diff options
| author | Eli Bendersky <eliben@chromium.org> | 2013-07-15 16:09:15 -0700 |
|---|---|---|
| committer | Eli Bendersky <eliben@chromium.org> | 2013-07-15 16:09:15 -0700 |
| commit | c6cf05cb5108f356dde97c01ee4188b0671d4542 (patch) | |
| tree | 436fdc2a55296d3c202e7ef11f31be3be53efb5f /lib/Target/Mips | |
| parent | c75199c649c739aade160289d93f257edc798cde (diff) | |
| parent | 7dfcb84fc16b3bf6b2379713b53090757f0a45f9 (diff) | |
Merge commit '7dfcb84fc16b3bf6b2379713b53090757f0a45f9'
Conflicts:
docs/LangRef.rst
include/llvm/CodeGen/CallingConvLower.h
include/llvm/IRReader/IRReader.h
include/llvm/Target/TargetMachine.h
lib/CodeGen/CallingConvLower.cpp
lib/IRReader/IRReader.cpp
lib/IRReader/LLVMBuild.txt
lib/IRReader/Makefile
lib/LLVMBuild.txt
lib/Makefile
lib/Support/MemoryBuffer.cpp
lib/Support/Unix/PathV2.inc
lib/Target/ARM/ARMBaseInstrInfo.cpp
lib/Target/ARM/ARMISelLowering.cpp
lib/Target/ARM/ARMInstrInfo.td
lib/Target/ARM/ARMSubtarget.cpp
lib/Target/ARM/ARMTargetMachine.cpp
lib/Target/Mips/CMakeLists.txt
lib/Target/Mips/MipsDelaySlotFiller.cpp
lib/Target/Mips/MipsISelLowering.cpp
lib/Target/Mips/MipsInstrInfo.td
lib/Target/Mips/MipsSubtarget.cpp
lib/Target/Mips/MipsSubtarget.h
lib/Target/X86/X86FastISel.cpp
lib/Target/X86/X86ISelDAGToDAG.cpp
lib/Target/X86/X86ISelLowering.cpp
lib/Target/X86/X86InstrControl.td
lib/Target/X86/X86InstrFormats.td
lib/Transforms/IPO/ExtractGV.cpp
lib/Transforms/InstCombine/InstCombineCompares.cpp
lib/Transforms/Utils/SimplifyLibCalls.cpp
test/CodeGen/X86/fast-isel-divrem.ll
test/MC/ARM/data-in-code.ll
tools/Makefile
tools/llvm-extract/llvm-extract.cpp
tools/llvm-link/CMakeLists.txt
tools/opt/CMakeLists.txt
tools/opt/LLVMBuild.txt
tools/opt/Makefile
tools/opt/opt.cpp
Diffstat (limited to 'lib/Target/Mips')
58 files changed, 6538 insertions, 3380 deletions
diff --git a/lib/Target/Mips/AsmParser/MipsAsmParser.cpp b/lib/Target/Mips/AsmParser/MipsAsmParser.cpp index ade6084752..0795cb963b 100644 --- a/lib/Target/Mips/AsmParser/MipsAsmParser.cpp +++ b/lib/Target/Mips/AsmParser/MipsAsmParser.cpp @@ -63,7 +63,6 @@ class MipsAsmParser : public MCTargetAsmParser { MCAsmParser &Parser; MipsAssemblerOptions Options; - #define GET_ASSEMBLER_HEADER #include "MipsGenAsmMatcher.inc" @@ -101,6 +100,9 @@ class MipsAsmParser : public MCTargetAsmParser { MipsAsmParser::OperandMatchResultTy parseCCRRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands); + bool searchSymbolAlias(SmallVectorImpl<MCParsedAsmOperand*> &Operands, + unsigned RegisterClass); + bool ParseOperand(SmallVectorImpl<MCParsedAsmOperand*> &, StringRef Mnemonic); @@ -119,11 +121,17 @@ class MipsAsmParser : public MCTargetAsmParser { SmallVectorImpl<MCInst> &Instructions); void expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc, SmallVectorImpl<MCInst> &Instructions); + void expandMemInst(MCInst &Inst, SMLoc IDLoc, + SmallVectorImpl<MCInst> &Instructions, + bool isLoad,bool isImmOpnd); bool reportParseError(StringRef ErrorMsg); - bool parseMemOffset(const MCExpr *&Res); + bool parseMemOffset(const MCExpr *&Res, bool isParenExpr); bool parseRelocOperand(const MCExpr *&Res); + const MCExpr* evaluateRelocExpr(const MCExpr *Expr, StringRef RelocStr); + + bool isEvaluated(const MCExpr *Expr); bool parseDirectiveSet(); bool parseSetAtDirective(); @@ -133,6 +141,8 @@ class MipsAsmParser : public MCTargetAsmParser { bool parseSetReorderDirective(); bool parseSetNoReorderDirective(); + bool parseSetAssignment(); + bool parseDirectiveWord(unsigned Size, SMLoc L); MCSymbolRefExpr::VariantKind getVariantKind(StringRef Symbol); @@ -163,9 +173,12 @@ class MipsAsmParser : public MCTargetAsmParser { bool requestsDoubleOperand(StringRef Mnemonic); - unsigned getReg(int RC,int RegNo); + unsigned getReg(int RC, int RegNo); int getATReg(); + + bool processInstruction(MCInst &Inst, SMLoc IDLoc, + SmallVectorImpl<MCInst> &Instructions); public: MipsAsmParser(MCSubtargetInfo &sti, MCAsmParser &parser) : MCTargetAsmParser(), STI(sti), Parser(parser) { @@ -211,25 +224,30 @@ private: MipsOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {} + struct Token { + const char *Data; + unsigned Length; + }; + + struct RegOp { + unsigned RegNum; + RegisterKind Kind; + }; + + struct ImmOp { + const MCExpr *Val; + }; + + struct MemOp { + unsigned Base; + const MCExpr *Off; + }; + union { - struct { - const char *Data; - unsigned Length; - } Tok; - - struct { - unsigned RegNum; - RegisterKind Kind; - } Reg; - - struct { - const MCExpr *Val; - } Imm; - - struct { - unsigned Base; - const MCExpr *Off; - } Mem; + struct Token Tok; + struct RegOp Reg; + struct ImmOp Imm; + struct MemOp Mem; }; SMLoc StartLoc, EndLoc; @@ -253,7 +271,7 @@ public: void addImmOperands(MCInst &Inst, unsigned N) const { assert(N == 1 && "Invalid number of operands!"); const MCExpr *Expr = getImm(); - addExpr(Inst,Expr); + addExpr(Inst, Expr); } void addMemOperands(MCInst &Inst, unsigned N) const { @@ -262,7 +280,7 @@ public: Inst.addOperand(MCOperand::CreateReg(getMemBase())); const MCExpr *Expr = getMemOff(); - addExpr(Inst,Expr); + addExpr(Inst, Expr); } bool isReg() const { return Kind == k_Register; } @@ -375,42 +393,112 @@ public: } /// getStartLoc - Get the location of the first token of this operand. - SMLoc getStartLoc() const { return StartLoc; } + SMLoc getStartLoc() const { + return StartLoc; + } /// getEndLoc - Get the location of the last token of this operand. - SMLoc getEndLoc() const { return EndLoc; } + SMLoc getEndLoc() const { + return EndLoc; + } virtual void print(raw_ostream &OS) const { llvm_unreachable("unimplemented!"); } -}; +}; // class MipsOperand +} // namespace + +namespace llvm { +extern const MCInstrDesc MipsInsts[]; +} +static const MCInstrDesc &getInstDesc(unsigned Opcode) { + return MipsInsts[Opcode]; +} + +bool MipsAsmParser::processInstruction(MCInst &Inst, SMLoc IDLoc, + SmallVectorImpl<MCInst> &Instructions) { + const MCInstrDesc &MCID = getInstDesc(Inst.getOpcode()); + Inst.setLoc(IDLoc); + if (MCID.hasDelaySlot() && Options.isReorder()) { + // If this instruction has a delay slot and .set reorder is active, + // emit a NOP after it. + Instructions.push_back(Inst); + MCInst NopInst; + NopInst.setOpcode(Mips::SLL); + NopInst.addOperand(MCOperand::CreateReg(Mips::ZERO)); + NopInst.addOperand(MCOperand::CreateReg(Mips::ZERO)); + NopInst.addOperand(MCOperand::CreateImm(0)); + Instructions.push_back(NopInst); + return false; + } + + if (MCID.mayLoad() || MCID.mayStore()) { + // Check the offset of memory operand, if it is a symbol + // reference or immediate we may have to expand instructions. + for (unsigned i = 0; i < MCID.getNumOperands(); i++) { + const MCOperandInfo &OpInfo = MCID.OpInfo[i]; + if ((OpInfo.OperandType == MCOI::OPERAND_MEMORY) + || (OpInfo.OperandType == MCOI::OPERAND_UNKNOWN)) { + MCOperand &Op = Inst.getOperand(i); + if (Op.isImm()) { + int MemOffset = Op.getImm(); + if (MemOffset < -32768 || MemOffset > 32767) { + // Offset can't exceed 16bit value. + expandMemInst(Inst, IDLoc, Instructions, MCID.mayLoad(), true); + return false; + } + } else if (Op.isExpr()) { + const MCExpr *Expr = Op.getExpr(); + if (Expr->getKind() == MCExpr::SymbolRef) { + const MCSymbolRefExpr *SR = + static_cast<const MCSymbolRefExpr*>(Expr); + if (SR->getKind() == MCSymbolRefExpr::VK_None) { + // Expand symbol. + expandMemInst(Inst, IDLoc, Instructions, MCID.mayLoad(), false); + return false; + } + } else if (!isEvaluated(Expr)) { + expandMemInst(Inst, IDLoc, Instructions, MCID.mayLoad(), false); + return false; + } + } + } + } // for + } // if load/store + + if (needsExpansion(Inst)) + expandInstruction(Inst, IDLoc, Instructions); + else + Instructions.push_back(Inst); + + return false; } bool MipsAsmParser::needsExpansion(MCInst &Inst) { - switch(Inst.getOpcode()) { - case Mips::LoadImm32Reg: - case Mips::LoadAddr32Imm: - case Mips::LoadAddr32Reg: - return true; - default: - return false; + switch (Inst.getOpcode()) { + case Mips::LoadImm32Reg: + case