0], AddrOps.size()); + VT, MVT::Other, AddrOps); NewNodes.push_back(Load); // Preserve memory reference information. @@ -4294,8 +4303,7 @@ X86InstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N, if (Load) BeforeOps.push_back(SDValue(Load, 0)); std::copy(AfterOps.begin(), AfterOps.end(), std::back_inserter(BeforeOps)); - SDNode *NewNode= DAG.getMachineNode(Opc, dl, VTs, &BeforeOps[0], - BeforeOps.size()); + SDNode *NewNode= DAG.getMachineNode(Opc, dl, VTs, BeforeOps); NewNodes.push_back(NewNode); // Emit the store instruction. @@ -4317,8 +4325,7 @@ X86InstrInfo::unfoldMemoryOperand(SelectionDAG &DAG, SDNode *N, (*MMOs.first)->getAlignment() >= Alignment; SDNode *Store = DAG.getMachineNode(getStoreRegOpcode(0, DstRC, isAligned, TM), - dl, MVT::Other, - &AddrOps[0], AddrOps.size()); + dl, MVT::Other, AddrOps); NewNodes.push_back(Store); // Preserve memory reference information. diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td index d989ec7..3380d8c 100644 --- a/lib/Target/X86/X86InstrInfo.td +++ b/lib/Target/X86/X86InstrInfo.td @@ -142,6 +142,9 @@ def X86sahf : SDNode<"X86ISD::SAHF", SDTX86sahf>; def X86rdrand : SDNode<"X86ISD::RDRAND", SDTX86rdrand, [SDNPHasChain, SDNPSideEffect]>; +def X86rdseed : SDNode<"X86ISD::RDSEED", SDTX86rdrand, + [SDNPHasChain, SDNPSideEffect]>; + def X86cas : SDNode<"X86ISD::LCMPXCHG_DAG", SDTX86cas, [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore, SDNPMayLoad, SDNPMemOperand]>; @@ -603,7 +606,12 @@ def HasLZCNT : Predicate<"Subtarget->hasLZCNT()">; def HasBMI : Predicate<"Subtarget->hasBMI()">; def HasBMI2 : Predicate<"Subtarget->hasBMI2()">; def HasRTM : Predicate<"Subtarget->hasRTM()">; +def HasHLE : Predicate<"Subtarget->hasHLE()">; +def HasTSX : Predicate<"Subtarget->hasRTM() || Subtarget->hasHLE()">; def HasADX : Predicate<"Subtarget->hasADX()">; +def HasPRFCHW : Predicate<"Subtarget->hasPRFCHW()">; +def HasRDSEED : Predicate<"Subtarget->hasRDSEED()">; +def HasPrefetchW : Predicate<"Subtarget->has3DNow() || Subtarget->hasPRFCHW()">; def FPStackf32 : Predicate<"!Subtarget->hasSSE1()">; def FPStackf64 : Predicate<"!Subtarget->hasSSE2()">; def HasCmpxchg16b: Predicate<"Subtarget->hasCmpxchg16b()">; @@ -626,6 +634,7 @@ def OptForSize : Predicate<"OptForSize">; def OptForSpeed : Predicate<"!OptForSize">; def FastBTMem : Predicate<"!Subtarget->isBTMemSlow()">; def CallImmAddr : Predicate<"Subtarget->IsLegalToCallImmediateAddr(TM)">; +def FavorMemIndirectCall : Predicate<"!Subtarget->callRegIndirect()">; //===----------------------------------------------------------------------===// // X86 Instruction Format Definitions. @@ -758,7 +767,7 @@ def trunc_su : PatFrag<(ops node:$src), (trunc node:$src), [{ // // Nop -let neverHasSideEffects = 1 in { +let neverHasSideEffects = 1, SchedRW = [WriteZero] in { def NOOP : I<0x90, RawFrm, (outs), (ins), "nop", [], IIC_NOP>; def NOOPW : I<0x1f, MRM0m, (outs), (ins i16mem:$zero), "nop{w}\t$zero", [], IIC_NOP>, TB, OpSize; @@ -769,8 +778,9 @@ let neverHasSideEffects = 1 in { // Constructing a stack frame. def ENTER : Ii16<0xC8, RawFrmImm8, (outs), (ins i16imm:$len, i8imm:$lvl), - "enter\t$len, $lvl", [], IIC_ENTER>; + "enter\t$len, $lvl", [], IIC_ENTER>, Sched<[WriteMicrocoded]>; +let SchedRW = [WriteALU] in { let Defs = [EBP, ESP], Uses = [EBP, ESP], mayLoad = 1, neverHasSideEffects=1 in def LEAVE : I<0xC9, RawFrm, (outs), (ins), "leave", [], IIC_LEAVE>, @@ -780,13 +790,14 @@ let Defs = [RBP,RSP], Uses = [RBP,RSP], mayLoad = 1, neverHasSideEffects = 1 in def LEAVE64 : I<0xC9, RawFrm, (outs), (ins), "leave", [], IIC_LEAVE>, Requires<[In64BitMode]>; +} // SchedRW //===----------------------------------------------------------------------===// // Miscellaneous Instructions. // let Defs = [ESP], Uses = [ESP], neverHasSideEffects=1 in { -let mayLoad = 1 in { +let mayLoad = 1, SchedRW = [WriteLoad] in { def POP16r : I<0x58, AddRegFrm, (outs GR16:$reg), (ins), "pop{w}\t$reg", [], IIC_POP_REG16>, OpSize; def POP32r : I<0x58, AddRegFrm, (outs GR32:$reg), (ins), "pop{l}\t$reg", [], @@ -803,9 +814,9 @@ def POP32rmm: I<0x8F, MRM0m, (outs i32mem:$dst), (ins), "pop{l}\t$dst", [], def POPF16 : I<0x9D, RawFrm, (outs), (ins), "popf{w}", [], IIC_POP_F>, OpSize; def POPF32 : I<0x9D, RawFrm, (outs), (ins), "popf{l|d}", [], IIC_POP_FD>, Requires<[In32BitMode]>; -} +} // mayLoad, SchedRW -let mayStore = 1 in { +let mayStore = 1, SchedRW = [WriteStore] in { def PUSH16r : I<0x50, AddRegFrm, (outs), (ins GR16:$reg), "push{w}\t$reg",[], IIC_PUSH_REG>, OpSize; def PUSH32r : I<0x50, AddRegFrm, (outs), (ins GR32:$reg), "push{l}\t$reg",[], @@ -832,29 +843,30 @@ def PUSHF16 : I<0x9C, RawFrm, (outs), (ins), "pushf{w}", [], IIC_PUSH_F>, def PUSHF32 : I<0x9C, RawFrm, (outs), (ins), "pushf{l|d}", [], IIC_PUSH_F>, Requires<[In32BitMode]>; -} +} // mayStore, SchedRW } let Defs = [RSP], Uses = [RSP], neverHasSideEffects=1 in { -let mayLoad = 1 in { +let mayLoad = 1, SchedRW = [WriteLoad] in { def POP64r : I<0x58, AddRegFrm, (outs GR64:$reg), (ins), "pop{q}\t$reg", [], IIC_POP_REG>; def POP64rmr: I<0x8F, MRM0r, (outs GR64:$reg), (ins), "pop{q}\t$reg", [], IIC_POP_REG>; def POP64rmm: I<0x8F, MRM0m, (outs i64mem:$dst), (ins), "pop{q}\t$dst", [], IIC_POP_MEM>; -} -let mayStore = 1 in { +} // mayLoad, SchedRW +let mayStore = 1, SchedRW = [WriteStore] in { def PUSH64r : I<0x50, AddRegFrm, (outs), (ins GR64:$reg), "push{q}\t$reg", [], IIC_PUSH_REG>; def PUSH64rmr: I<0xFF, MRM6r, (outs), (ins GR64:$reg), "push{q}\t$reg", [], IIC_PUSH_REG>; def PUSH64rmm: I<0xFF, MRM6m, (outs), (ins i64mem:$src), "push{q}\t$src", [], IIC_PUSH_MEM>; -} +} // mayStore, SchedRW } -let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1 in { +let Defs = [RSP], Uses = [RSP], neverHasSideEffects = 1, mayStore = 1, + SchedRW = [WriteStore] in { def PUSH64i8 : Ii8<0x6a, RawFrm, (outs), (ins i64i8imm:$imm), "push{q}\t$imm", [], IIC_PUSH_IMM>; def PUSH64i16 : Ii16<0x68, RawFrm, (outs), (ins i16imm:$imm), @@ -865,23 +877,24 @@ def PUSH64i32 : Ii32<0x68, RawFrm, (outs), (ins i64i32imm:$imm), let Defs = [RSP, EFLAGS], Uses = [RSP], mayLoad = 1, neverHasSideEffects=1 in def POPF64 : I<0x9D, RawFrm, (outs), (ins), "popfq", [], IIC_POP_FD>, - Requires<[In64BitMode]>; + Requires<[In64BitMode]>, Sched<[WriteLoad]>; let Defs = [RSP], Uses = [RSP, EFLAGS], mayStore = 1, neverHasSideEffects=1 in def PUSHF64 : I<0x9C, RawFrm, (outs), (ins), "pushfq", [], IIC_PUSH_F>, - Requires<[In64BitMode]>; + Requires<[In64BitMode]>, Sched<[WriteStore]>; let Defs = [EDI, ESI, EBP, EBX, EDX, ECX, EAX, ESP], Uses = [ESP], - mayLoad=1, neverHasSideEffects=1 in { + mayLoad = 1, neverHasSideEffects = 1, SchedRW = [WriteLoad] in { def POPA32 : I<0x61, RawFrm, (outs), (ins), "popa{l|d}", [], IIC_POP_A>, Requires<[In32BitMode]>; } let Defs = [ESP], Uses = [EDI, ESI, EBP, EBX, EDX, ECX, EAX, ESP], - mayStore=1, neverHasSideEffects=1 in { + mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in { def PUSHA32 : I<0x60, RawFrm, (outs), (ins), "pusha{l|d}", [], IIC_PUSH_A>, Requires<[In32BitMode]>; } -let Constraints = "$src = $dst" in { // GR32 = bswap GR32 +let Constraints = "$src = $dst", SchedRW = [WriteALU] in { +// GR32 = bswap GR32 def BSWAP32r : I<0xC8, AddRegFrm, (outs GR32:$dst), (ins GR32:$src), "bswap{l}\t$dst", @@ -890,60 +903,63 @@ def BSWAP32r : I<0xC8, AddRegFrm, def BSWAP64r : RI<0xC8, AddRegFrm, (outs GR64:$dst), (ins GR64:$src), "bswap{q}\t$dst", [(set GR64:$dst, (bswap GR64:$src))], IIC_BSWAP>, TB; -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW // Bit scan instructions. let Defs = [EFLAGS] in { def BSF16rr : I<0xBC, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src), "bsf{w}\t{$src, $dst|$dst, $src}", [(set GR16:$dst, EFLAGS, (X86bsf GR16:$src))], - IIC_BSF>, TB, OpSize; + IIC_BSF>, TB, OpSize, Sched<[WriteShift]>; def BSF16rm : I<0xBC, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), "bsf{w}\t{$src, $dst|$dst, $src}", [(set GR16:$dst, EFLAGS, (X86bsf (loadi16 addr:$src)))], - IIC_BSF>, TB, OpSize; + IIC_BSF>, TB, OpSize, Sched<[WriteShiftLd]>; def BSF32rr : I<0xBC, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src), "bsf{l}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, EFLAGS, (X86bsf GR32:$src))], IIC_BSF>, TB; + [(set GR32:$dst, EFLAGS, (X86bsf GR32:$src))], IIC_BSF>, TB, + Sched<[WriteShift]>; def BSF32rm : I<0xBC, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src), "bsf{l}\t{$src, $dst|$dst, $src}", [(set GR32:$dst, EFLAGS, (X86bsf (loadi32 addr:$src)))], - IIC_BSF>, TB; + IIC_BSF>, TB, Sched<[WriteShiftLd]>; def BSF64rr : RI<0xBC, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), "bsf{q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, EFLAGS, (X86bsf GR64:$src))], - IIC_BSF>, TB; + IIC_BSF>, TB, Sched<[WriteShift]>; def BSF64rm : RI<0xBC, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src), "bsf{q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, EFLAGS, (X86bsf (loadi64 addr:$src)))], - IIC_BSF>, TB; + IIC_BSF>, TB, Sched<[WriteShiftLd]>; def BSR16rr : I<0xBD, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src), "bsr{w}\t{$src, $dst|$dst, $src}", [(set GR16:$dst, EFLAGS, (X86bsr GR16:$src))], IIC_BSR>, - TB, OpSize; + TB, OpSize, Sched<[WriteShift]>; def BSR16rm : I<0xBD, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), "bsr{w}\t{$src, $dst|$dst, $src}", [(set GR16:$dst, EFLAGS, (X86bsr (loadi16 addr:$src)))], IIC_BSR>, TB, - OpSize; + OpSize, Sched<[WriteShiftLd]>; def BSR32rr : I<0xBD, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src), "bsr{l}\t{$src, $dst|$dst, $src}", - [(set GR32:$dst, EFLAGS, (X86bsr GR32:$src))], IIC_BSR>, TB; + [(set GR32:$dst, EFLAGS, (X86bsr GR32:$src))], IIC_BSR>, TB, + Sched<[WriteShift]>; def BSR32rm : I<0xBD, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src), "bsr{l}\t{$src, $dst|$dst, $src}", [(set GR32:$dst, EFLAGS, (X86bsr (loadi32 addr:$src)))], - IIC_BSR>, TB; + IIC_BSR>, TB, Sched<[WriteShiftLd]>; def BSR64rr : RI<0xBD, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), "bsr{q}\t{$src, $dst|$dst, $src}", - [(set GR64:$dst, EFLAGS, (X86bsr GR64:$src))], IIC_BSR>, TB; + [(set GR64:$dst, EFLAGS, (X86bsr GR64:$src))], IIC_BSR>, TB, + Sched<[WriteShift]>; def BSR64rm : RI<0xBD, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src), "bsr{q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, EFLAGS, (X86bsr (loadi64 addr:$src)))], - IIC_BSR>, TB; + IIC_BSR>, TB, Sched<[WriteShiftLd]>; } // Defs = [EFLAGS] - +let SchedRW = [WriteMicrocoded] in { // These uses the DF flag in the EFLAGS register to inc or dec EDI and ESI let Defs = [EDI,ESI], Uses = [EDI,ESI,EFLAGS] in { def MOVSB : I<0xA4, RawFrm, (outs), (ins), "movsb", [], IIC_MOVS>; @@ -971,12 +987,12 @@ def CMPS8 : I<0xA6, RawFrm, (outs), (ins), "cmpsb", [], IIC_CMPS>; def CMPS16 : I<0xA7, RawFrm, (outs), (ins), "cmpsw", [], IIC_CMPS>, OpSize; def CMPS32 : I<0xA7, RawFrm, (outs), (ins), "cmps{l|d}", [], IIC_CMPS>; def CMPS64 : RI<0xA7, RawFrm, (outs), (ins), "cmpsq", [], IIC_CMPS>; - +} // SchedRW //===----------------------------------------------------------------------===// // Move Instructions. // - +let SchedRW = [WriteMove] in { let neverHasSideEffects = 1 in { def MOV8rr : I<0x88, MRMDestReg, (outs GR8 :$dst), (ins GR8 :$src), "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; @@ -987,6 +1003,7 @@ def MOV32rr : I<0x89, MRMDestReg, (outs GR32:$dst), (ins GR32:$src), def MOV64rr : RI<0x89, MRMDestReg, (outs GR64:$dst), (ins GR64:$src), "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; } + let isReMaterializable = 1, isAsCheapAsAMove = 1 in { def MOV8ri : Ii8 <0xB0, AddRegFrm, (outs GR8 :$dst), (ins i8imm :$src), "mov{b}\t{$src, $dst|$dst, $src}", @@ -1004,7 +1021,9 @@ def MOV64ri32 : RIi32<0xC7, MRM0r, (outs GR64:$dst), (ins i64i32imm:$src), "mov{q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, i64immSExt32:$src)], IIC_MOV>; } +} // SchedRW +let SchedRW = [WriteStore] in { def MOV8mi : Ii8 <0xC6, MRM0m, (outs), (ins i8mem :$dst, i8imm :$src), "mov{b}\t{$src, $dst|$dst, $src}", [(store (i8 imm:$src), addr:$dst)], IIC_MOV_MEM>; @@ -1017,9 +1036,11 @@ def MOV32mi : Ii32<0xC7, MRM0m, (outs), (ins i32mem:$dst, i32imm:$src), def MOV64mi32 : RIi32<0xC7, MRM0m, (outs), (ins i64mem:$dst, i64i32imm:$src), "mov{q}\t{$src, $dst|$dst, $src}", [(store i64immSExt32:$src, addr:$dst)], IIC_MOV_MEM>; +} // SchedRW /// moffs8, moffs16 and moffs32 versions of moves. The immediate is a /// 32-bit offset from the PC. These are only valid in x86-32 mode. +let SchedRW = [WriteALU] in { def MOV8o8a : Ii32 <0xA0, RawFrm, (outs), (ins offset8:$src), "mov{b}\t{$src, %al|AL, $src}", [], IIC_MOV_MEM>, Requires<[In32BitMode]>; @@ -1038,6 +1059,7 @@ def MOV16ao16 : Ii32 <0xA3, RawFrm, (outs offset16:$dst), (ins), def MOV32ao32 : Ii32 <0xA3, RawFrm, (outs offset32:$dst), (ins), "mov{l}\t{%eax, $dst|$dst, EAX}", [], IIC_MOV_MEM>, Requires<[In32BitMode]>; +} // FIXME: These definitions are utterly broken // Just leave them commented out for now because they're useless outside @@ -1055,7 +1077,7 @@ def MOV64ao64 : RIi32<0xA3, RawFrm, (outs offset64:$dst), (ins), */ -let isCodeGenOnly = 1, hasSideEffects = 0 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def MOV8rr_REV : I<0x8A, MRMSrcReg, (outs GR8:$dst), (ins GR8:$src), "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; def MOV16rr_REV : I<0x8B, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src), @@ -1066,7 +1088,7 @@ def MOV64rr_REV : RI<0x8B, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; } -let canFoldAsLoad = 1, isReMaterializable = 1 in { +let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in { def MOV8rm : I<0x8A, MRMSrcMem, (outs GR8 :$dst), (ins i8mem :$src), "mov{b}\t{$src, $dst|$dst, $src}", [(set GR8:$dst, (loadi8 addr:$src))], IIC_MOV_MEM>; @@ -1081,6 +1103,7 @@ def MOV64rm : RI<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src), [(set GR64:$dst, (load addr:$src))], IIC_MOV_MEM>; } +let SchedRW = [WriteStore] in { def MOV8mr : I<0x88, MRMDestMem, (outs), (ins i8mem :$dst, GR8 :$src), "mov{b}\t{$src, $dst|$dst, $src}", [(store GR8:$src, addr:$dst)], IIC_MOV_MEM>; @@ -1093,6 +1116,7 @@ def MOV32mr : I<0x89, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src), def MOV64mr : RI<0x89, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src), "mov{q}\t{$src, $dst|$dst, $src}", [(store GR64:$src, addr:$dst)], IIC_MOV_MEM>; +} // SchedRW // Versions of MOV8rr, MOV8mr, and MOV8rm that use i8mem_NOREX and GR8_NOREX so // that they can be used for copying and storing h registers, which can't be @@ -1101,34 +1125,37 @@ let isCodeGenOnly = 1 in { let neverHasSideEffects = 1 in def MOV8rr_NOREX : I<0x88, MRMDestReg, (outs GR8_NOREX:$dst), (ins GR8_NOREX:$src), - "mov{b}\t{$src, $dst|$dst, $src} # NOREX", [], IIC_MOV>; + "mov{b}\t{$src, $dst|$dst, $src} # NOREX", [], IIC_MOV>, + Sched<[WriteMove]>; let mayStore = 1 in def MOV8mr_NOREX : I<0x88, MRMDestMem, (outs), (ins i8mem_NOREX:$dst, GR8_NOREX:$src), "mov{b}\t{$src, $dst|$dst, $src} # NOREX", [], - IIC_MOV_MEM>; + IIC_MOV_MEM>, Sched<[WriteStore]>; let mayLoad = 1, neverHasSideEffects = 1, canFoldAsLoad = 1, isReMaterializable = 1 in def MOV8rm_NOREX : I<0x8A, MRMSrcMem, (outs GR8_NOREX:$dst), (ins i8mem_NOREX:$src), "mov{b}\t{$src, $dst|$dst, $src} # NOREX", [], - IIC_MOV_MEM>; + IIC_MOV_MEM>, Sched<[WriteLoad]>; } // Condition code ops, incl. set if equal/not equal/... +let SchedRW = [WriteALU] in { let Defs = [EFLAGS], Uses = [AH] in def SAHF : I<0x9E, RawFrm, (outs), (ins), "sahf", [(set EFLAGS, (X86sahf AH))], IIC_AHF>; let Defs = [AH], Uses = [EFLAGS], neverHasSideEffects = 1 in def LAHF : I<0x9F, RawFrm, (outs), (ins), "lahf", [], IIC_AHF>; // AH = flags - +} // SchedRW //===----------------------------------------------------------------------===// // Bit tests instructions: BT, BTS, BTR, BTC. let Defs = [EFLAGS] in { +let SchedRW = [WriteALU] in { def BT16rr : I<0xA3, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2), "bt{w}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt GR16:$src1, GR16:$src2))], IIC_BT_RR>, @@ -1139,13 +1166,14 @@ def BT32rr : I<0xA3, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2), def BT64rr : RI<0xA3, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2), "bt{q}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt GR64:$src1, GR64:$src2))], IIC_BT_RR>, TB; +} // SchedRW // Unlike with the register+register form, the memory+register form of the // bt instruction does not ignore the high bits of the index. From ISel's // perspective, this is pretty bizarre. Make these instructions disassembly // only for now. -let mayLoad = 1, hasSideEffects = 0 in { +let mayLoad = 1, hasSideEffects = 0, SchedRW = [WriteALULd] in { def BT16mr : I<0xA3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2), "bt{w}\t{$src2, $src1|$src1, $src2}", // [(X86bt (loadi16 addr:$src1), GR16:$src2), @@ -1166,6 +1194,7 @@ let mayLoad = 1, hasSideEffects = 0 in { >, TB; } +let SchedRW = [WriteALU] in { def BT16ri8 : Ii8<0xBA, MRM4r, (outs), (ins GR16:$src1, i16i8imm:$src2), "bt{w}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt GR16:$src1, i16immSExt8:$src2))], @@ -1178,10 +1207,12 @@ def BT64ri8 : RIi8<0xBA, MRM4r, (outs), (ins GR64:$src1, i64i8imm:$src2), "bt{q}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt GR64:$src1, i64immSExt8:$src2))], IIC_BT_RI>, TB; +} // SchedRW // Note that these instructions don't need FastBTMem because that // only applies when the other operand is in a register. When it's // an immediate, bt is still fast. +let SchedRW = [WriteALU] in { def BT16mi8 : Ii8<0xBA, MRM4m, (outs), (ins i16mem:$src1, i16i8imm:$src2), "bt{w}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt (loadi16 addr:$src1), i16immSExt8:$src2)) @@ -1194,8 +1225,10 @@ def BT64mi8 : RIi8<0xBA, MRM4m, (outs), (ins i64mem:$src1, i64i8imm:$src2), "bt{q}\t{$src2, $src1|$src1, $src2}", [(set EFLAGS, (X86bt (loadi64 addr:$src1), i64immSExt8:$src2))], IIC_BT_MI>, TB; +} // SchedRW let hasSideEffects = 0 in { +let SchedRW = [WriteALU] in { def BTC16rr : I<0xBB, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2), "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, OpSize, TB; @@ -1203,8 +1236,9 @@ def BTC32rr : I<0xBB, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2), "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB; def BTC64rr : RI<0xBB, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2), "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTC16mr : I<0xBB, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2), "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, OpSize, TB; @@ -1214,6 +1248,7 @@ def BTC64mr : RI<0xBB, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2), "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB; } +let SchedRW = [WriteALU] in { def BTC16ri8 : Ii8<0xBA, MRM7r, (outs), (ins GR16:$src1, i16i8imm:$src2), "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, OpSize, TB; @@ -1221,8 +1256,9 @@ def BTC32ri8 : Ii8<0xBA, MRM7r, (outs), (ins GR32:$src1, i32i8imm:$src2), "btc{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; def BTC64ri8 : RIi8<0xBA, MRM7r, (outs), (ins GR64:$src1, i64i8imm:$src2), "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTC16mi8 : Ii8<0xBA, MRM7m, (outs), (ins i16mem:$src1, i16i8imm:$src2), "btc{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, OpSize, TB; @@ -1232,6 +1268,7 @@ def BTC64mi8 : RIi8<0xBA, MRM7m, (outs), (ins i64mem:$src1, i64i8imm:$src2), "btc{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB; } +let SchedRW = [WriteALU] in { def BTR16rr : I<0xB3, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2), "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, OpSize, TB; @@ -1239,8 +1276,9 @@ def BTR32rr : I<0xB3, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2), "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB; def BTR64rr : RI<0xB3, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2), "btr{q}\t{$src2, $src1|$src1, $src2}", []>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTR16mr : I<0xB3, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2), "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, OpSize, TB; @@ -1250,6 +1288,7 @@ def BTR64mr : RI<0xB3, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2), "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB; } +let SchedRW = [WriteALU] in { def BTR16ri8 : Ii8<0xBA, MRM6r, (outs), (ins GR16:$src1, i16i8imm:$src2), "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, OpSize, TB; @@ -1257,8 +1296,9 @@ def BTR32ri8 : Ii8<0xBA, MRM6r, (outs), (ins GR32:$src1, i32i8imm:$src2), "btr{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; def BTR64ri8 : RIi8<0xBA, MRM6r, (outs), (ins GR64:$src1, i64i8imm:$src2), "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTR16mi8 : Ii8<0xBA, MRM6m, (outs), (ins i16mem:$src1, i16i8imm:$src2), "btr{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, OpSize, TB; @@ -1268,6 +1308,7 @@ def BTR64mi8 : RIi8<0xBA, MRM6m, (outs), (ins i64mem:$src1, i64i8imm:$src2), "btr{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, TB; } +let SchedRW = [WriteALU] in { def BTS16rr : I<0xAB, MRMDestReg, (outs), (ins GR16:$src1, GR16:$src2), "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, OpSize, TB; @@ -1275,8 +1316,9 @@ def BTS32rr : I<0xAB, MRMDestReg, (outs), (ins GR32:$src1, GR32:$src2), "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB; def BTS64rr : RI<0xAB, MRMDestReg, (outs), (ins GR64:$src1, GR64:$src2), "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RR>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTS16mr : I<0xAB, MRMDestMem, (outs), (ins i16mem:$src1, GR16:$src2), "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, OpSize, TB; @@ -1286,6 +1328,7 @@ def BTS64mr : RI<0xAB, MRMDestMem, (outs), (ins i64mem:$src1, GR64:$src2), "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MR>, TB; } +let SchedRW = [WriteALU] in { def BTS16ri8 : Ii8<0xBA, MRM5r, (outs), (ins GR16:$src1, i16i8imm:$src2), "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, OpSize, TB; @@ -1293,8 +1336,9 @@ def BTS32ri8 : Ii8<0xBA, MRM5r, (outs), (ins GR32:$src1, i32i8imm:$src2), "bts{l}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; def BTS64ri8 : RIi8<0xBA, MRM5r, (outs), (ins GR64:$src1, i64i8imm:$src2), "bts{q}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_RI>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def BTS16mi8 : Ii8<0xBA, MRM5m, (outs), (ins i16mem:$src1, i16i8imm:$src2), "bts{w}\t{$src2, $src1|$src1, $src2}", [], IIC_BTX_MI>, OpSize, TB; @@ -1315,7 +1359,7 @@ def BTS64mi8 : RIi8<0xBA, MRM5m, (outs), (ins i64mem:$src1, i64i8imm:$src2), // operand is referenced, the atomicity is ensured. multiclass ATOMIC_SWAP<bits<8> opc8, bits<8> opc, string mnemonic, string frag, InstrItinClass itin> { - let Constraints = "$val = $dst" in { + let Constraints = "$val = $dst", SchedRW = [WriteALULd, WriteRMW] in { def NAME#8rm : I<opc8, MRMSrcMem, (outs GR8:$dst), (ins GR8:$val, i8mem:$ptr), !strconcat(mnemonic, "{b}\t{$val, $ptr|$ptr, $val}"), @@ -1350,6 +1394,7 @@ multiclass ATOMIC_SWAP<bits<8> opc8, bits<8> opc, string mnemonic, string frag, defm XCHG : ATOMIC_SWAP<0x86, 0x87, "xchg", "atomic_swap", IIC_XCHG_MEM>; // Swap between registers. +let SchedRW = [WriteALU] in { let Constraints = "$val = $dst" in { def XCHG8rr : I<0x86, MRMSrcReg, (outs GR8:$dst), (ins GR8:$val, GR8:$src), "xchg{b}\t{$val, $src|$src, $val}", [], IIC_XCHG_REG>; @@ -1374,9 +1419,9 @@ def XCHG32ar64 : I<0x90, AddRegFrm, (outs), (ins GR32_NOAX:$src), Requires<[In64BitMode]>; def XCHG64ar : RI<0x90, AddRegFrm, (outs), (ins GR64:$src), "xchg{q}\t{$src, %rax|RAX, $src}", [], IIC_XCHG_REG>; +} // SchedRW - - +let SchedRW = [WriteALU] in { def XADD8rr : I<0xC0, MRMDestReg, (outs GR8:$dst), (ins GR8:$src), "xadd{b}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB; def XADD16rr : I<0xC1, MRMDestReg, (outs GR16:$dst), (ins GR16:$src), @@ -1386,8 +1431,9 @@ def XADD32rr : I<0xC1, MRMDestReg, (outs GR32:$dst), (ins GR32:$src), "xadd{l}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB; def XADD64rr : RI<0xC1, MRMDestReg, (outs GR64:$dst), (ins GR64:$src), "xadd{q}\t{$src, $dst|$dst, $src}", [], IIC_XADD_REG>, TB; +} // SchedRW -let mayLoad = 1, mayStore = 1 in { +let mayLoad = 1, mayStore = 1, SchedRW = [WriteALULd, WriteRMW] in { def XADD8rm : I<0xC0, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src), "xadd{b}\t{$src, $dst|$dst, $src}", [], IIC_XADD_MEM>, TB; def XADD16rm : I<0xC1, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src), @@ -1400,6 +1446,7 @@ def XADD64rm : RI<0xC1, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src), } +let SchedRW = [WriteALU] in { def CMPXCHG8rr : I<0xB0, MRMDestReg, (outs GR8:$dst), (ins GR8:$src), "cmpxchg{b}\t{$src, $dst|$dst, $src}", [], IIC_CMPXCHG_REG8>, TB; @@ -1412,7 +1459,9 @@ def CMPXCHG32rr : I<0xB1, MRMDestReg, (outs GR32:$dst), (ins GR32:$src), def CMPXCHG64rr : RI<0xB1, MRMDestReg, (outs GR64:$dst), (ins GR64:$src), "cmpxchg{q}\t{$src, $dst|$dst, $src}", [], IIC_CMPXCHG_REG>, TB; +} // SchedRW +let SchedRW = [WriteALULd, WriteRMW] in { let mayLoad = 1, mayStore = 1 in { def CMPXCHG8rm : I<0xB0, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src), "cmpxchg{b}\t{$src, $dst|$dst, $src}", [], @@ -1436,7 +1485,7 @@ let Defs = [RAX, RDX, EFLAGS], Uses = [RAX, RBX, RCX, RDX] in def CMPXCHG16B : RI<0xC7, MRM1m, (outs), (ins i128mem:$dst), "cmpxchg16b\t$dst", [], IIC_CMPXCHG_16B>, TB, Requires<[HasCmpxchg16b]>; - +} // SchedRW // Lock instruction prefix @@ -1459,17 +1508,21 @@ def REPNE_PREFIX : I<0xF2, RawFrm, (outs), (ins), "repne", []>; // String manipulation instructions +let SchedRW = [WriteMicrocoded] in { def LODSB : I<0xAC, RawFrm, (outs), (ins), "lodsb", [], IIC_LODS>; def LODSW : I<0xAD, RawFrm, (outs), (ins), "lodsw", [], IIC_LODS>, OpSize; def LODSD : I<0xAD, RawFrm, (outs), (ins), "lods{l|d}", [], IIC_LODS>; def LODSQ : RI<0xAD, RawFrm, (outs), (ins), "lodsq", [], IIC_LODS>; +} +let SchedRW = [WriteSystem] in { def OUTSB : I<0x6E, RawFrm, (outs), (ins), "outsb", [], IIC_OUTS>; def OUTSW : I<0x6F, RawFrm, (outs), (ins), "outsw", [], IIC_OUTS>, OpSize; def OUTSD : I<0x6F, RawFrm, (outs), (ins), "outs{l|d}", [], IIC_OUTS>; - +} // Flag instructions +let SchedRW = [WriteALU] in { def CLC : I<0xF8, RawFrm, (outs), (ins), "clc", [], IIC_CLC>; def STC : I<0xF9, RawFrm, (outs), (ins), "stc", [], IIC_STC>; def CLI : I<0xFA, RawFrm, (outs), (ins), "cli", [], IIC_CLI>; @@ -1479,10 +1532,13 @@ def STD : I<0xFD, RawFrm, (outs), (ins), "std", [], IIC_STD>; def CMC : I<0xF5, RawFrm, (outs), (ins), "cmc", [], IIC_CMC>; def CLTS : I<0x06, RawFrm, (outs), (ins), "clts", [], IIC_CLTS>, TB; +} // Table lookup instructions -def XLAT : I<0xD7, RawFrm, (outs), (ins), "xlatb", [], IIC_XLAT>; +def XLAT : I<0xD7, RawFrm, (outs), (ins), "xlatb", [], IIC_XLAT>, + Sched<[WriteLoad]>; +let SchedRW = [WriteMicrocoded] in { // ASCII Adjust After Addition // sets AL, AH and CF and AF of EFLAGS and uses AL and AF of EFLAGS def AAA : I<0x37, RawFrm, (outs), (ins), "aaa", [], IIC_AAA>, @@ -1512,7 +1568,9 @@ def DAA : I<0x27, RawFrm, (outs), (ins), "daa", [], IIC_DAA>, // sets AL, CF and AF of EFLAGS and uses AL, CF and AF of EFLAGS def DAS : I<0x2F, RawFrm, (outs), (ins), "das", [], IIC_DAS>, Requires<[In32BitMode]>; +} // SchedRW +let SchedRW = [WriteSystem] in { // Check Array Index Against Bounds def BOUNDS16rm : I<0x62, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), "bound\t{$src, $dst|$dst, $src}", [], IIC_BOUND>, OpSize, @@ -1528,11 +1586,13 @@ def ARPL16rr : I<0x63, MRMDestReg, (outs GR16:$dst), (ins GR16:$src), def ARPL16mr : I<0x63, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src), "arpl\t{$src, $dst|$dst, $src}", [], IIC_ARPL_MEM>, Requires<[In32BitMode]>; +} // SchedRW //===----------------------------------------------------------------------===// // MOVBE Instructions // let Predicates = [HasMOVBE] in { + let SchedRW = [WriteALULd] in { def MOVBE16rm : I<0xF0, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), "movbe{w}\t{$src, $dst|$dst, $src}", [(set GR16:$dst, (bswap (loadi16 addr:$src)))], IIC_MOVBE>, @@ -1545,6 +1605,8 @@ let Predicates = [HasMOVBE] in { "movbe{q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (bswap (loadi64 addr:$src)))], IIC_MOVBE>, T8; + } + let SchedRW = [WriteStore] in { def MOVBE16mr : I<0xF1, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src), "movbe{w}\t{$src, $dst|$dst, $src}", [(store (bswap GR16:$src), addr:$dst)], IIC_MOVBE>, @@ -1557,6 +1619,7 @@ let Predicates = [HasMOVBE] in { "movbe{q}\t{$src, $dst|$dst, $src}", [(store (bswap GR64:$src), addr:$dst)], IIC_MOVBE>, T8; + } } //===----------------------------------------------------------------------===// @@ -1575,6 +1638,21 @@ let Predicates = [HasRDRAND], Defs = [EFLAGS] in { } //===----------------------------------------------------------------------===// +// RDSEED Instruction +// +let Predicates = [HasRDSEED], Defs = [EFLAGS] in { + def RDSEED16r : I<0xC7, MRM7r, (outs GR16:$dst), (ins), + "rdseed{w}\t$dst", + [(set GR16:$dst, EFLAGS, (X86rdseed))]>, OpSize, TB; + def RDSEED32r : I<0xC7, MRM7r, (outs GR32:$dst), (ins), + "rdseed{l}\t$dst", + [(set GR32:$dst, EFLAGS, (X86rdseed))]>, TB; + def RDSEED64r : RI<0xC7, MRM7r, (outs GR64:$dst), (ins), + "rdseed{q}\t$dst", + [(set GR64:$dst, EFLAGS, (X86rdseed))]>, TB; +} + +//===----------------------------------------------------------------------===// // LZCNT Instruction // let Predicates = [HasLZCNT], Defs = [EFLAGS] in { @@ -1755,90 +1833,90 @@ include "X86InstrCompiler.td" // Assembler Mnemonic Aliases //===----------------------------------------------------------------------===// -def : MnemonicAlias<"call", "calll">, Requires<[In32BitMode]>; -def : MnemonicAlias<"call", "callq">, Requires<[In64BitMode]>; +def : MnemonicAlias<"call", "calll", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"call", "callq", "att">, Requires<[In64BitMode]>; -def : MnemonicAlias<"cbw", "cbtw">; -def : MnemonicAlias<"cwde", "cwtl">; -def : MnemonicAlias<"cwd", "cwtd">; -def : MnemonicAlias<"cdq", "cltd">; -def : MnemonicAlias<"cdqe", "cltq">; -def : MnemonicAlias<"cqo", "cqto">; +def : MnemonicAlias<"cbw", "cbtw", "att">; +def : MnemonicAlias<"cwde", "cwtl", "att">; +def : MnemonicAlias<"cwd", "cwtd", "att">; +def : MnemonicAlias<"cdq", "cltd", "att">; +def : MnemonicAlias<"cdqe", "cltq", "att">; +def : MnemonicAlias<"cqo", "cqto", "att">; // lret maps to lretl, it is not ambiguous with lretq. -def : MnemonicAlias<"lret", "lretl">; +def : MnemonicAlias<"lret", "lretl", "att">; -def : MnemonicAlias<"leavel", "leave">, Requires<[In32BitMode]>; -def : MnemonicAlias<"leaveq", "leave">, Requires<[In64BitMode]>; +def : MnemonicAlias<"leavel", "leave", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"leaveq", "leave", "att">, Requires<[In64BitMode]>; -def : MnemonicAlias<"loopz", "loope">; -def : MnemonicAlias<"loopnz", "loopne">; +def : MnemonicAlias<"loopz", "loope", "att">; +def : MnemonicAlias<"loopnz", "loopne", "att">; -def : MnemonicAlias<"pop", "popl">, Requires<[In32BitMode]>; -def : MnemonicAlias<"pop", "popq">, Requires<[In64BitMode]>; -def : MnemonicAlias<"popf", "popfl">, Requires<[In32BitMode]>; -def : MnemonicAlias<"popf", "popfq">, Requires<[In64BitMode]>; -def : MnemonicAlias<"popfd", "popfl">; +def : MnemonicAlias<"pop", "popl", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"pop", "popq", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"popf", "popfl", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"popf", "popfq", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"popfd", "popfl", "att">; // FIXME: This is wrong for "push reg". "push %bx" should turn into pushw in // all modes. However: "push (addr)" and "push $42" should default to // pushl/pushq depending on the current mode. Similar for "pop %bx" -def : MnemonicAlias<"push", "pushl">, Requires<[In32BitMode]>; -def : MnemonicAlias<"push", "pushq">, Requires<[In64BitMode]>; -def : MnemonicAlias<"pushf", "pushfl">, Requires<[In32BitMode]>; -def : MnemonicAlias<"pushf", "pushfq">, Requires<[In64BitMode]>; -def : MnemonicAlias<"pushfd", "pushfl">; +def : MnemonicAlias<"push", "pushl", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"push", "pushq", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"pushf", "pushfl", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"pushf", "pushfq", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"pushfd", "pushfl", "att">; -def : MnemonicAlias<"repe", "rep">; -def : MnemonicAlias<"repz", "rep">; -def : MnemonicAlias<"repnz", "repne">; +def : MnemonicAlias<"repe", "rep", "att">; +def : MnemonicAlias<"repz", "rep", "att">; +def : MnemonicAlias<"repnz", "repne", "att">; -def : MnemonicAlias<"retl", "ret">, Requires<[In32BitMode]>; -def : MnemonicAlias<"retq", "ret">, Requires<[In64BitMode]>; +def : MnemonicAlias<"retl", "ret", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"retq", "ret", "att">, Requires<[In64BitMode]>; -def : MnemonicAlias<"salb", "shlb">; -def : MnemonicAlias<"salw", "shlw">; -def : MnemonicAlias<"sall", "shll">; -def : MnemonicAlias<"salq", "shlq">; +def : MnemonicAlias<"salb", "shlb", "att">; +def : MnemonicAlias<"salw", "shlw", "att">; +def : MnemonicAlias<"sall", "shll", "att">; +def : MnemonicAlias<"salq", "shlq", "att">; -def : MnemonicAlias<"smovb", "movsb">; -def : MnemonicAlias<"smovw", "movsw">; -def : MnemonicAlias<"smovl", "movsl">; -def : MnemonicAlias<"smovq", "movsq">; +def : MnemonicAlias<"smovb", "movsb", "att">; +def : MnemonicAlias<"smovw", "movsw", "att">; +def : MnemonicAlias<"smovl", "movsl", "att">; +def : MnemonicAlias<"smovq", "movsq", "att">; -def : MnemonicAlias<"ud2a", "ud2">; -def : MnemonicAlias<"verrw", "verr">; +def : MnemonicAlias<"ud2a", "ud2", "att">; +def : MnemonicAlias<"verrw", "verr", "att">; // System instruction aliases. -def : MnemonicAlias<"iret", "iretl">; -def : MnemonicAlias<"sysret", "sysretl">; -def : MnemonicAlias<"sysexit", "sysexitl">; +def : MnemonicAlias<"iret", "iretl", "att">; +def : MnemonicAlias<"sysret", "sysretl", "att">; +def : MnemonicAlias<"sysexit", "sysexitl", "att">; -def : MnemonicAlias<"lgdtl", "lgdt">, Requires<[In32BitMode]>; -def : MnemonicAlias<"lgdtq", "lgdt">, Requires<[In64BitMode]>; -def : MnemonicAlias<"lidtl", "lidt">, Requires<[In32BitMode]>; -def : MnemonicAlias<"lidtq", "lidt">, Requires<[In64BitMode]>; -def : MnemonicAlias<"sgdtl", "sgdt">, Requires<[In32BitMode]>; -def : MnemonicAlias<"sgdtq", "sgdt">, Requires<[In64BitMode]>; -def : MnemonicAlias<"sidtl", "sidt">, Requires<[In32BitMode]>; -def : MnemonicAlias<"sidtq", "sidt">, Requires<[In64BitMode]>; +def : MnemonicAlias<"lgdtl", "lgdt", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"lgdtq", "lgdt", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"lidtl", "lidt", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"lidtq", "lidt", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"sgdtl", "sgdt", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"sgdtq", "sgdt", "att">, Requires<[In64BitMode]>; +def : MnemonicAlias<"sidtl", "sidt", "att">, Requires<[In32BitMode]>; +def : MnemonicAlias<"sidtq", "sidt", "att">, Requires<[In64BitMode]>; // Floating point stack aliases. -def : MnemonicAlias<"fcmovz", "fcmove">; -def : MnemonicAlias<"fcmova", "fcmovnbe">; -def : MnemonicAlias<"fcmovnae", "fcmovb">; -def : MnemonicAlias<"fcmovna", "fcmovbe">; -def : MnemonicAlias<"fcmovae", "fcmovnb">; -def : MnemonicAlias<"fcomip", "fcompi">; -def : MnemonicAlias<"fildq", "fildll">; -def : MnemonicAlias<"fistpq", "fistpll">; -def : MnemonicAlias<"fisttpq", "fisttpll">; -def : MnemonicAlias<"fldcww", "fldcw">; -def : MnemonicAlias<"fnstcww", "fnstcw">; -def : MnemonicAlias<"fnstsww", "fnstsw">; -def : MnemonicAlias<"fucomip", "fucompi">; -def : MnemonicAlias<"fwait", "wait">; +def : MnemonicAlias<"fcmovz", "fcmove", "att">; +def : MnemonicAlias<"fcmova", "fcmovnbe", "att">; +def : MnemonicAlias<"fcmovnae", "fcmovb", "att">; +def : MnemonicAlias<"fcmovna", "fcmovbe", "att">; +def : MnemonicAlias<"fcmovae", "fcmovnb", "att">; +def : MnemonicAlias<"fcomip", "fcompi", "att">; +def : MnemonicAlias<"fildq", "fildll", "att">; +def : MnemonicAlias<"fistpq", "fistpll", "att">; +def : MnemonicAlias<"fisttpq", "fisttpll", "att">; +def : MnemonicAlias<"fldcww", "fldcw", "att">; +def : MnemonicAlias<"fnstcww", "fnstcw", "att">; +def : MnemonicAlias<"fnstsww", "fnstsw", "att">; +def : MnemonicAlias<"fucomip", "fucompi", "att">; +def : MnemonicAlias<"fwait", "wait", "att">; class CondCodeAlias<string Prefix,string Suffix, string OldCond, string NewCond> diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td index 127af6f..49721df 100644 --- a/lib/Target/X86/X86InstrMMX.td +++ b/lib/Target/X86/X86InstrMMX.td @@ -20,6 +20,7 @@ // MMX Multiclasses //===----------------------------------------------------------------------===// +let Sched = WriteVecALU in { def MMX_INTALU_ITINS : OpndItins< IIC_MMX_ALU_RR, IIC_MMX_ALU_RM >; @@ -35,11 +36,14 @@ def MMX_PHADDSUBW : OpndItins< def MMX_PHADDSUBD : OpndItins< IIC_MMX_PHADDSUBD_RR, IIC_MMX_PHADDSUBD_RM >; +} +let Sched = WriteVecIMul in def MMX_PMUL_ITINS : OpndItins< IIC_MMX_PMUL, IIC_MMX_PMUL >; +let Sched = WriteVecALU in { def MMX_PSADBW_ITINS : OpndItins< IIC_MMX_PSADBW, IIC_MMX_PSADBW >; @@ -47,11 +51,13 @@ def MMX_PSADBW_ITINS : OpndItins< def MMX_MISC_FUNC_ITINS : OpndItins< IIC_MMX_MISC_FUNC_MEM, IIC_MMX_MISC_FUNC_REG >; +} def MMX_SHIFT_ITINS : ShiftOpndItins< IIC_MMX_SHIFT_RR, IIC_MMX_SHIFT_RM, IIC_MMX_SHIFT_RI >; +let Sched = WriteShuffle in { def MMX_UNPCK_H_ITINS : OpndItins< IIC_MMX_UNPCK_H_RR, IIC_MMX_UNPCK_H_RM >; @@ -67,7 +73,9 @@ def MMX_PCK_ITINS : OpndItins< def MMX_PSHUF_ITINS : OpndItins< IIC_MMX_PSHUF, IIC_MMX_PSHUF >; +} // Sched +let Sched = WriteCvtF2I in { def MMX_CVT_PD_ITINS : OpndItins< IIC_MMX_CVT_PD_RR, IIC_MMX_CVT_PD_RM >; @@ -75,6 +83,7 @@ def MMX_CVT_PD_ITINS : OpndItins< def MMX_CVT_PS_ITINS : OpndItins< IIC_MMX_CVT_PS_RR, IIC_MMX_CVT_PS_RM >; +} let Constraints = "$src1 = $dst" in { // MMXI_binop_rm_int - Simple MMX binary operator based on intrinsic. @@ -84,7 +93,8 @@ let Constraints = "$src1 = $dst" in { def irr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src1, VR64:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), - [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr> { + [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>, + Sched<[itins.Sched]> { let isCommutable = Commutable; } def irm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), @@ -92,7 +102,7 @@ let Constraints = "$src1 = $dst" in { !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), [(set VR64:$dst, (IntId VR64:$src1, (bitconvert (load_mmx addr:$src2))))], - itins.rm>; + itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>; } multiclass MMXI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm, @@ -101,17 +111,19 @@ let Constraints = "$src1 = $dst" in { def rr : MMXI<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src1, VR64:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), - [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>; + [(set VR64:$dst, (IntId VR64:$src1, VR64:$src2))], itins.rr>, + Sched<[WriteVecShift]>; def rm : MMXI<opc, MRMSrcMem, (outs VR64:$dst), (ins VR64:$src1, i64mem:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), [(set VR64:$dst, (IntId VR64:$src1, (bitconvert (load_mmx addr:$src2))))], - itins.rm>; + itins.rm>, Sched<[WriteVecShiftLd, ReadAfterLd]>; def ri : MMXIi8<opc2, ImmForm, (outs VR64:$dst), (ins VR64:$src1, i32i8imm:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), - [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>; + [(set VR64:$dst, (IntId2 VR64:$src1, (i32 imm:$src2)))], itins.ri>, + Sched<[WriteVecShift]>; } } @@ -120,13 +132,14 @@ multiclass SS3I_unop_rm_int_mm<bits<8> opc, string OpcodeStr, Intrinsic IntId64, OpndItins itins> { def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>; + [(set VR64:$dst, (IntId64 VR64:$src))], itins.rr>, + Sched<[itins.Sched]>; def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR64:$dst, (IntId64 (bitconvert (memopmmx addr:$src))))], - itins.rm>; + itins.rm>, Sched<[itins.Sched.Folded]>; } /// Binary MMX instructions requiring SSSE3. @@ -137,13 +150,15 @@ multiclass SS3I_binop_rm_int_mm<bits<8> opc, string OpcodeStr, def rr64 : MMXSS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src1, VR64:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), - [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>; + [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm64 : MMXSS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins VR64:$src1, i64mem:$src2), !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), [(set VR64:$dst, (IntId64 VR64:$src1, - (bitconvert (memopmmx addr:$src2))))], itins.rm>; + (bitconvert (memopmmx addr:$src2))))], itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } } @@ -164,9 +179,11 @@ multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag, string asm, OpndItins itins, Domain d> { def irr : MMXPI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, - [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>; + [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr, d>, + Sched<[itins.Sched]>; def irm : MMXPI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, - [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>; + [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm, d>, + Sched<[itins.Sched.Folded]>; } multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC, @@ -174,11 +191,11 @@ multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC, PatFrag ld_frag, string asm, Domain d> { def irr : PI<opc, MRMSrcReg, (outs DstRC:$dst),(ins DstRC:$src1, SrcRC:$src2), asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], - IIC_DEFAULT, d>; + NoItinerary, d>; def irm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins DstRC:$src1, x86memop:$src2), asm, [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], - IIC_DEFAULT, d>; + NoItinerary, d>; } //===----------------------------------------------------------------------===// @@ -197,16 +214,17 @@ def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR64:$dst, (x86mmx (scalar_to_vector GR32:$src)))], - IIC_MMX_MOV_MM_RM>; + IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>; let canFoldAsLoad = 1 in def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR64:$dst, (x86mmx (scalar_to_vector (loadi32 addr:$src))))], - IIC_MMX_MOV_MM_RM>; + IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>; let mayStore = 1 in def MMX_MOVD64mr : MMXI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR64:$src), - "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>; + "movd\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOV_MM_RM>, + Sched<[WriteStore]>; // Low word of MMX to GPR. def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1, @@ -214,16 +232,18 @@ def MMX_X86movd2w : SDNode<"X86ISD::MMX_MOVD2W", SDTypeProfile<1, 1, def MMX_MOVD64grr : MMXI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR64:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, - (MMX_X86movd2w (x86mmx VR64:$src)))], IIC_MMX_MOV_REG_MM>; + (MMX_X86movd2w (x86mmx VR64:$src)))], + IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>; let neverHasSideEffects = 1 in def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src), "movd\t{$src, $dst|$dst, $src}", - [], IIC_MMX_MOV_MM_RM>; + [], IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>; // These are 64 bit moves, but since the OS X assembler doesn't // recognize a register-register movq, we write them as // movd. +let SchedRW = [WriteMove] in { def MMX_MOVD64from64rr : MMXRI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR64:$src), "movd\t{$src, $dst|$dst, $src}", @@ -237,6 +257,9 @@ let neverHasSideEffects = 1 in def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src), "movq\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOVQ_RR>; +} // SchedRW + +let SchedRW = [WriteLoad] in { let canFoldAsLoad = 1 in def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src), "movq\t{$src, $dst|$dst, $src}", @@ -246,7 +269,9 @@ def MMX_MOVQ64mr : MMXI<0x7F, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src), "movq\t{$src, $dst|$dst, $src}", [(store (x86mmx VR64:$src), addr:$dst)], IIC_MMX_MOVQ_RM>; +} // SchedRW +let SchedRW = [WriteMove] in { def MMX_MOVDQ2Qrr : MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src), "movdq2q\t{$src, $dst|$dst, $src}", [(set VR64:$dst, @@ -271,11 +296,12 @@ def MMX_MOVQ2FR64rr: MMXS2SIi8<0xD6, MRMSrcReg, (outs FR64:$dst), def MMX_MOVFR642Qrr: MMXSDIi8<0xD6, MRMSrcReg, (outs VR64:$dst), (ins FR64:$src), "movdq2q\t{$src, $dst|$dst, $src}", [], IIC_MMX_MOVQ_RR>; +} // SchedRW def MMX_MOVNTQmr : MMXI<0xE7, MRMDestMem, (outs), (ins i64mem:$dst, VR64:$src), "movntq\t{$src, $dst|$dst, $src}", [(int_x86_mmx_movnt_dq addr:$dst, VR64:$src)], - IIC_MMX_MOVQ_RM>; + IIC_MMX_MOVQ_RM>, Sched<[WriteStore]>; let AddedComplexity = 15 in // movd to MMX register zero-extends @@ -283,7 +309,7 @@ def MMX_MOVZDI2PDIrr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR64:$dst, (x86mmx (X86vzmovl (x86mmx (scalar_to_vector GR32:$src)))))], - IIC_MMX_MOV_MM_RM>; + IIC_MMX_MOV_MM_RM>, Sched<[WriteMove]>; let AddedComplexity = 20 in def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src), @@ -291,7 +317,7 @@ def MMX_MOVZDI2PDIrm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), [(set VR64:$dst, (x86mmx (X86vzmovl (x86mmx (scalar_to_vector (loadi32 addr:$src))))))], - IIC_MMX_MOV_MM_RM>; + IIC_MMX_MOV_MM_RM>, Sched<[WriteLoad]>; // Arithmetic Instructions defm MMX_PABSB : SS3I_unop_rm_int_mm<0x1C, "pabsb", int_x86_ssse3_pabs_b, @@ -491,14 +517,14 @@ def MMX_PSHUFWri : MMXIi8<0x70, MRMSrcReg, "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR64:$dst, (int_x86_sse_pshuf_w VR64:$src1, imm:$src2))], - IIC_MMX_PSHUF>; + IIC_MMX_PSHUF>, Sched<[WriteShuffle]>; def MMX_PSHUFWmi : MMXIi8<0x70, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src1, i8imm:$src2), "pshufw\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR64:$dst, (int_x86_sse_pshuf_w (load_mmx addr:$src1), imm:$src2))], - IIC_MMX_PSHUF>; + IIC_MMX_PSHUF>, Sched<[WriteShuffleLd]>; @@ -532,7 +558,7 @@ def MMX_PEXTRWirri: MMXIi8<0xC5, MRMSrcReg, "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set GR32:$dst, (int_x86_mmx_pextr_w VR64:$src1, (iPTR imm:$src2)))], - IIC_MMX_PEXTR>; + IIC_MMX_PEXTR>, Sched<[WriteShuffle]>; let Constraints = "$src1 = $dst" in { def MMX_PINSRWirri : MMXIi8<0xC4, MRMSrcReg, (outs VR64:$dst), @@ -540,7 +566,7 @@ let Constraints = "$src1 = $dst" in { "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}", [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1, GR32:$src2, (iPTR imm:$src3)))], - IIC_MMX_PINSRW>; + IIC_MMX_PINSRW>, Sched<[WriteShuffle]>; def MMX_PINSRWirmi : MMXIi8<0xC4, MRMSrcMem, (outs VR64:$dst), @@ -549,7 +575,7 @@ let Constraints = "$src1 = $dst" in { [(set VR64:$dst, (int_x86_mmx_pinsr_w VR64:$src1, (i32 (anyext (loadi16 addr:$src2))), (iPTR imm:$src3)))], - IIC_MMX_PINSRW>; + IIC_MMX_PINSRW>, Sched<[WriteShuffleLd, ReadAfterLd]>; } // Mask creation @@ -570,6 +596,7 @@ def : Pat<(x86mmx (MMX_X86movdq2q (loadv2i64 addr:$src))), (x86mmx (MMX_MOVQ64rm addr:$src))>; // Misc. +let SchedRW = [WriteShuffle] in { let Uses = [EDI] in def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask), "maskmovq\t{$mask, $src|$src, $mask}", @@ -580,6 +607,7 @@ def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask), "maskmovq\t{$mask, $src|$src, $mask}", [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)], IIC_MMX_MASKMOV>; +} // 64-bit bit convert. let Predicates = [HasSSE2] in { diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 105963f..cce938b 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -35,6 +35,7 @@ class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm, // scalar +let Sched = WriteFAdd in { def SSE_ALU_F32S : OpndItins< IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM >; @@ -42,6 +43,7 @@ def SSE_ALU_F32S : OpndItins< def SSE_ALU_F64S : OpndItins< IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM >; +} def SSE_ALU_ITINS_S : SizeItins< SSE_ALU_F32S, SSE_ALU_F64S @@ -76,6 +78,7 @@ def SSE_DIV_ITINS_S : SizeItins< >; // parallel +let Sched = WriteFAdd in { def SSE_ALU_F32P : OpndItins< IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM >; @@ -83,6 +86,7 @@ def SSE_ALU_F32P : OpndItins< def SSE_ALU_F64P : OpndItins< IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM >; +} def SSE_ALU_ITINS_P : SizeItins< SSE_ALU_F32P, SSE_ALU_F64P @@ -184,14 +188,16 @@ multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC, !