Mips::LoadAddr32Imm: + case Mips::LoadAddr32Reg: + return true; + default: + return false; } } void MipsAsmParser::expandInstruction(MCInst &Inst, SMLoc IDLoc, - SmallVectorImpl<MCInst> &Instructions){ - switch(Inst.getOpcode()) { - case Mips::LoadImm32Reg: - return expandLoadImm(Inst, IDLoc, Instructions); - case Mips::LoadAddr32Imm: - return expandLoadAddressImm(Inst,IDLoc,Instructions); - case Mips::LoadAddr32Reg: - return expandLoadAddressReg(Inst,IDLoc,Instructions); - } + SmallVectorImpl<MCInst> &Instructions) { + switch (Inst.getOpcode()) { + case Mips::LoadImm32Reg: + return expandLoadImm(Inst, IDLoc, Instructions); + case Mips::LoadAddr32Imm: + return expandLoadAddressImm(Inst, IDLoc, Instructions); + case Mips::LoadAddr32Reg: + return expandLoadAddressReg(Inst, IDLoc, Instructions); + } } void MipsAsmParser::expandLoadImm(MCInst &Inst, SMLoc IDLoc, - SmallVectorImpl<MCInst> &Instructions){ + SmallVectorImpl<MCInst> &Instructions) { MCInst tmpInst; const MCOperand &ImmOp = Inst.getOperand(1); assert(ImmOp.isImm() && "expected immediate operand kind"); @@ -419,26 +507,24 @@ void MipsAsmParser::expandLoadImm(MCInst &Inst, SMLoc IDLoc, int ImmValue = ImmOp.getImm(); tmpInst.setLoc(IDLoc); - if ( 0 <= ImmValue && ImmValue <= 65535) { - // for 0 <= j <= 65535. + if (0 <= ImmValue && ImmValue <= 65535) { + // For 0 <= j <= 65535. // li d,j => ori d,$zero,j tmpInst.setOpcode(Mips::ORi); tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg())); - tmpInst.addOperand( - MCOperand::CreateReg(Mips::ZERO)); + tmpInst.addOperand(MCOperand::CreateReg(Mips::ZERO)); tmpInst.addOperand(MCOperand::CreateImm(ImmValue)); Instructions.push_back(tmpInst); - } else if ( ImmValue < 0 && ImmValue >= -32768) { - // for -32768 <= j < 0. + } else if (ImmValue < 0 && ImmValue >= -32768) { + // For -32768 <= j < 0. // li d,j => addiu d,$zero,j tmpInst.setOpcode(Mips::ADDiu); tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg())); - tmpInst.addOperand( - MCOperand::CreateReg(Mips::ZERO)); + tmpInst.addOperand(MCOperand::CreateReg(Mips::ZERO)); tmpInst.addOperand(MCOperand::CreateImm(ImmValue)); Instructions.push_back(tmpInst); } else { - // for any other value of j that is representable as a 32-bit integer. + // For any other value of j that is representable as a 32-bit integer. // li d,j => lui d,hi16(j) // ori d,d,lo16(j) tmpInst.setOpcode(Mips::LUi); @@ -456,7 +542,7 @@ void MipsAsmParser::expandLoadImm(MCInst &Inst, SMLoc IDLoc, } void MipsAsmParser::expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc, - SmallVectorImpl<MCInst> &Instructions){ + SmallVectorImpl<MCInst> &Instructions) { MCInst tmpInst; const MCOperand &ImmOp = Inst.getOperand(2); assert(ImmOp.isImm() && "expected immediate operand kind"); @@ -465,19 +551,19 @@ void MipsAsmParser::expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc, const MCOperand &DstRegOp = Inst.getOperand(0); assert(DstRegOp.isReg() && "expected register operand kind"); int ImmValue = ImmOp.getImm(); - if ( -32768 <= ImmValue && ImmValue <= 65535) { - //for -32768 <= j <= 65535. - //la d,j(s) => addiu d,s,j + if (-32768 <= ImmValue && ImmValue <= 65535) { + // For -32768 <= j <= 65535. + // la d,j(s) => addiu d,s,j tmpInst.setOpcode(Mips::ADDiu); tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg())); tmpInst.addOperand(MCOperand::CreateReg(SrcRegOp.getReg())); tmpInst.addOperand(MCOperand::CreateImm(ImmValue)); Instructions.push_back(tmpInst); } else { - //for any other value of j that is representable as a 32-bit integer. - //la d,j(s) => lui d,hi16(j) - // ori d,d,lo16(j) - // addu d,d,s + // For any other value of j that is representable as a 32-bit integer. + // la d,j(s) => lui d,hi16(j) + // ori d,d,lo16(j) + // addu d,d,s tmpInst.setOpcode(Mips::LUi); tmpInst.addOperand(MCOperand::CreateReg(DstRegOp.getReg())); tmpInst.addOperand(MCOperand::CreateImm((ImmValue & 0xffff0000) >> 16)); @@ -498,26 +584,25 @@ void MipsAsmParser::expandLoadAddressReg(MCInst &Inst, SMLoc IDLoc, } void MipsAsmParser::expandLoadAddressImm(MCInst &Inst, SMLoc IDLoc, - SmallVectorImpl<MCInst> &Instructions){ + SmallVectorImpl<MCInst> &Instructions) { MCInst tmpInst; const MCOperand &ImmOp = Inst.getOperand(1); assert(ImmOp.isImm() && "expected immediate operand kind"); const MCOperand &RegOp = Inst.getOperand(0); assert(RegOp.isReg() && "expected register operand kind"); int ImmValue = ImmOp.getImm(); - if ( -32768 <= ImmValue && ImmValue <= 65535) { - //for -32768 <= j <= 65535. - //la d,j => addiu d,$zero,j + if (-32768 <= ImmValue && ImmValue <= 65535) { + // For -32768 <= j <= 65535. + // la d,j => addiu d,$zero,j tmpInst.setOpcode(Mips::ADDiu); tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg())); - tmpInst.addOperand( - MCOperand::CreateReg(Mips::ZERO)); + tmpInst.addOperand(MCOperand::CreateReg(Mips::ZERO)); tmpInst.addOperand(MCOperand::CreateImm(ImmValue)); Instructions.push_back(tmpInst); } else { - //for any other value of j that is representable as a 32-bit integer. - //la d,j => lui d,hi16(j) - // ori d,d,lo16(j) + // For any other value of j that is representable as a 32-bit integer. + // la d,j => lui d,hi16(j) + // ori d,d,lo16(j) tmpInst.setOpcode(Mips::LUi); tmpInst.addOperand(MCOperand::CreateReg(RegOp.getReg())); tmpInst.addOperand(MCOperand::CreateImm((ImmValue & 0xffff0000) >> 16)); @@ -531,28 +616,105 @@ void MipsAsmParser::expandLoadAddressImm(MCInst &Inst, SMLoc IDLoc, } } +void MipsAsmParser::expandMemInst(MCInst &Inst, SMLoc IDLoc, + SmallVectorImpl<MCInst> &Instructions, bool isLoad, bool isImmOpnd) { + const MCSymbolRefExpr *SR; + MCInst TempInst; + unsigned ImmOffset, HiOffset, LoOffset; + const MCExpr *ExprOffset; + unsigned TmpRegNum; + unsigned AtRegNum = getReg((isMips64()) ? Mips::CPU64RegsRegClassID + : Mips::CPURegsRegClassID, getATReg()); + // 1st operand is either the source or destination register. + assert(Inst.getOperand(0).isReg() && "expected register operand kind"); + unsigned RegOpNum = Inst.getOperand(0).getReg(); + // 2nd operand is the base register. + assert(Inst.getOperand(1).isReg() && "expected register operand kind"); + unsigned BaseRegNum = Inst.getOperand(1).getReg(); + // 3rd operand is either an immediate or expression. + if (isImmOpnd) { + assert(Inst.getOperand(2).isImm() && "expected immediate operand kind"); + ImmOffset = Inst.getOperand(2).getImm(); + LoOffset = ImmOffset & 0x0000ffff; + HiOffset = (ImmOffset & 0xffff0000) >> 16; + // If msb of LoOffset is 1(negative number) we must increment HiOffset. + if (LoOffset & 0x8000) + HiOffset++; + } else + ExprOffset = Inst.getOperand(2).getExpr(); + // All instructions will have the same location. + TempInst.setLoc(IDLoc); + // 1st instruction in expansion is LUi. For load instruction we can use + // the dst register as a temporary if base and dst are different, + // but for stores we must use $at. + TmpRegNum = (isLoad && (BaseRegNum != RegOpNum)) ? RegOpNum : AtRegNum; + TempInst.setOpcode(Mips::LUi); + TempInst.addOperand(MCOperand::CreateReg(TmpRegNum)); + if (isImmOpnd) + TempInst.addOperand(MCOperand::CreateImm(HiOffset)); + else { + if (ExprOffset->getKind() == MCExpr::SymbolRef) { + SR = static_cast<const MCSymbolRefExpr*>(ExprOffset); + const MCSymbolRefExpr *HiExpr = MCSymbolRefExpr::Create( + SR->getSymbol().getName(), MCSymbolRefExpr::VK_Mips_ABS_HI, + getContext()); + TempInst.addOperand(MCOperand::CreateExpr(HiExpr)); + } else { + const MCExpr *HiExpr = evaluateRelocExpr(ExprOffset, "hi"); + TempInst.addOperand(MCOperand::CreateExpr(HiExpr)); + } + } + // Add the instruction to the list. + Instructions.push_back(TempInst); + // Prepare TempInst for next instruction. + TempInst.clear(); + // Add temp register to base. + TempInst.setOpcode(Mips::ADDu); + TempInst.addOperand(MCOperand::CreateReg(TmpRegNum)); + TempInst.addOperand(MCOperand::CreateReg(TmpRegNum)); + TempInst.addOperand(MCOperand::CreateReg(BaseRegNum)); + Instructions.push_back(TempInst); + TempInst.clear(); + // And finaly, create original instruction with low part + // of offset and new base. + TempInst.setOpcode(Inst.getOpcode()); + TempInst.addOperand(MCOperand::CreateReg(RegOpNum)); + TempInst.addOperand(MCOperand::CreateReg(TmpRegNum)); + if (isImmOpnd) + TempInst.addOperand(MCOperand::CreateImm(LoOffset)); + else { + if (ExprOffset->getKind() == MCExpr::SymbolRef) { + const MCSymbolRefExpr *LoExpr = MCSymbolRefExpr::Create( + SR->getSymbol().getName(), MCSymbolRefExpr::VK_Mips_ABS_LO, + getContext()); + TempInst.addOperand(MCOperand::CreateExpr(LoExpr)); + } else { + const MCExpr *LoExpr = evaluateRelocExpr(ExprOffset, "lo"); + TempInst.addOperand(MCOperand::CreateExpr(LoExpr)); + } + } + Instructions.push_back(TempInst); + TempInst.clear(); +} + bool MipsAsmParser:: MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, SmallVectorImpl<MCParsedAsmOperand*> &Operands, MCStreamer &Out, unsigned &ErrorInfo, bool MatchingInlineAsm) { MCInst Inst; + SmallVector<MCInst, 8> Instructions; unsigned MatchResult = MatchInstructionImpl(Operands, Inst, ErrorInfo, MatchingInlineAsm); switch (MatchResult) { - default: break; + default: + break; case Match_Success: { - if (needsExpansion(Inst)) { - SmallVector<MCInst, 4> Instructions; - expandInstruction(Inst, IDLoc, Instructions); - for(unsigned i =0; i < Instructions.size(); i++){ - Out.EmitInstruction(Instructions[i]); - } - } else { - Inst.setLoc(IDLoc); - Out.EmitInstruction(Inst); - } + if (processInstruction(Inst, IDLoc, Instructions)) + return true; + for (unsigned i = 0; i < Instructions.size(); i++) + Out.EmitInstruction(Instructions[i]); return false; } case Match_MissingFeature: @@ -564,8 +726,9 @@ MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, if (ErrorInfo >= Operands.size()) return Error(IDLoc, "too few operands for instruction"); - ErrorLoc = ((MipsOperand*)Operands[ErrorInfo])->getStartLoc(); - if (ErrorLoc == SMLoc()) ErrorLoc = IDLoc; + ErrorLoc = ((MipsOperand*) Operands[ErrorInfo])->getStartLoc(); + if (ErrorLoc == SMLoc()) + ErrorLoc = IDLoc; } return Error(ErrorLoc, "invalid operand for instruction"); @@ -616,10 +779,10 @@ int MipsAsmParser::matchCPURegisterName(StringRef Name) { .Case("t9", 25) .Default(-1); - // Although SGI documentation just cut out t0-t3 for n32/n64, + // Although SGI documentation just cuts out t0-t3 for n32/n64, // GNU pushes the values of t0-t3 to override the o32/o64 values for t4-t7 // We are supporting both cases, so for t0-t3 we'll just push them to t4-t7. - if (isMips64() && 8 <= CC && CC <= 11) + if (isMips64() && 8 <= CC && CC <= 11) CC += 4; if (CC == -1 && isMips64()) @@ -635,19 +798,23 @@ int MipsAsmParser::matchCPURegisterName(StringRef Name) { return CC; } + int MipsAsmParser::matchRegisterName(StringRef Name, bool is64BitReg) { + if (Name.equals("fcc0")) + return Mips::FCC0; + int CC; CC = matchCPURegisterName(Name); if (CC != -1) - return matchRegisterByNumber(CC,is64BitReg?Mips::CPU64RegsRegClassID: - Mips::CPURegsRegClassID); + return matchRegisterByNumber(CC, is64BitReg ? Mips::CPU64RegsRegClassID + : Mips::CPURegsRegClassID); if (Name[0] == 'f') { StringRef NumString = Name.substr(1); unsigned IntVal; - if( NumString.getAsInteger(10, IntVal)) - return -1; // not integer + if (NumString.getAsInteger(10, IntVal)) + return -1; // This is not an integer. if (IntVal > 31) return -1; @@ -656,18 +823,19 @@ int MipsAsmParser::matchRegisterName(StringRef Name, bool is64BitReg) { if (Format == FP_FORMAT_S || Format == FP_FORMAT_W) return getReg(Mips::FGR32RegClassID, IntVal); if (Format == FP_FORMAT_D) { - if(isFP64()) { + if (isFP64()) { return getReg(Mips::FGR64RegClassID, IntVal); } - // only even numbers available as register pairs - if (( IntVal > 31) || (IntVal%2 != 0)) + // Only even numbers available as register pairs. + if ((IntVal > 31) || (IntVal % 2 != 0)) return -1; - return getReg(Mips::AFGR64RegClassID, IntVal/2); + return getReg(Mips::AFGR64RegClassID, IntVal / 2); } } return -1; } + void MipsAsmParser::setDefaultFpFormat() { if (isMips64() || isFP64()) @@ -687,6 +855,7 @@ bool MipsAsmParser::requestsDoubleOperand(StringRef Mnemonic){ return IsDouble; } + void MipsAsmParser::setFpFormat(StringRef Format) { FpFormat = StringSwitch<FpFormatTy>(Format.lower()) @@ -709,7 +878,7 @@ int MipsAsmParser::getATReg() { return Options.getATRegNum(); } -unsigned MipsAsmParser::getReg(int RC,int RegNo) { +unsigned MipsAsmParser::getReg(int RC, int RegNo) { return *(getContext().getRegisterInfo().getRegClass(RC).begin() + RegNo); } @@ -730,14 +899,12 @@ int MipsAsmParser::tryParseRegister(bool is64BitReg) { RegNum = matchRegisterName(lowerCase, is64BitReg); } else if (Tok.is(AsmToken::Integer)) RegNum = matchRegisterByNumber(static_cast<unsigned>(Tok.getIntVal()), - is64BitReg ? Mips::CPU64RegsRegClassID - : Mips::CPURegsRegClassID); + is64BitReg ? Mips::CPU64RegsRegClassID : Mips::CPURegsRegClassID); return RegNum; } -bool MipsAsmParser:: - tryParseRegisterOperand(SmallVectorImpl<MCParsedAsmOperand*> &Operands, - bool is64BitReg){ +bool MipsAsmParser::tryParseRegisterOperand( + SmallVectorImpl<MCParsedAsmOperand*> &Operands, bool is64BitReg) { SMLoc S = Parser.getTok().getLoc(); int RegNo = -1; @@ -747,7 +914,7 @@ bool MipsAsmParser:: return true; Operands.push_back(MipsOperand::CreateReg(RegNo, S, - Parser.getTok().getLoc())); + Parser.getTok().getLoc())); Parser.Lex(); // Eat register token. return false; } @@ -770,19 +937,19 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, Error(Parser.getTok().getLoc(), "unexpected token in operand"); return true; case AsmToken::Dollar: { - // parse register + // Parse the register. SMLoc S = Parser.getTok().getLoc(); Parser.Lex(); // Eat dollar token. - // parse register operand + // Parse the register operand. if (!tryParseRegisterOperand(Operands, isMips64())) { if (getLexer().is(AsmToken::LParen)) { - // check if it is indexed addressing operand + // Check if it is indexed addressing operand. Operands.push_back(MipsOperand::CreateToken("(", S)); - Parser.Lex(); // eat parenthesis + Parser.Lex(); // Eat the parenthesis. if (getLexer().isNot(AsmToken::Dollar)) return true; - Parser.Lex(); // eat dollar + Parser.Lex(); // Eat the dollar if (tryParseRegisterOperand(Operands, isMips64())) return true; @@ -795,7 +962,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, } return false; } - // maybe it is a symbol reference + // Maybe it is a symbol reference. StringRef Identifier; if (Parser.parseIdentifier(Identifier)) return true; @@ -804,7 +971,7 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, MCSymbol *Sym = getContext().GetOrCreateSymbol("$" + Identifier); - // Otherwise create a symbol ref. + // Otherwise create a symbol reference. const MCExpr *Res = MCSymbolRefExpr::Create(Sym, MCSymbolRefExpr::VK_None, getContext()); @@ -812,12 +979,17 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, return false; } case AsmToken::Identifier: + // Look for the existing symbol, we should check if + // we need to assigne the propper RegisterKind. + if (searchSymbolAlias(Operands, MipsOperand::Kind_None)) + return false; + // Else drop to expression parsing. case AsmToken::LParen: case AsmToken::Minus: case AsmToken::Plus: case AsmToken::Integer: case AsmToken::String: { - // quoted label names + // Quoted label names. const MCExpr *IdVal; SMLoc S = Parser.getTok().getLoc(); if (getParser().parseExpression(IdVal)) @@ -827,9 +999,9 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, return false; } case AsmToken::Percent: { - // it is a symbol reference or constant expression + // It is a symbol reference or constant expression. const MCExpr *IdVal; - SMLoc S = Parser.getTok().getLoc(); // start location of the operand + SMLoc S = Parser.getTok().getLoc(); // Start location of the operand. if (parseRelocOperand(IdVal)) return true; @@ -842,129 +1014,200 @@ bool MipsAsmParser::ParseOperand(SmallVectorImpl<MCParsedAsmOperand*>&Operands, return true; } -bool MipsAsmParser::parseRelocOperand(const