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (!cast<Intrinsic>( !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, RC:$src2))], itins.rr>; + RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2), !if(Is2Addr, !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, mem_cpat:$src2))], itins.rm>; + RC:$src1, mem_cpat:$src2))], itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// sse12_fp_packed - SSE 1 & 2 packed instructions class @@ -226,13 +232,13 @@ multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - pat_rr, IIC_DEFAULT, d>, + pat_rr, NoItinerary, d>, Sched<[WriteVecLogic]>; def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - pat_rm, IIC_DEFAULT, d>, + pat_rm, NoItinerary, d>, Sched<[WriteVecLogicLd, ReadAfterLd]>; } @@ -364,7 +370,7 @@ let Predicates = [HasAVX] in { // Alias instructions that map fld0 to xorps for sse or vxorps for avx. // This is expanded by ExpandPostRAPseudos. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "", [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>; def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "", @@ -381,7 +387,7 @@ let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, // We set canFoldAsLoad because this can be converted to a constant-pool // load of an all-zeros value if folding it would be beneficial. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", [(set VR128:$dst, (v4f32 immAllZerosV))]>; } @@ -398,7 +404,7 @@ def : Pat<(v16i8 immAllZerosV), (V_SET0)>; // at the rename stage without using any execution unit, so SET0PSY // and SET0PDY can be used for vector int instructions without penalty let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1, Predicates = [HasAVX] in { + isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in { def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "", [(set VR256:$dst, (v8f32 immAllZerosV))]>; } @@ -436,7 +442,7 @@ def : Pat<(bc_v4i64 (v8f32 immAllZerosV)), // We set canFoldAsLoad because this can be converted to a constant-pool // load of an all-ones value if folding it would be beneficial. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "", [(set VR128:$dst, (v4i32 immAllOnesV))]>; let Predicates = [HasAVX2] in @@ -470,7 +476,7 @@ multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), !strconcat(base_opc, asm_opr), - [], IIC_SSE_MOV_S_RR>; + [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>; } multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt, @@ -848,7 +854,7 @@ def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src), } // SchedRW // For disassembler -let isCodeGenOnly = 1, hasSideEffects = 0 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [], @@ -924,7 +930,7 @@ def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), } // SchedRW // For disassembler -let isCodeGenOnly = 1, hasSideEffects = 0 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>; @@ -1070,7 +1076,7 @@ let Predicates = [UseSSE1] in { // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper // bits are disregarded. FIXME: Set encoding to pseudo! -let neverHasSideEffects = 1 in { +let neverHasSideEffects = 1, SchedRW = [WriteMove] in { def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src), "movaps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, VEX; @@ -1087,7 +1093,7 @@ def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src), // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper // bits are disregarded. FIXME: Set encoding to pseudo! -let canFoldAsLoad = 1, isReMaterializable = 1 in { +let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in { let isCodeGenOnly = 1 in { def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src), "movaps\t{$src, $dst|$dst, $src}", @@ -1436,11 +1442,13 @@ multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, X86MemOperand x86memop, string asm> { let neverHasSideEffects = 1 in { def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src), - !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>; + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + Sched<[WriteCvtI2F]>; let mayLoad = 1 in def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins DstRC:$src1, x86memop:$src), - !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>; + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + Sched<[WriteCvtI2FLd, ReadAfterLd]>; } // neverHasSideEffects = 1 } @@ -1740,13 +1748,15 @@ let neverHasSideEffects = 1 in { def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src1, FR64:$src2), "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG; + IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2F]>; let mayLoad = 1 in def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins FR64:$src1, f64mem:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RM>, - XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG; + XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>, @@ -1755,26 +1765,28 @@ def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>, def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src), "cvtsd2ss\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (fround FR64:$src))], - IIC_SSE_CVT_Scalar_RR>; + IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>; def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src), "cvtsd2ss\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (fround (loadf64 addr:$src)))], IIC_SSE_CVT_Scalar_RM>, XD, - Requires<[UseSSE2, OptForSize]>; + Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>; def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2F]>; def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, sse_load_f64:$src2))], - IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; let Constraints = "$src1 = $dst" in { def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg, @@ -1782,13 +1794,15 @@ def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg, "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>, + Sched<[WriteCvtF2F]>; def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2), "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, sse_load_f64:$src2))], - IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } // Convert scalar single to scalar double @@ -1798,13 +1812,15 @@ def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src1, FR32:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RR>, - XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG; + XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2F]>; let mayLoad = 1 in def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins FR32:$src1, f32mem:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RM>, - XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>; + XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } def : Pat<(f64 (fextend FR32:$src)), @@ -1823,12 +1839,12 @@ def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src), "cvtss2sd\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (fextend FR32:$src))], IIC_SSE_CVT_Scalar_RR>, XS, - Requires<[UseSSE2]>; + Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>; def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src), "cvtss2sd\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (extloadf32 addr:$src))], IIC_SSE_CVT_Scalar_RM>, XS, - Requires<[UseSSE2, OptForSize]>; + Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>; // extload f32 -> f64. This matches load+fextend because we have a hack in // the isel (PreprocessForFPConvert) that can introduce loads after dag @@ -1845,57 +1861,61 @@ def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg, "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2F]>; def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))], - IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "cvtss2sd\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>, + Sched<[WriteCvtF2F]>; def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2), "cvtss2sd\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))], - IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } // Convert packed single/double fp to doubleword def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>, VEX; + IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>; def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX; + IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>; def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2dq_256 VR256:$src))], - IIC_SSE_CVT_PS_RR>, VEX, VEX_L; + IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX, VEX_L; + IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>; + IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>; def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>; + IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>; // Convert Packed Double FP to Packed DW Integers @@ -1906,7 +1926,7 @@ let Predicates = [HasAVX] in { def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>, - VEX; + VEX, Sched<[WriteCvtF2I]>; // XMM only def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}", @@ -1914,18 +1934,20 @@ def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}", def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "vcvtpd2dqx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX; + (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX, + Sched<[WriteCvtF2ILd]>; // YMM only def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L; + (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L, + Sched<[WriteCvtF2I]>; def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>, - VEX, VEX_L; + VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}", (VCVTPD2DQYrr VR128:$dst, VR256:$src)>; } @@ -1934,11 +1956,11 @@ def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>; def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>; // Convert with truncation packed single/double fp to doubleword // SSE2 packed instructions with XS prefix @@ -1946,32 +1968,33 @@ def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>, VEX; + IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>; def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX; + IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>; def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256 VR256:$src))], - IIC_SSE_CVT_PS_RR>, VEX, VEX_L; + IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX, VEX_L; + IIC_SSE_CVT_PS_RM>, VEX, VEX_L, + Sched<[WriteCvtF2ILd]>; def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>; + IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>; def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>; + IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>; let Predicates = [HasAVX] in { def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))), @@ -2021,7 +2044,7 @@ def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>, VEX; + IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>; // The assembler can recognize rr 256-bit instructions by seeing a ymm // register, but the same isn't true when using memory operands instead. @@ -2034,19 +2057,19 @@ def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttpd2dqx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX; + IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>; // YMM only def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvtt_pd2dq_256 VR256:$src))], - IIC_SSE_CVT_PD_RR>, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}", (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>; @@ -2060,12 +2083,13 @@ let Predicates = [HasAVX] in { def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>; def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, + Sched<[WriteCvtF2ILd]>; // Convert packed single to packed double let Predicates = [HasAVX] in { @@ -2073,32 +2097,32 @@ let Predicates = [HasAVX] in { def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB, VEX; + IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>; def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB, VEX; + IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>; def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2_pd_256 VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>; def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>; } let Predicates = [UseSSE2] in { def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB; + IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>; def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), "cvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB; + IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>; } // Convert Packed DW Integers to Packed Double FP @@ -2106,30 +2130,33 @@ let Predicates = [HasAVX] in { let neverHasSideEffects = 1, mayLoad = 1 in def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", - []>, VEX; + []>, VEX, Sched<[WriteCvtI2FLd]>; def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX; + (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX, + Sched<[WriteCvtI2F]>; def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtdq2_pd_256 - (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L; + (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L, + Sched<[WriteCvtI2FLd]>; def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, - (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L; + (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L, + Sched<[WriteCvtI2F]>; } let neverHasSideEffects = 1, mayLoad = 1 in def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "cvtdq2pd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>; def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>; // AVX 256-bit register conversion intrinsics let Predicates = [HasAVX] in { @@ -2146,7 +2173,7 @@ let Predicates = [HasAVX] in { def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))], - IIC_SSE_CVT_PD_RR>, VEX; + IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>; // XMM only def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}", @@ -2155,31 +2182,31 @@ def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2psx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX; + IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>; // YMM only def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2_ps_256 VR256:$src))], - IIC_SSE_CVT_PD_RR>, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>; def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>; def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}", (VCVTPD2PSYrr VR128:$dst, VR256:$src)>; def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>; def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>; // AVX 256-bit register conversion intrinsics @@ -2244,11 +2271,12 @@ multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop, let neverHasSideEffects = 1 in { def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [], - IIC_SSE_ALU_F32S_RR>; + IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>; let mayLoad = 1 in def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [], - IIC_SSE_ALU_F32S_RM>; + IIC_SSE_ALU_F32S_RM>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } } @@ -2394,10 +2422,11 @@ multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop, let neverHasSideEffects = 1 in { def rri_alt : PIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), - asm_alt, [], IIC_SSE_CMPP_RR, d>; + asm_alt, [], IIC_SSE_CMPP_RR, d>, Sched<[WriteFAdd]>; def rmi_alt : PIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), - asm_alt, [], IIC_SSE_CMPP_RM, d>; + asm_alt, [], IIC_SSE_CMPP_RM, d>, + Sched<[WriteFAddLd, ReadAfterLd]>; } } @@ -2694,18 +2723,18 @@ let Predicates = [HasAVX] in { // Assembler Only def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src), "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, - SSEPackedSingle>, TB, VEX; + SSEPackedSingle>, TB, VEX, Sched<[WriteVecLogic]>; def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src), "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, SSEPackedDouble>, TB, - OpSize, VEX; + OpSize, VEX, Sched<[WriteVecLogic]>; def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src), "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, - SSEPackedSingle>, TB, VEX, VEX_L; + SSEPackedSingle>, TB, VEX, VEX_L, Sched<[WriteVecLogic]>; def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src), "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, SSEPackedDouble>, TB, - OpSize, VEX, VEX_L; + OpSize, VEX, VEX_L, Sched<[WriteVecLogic]>; } defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps", @@ -3458,7 +3487,7 @@ def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), //===----------------------------------------------------------------------===// // Prefetch intrinsic. -let Predicates = [HasSSE1] in { +let Predicates = [HasSSE1], SchedRW = [WriteLoad] in { def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src), "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))], IIC_SSE_PREFETCH>, TB; @@ -3473,6 +3502,8 @@ def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src), IIC_SSE_PREFETCH>, TB; } +// FIXME: How should these memory instructions be modeled? +let SchedRW = [WriteLoad] in { // Flush cache def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src), "clflush\t$src", [(int_x86_sse2_clflush addr:$src)], @@ -3492,6 +3523,7 @@ def LFENCE : I<0xAE, MRM_E8, (outs), (ins), def MFENCE : I<0xAE, MRM_F0, (outs), (ins), "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>, TB, Requires<[HasSSE2]>; +} // SchedRW def : Pat<(X86SFence), (SFENCE)>; def : Pat<(X86LFence), (LFENCE)>; @@ -3503,17 +3535,17 @@ def : Pat<(X86MFence), (MFENCE)>; def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src), "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)], - IIC_SSE_LDMXCSR>, VEX; + IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>; def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst), "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)], - IIC_SSE_STMXCSR>, VEX; + IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>; def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src), "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)], - IIC_SSE_LDMXCSR>; + IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>; def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst), "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)], - IIC_SSE_STMXCSR>; + IIC_SSE_STMXCSR>, Sched<[WriteStore]>; //===---------------------------------------------------------------------===// // SSE2 - Move Aligned/Unaligned Packed Integer Instructions @@ -4430,12 +4462,12 @@ def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src), // Move Packed Doubleword Int first element to Doubleword Int // let SchedRW = [WriteMove] in { -def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), - "vmov{d|q}\t{$src, $dst|$dst, $src}", +def VMOVPQIto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), + "mov{d|q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (vector_extract (v2i64 VR128:$src), (iPTR 0)))], IIC_SSE_MOVD_ToGP>, - TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>; + VEX; def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", @@ -4480,23 +4512,24 @@ def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src), def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (bitconvert FR32:$src))], - IIC_SSE_MOVD_ToGP>, VEX; + IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>; def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (bitconvert FR32:$src)), addr:$dst)], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>; def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (bitconvert FR32:$src))], - IIC_SSE_MOVD_ToGP>; + IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>; def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (bitconvert FR32:$src)), addr:$dst)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteStore]>; //===---------------------------------------------------------------------===// // Patterns and instructions to describe movd/movq to XMM register zero-extends // +let SchedRW = [WriteMove] in { let AddedComplexity = 15 in { def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", @@ -4522,8 +4555,9 @@ def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), (v2i64 (scalar_to_vector GR64:$src)))))], IIC_SSE_MOVDQ>; } +} // SchedRW -let AddedComplexity = 20 in { +let AddedComplexity = 20, SchedRW = [WriteLoad] in { def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, @@ -4536,7 +4570,7 @@ def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), (v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))))], IIC_SSE_MOVDQ>; -} +} // AddedComplexity, SchedRW let Predicates = [HasAVX] in { // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part. @@ -4585,6 +4619,8 @@ def : InstAlias<"movq\t{$src, $dst|$dst, $src}", //===---------------------------------------------------------------------===// // Move Quadword Int to Packed Quadword Int // + +let SchedRW = [WriteLoad] in { def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "vmovq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, @@ -4596,10 +4632,12 @@ def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (scalar_to_vector (loadi64 addr:$src))))], IIC_SSE_MOVDQ>, XS, Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix +} // SchedRW //===---------------------------------------------------------------------===// // Move Packed Quadword Int to Quadword Int // +let SchedRW = [WriteStore] in { def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", [(store (i64 (vector_extract (v2i64 VR128:$src), @@ -4610,17 +4648,19 @@ def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), [(store (i64 (vector_extract (v2i64 VR128:$src), (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>; +} // SchedRW //===---------------------------------------------------------------------===// // Store / copy lower 64-bits of a XMM register. // def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX; + [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX, + Sched<[WriteStore]>; def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteStore]>; let AddedComplexity = 20 in def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), @@ -4629,7 +4669,7 @@ def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (X86vzmovl (v2i64 (scalar_to_vector (loadi64 addr:$src))))))], IIC_SSE_MOVDQ>, - XS, VEX, Requires<[HasAVX]>; + XS, VEX, Requires<[HasAVX]>, Sched<[WriteLoad]>; let AddedComplexity = 20 in def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), @@ -4638,7 +4678,7 @@ def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (X86vzmovl (v2i64 (scalar_to_vector (loadi64 addr:$src))))))], IIC_SSE_MOVDQ>, - XS, Requires<[UseSSE2]>; + XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>; let Predicates = [HasAVX], AddedComplexity = 20 in { def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))), @@ -4668,6 +4708,7 @@ def : Pat<(v4i64 (X86vzload addr:$src)), // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in // IA32 document. movq xmm1, xmm2 does clear the high bits. // +let SchedRW = [WriteVecLogic] in { let AddedComplexity = 15 in def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vmovq\t{$src, $dst|$dst, $src}", @@ -4680,7 +4721,9 @@ def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))], IIC_SSE_MOVQ_RR>, XS, Requires<[UseSSE2]>; +} // SchedRW +let SchedRW = [WriteVecLogicLd] in { let AddedComplexity = 20 in def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "vmovq\t{$src, $dst|$dst, $src}", @@ -4696,6 +4739,7 @@ def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), IIC_SSE_MOVDQ>, XS, Requires<[UseSSE2]>; } +} // SchedRW let AddedComplexity = 20 in { let Predicates = [HasAVX] in { @@ -4713,6 +4757,7 @@ let AddedComplexity = 20 in { } // Instructions to match in the assembler +let SchedRW = [WriteMove] in { def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVDQ>, VEX, VEX_W; @@ -4723,16 +4768,19 @@ def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), "movd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVDQ>, VEX, VEX_W; +} // SchedRW // Instructions for the disassembler // xr = XMM register // xm = mem64 +let SchedRW = [WriteMove] in { let Predicates = [HasAVX] in def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS; def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS; +} // SchedRW //===---------------------------------------------------------------------===// // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP @@ -4743,11 +4791,11 @@ multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr, def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set RC:$dst, (vt (OpNode RC:$src)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set RC:$dst, (OpNode (mem_frag addr:$src)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>; } let Predicates = [HasAVX] in { @@ -4803,25 +4851,27 @@ multiclass sse3_replicate_dfp<string OpcodeStr> { let neverHasSideEffects = 1 in def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [], IIC_SSE_MOV_LH>; + [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (v2f64 (X86Movddup (scalar_to_vector (loadf64 addr:$src)))))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>; } // FIXME: Merge with above classe when there're patterns for the ymm version multiclass sse3_replicate_dfp_y<string OpcodeStr> { def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>; + [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>, + Sched<[WriteShuffle]>; def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (v4f64 (X86Movddup - (scalar_to_vector (loadf64 addr:$src)))))]>; + (scalar_to_vector (loadf64 addr:$src)))))]>, + Sched<[WriteShuffleLd]>; } let Predicates = [HasAVX] in { @@ -4869,6 +4919,7 @@ let Predicates = [UseSSE3] in { // SSE3 - Move Unaligned Integer //===---------------------------------------------------------------------===// +let SchedRW = [WriteLoad] in { let Predicates = [HasAVX] in { def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "vlddqu\t{$src, $dst|$dst, $src}", @@ -4882,6 +4933,7 @@ def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "lddqu\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))], IIC_SSE_LDDQU>; +} //===---------------------------------------------------------------------===// // SSE3 - Arithmetic @@ -4895,13 +4947,15 @@ multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>; + [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm : I<0xD0, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>; + [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } let Predicates = [HasAVX] in { @@ -4938,14 +4992,15 @@ multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>; + [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>, + Sched<[WriteFAdd]>; def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))], - IIC_SSE_HADDSUB_RM>; + IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>; } multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> { @@ -4953,14 +5008,15 @@ multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>; + [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>, + Sched<[WriteFAdd]>; def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))], - IIC_SSE_HADDSUB_RM>; + IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>; } let Predicates = [HasAVX] in { @@ -5009,7 +5065,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr, (ins VR128:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>, - OpSize; + OpSize, Sched<[WriteVecALU]>; def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), @@ -5017,7 +5073,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr, [(set VR128:$dst, (IntId128 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>, - OpSize; + OpSize, Sched<[WriteVecALULd]>; } /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}. @@ -5027,16 +5083,27 @@ multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr, (ins VR256:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (IntId256 VR256:$src))]>, - OpSize; + OpSize, Sched<[WriteVecALU]>; def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (IntId256 - (bitconvert (memopv4i64 addr:$src))))]>, OpSize; + (bitconvert (memopv4i64 addr:$src))))]>, OpSize, + Sched<[WriteVecALULd]>; } +// Helper fragments to match sext vXi1 to vXiY. +def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)), + VR128:$src))>; +def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i32 15)))>; +def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i32 31)))>; +def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)), + VR256:$src))>; +def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i32 15)))>; +def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i32 31)))>; + let Predicates = [HasAVX] in { defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", int_x86_ssse3_pabs_b_128>, VEX; @@ -5044,6 +5111,19 @@ let Predicates = [HasAVX] in { int_x86_ssse3_pabs_w_128>, VEX; defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", int_x86_ssse3_pabs_d_128>, VEX; + + def : Pat<(xor + (bc_v2i64 (v16i1sextv16i8)), + (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))), + (VPABSBrr128 VR128:$src)>; + def : Pat<(xor + (bc_v2i64 (v8i1sextv8i16)), + (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))), + (VPABSWrr128 VR128:$src)>; + def : Pat<(xor + (bc_v2i64 (v4i1sextv4i32)), + (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))), + (VPABSDrr128 VR128:$src)>; } let Predicates = [HasAVX2] in { @@ -5053,6 +5133,19 @@ let Predicates = [HasAVX2] in { int_x86_avx2_pabs_w>, VEX, VEX_L; defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd", int_x86_avx2_pabs_d>, VEX, VEX_L; + + def : Pat<(xor + (bc_v4i64 (v32i1sextv32i8)), + (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))), + (VPABSBrr256 VR256:$src)>; + def : Pat<(xor + (bc_v4i64 (v16i1sextv16i16)), + (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))), + (VPABSWrr256 VR256:$src)>; + def : Pat<(xor + (bc_v4i64 (v8i1sextv8i32)), + (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))), + (VPABSDrr256 VR256:$src)>; } defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", @@ -5062,10 +5155,26 @@ defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", int_x86_ssse3_pabs_d_128>; +let Predicates = [HasSSSE3] in { + def : Pat<(xor + (bc_v2i64 (v16i1sextv16i8)), + (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))), + (PABSBrr128 VR128:$src)>; + def : Pat<(xor + (bc_v2i64 (v8i1sextv8i16)), + (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))), + (PABSWrr128 VR128:$src)>; + def : Pat<(xor + (bc_v2i64 (v4i1sextv4i32)), + (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))), + (PABSDrr128 VR128:$src)>; +} + //===---------------------------------------------------------------------===// // SSSE3 - Packed Binary Operator Instructions //===---------------------------------------------------------------------===// +let Sched = WriteVecALU in { def SSE_PHADDSUBD : OpndItins< IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM >; @@ -5075,12 +5184,16 @@ def SSE_PHADDSUBSW : OpndItins< def SSE_PHADDSUBW : OpndItins< IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM >; +} +let Sched = WriteShuffle in def SSE_PSHUFB : OpndItins< IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM >; +let Sched = WriteVecALU in def SSE_PSIGN : OpndItins< IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM >; +let Sched = WriteVecIMul in def SSE_PMULHRSW : OpndItins< IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW >; @@ -5097,7 +5210,7 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>, - OpSize; + OpSize, Sched<[itins.Sched]>; def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, @@ -5105,7 +5218,8 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpVT (OpNode RC:$src1, - (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize; + (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}. @@ -5119,7 +5233,7 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, - OpSize; + OpSize, Sched<[itins.Sched]>; def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2), !if(Is2Addr, @@ -5127,7 +5241,8 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set VR128:$dst, (IntId128 VR128:$src1, - (bitconvert (memopv2i64 addr:$src2))))]>, OpSize; + (bitconvert (memopv2i64 addr:$src2))))]>, OpSize, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr, @@ -5269,7 +5384,7 @@ multiclass ssse3_palignr<string asm, bit Is2Addr = 1> { !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [], IIC_SSE_PALIGNR>, OpSize; + [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffle]>; let mayLoad = 1 in def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2, i8imm:$src3), @@ -5277,7 +5392,7 @@ multiclass ssse3_palignr<string asm, bit Is2Addr = 1> { !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [], IIC_SSE_PALIGNR>, OpSize; + [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>; } } @@ -5287,13 +5402,13 @@ multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> { (ins VR256:$src1, VR256:$src2, i8imm:$src3), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, OpSize; + []>, OpSize, Sched<[WriteShuffle]>; let mayLoad = 1 in def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2, i8imm:$src3), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, OpSize; + []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>; } } @@ -5341,6 +5456,7 @@ def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), // SSSE3 - Thread synchronization //===---------------------------------------------------------------------===// +let SchedRW = [WriteSystem] in { let usesCustomInserter = 1 in { def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3), [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>, @@ -5354,6 +5470,7 @@ let Uses = [ECX, EAX] in def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>, TB, Requires<[HasSSE3]>; +} // SchedRW def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>; def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>; @@ -6773,7 +6890,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))], - IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; + NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, RC:$src3), @@ -6782,7 +6899,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr, [(set RC:$dst, (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)), RC:$src3))], - IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; + NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; } let Predicates = [HasAVX] in { diff --git a/lib/Target/X86/X86InstrShiftRotate.td b/lib/Target/X86/X86InstrShiftRotate.td index 1185941..5b6298b 100644 --- a/lib/Target/X86/X86InstrShiftRotate.td +++ b/lib/Target/X86/X86InstrShiftRotate.td @@ -15,7 +15,7 @@ let Defs = [EFLAGS] in { -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { let Uses = [CL] in { def SHL8rCL : I<0xD2, MRM4r, (outs GR8 :$dst), (ins GR8 :$src1), "shl{b}\t{%cl, $dst|$dst, CL}", @@ -62,9 +62,10 @@ def SHL64r1 : RI<0xD1, MRM4r, (outs GR64:$dst), (ins GR64:$src1), "shl{q}\t$dst", [], IIC_SR>; } // hasSideEffects = 0 } // isConvertibleToThreeAddress = 1 -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { // FIXME: Why do we need an explicit "Uses = [CL]" when the instr has a pattern // using CL? let Uses = [CL] in { @@ -118,8 +119,9 @@ def SHL64m1 : RI<0xD1, MRM4m, (outs), (ins i64mem:$dst), "shl{q}\t$dst", [(store (shl (loadi64 addr:$dst), (i8 1)), addr:$dst)], IIC_SR>; +} // SchedRW -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { let Uses = [CL] in { def SHR8rCL : I<0xD2, MRM5r, (outs GR8 :$dst), (ins GR8 :$src1), "shr{b}\t{%cl, $dst|$dst, CL}", @@ -163,9 +165,10 @@ def SHR32r1 : I<0xD1, MRM5r, (outs GR32:$dst), (ins GR32:$src1), def SHR64r1 : RI<0xD1, MRM5r, (outs GR64:$dst), (ins GR64:$src1), "shr{q}\t$dst", [(set GR64:$dst, (srl GR64:$src1, (i8 1)))], IIC_SR>; -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { let Uses = [CL] in { def SHR8mCL : I<0xD2, MRM5m, (outs), (ins i8mem :$dst), "shr{b}\t{%cl, $dst|$dst, CL}", @@ -216,8 +219,9 @@ def SHR64m1 : RI<0xD1, MRM5m, (outs), (ins i64mem:$dst), "shr{q}\t$dst", [(store (srl (loadi64 addr:$dst), (i8 1)), addr:$dst)], IIC_SR>; +} // SchedRW -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { let Uses = [CL] in { def SAR8rCL : I<0xD2, MRM7r, (outs GR8 :$dst), (ins GR8 :$src1), "sar{b}\t{%cl, $dst|$dst, CL}", @@ -273,9 +277,10 @@ def SAR64r1 : RI<0xD1, MRM7r, (outs GR64:$dst), (ins GR64:$src1), "sar{q}\t$dst", [(set GR64:$dst, (sra GR64:$src1, (i8 1)))], IIC_SR>; -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { let Uses = [CL] in { def SAR8mCL : I<0xD2, MRM7m, (outs), (ins i8mem :$dst), "sar{b}\t{%cl, $dst|$dst, CL}", @@ -330,13 +335,14 @@ def SAR64m1 : RI<0xD1, MRM7m, (outs), (ins i64mem:$dst), "sar{q}\t$dst", [(store (sra (loadi64 addr:$dst), (i8 1)), addr:$dst)], IIC_SR>; +} // SchedRW //===----------------------------------------------------------------------===// // Rotate instructions //===----------------------------------------------------------------------===// let hasSideEffects = 0 in { -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { def RCL8r1 : I<0xD0, MRM2r, (outs GR8:$dst), (ins GR8:$src1), "rcl{b}\t$dst", [], IIC_SR>; def RCL8ri : Ii8<0xC0, MRM2r, (outs GR8:$dst), (ins GR8:$src1, i8imm:$cnt), @@ -405,6 +411,7 @@ def RCR64rCL : RI<0xD3, MRM3r, (outs GR64:$dst), (ins GR64:$src1), } // Constraints = "$src = $dst" +let SchedRW = [WriteShiftLd, WriteRMW] in { def RCL8m1 : I<0xD0, MRM2m, (outs), (ins i8mem:$dst), "rcl{b}\t$dst", [], IIC_SR>; def RCL8mi : Ii8<0xC0, MRM2m, (outs), (ins i8mem:$dst, i8imm:$cnt), @@ -458,9 +465,10 @@ def RCR32mCL : I<0xD3, MRM3m, (outs), (ins i32mem:$dst), def RCR64mCL : RI<0xD3, MRM3m, (outs), (ins i64mem:$dst), "rcr{q}\t{%cl, $dst|$dst, CL}", [], IIC_SR>; } +} // SchedRW } // hasSideEffects = 0 -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { // FIXME: provide shorter instructions when imm8 == 1 let Uses = [CL] in { def ROL8rCL : I<0xD2, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1), @@ -512,8 +520,9 @@ def ROL64r1 : RI<0xD1, MRM0r, (outs GR64:$dst), (ins GR64:$src1), "rol{q}\t$dst", [(set GR64:$dst, (rotl GR64:$src1, (i8 1)))], IIC_SR>; -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { let Uses = [CL] in { def ROL8mCL : I<0xD2, MRM0m, (outs), (ins i8mem :$dst), "rol{b}\t{%cl, $dst|$dst, CL}", @@ -568,8 +577,9 @@ def ROL64m1 : RI<0xD1, MRM0m, (outs), (ins i64mem:$dst), "rol{q}\t$dst", [(store (rotl (loadi64 addr:$dst), (i8 1)), addr:$dst)], IIC_SR>; +} // SchedRW -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { let Uses = [CL] in { def ROR8rCL : I<0xD2, MRM1r, (outs GR8 :$dst), (ins GR8 :$src1), "ror{b}\t{%cl, $dst|$dst, CL}", @@ -620,8 +630,9 @@ def ROR64r1 : RI<0xD1, MRM1r, (outs GR64:$dst), (ins GR64:$src1), "ror{q}\t$dst", [(set GR64:$dst, (rotr GR64:$src1, (i8 1)))], IIC_SR>; -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { let Uses = [CL] in { def ROR8mCL : I<0xD2, MRM1m, (outs), (ins i8mem :$dst), "ror{b}\t{%cl, $dst|$dst, CL}", @@ -676,13 +687,14 @@ def ROR64m1 : RI<0xD1, MRM1m, (outs), (ins i64mem:$dst), "ror{q}\t$dst", [(store (rotr (loadi64 addr:$dst), (i8 1)), addr:$dst)], IIC_SR>; +} // SchedRW //===----------------------------------------------------------------------===// // Double shift instructions (generalizations of rotate) //===----------------------------------------------------------------------===// -let Constraints = "$src1 = $dst" in { +let Constraints = "$src1 = $dst", SchedRW = [WriteShift] in { let Uses = [CL] in { def SHLD16rrCL : I<0xA5, MRMDestReg, (outs GR16:$dst), @@ -765,8 +777,9 @@ def SHRD64rri8 : RIi8<0xAC, MRMDestReg, (i8 imm:$src3)))], IIC_SHD64_REG_IM>, TB; } -} // Constraints = "$src = $dst" +} // Constraints = "$src = $dst", SchedRW +let SchedRW = [WriteShiftLd, WriteRMW] in { let Uses = [CL] in { def SHLD16mrCL : I<0xA5, MRMDestMem, (outs), (ins i16mem:$dst, GR16:$src2), "shld{w}\t{%cl, $src2, $dst|$dst, $src2, CL}", @@ -840,6 +853,7 @@ def SHRD64mri8 : RIi8<0xAC, MRMDestMem, (i8 imm:$src3)), addr:$dst)], IIC_SHD64_MEM_IM>, TB; +} // SchedRW } // Defs = [EFLAGS] @@ -857,12 +871,12 @@ multiclass bmi_rotate<string asm, RegisterClass RC, X86MemOperand x86memop> { let neverHasSideEffects = 1 in { def ri : Ii8<0xF0, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, i8imm:$src2), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - []>, TAXD, VEX; + []>, TAXD, VEX, Sched<[WriteShift]>; let mayLoad = 1 in def mi : Ii8<0xF0, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src1, i8imm:$src2), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - []>, TAXD, VEX; + []>, TAXD, VEX, Sched<[WriteShiftLd]>; } } @@ -870,11 +884,17 @@ multiclass bmi_shift<string asm, RegisterClass RC, X86MemOperand x86memop> { let neverHasSideEffects = 1 in { def rr : I<0xF7, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, - VEX_4VOp3; + VEX_4VOp3, Sched<[WriteShift]>; let mayLoad = 1 in def rm : I<0xF7, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src1, RC:$src2), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, - VEX_4VOp3; + VEX_4VOp3, + Sched<[WriteShiftLd, + // x86memop:$src1 + ReadDefault, ReadDefault, ReadDefault, ReadDefault, + ReadDefault, + // RC:$src1 + ReadAfterLd]>; } } diff --git a/lib/Target/X86/X86InstrSystem.td b/lib/Target/X86/X86InstrSystem.td index 3caa1b5..bab3cdd 100644 --- a/lib/Target/X86/X86InstrSystem.td +++ b/lib/Target/X86/X86InstrSystem.td @@ -13,6 +13,7 @@ // //===----------------------------------------------------------------------===// +let SchedRW = [WriteSystem] in { let Defs = [RAX, RDX] in def RDTSC : I<0x31, RawFrm, (outs), (ins), "rdtsc", [(X86rdtsc)], IIC_RDTSC>, TB; @@ -35,6 +36,7 @@ let Uses = [EFLAGS] in def INTO : I<0xce, RawFrm, (outs), (ins), "into", []>; def INT3 : I<0xcc, RawFrm, (outs), (ins), "int3", [(int_x86_int (i8 3))], IIC_INT3>; +} // SchedRW def : Pat<(debugtrap), (INT3)>; @@ -43,6 +45,7 @@ def : Pat<(debugtrap), // FIXME: This doesn't work because InstAlias can't match immediate constants. //def : InstAlias<"int\t$3", (INT3)>; +let SchedRW = [WriteSystem] in { def INT : Ii8<0xcd, RawFrm, (outs), (ins i8imm:$trap), "int\t$trap", [(int_x86_int imm:$trap)], IIC_INT>; @@ -65,11 +68,13 @@ def IRET16 : I<0xcf, RawFrm, (outs), (ins), "iret{w}", [], IIC_IRET>, OpSize; def IRET32 : I<0xcf, RawFrm, (outs), (ins), "iret{l|d}", [], IIC_IRET>; def IRET64 : RI<0xcf, RawFrm, (outs), (ins), "iretq", [], IIC_IRET>, Requires<[In64BitMode]>; +} // SchedRW //===----------------------------------------------------------------------===// // Input/Output Instructions. // +let SchedRW = [WriteSystem] in { let Defs = [AL], Uses = [DX] in def IN8rr : I<0xEC, RawFrm, (outs), (ins), "in{b}\t{%dx, %al|AL, DX}", [], IIC_IN_RR>; @@ -113,10 +118,12 @@ def OUT32ir : Ii8<0xE7, RawFrm, (outs), (ins i8imm:$port), def IN8 : I<0x6C, RawFrm, (outs), (ins), "ins{b}", [], IIC_INS>; def IN16 : I<0x6D, RawFrm, (outs), (ins), "ins{w}", [], IIC_INS>, OpSize; def IN32 : I<0x6D, RawFrm, (outs), (ins), "ins{l}", [], IIC_INS>; +} // SchedRW //===----------------------------------------------------------------------===// // Moves to and from debug registers +let SchedRW = [WriteSystem] in { def MOV32rd : I<0x21, MRMDestReg, (outs GR32:$dst), (ins DEBUG_REG:$src), "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_DR>, TB; def MOV64rd : I<0x21, MRMDestReg, (outs GR64:$dst), (ins DEBUG_REG:$src), @@ -126,10 +133,12 @@ def MOV32dr : I<0x23, MRMSrcReg, (outs DEBUG_REG:$dst), (ins GR32:$src), "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_DR_REG>, TB; def MOV64dr : I<0x23, MRMSrcReg, (outs DEBUG_REG:$dst), (ins GR64:$src), "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_DR_REG>, TB; +} // SchedRW //===----------------------------------------------------------------------===// // Moves to and from control registers +let SchedRW = [WriteSystem] in { def MOV32rc : I<0x20, MRMDestReg, (outs GR32:$dst), (ins CONTROL_REG:$src), "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_CR>, TB; def MOV64rc : I<0x20, MRMDestReg, (outs GR64:$dst), (ins CONTROL_REG:$src), @@ -139,6 +148,7 @@ def MOV32cr : I<0x22, MRMSrcReg, (outs CONTROL_REG:$dst), (ins GR32:$src), "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_CR_REG>, TB; def MOV64cr : I<0x22, MRMSrcReg, (outs CONTROL_REG:$dst), (ins GR64:$src), "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_CR_REG>, TB; +} // SchedRW //===----------------------------------------------------------------------===// // Segment override instruction prefixes @@ -155,6 +165,7 @@ def GS_PREFIX : I<0x65, RawFrm, (outs), (ins), "gs", []>; // Moves to and from segment registers. // +let SchedRW = [WriteMove] in { def MOV16rs : I<0x8C, MRMDestReg, (outs GR16:$dst), (ins SEGMENT_REG:$src), "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV_REG_SR>, OpSize; def MOV32rs : I<0x8C, MRMDestReg, (outs GR32:$dst), (ins SEGMENT_REG:$src), @@ -182,10 +193,12 @@ def MOV32sm : I<0x8E, MRMSrcMem, (outs SEGMENT_REG:$dst), (ins i32mem:$src), "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV_SR_MEM>; def MOV64sm : RI<0x8E, MRMSrcMem, (outs SEGMENT_REG:$dst), (ins i64mem:$src), "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV_SR_MEM>; +} // SchedRW //===----------------------------------------------------------------------===// // Segmentation