MCExpr *&Res) { +const MCExpr* MipsAsmParser::evaluateRelocExpr(const MCExpr *Expr, + StringRef RelocStr) { + const MCExpr *Res; + // Check the type of the expression. + if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(Expr)) { + // It's a constant, evaluate lo or hi value. + if (RelocStr == "lo") { + short Val = MCE->getValue(); + Res = MCConstantExpr::Create(Val, getContext()); + } else if (RelocStr == "hi") { + int Val = MCE->getValue(); + int LoSign = Val & 0x8000; + Val = (Val & 0xffff0000) >> 16; + // Lower part is treated as a signed int, so if it is negative + // we must add 1 to the hi part to compensate. + if (LoSign) + Val++; + Res = MCConstantExpr::Create(Val, getContext()); + } else { + llvm_unreachable("Invalid RelocStr value"); + } + return Res; + } + + if (const MCSymbolRefExpr *MSRE = dyn_cast<MCSymbolRefExpr>(Expr)) { + // It's a symbol, create a symbolic expression from the symbol. + StringRef Symbol = MSRE->getSymbol().getName(); + MCSymbolRefExpr::VariantKind VK = getVariantKind(RelocStr); + Res = MCSymbolRefExpr::Create(Symbol, VK, getContext()); + return Res; + } + + if (const MCBinaryExpr *BE = dyn_cast<MCBinaryExpr>(Expr)) { + const MCExpr *LExp = evaluateRelocExpr(BE->getLHS(), RelocStr); + const MCExpr *RExp = evaluateRelocExpr(BE->getRHS(), RelocStr); + Res = MCBinaryExpr::Create(BE->getOpcode(), LExp, RExp, getContext()); + return Res; + } + + if (const MCUnaryExpr *UN = dyn_cast<MCUnaryExpr>(Expr)) { + const MCExpr *UnExp = evaluateRelocExpr(UN->getSubExpr(), RelocStr); + Res = MCUnaryExpr::Create(UN->getOpcode(), UnExp, getContext()); + return Res; + } + // Just return the original expression. + return Expr; +} + +bool MipsAsmParser::isEvaluated(const MCExpr *Expr) { - Parser.Lex(); // eat % token - const AsmToken &Tok = Parser.getTok(); // get next token, operation + switch (Expr->getKind()) { + case MCExpr::Constant: + return true; + case MCExpr::SymbolRef: + return (cast<MCSymbolRefExpr>(Expr)->getKind() != MCSymbolRefExpr::VK_None); + case MCExpr::Binary: + if (const MCBinaryExpr *BE = dyn_cast<MCBinaryExpr>(Expr)) { + if (!isEvaluated(BE->getLHS())) + return false; + return isEvaluated(BE->getRHS()); + } + case MCExpr::Unary: + return isEvaluated(cast<MCUnaryExpr>(Expr)->getSubExpr()); + default: + return false; + } + return false; +} + +bool MipsAsmParser::parseRelocOperand(const MCExpr *&Res) { + Parser.Lex(); // Eat the % token. + const AsmToken &Tok = Parser.getTok(); // Get next token, operation. if (Tok.isNot(AsmToken::Identifier)) return true; std::string Str = Tok.getIdentifier().str(); - Parser.Lex(); // eat identifier - // now make expression from the rest of the operand + Parser.Lex(); // Eat the identifier. + // Now make an expression from the rest of the operand. const MCExpr *IdVal; SMLoc EndLoc; if (getLexer().getKind() == AsmToken::LParen) { while (1) { - Parser.Lex(); // eat '(' token + Parser.Lex(); // Eat the '(' token. if (getLexer().getKind() == AsmToken::Percent) { - Parser.Lex(); // eat % token + Parser.Lex(); // Eat the % token. const AsmToken &nextTok = Parser.getTok(); if (nextTok.isNot(AsmToken::Identifier)) return true; Str += "(%"; Str += nextTok.getIdentifier(); - Parser.Lex(); // eat identifier + Parser.Lex(); // Eat the identifier. if (getLexer().getKind() != AsmToken::LParen) return true; } else break; } - if (getParser().parseParenExpression(IdVal,EndLoc)) + if (getParser().parseParenExpression(IdVal, EndLoc)) return true; while (getLexer().getKind() == AsmToken::RParen) - Parser.Lex(); // eat ')' token + Parser.Lex(); // Eat the ')' token. } else - return true; // parenthesis must follow reloc operand - - // Check the type of the expression - if (const MCConstantExpr *MCE = dyn_cast<MCConstantExpr>(IdVal)) { - // it's a constant, evaluate lo or hi value - int Val = MCE->getValue(); - if (Str == "lo") { - Val = Val & 0xffff; - } else if (Str == "hi") { - int LoSign = Val & 0x8000; - Val = (Val & 0xffff0000) >> 16; - //lower part is treated as signed int, so if it is negative - //we must add 1 to hi part to compensate - if (LoSign) - Val++; - } - Res = MCConstantExpr::Create(Val, getContext()); - return false; - } + return true; // Parenthesis must follow the relocation operand. - if (const MCSymbolRefExpr *MSRE = dyn_cast<MCSymbolRefExpr>(IdVal)) { - // it's a symbol, create symbolic expression from symbol - StringRef Symbol = MSRE->getSymbol().getName(); - MCSymbolRefExpr::VariantKind VK = getVariantKind(Str); - Res = MCSymbolRefExpr::Create(Symbol,VK,getContext()); - return false; - } - return true; + Res = evaluateRelocExpr(IdVal, Str); + return false; } bool MipsAsmParser::ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc) { - StartLoc = Parser.getTok().getLoc(); RegNo = tryParseRegister(isMips64()); EndLoc = Parser.getTok().getLoc(); - return (RegNo == (unsigned)-1); + return (RegNo == (unsigned) -1); } -bool MipsAsmParser::parseMemOffset(const MCExpr *&Res) { - +bool MipsAsmParser::parseMemOffset(const MCExpr *&Res, bool isParenExpr) { SMLoc S; + bool Result = true; + + while (getLexer().getKind() == AsmToken::LParen) + Parser.Lex(); - switch(getLexer().getKind()) { + switch (getLexer().getKind()) { default: return true; + case AsmToken::Identifier: + case AsmToken::LParen: case AsmToken::Integer: case AsmToken::Minus: case AsmToken::Plus: - return (getParser().parseExpression(Res)); + if (isParenExpr) + Result = getParser().parseParenExpression(Res, S); + else + Result = (getParser().parseExpression(Res)); + while (getLexer().getKind() == AsmToken::RParen) + Parser.Lex(); + break; case AsmToken::Percent: - return parseRelocOperand(Res); - case AsmToken::LParen: - return false; // it's probably assuming 0 + Result = parseRelocOperand(Res); } - return true; + return Result; } MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand( - SmallVectorImpl<MCParsedAsmOperand*>&Operands) { + SmallVectorImpl<MCParsedAsmOperand*>&Operands) { const MCExpr *IdVal = 0; SMLoc S; - // first operand is the offset + bool isParenExpr = false; + // First operand is the offset. S = Parser.getTok().getLoc(); - if (parseMemOffset(IdVal)) - return MatchOperand_ParseFail; + if (getLexer().getKind() == AsmToken::LParen) { + Parser.Lex(); + isParenExpr = true; + } - const AsmToken &Tok = Parser.getTok(); // get next token - if (Tok.isNot(AsmToken::LParen)) { - MipsOperand *Mnemonic = static_cast<MipsOperand*>(Operands[0]); - if (Mnemonic->getToken() == "la") { - SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() -1); - Operands.push_back(MipsOperand::CreateImm(IdVal, S, E)); - return MatchOperand_Success; + if (getLexer().getKind() != AsmToken::Dollar) { + if (parseMemOffset(IdVal, isParenExpr)) + return MatchOperand_ParseFail; + + const AsmToken &Tok = Parser.getTok(); // Get the next token. + if (Tok.isNot(AsmToken::LParen)) { + MipsOperand *Mnemonic = static_cast<MipsOperand*>(Operands[0]); + if (Mnemonic->getToken() == "la") { + SMLoc E = SMLoc::getFromPointer( + Parser.getTok().getLoc().getPointer() - 1); + Operands.push_back(MipsOperand::CreateImm(IdVal, S, E)); + return MatchOperand_Success; + } + if (Tok.is(AsmToken::EndOfStatement)) { + SMLoc E = SMLoc::getFromPointer( + Parser.getTok().getLoc().getPointer() - 1); + + // Zero register assumed, add a memory operand with ZERO as its base. + Operands.push_back(MipsOperand::CreateMem(isMips64() ? Mips::ZERO_64 + : Mips::ZERO, + IdVal, S, E)); + return MatchOperand_Success; + } + Error(Parser.getTok().getLoc(), "'(' expected"); + return MatchOperand_ParseFail; } - Error(Parser.getTok().getLoc(), "'(' expected"); - return MatchOperand_ParseFail; - } - Parser.Lex(); // Eat '(' token. + Parser.Lex(); // Eat the '(' token. + } - const AsmToken &Tok1 = Parser.getTok(); // get next token + const AsmToken &Tok1 = Parser.getTok(); // Get next token if (Tok1.is(AsmToken::Dollar)) { - Parser.Lex(); // Eat '$' token. + Parser.Lex(); // Eat the '$' token. if (tryParseRegisterOperand(Operands, isMips64())) { Error(Parser.getTok().getLoc(), "unexpected token in operand"); return MatchOperand_ParseFail; @@ -975,7 +1218,7 @@ MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand( return MatchOperand_ParseFail; } - const AsmToken &Tok2 = Parser.getTok(); // get next token + const AsmToken &Tok2 = Parser.getTok(); // Get next token. if (Tok2.isNot(AsmToken::RParen)) { Error(Parser.getTok().getLoc(), "')' expected"); return MatchOperand_ParseFail; @@ -983,17 +1226,26 @@ MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseMemOperand( SMLoc E = SMLoc::getFromPointer(Parser.getTok().getLoc().getPointer() - 1); - Parser.Lex(); // Eat ')' token. + Parser.Lex(); // Eat the ')' token. if (IdVal == 0) IdVal = MCConstantExpr::Create(0, getContext()); - // now replace register operand with the mem operand + // Replace the register operand with the memory operand. MipsOperand* op = static_cast<MipsOperand*>(Operands.back()); int RegNo = op->getReg(); - // remove register from operands + // Remove the register from the operands. Operands.pop_back(); - // and add memory operand + // Add the memory operand. + if (const MCBinaryExpr *BE = dyn_cast<MCBinaryExpr>(IdVal)) { + int64_t Imm; + if (IdVal->EvaluateAsAbsolute(Imm)) + IdVal = MCConstantExpr::Create(Imm, getContext()); + else if (BE->getLHS()->getKind() != MCExpr::SymbolRef) + IdVal = MCBinaryExpr::Create(BE->getOpcode(), BE->getRHS(), BE->getLHS(), + getContext()); + } + Operands.push_back(MipsOperand::CreateMem(RegNo, IdVal, S, E)); delete op; return MatchOperand_Success; @@ -1004,13 +1256,18 @@ MipsAsmParser::parseCPU64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { if (!isMips64()) return MatchOperand_NoMatch; - // if the first token is not '$' we have an error + if (getLexer().getKind() == AsmToken::Identifier) { + if (searchSymbolAlias(Operands, MipsOperand::Kind_CPU64Regs)) + return MatchOperand_Success; + return MatchOperand_NoMatch; + } + // If the first token is not '$', we have an error. if (Parser.getTok().isNot(AsmToken::Dollar)) return MatchOperand_NoMatch; Parser.Lex(); // Eat $ - if(!tryParseRegisterOperand(Operands, true)) { - // set the proper register kind + if (!tryParseRegisterOperand(Operands, true)) { + // Set the proper register kind. MipsOperand* op = static_cast<MipsOperand*>(Operands.back()); op->setRegKind(MipsOperand::Kind_CPU64Regs); return MatchOperand_Success; @@ -1018,16 +1275,59 @@ MipsAsmParser::parseCPU64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { return MatchOperand_NoMatch; } +bool MipsAsmParser::searchSymbolAlias( + SmallVectorImpl<MCParsedAsmOperand*> &Operands, unsigned RegisterKind) { + + MCSymbol *Sym = getContext().LookupSymbol(Parser.getTok().getIdentifier()); + if (Sym) { + SMLoc S = Parser.getTok().getLoc(); + const MCExpr *Expr; + if (Sym->isVariable()) + Expr = Sym->getVariableValue(); + else + return false; + if (Expr->getKind() == MCExpr::SymbolRef) { + const MCSymbolRefExpr *Ref = static_cast<const MCSymbolRefExpr*>(Expr); + const StringRef DefSymbol = Ref->getSymbol().getName(); + if (DefSymbol.startswith("$")) { + // Lookup for the register with the corresponding name. + int RegNum = matchRegisterName(DefSymbol.substr(1), isMips64()); + if (RegNum > -1) { + Parser.Lex(); + MipsOperand *op = MipsOperand::CreateReg(RegNum, S, + Parser.getTok().getLoc()); + op->setRegKind((MipsOperand::RegisterKind) RegisterKind); + Operands.push_back(op); + return true; + } + } + } else if (Expr->getKind() == MCExpr::Constant) { + Parser.Lex(); + const MCConstantExpr *Const = static_cast<const MCConstantExpr*>(Expr); + MipsOperand *op = MipsOperand::CreateImm(Const, S, + Parser.getTok().getLoc()); + Operands.push_back(op); + return true; + } + } + return false; +} + MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseCPURegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { - // if the first token is not '$' we have an error + if (getLexer().getKind() == AsmToken::Identifier) { + if (searchSymbolAlias(Operands, MipsOperand::Kind_CPURegs)) + return MatchOperand_Success; + return MatchOperand_NoMatch; + } + // If the first token is not '$' we have an error. if (Parser.getTok().isNot(AsmToken::Dollar)) return MatchOperand_NoMatch; Parser.Lex(); // Eat $ - if(!tryParseRegisterOperand(Operands, false)) { - // set the propper register kind + if (!tryParseRegisterOperand(Operands, false)) { + // Set the proper register kind. MipsOperand* op = static_cast<MipsOperand*>(Operands.back()); op->setRegKind(MipsOperand::Kind_CPURegs); return MatchOperand_Success; @@ -1041,87 +1341,88 @@ MipsAsmParser::parseHWRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { if (isMips64()) return MatchOperand_NoMatch; - // if the first token is not '$' we have error + // If the first token is not '$' we have error. if (Parser.getTok().isNot(AsmToken::Dollar)) return MatchOperand_NoMatch; SMLoc S = Parser.getTok().getLoc(); - Parser.Lex(); // Eat $ + Parser.Lex(); // Eat the '$'. - const AsmToken &Tok = Parser.getTok(); // get next token + const AsmToken &Tok = Parser.getTok(); // Get the next token. if (Tok.isNot(AsmToken::Integer)) return MatchOperand_NoMatch; unsigned RegNum = Tok.getIntVal(); - // at the moment only hwreg29 is supported + // At the moment only hwreg29 is supported. if (RegNum != 29) return MatchOperand_ParseFail; MipsOperand *op = MipsOperand::CreateReg(Mips::HWR29, S, - Parser.getTok().getLoc()); + Parser.getTok().getLoc()); op->setRegKind(MipsOperand::Kind_HWRegs); Operands.push_back(op); - Parser.Lex(); // Eat reg number + Parser.Lex(); // Eat the register number. return MatchOperand_Success; } MipsAsmParser::OperandMatchResultTy -MipsAsmParser::parseHW64Regs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { +MipsAsmParser::parseHW64Regs( + SmallVectorImpl<MCParsedAsmOperand*> &Operands) { if (!isMips64()) return MatchOperand_NoMatch; - //if the first token is not '$' we have error + // If the first token is not '$' we have an error. if (Parser.getTok().isNot(AsmToken::Dollar)) return MatchOperand_NoMatch; SMLoc S = Parser.getTok().getLoc(); Parser.Lex(); // Eat $ - const AsmToken &Tok = Parser.getTok(); // get next token + const AsmToken &Tok = Parser.getTok(); // Get the next token. if (Tok.isNot(AsmToken::Integer)) return MatchOperand_NoMatch; unsigned RegNum = Tok.getIntVal(); - // at the moment only hwreg29 is supported + // At the moment only hwreg29 is supported. if (RegNum != 29) return MatchOperand_ParseFail; MipsOperand *op = MipsOperand::CreateReg(Mips::HWR29_64, S, - Parser.getTok().getLoc()); + Parser.getTok().getLoc()); op->setRegKind(MipsOperand::Kind_HW64Regs); Operands.push_back(op); - Parser.Lex(); // Eat reg number + Parser.Lex(); // Eat the register number. return MatchOperand_Success; } MipsAsmParser::OperandMatchResultTy MipsAsmParser::parseCCRRegs(SmallVectorImpl<MCParsedAsmOperand*> &Operands) { unsigned RegNum; - //if the first token is not '$' we have error + // If the first token is not '$' we have an error. if (Parser.getTok().isNot(AsmToken::Dollar)) return MatchOperand_NoMatch; SMLoc S = Parser.getTok().getLoc(); - Parser.Lex(); // Eat $ + Parser.Lex(); // Eat the '$' - const AsmToken &Tok = Parser.getTok(); // get next token + const AsmToken &Tok = Parser.getTok(); // Get next token. if (Tok.is(AsmToken::Integer)) { RegNum = Tok.getIntVal(); - // at the moment only fcc0 is supported + // At the moment only fcc0 is supported. if (RegNum != 0) return MatchOperand_ParseFail; } else if (Tok.is(AsmToken::Identifier)) { - // at the moment only fcc0 is supported + // At the moment only fcc0 is supported. if (Tok.getIdentifier() != "fcc0") return MatchOperand_ParseFail; } else return MatchOperand_NoMatch; MipsOperand *op = MipsOperand::CreateReg(Mips::FCC0, S, - Parser.getTok().getLoc()); + Parser.getTok().getLoc()); op->setRegKind(MipsOperand::Kind_CCRRegs); Operands.push_back(op); - Parser.Lex(); // Eat reg number + Parser.Lex(); // Eat the register number. return MatchOperand_Success; } @@ -1153,23 +1454,23 @@ MCSymbolRefExpr::VariantKind MipsAsmParser::getVariantKind(StringRef Symbol) { static int ConvertCcString(StringRef CondString) { int CC = StringSwitch<unsigned>(CondString) - .Case(".f", 0) - .Case(".un", 1) - .Case(".eq", 2) - .Case(".ueq", 3) - .Case(".olt", 4) - .Case(".ult", 5) - .Case(".ole", 6) - .Case(".ule", 7) - .Case(".sf", 8) - .Case(".ngle", 9) - .Case(".seq", 10) - .Case(".ngl", 11) - .Case(".lt", 12) - .Case(".nge", 13) - .Case(".le", 14) - .Case(".ngt", 15) - .Default(-1); + .Case(".f", 0) + .Case(".un", 1) + .Case(".eq", 2) + .Case(".ueq", 3) + .Case(".olt", 4) + .Case(".ult", 5) + .Case(".ole", 6) + .Case(".ule", 7) + .Case(".sf", 8) + .Case(".ngle", 9) + .Case(".seq", 10) + .Case(".ngl", 11) + .Case(".lt", 12) + .Case(".nge", 13) + .Case(".le", 14) + .Case(".ngt", 15) + .Default(-1); return CC; } @@ -1177,16 +1478,16 @@ static int ConvertCcString(StringRef CondString) { bool MipsAsmParser:: parseMathOperation(StringRef Name, SMLoc NameLoc, SmallVectorImpl<MCParsedAsmOperand*> &Operands) { - // split the format + // Split the format. size_t Start = Name.find('.'), Next = Name.rfind('.'); StringRef Format1 = Name.slice(Start, Next); - // and add the first format to the operands + // Add the first format to the operands. Operands.push_back(MipsOperand::CreateToken(Format1, NameLoc)); - // now for the second format + // Now for the second format. StringRef Format2 = Name.slice(Next, StringRef::npos); Operands.push_back(MipsOperand::CreateToken(Format2, NameLoc)); - // set the format for the first register + // Set the format for the first register. setFpFormat(Format1); // Read the remaining operands. @@ -1202,11 +1503,10 @@ parseMathOperation(StringRef Name, SMLoc NameLoc, SMLoc Loc = getLexer().getLoc(); Parser.eatToEndOfStatement(); return Error(Loc, "unexpected token in argument list"); - } - Parser.Lex(); // Eat the comma. + Parser.Lex(); // Eat the comma. - //set the format for the first register + // Set the format for the first register setFpFormat(Format2); // Parse and remember the operand. @@ -1223,7 +1523,7 @@ parseMathOperation(StringRef Name, SMLoc NameLoc, return Error(Loc, "unexpected token in argument list"); } - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } @@ -1231,13 +1531,12 @@ bool MipsAsmParser:: ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc, SmallVectorImpl<MCParsedAsmOperand*> &Operands) { StringRef Mnemonic; - // floating point instructions: should register be treated as double? + // Floating point instructions: Should the register be treated as a double? if (requestsDoubleOperand(Name)) { setFpFormat(FP_FORMAT_D); - Operands.push_back(MipsOperand::CreateToken(Name, NameLoc)); - Mnemonic = Name; - } - else { + Operands.push_back(MipsOperand::CreateToken(Name, NameLoc)); + Mnemonic = Name; + } else { setDefaultFpFormat(); // Create the leading tokens for the mnemonic, split by '.' characters. size_t Start = 0, Next = Name.find('.'); @@ -1246,30 +1545,30 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc, Operands.push_back(MipsOperand::CreateToken(Mnemonic, NameLoc)); if (Next != StringRef::npos) { - // there is a format token in mnemonic - // StringRef Rest = Name.slice(Next, StringRef::npos); - size_t Dot = Name.find('.', Next+1); + // There is a format token in mnemonic. + size_t Dot = Name.find('.', Next + 1); StringRef Format = Name.slice(Next, Dot); - if (Dot == StringRef::npos) //only one '.' in a string, it's a format + if (Dot == StringRef::npos) // Only one '.' in a string, it's a format. Operands.push_back(MipsOperand::CreateToken(Format, NameLoc)); else { - if (Name.startswith("c.")){ - // floating point compare, add '.' and immediate represent for cc + if (Name.startswith("c.")) { + // Floating point compare, add '.' and immediate represent for cc. Operands.push_back(MipsOperand::CreateToken(".", NameLoc)); int Cc = ConvertCcString(Format); if (Cc == -1) { return Error(NameLoc, "Invalid conditional code"); } SMLoc E = SMLoc::getFromPointer( - Parser.getTok().getLoc().getPointer() -1 ); - Operands.push_back(MipsOperand::CreateImm( - MCConstantExpr::Create(Cc, getContext()), NameLoc, E)); + Parser.getTok().getLoc().getPointer() - 1); + Operands.push_back( + MipsOperand::CreateImm(MCConstantExpr::Create(Cc, getContext()), + NameLoc, E)); } else { // trunc, ceil, floor ... return parseMathOperation(Name, NameLoc, Operands); } - // the rest is a format + // The rest is a format. Format = Name.slice(Dot, StringRef::npos); Operands.push_back(MipsOperand::CreateToken(Format, NameLoc)); } @@ -1287,8 +1586,8 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc, return Error(Loc, "unexpected token in argument list"); } - while (getLexer().is(AsmToken::Comma) ) { - Parser.Lex(); // Eat the comma. + while (getLexer().is(AsmToken::Comma)) { + Parser.Lex(); // Eat the comma. // Parse and remember the operand. if (ParseOperand(Operands, Name)) { @@ -1305,48 +1604,47 @@ ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc, return Error(Loc, "unexpected token in argument list"); } - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } bool MipsAsmParser::reportParseError(StringRef ErrorMsg) { - SMLoc Loc = getLexer().getLoc(); - Parser.eatToEndOfStatement(); - return Error(Loc, ErrorMsg); + SMLoc Loc = getLexer().getLoc(); + Parser.eatToEndOfStatement(); + return Error(Loc, ErrorMsg); } bool MipsAsmParser::parseSetNoAtDirective() { - // line should look like: - // .set noat - // set at reg to 0 + // Line should look like: ".set noat". + // set at reg to 0. Options.setATReg(0); // eat noat Parser.Lex(); - // if this is not the end of the statement, report error + // If this is not the end of the statement, report an error. if (getLexer().isNot(AsmToken::EndOfStatement)) { reportParseError("unexpected token in statement"); return false; } - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } + bool MipsAsmParser::parseSetAtDirective() { - // line can be - // .set at - defaults to $1 + // Line can be .set at - defaults to $1 // or .set at=$reg int AtRegNo; getParser().Lex(); if (getLexer().is(AsmToken::EndOfStatement)) { Options.setATReg(1); - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } else if (getLexer().is(AsmToken::Equal)) { - getParser().Lex(); //eat '=' + getParser().Lex(); // Eat the '='. if (getLexer().isNot(AsmToken::Dollar)) { reportParseError("unexpected token in statement"); return false; } - Parser.Lex(); // eat '$' + Parser.Lex(); // Eat the '$'. const AsmToken &Reg = Parser.getTok(); if (Reg.is(AsmToken::Identifier)) { AtRegNo = matchCPURegisterName(Reg.getIdentifier()); @@ -1357,7 +1655,7 @@ bool MipsAsmParser::parseSetAtDirective() { return false; } - if ( AtRegNo < 1 || AtRegNo > 31) { + if (AtRegNo < 1 || AtRegNo > 31) { reportParseError("unexpected token in statement"); return false; } @@ -1366,13 +1664,13 @@ bool MipsAsmParser::parseSetAtDirective() { reportParseError("unexpected token in statement"); return false; } - getParser().Lex(); //eat reg + getParser().Lex(); // Eat the register. if (getLexer().isNot(AsmToken::EndOfStatement)) { reportParseError("unexpected token in statement"); return false; - } - Parser.Lex(); // Consume the EndOfStatement + } + Parser.Lex(); // Consume the EndOfStatement. return false; } else { reportParseError("unexpected token in statement"); @@ -1382,43 +1680,43 @@ bool MipsAsmParser::parseSetAtDirective() { bool MipsAsmParser::parseSetReorderDirective() { Parser.Lex(); - // if this is not the end of the statement, report error + // If this is not the end of the statement, report an error. if (getLexer().isNot(AsmToken::EndOfStatement)) { reportParseError("unexpected token in statement"); return false; } Options.setReorder(); - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } bool MipsAsmParser::parseSetNoReorderDirective() { - Parser.Lex(); - // if this is not the end of the statement, report error - if (getLexer().isNot(AsmToken::EndOfStatement)) { - reportParseError("unexpected token in statement"); - return