support instructions. +let SchedRW = [WriteSystem] in { def SWAPGS : I<0x01, MRM_F8, (outs), (ins), "swapgs", [], IIC_SWAPGS>, TB; def LAR16rm : I<0x02, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src), @@ -347,10 +360,12 @@ def VERWr : I<0x00, MRM5r, (outs), (ins GR16:$seg), "verw\t$seg", [], IIC_VERW_MEM>, TB; def VERWm : I<0x00, MRM5m, (outs), (ins i16mem:$seg), "verw\t$seg", [], IIC_VERW_REG>, TB; +} // SchedRW //===----------------------------------------------------------------------===// // Descriptor-table support instructions +let SchedRW = [WriteSystem] in { def SGDT16m : I<0x01, MRM0m, (outs opaque48mem:$dst), (ins), "sgdt{w}\t$dst", [], IIC_SGDT>, TB, OpSize, Requires<[In32BitMode]>; def SGDTm : I<0x01, MRM0m, (outs opaque48mem:$dst), (ins), @@ -385,9 +400,11 @@ def LLDT16r : I<0x00, MRM2r, (outs), (ins GR16:$src), "lldt{w}\t$src", [], IIC_LLDT_REG>, TB; def LLDT16m : I<0x00, MRM2m, (outs), (ins i16mem:$src), "lldt{w}\t$src", [], IIC_LLDT_MEM>, TB; - +} // SchedRW + //===----------------------------------------------------------------------===// // Specialized register support +let SchedRW = [WriteSystem] in { def WRMSR : I<0x30, RawFrm, (outs), (ins), "wrmsr", [], IIC_WRMSR>, TB; def RDMSR : I<0x32, RawFrm, (outs), (ins), "rdmsr", [], IIC_RDMSR>, TB; def RDPMC : I<0x33, RawFrm, (outs), (ins), "rdpmc", [], IIC_RDPMC>, TB; @@ -410,14 +427,18 @@ def LMSW16m : I<0x01, MRM6m, (outs), (ins i16mem:$src), "lmsw{w}\t$src", [], IIC_LMSW_REG>, TB; def CPUID : I<0xA2, RawFrm, (outs), (ins), "cpuid", [], IIC_CPUID>, TB; +} // SchedRW //===----------------------------------------------------------------------===// // Cache instructions +let SchedRW = [WriteSystem] in { def INVD : I<0x08, RawFrm, (outs), (ins), "invd", [], IIC_INVD>, TB; def WBINVD : I<0x09, RawFrm, (outs), (ins), "wbinvd", [], IIC_INVD>, TB; +} // SchedRW //===----------------------------------------------------------------------===// // XSAVE instructions +let SchedRW = [WriteSystem] in { let Defs = [RDX, RAX], Uses = [RCX] in def XGETBV : I<0x01, MRM_D0, (outs), (ins), "xgetbv", []>, TB; @@ -428,16 +449,17 @@ let Uses = [RDX, RAX] in { def XSAVE : I<0xAE, MRM4m, (outs opaque512mem:$dst), (ins), "xsave\t$dst", []>, TB; def XSAVE64 : I<0xAE, MRM4m, (outs opaque512mem:$dst), (ins), - "xsaveq\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; + "xsave{q|64}\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; def XRSTOR : I<0xAE, MRM5m, (outs), (ins opaque512mem:$dst), "xrstor\t$dst", []>, TB; def XRSTOR64 : I<0xAE, MRM5m, (outs), (ins opaque512mem:$dst), - "xrstorq\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; + "xrstor{q|64}\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; def XSAVEOPT : I<0xAE, MRM6m, (outs opaque512mem:$dst), (ins), "xsaveopt\t$dst", []>, TB; def XSAVEOPT64 : I<0xAE, MRM6m, (outs opaque512mem:$dst), (ins), - "xsaveoptq\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; + "xsaveopt{q|64}\t$dst", []>, TB, REX_W, Requires<[In64BitMode]>; } +} // SchedRW //===----------------------------------------------------------------------===// // VIA PadLock crypto instructions @@ -493,8 +515,15 @@ let Predicates = [HasFSGSBase, In64BitMode] in { //===----------------------------------------------------------------------===// // INVPCID Instruction def INVPCID32 : I<0x82, MRMSrcMem, (outs), (ins GR32:$src1, i128mem:$src2), - "invpcid {$src2, $src1|$src1, $src2}", []>, OpSize, T8, + "invpcid\t{$src2, $src1|$src1, $src2}", []>, OpSize, T8, Requires<[In32BitMode]>; def INVPCID64 : I<0x82, MRMSrcMem, (outs), (ins GR64:$src1, i128mem:$src2), - "invpcid {$src2, $src1|$src1, $src2}", []>, OpSize, T8, + "invpcid\t{$src2, $src1|$src1, $src2}", []>, OpSize, T8, Requires<[In64BitMode]>; + +//===----------------------------------------------------------------------===// +// SMAP Instruction +let Defs = [EFLAGS], Uses = [EFLAGS] in { + def CLAC : I<0x01, MRM_CA, (outs), (ins), "clac", []>, TB; + def STAC : I<0x01, MRM_CB, (outs), (ins), "stac", []>, TB; +} diff --git a/lib/Target/X86/X86InstrTSX.td b/lib/Target/X86/X86InstrTSX.td index a37a8cc..363a190 100644 --- a/lib/Target/X86/X86InstrTSX.td +++ b/lib/Target/X86/X86InstrTSX.td @@ -15,6 +15,9 @@ //===----------------------------------------------------------------------===// // TSX instructions +def X86xtest: SDNode<"X86ISD::XTEST", SDTypeProfile<1, 0, [SDTCisVT<0, i32>]>, + [SDNPHasChain, SDNPSideEffect]>; + let usesCustomInserter = 1 in def XBEGIN : I<0, Pseudo, (outs GR32:$dst), (ins), "# XBEGIN", [(set GR32:$dst, (int_x86_xbegin))]>, @@ -27,6 +30,10 @@ def XBEGIN_4 : Ii32PCRel<0xc7, MRM_F8, (outs), (ins brtarget:$dst), def XEND : I<0x01, MRM_D5, (outs), (ins), "xend", [(int_x86_xend)]>, TB, Requires<[HasRTM]>; +let Defs = [EFLAGS] in +def XTEST : I<0x01, MRM_D6, (outs), (ins), + "xtest", [(set EFLAGS, (X86xtest))]>, TB, Requires<[HasTSX]>; + def XABORT : Ii8<0xc6, MRM_F8, (outs), (ins i8imm:$imm), "xabort\t$imm", [(int_x86_xabort imm:$imm)]>, Requires<[HasRTM]>; diff --git a/lib/Target/X86/X86SchedHaswell.td b/lib/Target/X86/X86SchedHaswell.td new file mode 100644 index 0000000..84c9203 --- /dev/null +++ b/lib/Target/X86/X86SchedHaswell.td @@ -0,0 +1,126 @@ +//=- X86SchedHaswell.td - X86 Haswell Scheduling -------------*- tablegen -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines the machine model for Haswell to support instruction +// scheduling and other instruction cost heuristics. +// +//===----------------------------------------------------------------------===// + +def HaswellModel : SchedMachineModel { + // All x86 instructions are modeled as a single micro-op, and HW can decode 4 + // instructions per cycle. + let IssueWidth = 4; + let MinLatency = 0; // 0 = Out-of-order execution. + let LoadLatency = 4; + let ILPWindow = 30; + let MispredictPenalty = 16; +} + +let SchedModel = HaswellModel in { + +// Haswell can issue micro-ops to 8 different ports in one cycle. + +// Ports 0, 1, 5, 6 and 7 handle all computation. +// Port 4 gets the data half of stores. Store data can be available later than +// the store address, but since we don't model the latency of stores, we can +// ignore that. +// Ports 2 and 3 are identical. They handle loads and the address half of +// stores. Port 7 can handle address calculations. +def HWPort0 : ProcResource<1>; +def HWPort1 : ProcResource<1>; +def HWPort2 : ProcResource<1>; +def HWPort3 : ProcResource<1>; +def HWPort4 : ProcResource<1>; +def HWPort5 : ProcResource<1>; +def HWPort6 : ProcResource<1>; +def HWPort7 : ProcResource<1>; + +// Many micro-ops are capable of issuing on multiple ports. +def HWPort23 : ProcResGroup<[HWPort2, HWPort3]>; +def HWPort237 : ProcResGroup<[HWPort2, HWPort3, HWPort7]>; +def HWPort05 : ProcResGroup<[HWPort0, HWPort5]>; +def HWPort056 : ProcResGroup<[HWPort0, HWPort5, HWPort6]>; +def HWPort15 : ProcResGroup<[HWPort1, HWPort5]>; +def HWPort015 : ProcResGroup<[HWPort0, HWPort1, HWPort5]>; +def HWPort0156: ProcResGroup<[HWPort0, HWPort1, HWPort5, HWPort6]>; + +// Integer division issued on port 0. +def HWDivider : ProcResource<1>; + +// Loads are 4 cycles, so ReadAfterLd registers needn't be available until 4 +// cycles after the memory operand. +def : ReadAdvance<ReadAfterLd, 4>; + +// Many SchedWrites are defined in pairs with and without a folded load. +// Instructions with folded loads are usually micro-fused, so they only appear +// as two micro-ops when queued in the reservation station. +// This multiclass defines the resource usage for variants with and without +// folded loads. +multiclass HWWriteResPair<X86FoldableSchedWrite SchedRW, + ProcResourceKind ExePort, + int Lat> { + // Register variant is using a single cycle on ExePort. + def : WriteRes<SchedRW, [ExePort]> { let Latency = Lat; } + + // Memory variant also uses a cycle on port 2/3 and adds 4 cycles to the + // latency. + def : WriteRes<SchedRW.Folded, [HWPort23, ExePort]> { + let Latency = !add(Lat, 4); + } +} + +// A folded store needs a cycle on port 4 for the store data, but it does not +// need an extra port 2/3 cycle to recompute the address. +def : WriteRes<WriteRMW, [HWPort4]>; + +def : WriteRes<WriteStore, [HWPort237, HWPort4]>; +def : WriteRes<WriteLoad, [HWPort23]> { let Latency = 4; } +def : WriteRes<WriteMove, [HWPort0156]>; +def : WriteRes<WriteZero, []>; + +defm : HWWriteResPair<WriteALU, HWPort0156, 1>; +defm : HWWriteResPair<WriteIMul, HWPort1, 3>; +defm : HWWriteResPair<WriteShift, HWPort056, 1>; +defm : HWWriteResPair<WriteJump, HWPort5, 1>; + +// This is for simple LEAs with one or two input operands. +// The complex ones can only execute on port 1, and they require two cycles on +// the port to read all inputs. We don't model that. +def : WriteRes<WriteLEA, [HWPort15]>; + +// This is quite rough, latency depends on the dividend. +def : WriteRes<WriteIDiv, [HWPort0, HWDivider]> { + let Latency = 25; + let ResourceCycles = [1, 10]; +} +def : WriteRes<WriteIDivLd, [HWPort23, HWPort0, HWDivider]> { + let Latency = 29; + let ResourceCycles = [1, 1, 10]; +} + +// Scalar and vector floating point. +defm : HWWriteResPair<WriteFAdd, HWPort1, 3>; +defm : HWWriteResPair<WriteFMul, HWPort0, 5>; +defm : HWWriteResPair<WriteFDiv, HWPort0, 12>; // 10-14 cycles. +defm : HWWriteResPair<WriteFRcp, HWPort0, 5>; +defm : HWWriteResPair<WriteFSqrt, HWPort0, 15>; +defm : HWWriteResPair<WriteCvtF2I, HWPort1, 3>; +defm : HWWriteResPair<WriteCvtI2F, HWPort1, 4>; +defm : HWWriteResPair<WriteCvtF2F, HWPort1, 3>; + +// Vector integer operations. +defm : HWWriteResPair<WriteVecShift, HWPort05, 1>; +defm : HWWriteResPair<WriteVecLogic, HWPort015, 1>; +defm : HWWriteResPair<WriteVecALU, HWPort15, 1>; +defm : HWWriteResPair<WriteVecIMul, HWPort0, 5>; +defm : HWWriteResPair<WriteShuffle, HWPort15, 1>; + +def : WriteRes<WriteSystem, [HWPort0156]> { let Latency = 100; } +def : WriteRes<WriteMicrocoded, [HWPort0156]> { let Latency = 100; } +} // SchedModel diff --git a/lib/Target/X86/X86SchedSandyBridge.td b/lib/Target/X86/X86SchedSandyBridge.td new file mode 100644 index 0000000..b36b3ad --- /dev/null +++ b/lib/Target/X86/X86SchedSandyBridge.td @@ -0,0 +1,122 @@ +//=- X86SchedSandyBridge.td - X86 Sandy Bridge Scheduling ----*- tablegen -*-=// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines the machine model for Sandy Bridge to support instruction +// scheduling and other instruction cost heuristics. +// +//===----------------------------------------------------------------------===// + +def SandyBridgeModel : SchedMachineModel { + // All x86 instructions are modeled as a single micro-op, and SB can decode 4 + // instructions per cycle. + // FIXME: Identify instructions that aren't a single fused micro-op. + let IssueWidth = 4; + let MinLatency = 0; // 0 = Out-of-order execution. + let LoadLatency = 4; + let ILPWindow = 20; + let MispredictPenalty = 16; +} + +let SchedModel = SandyBridgeModel in { + +// Sandy Bridge can issue micro-ops to 6 different ports in one cycle. + +// Ports 0, 1, and 5 handle all computation. +def SBPort0 : ProcResource<1>; +def SBPort1 : ProcResource<1>; +def SBPort5 : ProcResource<1>; + +// Ports 2 and 3 are identical. They handle loads and the address half of +// stores. +def SBPort23 : ProcResource<2>; + +// Port 4 gets the data half of stores. Store data can be available later than +// the store address, but since we don't model the latency of stores, we can +// ignore that. +def SBPort4 : ProcResource<1>; + +// Many micro-ops are capable of issuing on multiple ports. +def SBPort05 : ProcResGroup<[SBPort0, SBPort5]>; +def SBPort15 : ProcResGroup<[SBPort1, SBPort5]>; +def SBPort015 : ProcResGroup<[SBPort0, SBPort1, SBPort5]>; + +// Integer division issued on port 0. +def SBDivider : ProcResource<1>; + +// Loads are 4 cycles, so ReadAfterLd registers needn't be available until 4 +// cycles after the memory operand. +def : ReadAdvance<ReadAfterLd, 4>; + +// Many SchedWrites are defined in pairs with and without a folded load. +// Instructions with folded loads are usually micro-fused, so they only appear +// as two micro-ops when queued in the reservation station. +// This multiclass defines the resource usage for variants with and without +// folded loads. +multiclass SBWriteResPair<X86FoldableSchedWrite SchedRW, + ProcResourceKind ExePort, + int Lat> { + // Register variant is using a single cycle on ExePort. + def : WriteRes<SchedRW, [ExePort]> { let Latency = Lat; } + + // Memory variant also uses a cycle on port 2/3 and adds 4 cycles to the + // latency. + def : WriteRes<SchedRW.Folded, [SBPort23, ExePort]> { + let Latency = !add(Lat, 4); + } +} + +// A folded store needs a cycle on port 4 for the store data, but it does not +// need an extra port 2/3 cycle to recompute the address. +def : WriteRes<WriteRMW, [SBPort4]>; + +def : WriteRes<WriteStore, [SBPort23, SBPort4]>; +def : WriteRes<WriteLoad, [SBPort23]> { let Latency = 4; } +def : WriteRes<WriteMove, [SBPort015]>; +def : WriteRes<WriteZero, []>; + +defm : SBWriteResPair<WriteALU, SBPort015, 1>; +defm : SBWriteResPair<WriteIMul, SBPort1, 3>; +defm : SBWriteResPair<WriteShift, SBPort05, 1>; +defm : SBWriteResPair<WriteJump, SBPort5, 1>; + +// This is for simple LEAs with one or two input operands. +// The complex ones can only execute on port 1, and they require two cycles on +// the port to read all inputs. We don't model that. +def : WriteRes<WriteLEA, [SBPort15]>; + +// This is quite rough, latency depends on the dividend. +def : WriteRes<WriteIDiv, [SBPort0, SBDivider]> { + let Latency = 25; + let ResourceCycles = [1, 10]; +} +def : WriteRes<WriteIDivLd, [SBPort23, SBPort0, SBDivider]> { + let Latency = 29; + let ResourceCycles = [1, 1, 10]; +} + +// Scalar and vector floating point. +defm : SBWriteResPair<WriteFAdd, SBPort1, 3>; +defm : SBWriteResPair<WriteFMul, SBPort0, 5>; +defm : SBWriteResPair<WriteFDiv, SBPort0, 12>; // 10-14 cycles. +defm : SBWriteResPair<WriteFRcp, SBPort0, 5>; +defm : SBWriteResPair<WriteFSqrt, SBPort0, 15>; +defm : SBWriteResPair<WriteCvtF2I, SBPort1, 3>; +defm : SBWriteResPair<WriteCvtI2F, SBPort1, 4>; +defm : SBWriteResPair<WriteCvtF2F, SBPort1, 3>; + +// Vector integer operations. +defm : SBWriteResPair<WriteVecShift, SBPort05, 1>; +defm : SBWriteResPair<WriteVecLogic, SBPort015, 1>; +defm : SBWriteResPair<WriteVecALU, SBPort15, 1>; +defm : SBWriteResPair<WriteVecIMul, SBPort0, 5>; +defm : SBWriteResPair<WriteShuffle, SBPort15, 1>; + +def : WriteRes<WriteSystem, [SBPort015]> { let Latency = 100; } +def : WriteRes<WriteMicrocoded, [SBPort015]> { let Latency = 100; } +} // SchedModel diff --git a/lib/Target/X86/X86Schedule.td b/lib/Target/X86/X86Schedule.td index da0ca7d..9fbde88 100644 --- a/lib/Target/X86/X86Schedule.td +++ b/lib/Target/X86/X86Schedule.td @@ -53,6 +53,10 @@ def WriteLoad : SchedWrite; def WriteStore : SchedWrite; def WriteMove : SchedWrite; +// Idioms that clear a register, like xorps %xmm0, %xmm0. +// These can often bypass execution ports completely. +def WriteZero : SchedWrite; + // Branches don't produce values, so they have no latency, but they still // consume resources. Indirect branches can fold loads. defm WriteJump : X86SchedWritePair; @@ -63,6 +67,10 @@ defm WriteFMul : X86SchedWritePair; // Floating point multiplication. defm WriteFDiv : X86SchedWritePair; // Floating point division. defm WriteFSqrt : X86SchedWritePair; // Floating point square root. defm WriteFRcp : X86SchedWritePair; // Floating point reciprocal. +defm WriteFMA : X86SchedWritePair; // Fused Multiply Add. + +// FMA Scheduling helper class. +class FMASC { X86FoldableSchedWrite Sched = WriteFAdd; } // Vector integer operations. defm WriteVecALU : X86SchedWritePair; // Vector integer ALU op, no logicals. @@ -79,9 +87,14 @@ defm WriteCvtF2I : X86SchedWritePair; // Float -> Integer. defm WriteCvtI2F : X86SchedWritePair; // Integer -> Float. defm WriteCvtF2F : X86SchedWritePair; // Float -> Float size conversion. +// Catch-all for expensive system instructions. +def WriteSystem : SchedWrite; + +// Old microcoded instructions that nobody use. +def WriteMicrocoded : SchedWrite; + //===----------------------------------------------------------------------===// // Instruction Itinerary classes used for X86 -def IIC_DEFAULT : InstrItinClass; def IIC_ALU_MEM : InstrItinClass; def IIC_ALU_NONMEM : InstrItinClass; def IIC_LEA : InstrItinClass; @@ -556,3 +569,5 @@ def GenericModel : SchedMachineModel { } include "X86ScheduleAtom.td" +include "X86SchedSandyBridge.td" +include "X86SchedHaswell.td" diff --git a/lib/Target/X86/X86ScheduleAtom.td b/lib/Target/X86/X86ScheduleAtom.td index 1e5f2d6..cce8f1b 100644 --- a/lib/Target/X86/X86ScheduleAtom.td +++ b/lib/Target/X86/X86ScheduleAtom.td @@ -33,7 +33,6 @@ def AtomItineraries : ProcessorItineraries< // InstrItinData<class, [InstrStage<N, [P0], 0>, InstrStage<N, [P1]>] >, // // Default is 1 cycle, port0 or port1 - InstrItinData<IIC_DEFAULT, [InstrStage<1, [Port0, Port1]>] >, InstrItinData<IIC_ALU_MEM, [InstrStage<1, [Port0]>] >, InstrItinData<IIC_ALU_NONMEM, [InstrStage<1, [Port0, Port1]>] >, InstrItinData<IIC_LEA, [InstrStage<1, [Port1]>] >, diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp index 0f2c008..448d2e6 100644 --- a/lib/Target/X86/X86Subtarget.cpp +++ b/lib/Target/X86/X86Subtarget.cpp @@ -37,8 +37,7 @@ using namespace llvm; /// ClassifyBlockAddressReference - Classify a blockaddress reference for the /// current subtarget according to how we should reference it in a non-pcrel /// context. -unsigned char X86Subtarget:: -ClassifyBlockAddressReference() const { +unsigned char X86Subtarget::ClassifyBlockAddressReference() const { if (isPICStyleGOT()) // 32-bit ELF targets. return X86II::MO_GOTOFF; @@ -283,6 +282,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() { HasLZCNT = true; ToggleFeature(X86::FeatureLZCNT); } + if (IsIntel && ((ECX >> 8) & 0x1)) { + HasPRFCHW = true; + ToggleFeature(X86::FeaturePRFCHW); + } if (IsAMD) { if ((ECX >> 6) & 0x1) { HasSSE4A = true; @@ -310,6 +313,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() { HasBMI = true; ToggleFeature(X86::FeatureBMI); } + if ((EBX >> 4) & 0x1) { + HasHLE = true; + ToggleFeature(X86::FeatureHLE); + } if (IsIntel && ((EBX >> 5) & 0x1)) { X86SSELevel = AVX2; ToggleFeature(X86::FeatureAVX2); @@ -322,6 +329,14 @@ void X86Subtarget::AutoDetectSubtargetFeatures() { HasRTM = true; ToggleFeature(X86::FeatureRTM); } + if (IsIntel && ((EBX >> 19) & 0x1)) { + HasADX = true; + ToggleFeature(X86::FeatureADX); + } + if (IsIntel && ((EBX >> 18) & 0x1)) { + HasRDSEED = true; + ToggleFeature(X86::FeatureRDSEED); + } } } } @@ -439,7 +454,10 @@ void X86Subtarget::initializeEnvironment() { HasBMI = false; HasBMI2 = false; HasRTM = false; + HasHLE = false; HasADX = false; + HasPRFCHW = false; + HasRDSEED = false; IsBTMemSlow = false; IsUAMemFast = false; HasVectorUAMem = false; @@ -448,6 +466,8 @@ void X86Subtarget::initializeEnvironment() { HasSlowDivide = false; PostRAScheduler = false; PadShortFunctions = false; + CallRegIndirect = false; + LEAUsesAG = false; stackAlignment = 4; // FIXME: this is a known good value for Yonah. How about others? MaxInlineSizeThreshold = 128; diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h index e97da4b..66832b9 100644 --- a/lib/Target/X86/X86Subtarget.h +++ b/lib/Target/X86/X86Subtarget.h @@ -121,9 +121,18 @@ protected: /// HasRTM - Processor has RTM instructions. bool HasRTM; + /// HasHLE - Processor has HLE. + bool HasHLE; + /// HasADX - Processor has ADX instructions. bool HasADX; + /// HasPRFCHW - Processor has PRFCHW instructions. + bool HasPRFCHW; + + /// HasRDSEED - Processor has RDSEED instructions. + bool HasRDSEED; + /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow. bool IsBTMemSlow; @@ -153,6 +162,13 @@ protected: /// a stall when returning too early. bool PadShortFunctions; + /// CallRegIndirect - True if the Calls with memory reference should be converted + /// to a register-based indirect call. + bool CallRegIndirect; + /// LEAUsesAG - True if the LEA instruction inputs have to be ready at + /// address generation (AG) time. + bool LEAUsesAG; + /// stackAlignment - The minimum alignment known to hold of the stack frame on /// entry to the function and which must be maintained by every function. unsigned stackAlignment; @@ -253,7 +269,10 @@ public: bool hasBMI() const { return HasBMI; } bool hasBMI2() const { return HasBMI2; } bool hasRTM() const { return HasRTM; } + bool hasHLE() const { return HasHLE; } bool hasADX() const { return HasADX; } + bool hasPRFCHW() const { return HasPRFCHW; } + bool hasRDSEED() const { return HasRDSEED; } bool isBTMemSlow() const { return IsBTMemSlow; } bool isUnalignedMemAccessFast() const { return IsUAMemFast; } bool hasVectorUAMem() const { return HasVectorUAMem; } @@ -261,6 +280,8 @@ public: bool useLeaForSP() const { return UseLeaForSP; } bool hasSlowDivide() const { return HasSlowDivide; } bool padShortFunctions() const { return PadShortFunctions; } + bool callRegIndirect() const { return CallRegIndirect; } + bool LEAusesAG() const { return LEAUsesAG; } bool isAtom() const { return X86ProcFamily == IntelAtom; } diff --git a/lib/Target/X86/X86TargetMachine.cpp b/lib/Target/X86/X86TargetMachine.cpp index 8aa58a2..00fa47f 100644 --- a/lib/Target/X86/X86TargetMachine.cpp +++ b/lib/Target/X86/X86TargetMachine.cpp @@ -215,6 +215,11 @@ bool X86PassConfig::addPreEmitPass() { addPass(createX86PadShortFunctions()); ShouldPrint = true; } + if (getOptLevel() != CodeGenOpt::None && + getX86Subtarget().LEAusesAG()){ + addPass(createX86FixupLEAs()); + ShouldPrint = true; + } return ShouldPrint; } diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp index be2a997..eba9d78 100644 --- a/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/lib/Target/X86/X86TargetTransformInfo.cpp @@ -86,7 +86,9 @@ public: virtual unsigned getNumberOfRegisters(bool Vector) const; virtual unsigned getRegisterBitWidth(bool Vector) const; virtual unsigned getMaximumUnrollFactor() const; - virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty) const; + virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty, + OperandValueKind, + OperandValueKind) const; virtual unsigned getShuffleCost(ShuffleKind Kind, Type *Tp, int Index, Type *SubTp) const; virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst, @@ -162,13 +164,109 @@ unsigned X86TTI::getMaximumUnrollFactor() const { return 2; } -unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty) const { +unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty, + OperandValueKind Op1Info, + OperandValueKind Op2Info) const { // Legalize the type. std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Ty); int ISD = TLI->InstructionOpcodeToISD(Opcode); assert(ISD && "Invalid opcode"); + static const CostTblEntry<MVT> AVX2CostTable[] = { + // Shifts on v4i64/v8i32 on AVX2 is legal even though we declare to + // customize them to detect the cases where shift amount is a scalar one. + { ISD::SHL, MVT::v4i32, 1 }, + { ISD::SRL, MVT::v4i32, 1 }, + { ISD::SRA, MVT::v4i32, 1 }, + { ISD::SHL, MVT::v8i32, 1 }, + { ISD::SRL, MVT::v8i32, 1 }, + { ISD::SRA, MVT::v8i32, 1 }, + { ISD::SHL, MVT::v2i64, 1 }, + { ISD::SRL, MVT::v2i64, 1 }, + { ISD::SHL, MVT::v4i64, 1 }, + { ISD::SRL, MVT::v4i64, 1 }, + + { ISD::SHL, MVT::v32i8, 42 }, // cmpeqb sequence. + { ISD::SHL, MVT::v16i16, 16*10 }, // Scalarized. + + { ISD::SRL, MVT::v32i8, 32*10 }, // Scalarized. + { ISD::SRL, MVT::v16i16, 8*10 }, // Scalarized. + + { ISD::SRA, MVT::v32i8, 32*10 }, // Scalarized. + { ISD::SRA, MVT::v16i16, 16*10 }, // Scalarized. + { ISD::SRA, MVT::v4i64, 4*10 }, // Scalarized. + }; + + // Look for AVX2 lowering tricks. + if (ST->hasAVX2()) { + int Idx = CostTableLookup<MVT>(AVX2CostTable, array_lengthof(AVX2CostTable), + ISD, LT.second); + if (Idx != -1) + return LT.first * AVX2CostTable[Idx].Cost; + } + + static const CostTblEntry<MVT> SSE2UniformConstCostTable[] = { + // We don't correctly identify costs of casts because they are marked as + // custom. + // Constant splats are cheaper for the following instructions. + { ISD::SHL, MVT::v16i8, 1 }, // psllw. + { ISD::SHL, MVT::v8i16, 1 }, // psllw. + { ISD::SHL, MVT::v4i32, 1 }, // pslld + { ISD::SHL, MVT::v2i64, 1 }, // psllq. + + { ISD::SRL, MVT::v16i8, 1 }, // psrlw. + { ISD::SRL, MVT::v8i16, 1 }, // psrlw. + { ISD::SRL, MVT::v4i32, 1 }, // psrld. + { ISD::SRL, MVT::v2i64, 1 }, // psrlq. + + { ISD::SRA, MVT::v16i8, 4 }, // psrlw, pand, pxor, psubb. + { ISD::SRA, MVT::v8i16, 1 }, // psraw. + { ISD::SRA, MVT::v4i32, 1 }, // psrad. + }; + + if (Op2Info == TargetTransformInfo::OK_UniformConstantValue && + ST->hasSSE2()) { + int Idx = CostTableLookup<MVT>(SSE2UniformConstCostTable, + array_lengthof(SSE2UniformConstCostTable), + ISD, LT.second); + if (Idx != -1) + return LT.first * SSE2UniformConstCostTable[Idx].Cost; + } + + + static const CostTblEntry<MVT> SSE2CostTable[] = { + // We don't correctly identify costs of casts because they are marked as + // custom. + // For some cases, where the shift amount is a scalar we would be able + // to generate better code. Unfortunately, when this is the case the value + // (the splat) will get hoisted out of the loop, thereby making it invisible + // to ISel. The cost model must return worst case assumptions because it is + // used for vectorization and we don't want to make vectorized code worse + // than scalar code. + { ISD::SHL, MVT::v16i8, 30 }, // cmpeqb sequence. + { ISD::SHL, MVT::v8i16, 8*10 }, // Scalarized. + { ISD::SHL, MVT::v4i32, 2*5 }, // We optimized this using mul. + { ISD::SHL, MVT::v2i64, 2*10 }, // Scalarized. + + { ISD::SRL, MVT::v16i8, 16*10 }, // Scalarized. + { ISD::SRL, MVT::v8i16, 8*10 }, // Scalarized. + { ISD::SRL, MVT::v4i32, 4*10 }, // Scalarized. + { ISD::SRL, MVT::v2i64, 2*10 }, // Scalarized. + + { ISD::SRA, MVT::v16i8, 16*10 }, // Scalarized. + { ISD::SRA, MVT::v8i16, 8*10 }, // Scalarized. + { ISD::SRA, MVT::v4i32, 4*10 }, // Scalarized. + { ISD::SRA, MVT::v2i64, 2*10 }, // Scalarized. + }; + + if (ST->hasSSE2()) { + int Idx = CostTableLookup<MVT>(SSE2CostTable, array_lengthof(SSE2CostTable), + ISD, LT.second); + if (Idx != -1) + return LT.first * SSE2CostTable[Idx].Cost; + } + static const CostTblEntry<MVT> AVX1CostTable[] = { // We don't have to scalarize unsupported ops. We can issue two half-sized // operations and we only need to extract the upper YMM half. @@ -213,7 +311,8 @@ unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty) const { return 6; // Fallback to the default implementation. - return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty); + return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Op1Info, + Op2Info); } unsigned X86TTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index, @@ -235,9 +334,44 @@ unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const { int ISD = TLI->InstructionOpcodeToISD(Opcode); assert(ISD && "Invalid opcode"); + std::pair<unsigned, MVT> LTSrc = TLI->getTypeLegalizationCost(Src); + std::pair<unsigned, MVT> LTDest = TLI->getTypeLegalizationCost(Dst); + + static const TypeConversionCostTblEntry<MVT> SSE2ConvTbl[] = { + // These are somewhat magic numbers justified by looking at the output of + // Intel's IACA, running some kernels and making sure when we take + // legalization into account the throughput will be overestimated. + { ISD::UINT_TO_FP, MVT::v2f64, MVT::v2i64, 2*10 }, + { ISD::UINT_TO_FP, MVT::v2f64, MVT::v4i32, 4*10 }, + { ISD::UINT_TO_FP, MVT::v2f64, MVT::v8i16, 8*10 }, + { ISD::UINT_TO_FP, MVT::v2f64, MVT::v16i8, 16*10 }, + { ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i64, 2*10 }, + { ISD::SINT_TO_FP, MVT::v2f64, MVT::v4i32, 4*10 }, + { ISD::SINT_TO_FP, MVT::v2f64, MVT::v8i16, 8*10 }, + { ISD::SINT_TO_FP, MVT::v2f64, MVT::v16i8, 16*10 }, + // There are faster sequences for float conversions. + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v2i64, 15 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i32, 15 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v8i16, 15 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v16i8, 8 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v2i64, 15 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i32, 15 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v8i16, 15 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v16i8, 8 }, + }; + + if (ST->hasSSE2() && !ST->hasAVX()) { + int Idx = ConvertCostTableLookup<MVT>(SSE2ConvTbl, + array_lengthof(SSE2ConvTbl), + ISD, LTDest.second, LTSrc.second); + if (Idx != -1) + return LTSrc.first * SSE2ConvTbl[Idx].Cost; + } + EVT SrcTy = TLI->getValueType(Src); EVT DstTy = TLI->getValueType(Dst); + // The function getSimpleVT only handles simple value types. if (!SrcTy.isSimple() || !DstTy.isSimple()) return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src); @@ -248,17 +382,40 @@ unsigned X86TTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const { { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 1 }, { ISD::TRUNCATE, MVT::v4i32, MVT::v4i64, 1 }, { ISD::TRUNCATE, MVT::v8i16, MVT::v8i32, 1 }, - { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i8, 1 }, - { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i8, 1 }, - { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i8, 1 }, - { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i8, 1 }, + + { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i1, 8 }, + { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i8, 8 }, + { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i16, 5 }, + { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i32, 1 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i1, 3 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i8, 3 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i16, 3 }, + { ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 }, + { ISD::SINT_TO_FP, MVT::v4f64, MVT::v4i1, 3 }, + { ISD::SINT_TO_FP, MVT::v4f64, MVT::v4i8, 3 }, + { ISD::SINT_TO_FP, MVT::v4f64, MVT::v4i16, 3 }, + { ISD::SINT_TO_FP, MVT::v4f64, MVT::v4i32, 1 }, + + { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i1, 6 }, + { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i8, 5 }, + { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i16, 5 }, + { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i32, 9 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i1, 7 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i8, 2 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 }, + { ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i32, 6 }, + { ISD::UINT_TO_FP, MVT::v4f64, MVT::v4i1, 7 }, + { ISD::UINT_TO_FP, MVT::v4f64, MVT::v4i8, 2 }, + { ISD::UINT_TO_FP, MVT::v4f64, MVT::v4i16, 2 }, + { ISD::UINT_TO_FP, MVT::v4f64, MVT::v4i32, 6 }, + { ISD::FP_TO_SINT, MVT::v8i8, MVT::v8f32, 1 }, { ISD::FP_TO_SINT, MVT::v4i8, MVT::v4f32, 1 }, { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i1, 6 }, { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i1, 9 }, { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i1, 8 }, - { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i8, 8 }, - { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 8 }, + { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i8, 6 }, + { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 6 }, { ISD::TRUNCATE, MVT::v8i32, MVT::v8i64, 3 }, }; |