false; - } - Options.setNoreorder(); - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); + // If this is not the end of the statement, report an error. + if (getLexer().isNot(AsmToken::EndOfStatement)) { + reportParseError("unexpected token in statement"); return false; + } + Options.setNoreorder(); + Parser.Lex(); // Consume the EndOfStatement. + return false; } bool MipsAsmParser::parseSetMacroDirective() { Parser.Lex(); - // if this is not the end of the statement, report error + // If this is not the end of the statement, report an error. if (getLexer().isNot(AsmToken::EndOfStatement)) { reportParseError("unexpected token in statement"); return false; } Options.setMacro(); - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } bool MipsAsmParser::parseSetNoMacroDirective() { Parser.Lex(); - // if this is not the end of the statement, report error + // If this is not the end of the statement, report an error. if (getLexer().isNot(AsmToken::EndOfStatement)) { reportParseError("`noreorder' must be set before `nomacro'"); return false; @@ -1428,12 +1726,37 @@ bool MipsAsmParser::parseSetNoMacroDirective() { return false; } Options.setNomacro(); - Parser.Lex(); // Consume the EndOfStatement + Parser.Lex(); // Consume the EndOfStatement. return false; } + +bool MipsAsmParser::parseSetAssignment() { + StringRef Name; + const MCExpr *Value; + + if (Parser.parseIdentifier(Name)) + reportParseError("expected identifier after .set"); + + if (getLexer().isNot(AsmToken::Comma)) + return reportParseError("unexpected token in .set directive"); + Lex(); // Eat comma + + if (Parser.parseExpression(Value)) + reportParseError("expected valid expression after comma"); + + // Check if the Name already exists as a symbol. + MCSymbol *Sym = getContext().LookupSymbol(Name); + if (Sym) + return reportParseError("symbol already defined"); + Sym = getContext().GetOrCreateSymbol(Name); + Sym->setVariableValue(Value); + + return false; +} + bool MipsAsmParser::parseDirectiveSet() { - // get next token + // Get the next token. const AsmToken &Tok = Parser.getTok(); if (Tok.getString() == "noat") { @@ -1449,13 +1772,17 @@ bool MipsAsmParser::parseDirectiveSet() { } else if (Tok.getString() == "nomacro") { return parseSetNoMacroDirective(); } else if (Tok.getString() == "nomips16") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; } else if (Tok.getString() == "nomicromips") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; + } else { + // It is just an identifier, look for an assignment. + parseSetAssignment(); + return false; } return true; @@ -1490,20 +1817,20 @@ bool MipsAsmParser::ParseDirective(AsmToken DirectiveID) { StringRef IDVal = DirectiveID.getString(); - if ( IDVal == ".ent") { - // ignore this directive for now + if (IDVal == ".ent") { + // Ignore this directive for now. Parser.Lex(); return false; } if (IDVal == ".end") { - // ignore this directive for now + // Ignore this directive for now. Parser.Lex(); return false; } if (IDVal == ".frame") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; } @@ -1513,19 +1840,19 @@ bool MipsAsmParser::ParseDirective(AsmToken DirectiveID) { } if (IDVal == ".fmask") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; } if (IDVal == ".mask") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; } if (IDVal == ".gpword") { - // ignore this directive for now + // Ignore this directive for now. Parser.eatToEndOfStatement(); return false; } diff --git a/lib/Target/Mips/CMakeLists.txt b/lib/Target/Mips/CMakeLists.txt index 5f68791890..5b22c53453 100644 --- a/lib/Target/Mips/CMakeLists.txt +++ b/lib/Target/Mips/CMakeLists.txt @@ -16,6 +16,8 @@ add_public_tablegen_target(MipsCommonTableGen) add_llvm_target(MipsCodeGen Mips16FrameLowering.cpp Mips16InstrInfo.cpp + Mips16ISelDAGToDAG.cpp + Mips16ISelLowering.cpp Mips16RegisterInfo.cpp MipsAnalyzeImmediate.cpp MipsAsmPrinter.cpp @@ -30,6 +32,8 @@ add_llvm_target(MipsCodeGen MipsLongBranch.cpp MipsMCInstLower.cpp MipsMachineFunction.cpp + MipsModuleISelDAGToDAG.cpp + MipsOs16.cpp # LOCALMOD-BEGIN MipsNaClHeaders.cpp MipsNaClRewritePass.cpp @@ -37,6 +41,8 @@ add_llvm_target(MipsCodeGen MipsRegisterInfo.cpp MipsSEFrameLowering.cpp MipsSEInstrInfo.cpp + MipsSEISelDAGToDAG.cpp + MipsSEISelLowering.cpp MipsSERegisterInfo.cpp MipsSubtarget.cpp MipsTargetMachine.cpp diff --git a/lib/Target/Mips/Disassembler/MipsDisassembler.cpp b/lib/Target/Mips/Disassembler/MipsDisassembler.cpp index 025a783f93..0dba33a276 100644 --- a/lib/Target/Mips/Disassembler/MipsDisassembler.cpp +++ b/lib/Target/Mips/Disassembler/MipsDisassembler.cpp @@ -138,10 +138,20 @@ static DecodeStatus DecodeHWRegs64RegisterClass(MCInst &Inst, uint64_t Address, const void *Decoder); -static DecodeStatus DecodeACRegsRegisterClass(MCInst &Inst, - unsigned RegNo, - uint64_t Address, - const void *Decoder); +static DecodeStatus DecodeACRegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder); + +static DecodeStatus DecodeHIRegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder); + +static DecodeStatus DecodeLORegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder); static DecodeStatus DecodeBranchTarget(MCInst &Inst, unsigned Offset, @@ -484,14 +494,38 @@ static DecodeStatus DecodeHWRegs64RegisterClass(MCInst &Inst, return MCDisassembler::Success; } -static DecodeStatus DecodeACRegsRegisterClass(MCInst &Inst, - unsigned RegNo, - uint64_t Address, - const void *Decoder) { +static DecodeStatus DecodeACRegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder) { + if (RegNo >= 4) + return MCDisassembler::Fail; + + unsigned Reg = getReg(Decoder, Mips::ACRegsDSPRegClassID, RegNo); + Inst.addOperand(MCOperand::CreateReg(Reg)); + return MCDisassembler::Success; +} + +static DecodeStatus DecodeHIRegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder) { + if (RegNo >= 4) + return MCDisassembler::Fail; + + unsigned Reg = getReg(Decoder, Mips::HIRegsDSPRegClassID, RegNo); + Inst.addOperand(MCOperand::CreateReg(Reg)); + return MCDisassembler::Success; +} + +static DecodeStatus DecodeLORegsDSPRegisterClass(MCInst &Inst, + unsigned RegNo, + uint64_t Address, + const void *Decoder) { if (RegNo >= 4) return MCDisassembler::Fail; - unsigned Reg = getReg(Decoder, Mips::ACRegsRegClassID, RegNo); + unsigned Reg = getReg(Decoder, Mips::LORegsDSPRegClassID, RegNo); Inst.addOperand(MCOperand::CreateReg(Reg)); return MCDisassembler::Success; } diff --git a/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp b/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp index 96f93a0789..9460731c19 100644 --- a/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp +++ b/lib/Target/Mips/MCTargetDesc/MipsMCCodeEmitter.cpp @@ -27,6 +27,9 @@ #include "llvm/MC/MCSubtargetInfo.h" #include "llvm/Support/raw_ostream.h" +#define GET_INSTRMAP_INFO +#include "MipsGenInstrInfo.inc" + using namespace llvm; namespace { @@ -35,12 +38,13 @@ class MipsMCCodeEmitter : public MCCodeEmitter { void operator=(const MipsMCCodeEmitter &) LLVM_DELETED_FUNCTION; const MCInstrInfo &MCII; MCContext &Ctx; + const MCSubtargetInfo &STI; bool IsLittleEndian; public: MipsMCCodeEmitter(const MCInstrInfo &mcii, MCContext &Ctx_, const MCSubtargetInfo &sti, bool IsLittle) : - MCII(mcii), Ctx(Ctx_), IsLittleEndian(IsLittle) {} + MCII(mcii), Ctx(Ctx_), STI (sti), IsLittleEndian(IsLittle) {} ~MipsMCCodeEmitter() {} @@ -88,6 +92,9 @@ public: unsigned getSizeInsEncoding(const MCInst &MI, unsigned OpNo, SmallVectorImpl<MCFixup> &Fixups) const; + unsigned + getExprOpValue(const MCExpr *Expr,SmallVectorImpl<MCFixup> &Fixups) const; + }; // class MipsMCCodeEmitter } // namespace @@ -141,6 +148,15 @@ EncodeInstruction(const MCInst &MI, raw_ostream &OS, if ((Opcode != Mips::NOP) && (Opcode != Mips::SLL) && !Binary) llvm_unreachable("unimplemented opcode in EncodeInstruction()"); + if (STI.getFeatureBits() & Mips::FeatureMicroMips) { + int NewOpcode = Mips::Std2MicroMips (Opcode, Mips::Arch_micromips); + if (NewOpcode != -1) { + Opcode = NewOpcode; + TmpInst.setOpcode (NewOpcode); + Binary = getBinaryCodeForInstr(TmpInst, Fixups); + } + } + const MCInstrDesc &Desc = MCII.get(TmpInst.getOpcode()); // Get byte count of instruction @@ -160,8 +176,9 @@ getBranchTargetOpValue(const MCInst &MI, unsigned OpNo, const MCOperand &MO = MI.getOperand(OpNo); - // If the destination is an immediate, we have nothing to do. - if (MO.isImm()) return MO.getImm(); + // If the destination is an immediate, divide by 4. + if (MO.isImm()) return MO.getImm() >> 2; + assert(MO.isExpr() && "getBranchTargetOpValue expects only expressions or immediates"); @@ -179,8 +196,9 @@ getJumpTargetOpValue(const MCInst &MI, unsigned OpNo, SmallVectorImpl<MCFixup> &Fixups) const { const MCOperand &MO = MI.getOperand(OpNo); - // If the destination is an immediate, we have nothing to do. - if (MO.isImm()) return MO.getImm(); + // If the destination is an immediate, divide by 4. + if (MO.isImm()) return MO.getImm()>>2; + assert(MO.isExpr() && "getJumpTargetOpValue expects only expressions or an immediate"); @@ -190,35 +208,24 @@ getJumpTargetOpValue(const MCInst &MI, unsigned OpNo, return 0; } -/// getMachineOpValue - Return binary encoding of operand. If the machine -/// operand requires relocation, record the relocation and return zero. unsigned MipsMCCodeEmitter:: -getMachineOpValue(const MCInst &MI, const MCOperand &MO, - SmallVectorImpl<MCFixup> &Fixups) const { - if (MO.isReg()) { - unsigned Reg = MO.getReg(); - unsigned RegNo = Ctx.getRegisterInfo().getEncodingValue(Reg); - return RegNo; - } else if (MO.isImm()) { - return static_cast<unsigned>(MO.getImm()); - } else if (MO.isFPImm()) { - return static_cast<unsigned>(APFloat(MO.getFPImm()) - .bitcastToAPInt().getHiBits(32).getLimitedValue()); - } +getExprOpValue(const MCExpr *Expr,SmallVectorImpl<MCFixup> &Fixups) const { + int64_t Res; - // MO must be an Expr. - assert(MO.isExpr()); + if (Expr->EvaluateAsAbsolute(Res)) + return Res; - const MCExpr *Expr = MO.getExpr(); MCExpr::ExprKind Kind = Expr->getKind(); + if (Kind == MCExpr::Constant) { + return cast<MCConstantExpr>(Expr)->getValue(); + } if (Kind == MCExpr::Binary) { - Expr = static_cast<const MCBinaryExpr*>(Expr)->getLHS(); - Kind = Expr->getKind(); + unsigned Res = getExprOpValue(cast<MCBinaryExpr>(Expr)->getLHS(), Fixups); + Res += getExprOpValue(cast<MCBinaryExpr>(Expr)->getRHS(), Fixups); + return Res; } - - assert (Kind == MCExpr::SymbolRef); - + if (Kind == MCExpr::SymbolRef) { Mips::Fixups FixupKind = Mips::Fixups(0); switch(cast<MCSymbolRefExpr>(Expr)->getKind()) { @@ -298,12 +305,32 @@ getMachineOpValue(const MCInst &MI, const MCOperand &MO, break; } // switch - Fixups.push_back(MCFixup::Create(0, MO.getExpr(), MCFixupKind(FixupKind))); - - // All of the information is in the fixup. + Fixups.push_back(MCFixup::Create(0, Expr, MCFixupKind(FixupKind))); + return 0; + } return 0; } +/// getMachineOpValue - Return binary encoding of operand. If the machine +/// operand requires relocation, record the relocation and return zero. +unsigned MipsMCCodeEmitter:: +getMachineOpValue(const MCInst &MI, const MCOperand &MO, + SmallVectorImpl<MCFixup> &Fixups) const { + if (MO.isReg()) { + unsigned Reg = MO.getReg(); + unsigned RegNo = Ctx.getRegisterInfo().getEncodingValue(Reg); + return RegNo; + } else if (MO.isImm()) { + return static_cast<unsigned>(MO.getImm()); + } else if (MO.isFPImm()) { + return static_cast<unsigned>(APFloat(MO.getFPImm()) + .bitcastToAPInt().getHiBits(32).getLimitedValue()); + } + // MO must be an Expr. + assert(MO.isExpr()); + return getExprOpValue(MO.getExpr(),Fixups); +} + /// getMemEncoding - Return binary encoding of memory related operand. /// If the offset operand requires relocation, record the relocation. unsigned diff --git a/lib/Target/Mips/MicroMipsInstrFormats.td b/lib/Target/Mips/MicroMipsInstrFormats.td new file mode 100644 index 0000000000..665b4d2d8b --- /dev/null +++ b/lib/Target/Mips/MicroMipsInstrFormats.td @@ -0,0 +1,112 @@ +class MMArch { + string Arch = "micromips"; + list<dag> Pattern = []; +} + +class ADD_FM_MM<bits<6> op, bits<10> funct> : MMArch { + bits<5> rt; + bits<5> rs; + bits<5> rd; + + bits<32> Inst; + + let Inst{31-26} = op; + let Inst{25-21} = rt; + let Inst{20-16} = rs; + let Inst{15-11} = rd; + let Inst{10} = 0; + let Inst{9-0} = funct; +} + +class ADDI_FM_MM<bits<6> op> : MMArch { + bits<5> rs; + bits<5> rt; + bits<16> imm16; + + bits<32> Inst; + + let Inst{31-26} = op; + let Inst{25-21} = rt; + let Inst{20-16} = rs; + let Inst{15-0} = imm16; +} + +class SLTI_FM_MM<bits<6> op> : MMArch { + bits<5> rt; + bits<5> rs; + bits<16> imm16; + + bits<32> Inst; + + let Inst{31-26} = op; + let Inst{25-21} = rs; + let Inst{20-16} = rt; + let Inst{15-0} = imm16; +} + +class LUI_FM_MM : MMArch { + bits<5> rt; + bits<16> imm16; + + bits<32> Inst; + + let Inst{31-26} = 0x10; + let Inst{25-21} = 0xd; + let Inst{20-16} = rt; + let Inst{15-0} = imm16; +} + +class MULT_FM_MM<bits<10> funct> : MMArch { + bits<5> rs; + bits<5> rt; + + bits<32> Inst; + + let Inst{31-26} = 0x00; + let Inst{25-21} = rt; + let Inst{20-16} = rs; + let Inst{15-6} = funct; + let Inst{5-0} = 0x3c; +} + +class SRA_FM_MM<bits<10> funct, bit rotate> : MMArch { + bits<5> rd; + bits<5> rt; + bits<5> shamt; + + bits<32> Inst; + + let Inst{31-26} = 0; + let Inst{25-21} = rd; + let Inst{20-16} = rt; + let Inst{15-11} = shamt; + let Inst{10} = rotate; + let Inst{9-0} = funct; +} + +class SRLV_FM_MM<bits<10> funct, bit rotate> : MMArch { + bits<5> rd; + bits<5> rt; + bits<5> rs; + + bits<32> Inst; + + let Inst{31-26} = 0; + let Inst{25-21} = rt; + let Inst{20-16} = rs; + let Inst{15-11} = rd; + let Inst{10} = rotate; + let Inst{9-0} = funct; +} + +class LW_FM_MM<bits<6> op> : MMArch { + bits<5> rt; + bits<21> addr; + + bits<32> Inst; + + let Inst{31-26} = op; + let Inst{25-21} = rt; + let Inst{20-16} = addr{20-16}; + let Inst{15-0} = addr{15-0}; +} diff --git a/lib/Target/Mips/MicroMipsInstrInfo.td b/lib/Target/Mips/MicroMipsInstrInfo.td new file mode 100644 index 0000000000..74cdccd3ed --- /dev/null +++ b/lib/Target/Mips/MicroMipsInstrInfo.td @@ -0,0 +1,67 @@ +let isCodeGenOnly = 1 in { + /// Arithmetic Instructions (ALU Immediate) + def ADDiu_MM : MMRel, ArithLogicI<"addiu", simm16, CPURegsOpnd>, + ADDI_FM_MM<0xc>; + def ADDi_MM : MMRel, ArithLogicI<"addi", simm16, CPURegsOpnd>, + ADDI_FM_MM<0x4>; + def SLTi_MM : MMRel, SetCC_I<"slti", setlt, simm16, immSExt16, CPURegs>, + SLTI_FM_MM<0x24>; + def SLTiu_MM : MMRel, SetCC_I<"sltiu", setult, simm16, immSExt16, CPURegs>, + SLTI_FM_MM<0x2c>; + def ANDi_MM : MMRel, ArithLogicI<"andi", uimm16, CPURegsOpnd, immZExt16, and>, + ADDI_FM_MM<0x34>; + def ORi_MM : MMRel, ArithLogicI<"ori", uimm16, CPURegsOpnd, immZExt16, or>, + ADDI_FM_MM<0x14>; + def XORi_MM : MMRel, ArithLogicI<"xori", uimm16, CPURegsOpnd, immZExt16, xor>, + ADDI_FM_MM<0x1c>; + def LUi_MM : MMRel, LoadUpper<"lui", CPURegs, uimm16>, LUI_FM_MM; + + /// Arithmetic Instructions (3-Operand, R-Type) + def ADDu_MM : MMRel, ArithLogicR<"addu", CPURegsOpnd>, ADD_FM_MM<0, 0x150>; + def SUBu_MM : MMRel, ArithLogicR<"subu", CPURegsOpnd>, ADD_FM_MM<0, 0x1d0>; + def MUL_MM : MMRel, ArithLogicR<"mul", CPURegsOpnd>, ADD_FM_MM<0, 0x210>; + def ADD_MM : MMRel, ArithLogicR<"add", CPURegsOpnd>, ADD_FM_MM<0, 0x110>; + def SUB_MM : MMRel, ArithLogicR<"sub", CPURegsOpnd>, ADD_FM_MM<0, 0x190>; + def SLT_MM : MMRel, SetCC_R<"slt", setlt, CPURegs>, ADD_FM_MM<0, 0x350>; + def SLTu_MM : MMRel, SetCC_R<"sltu", setult, CPURegs>, + ADD_FM_MM<0, 0x390>; + def AND_MM : MMRel, ArithLogicR<"and", CPURegsOpnd, 1, IIAlu, and>, + ADD_FM_MM<0, 0x250>; + def OR_MM : MMRel, ArithLogicR<"or", CPURegsOpnd, 1, IIAlu, or>, + ADD_FM_MM<0, 0x290>; + def XOR_MM : MMRel, ArithLogicR<"xor", CPURegsOpnd, 1, IIAlu, xor>, + ADD_FM_MM<0, 0x310>; + def NOR_MM : MMRel, LogicNOR<"nor", CPURegsOpnd>, ADD_FM_MM<0, 0x2d0>; + def MULT_MM : MMRel, Mult<"mult", IIImul, CPURegsOpnd, [HI, LO]>, + MULT_FM_MM<0x22c>; + def MULTu_MM : MMRel, Mult<"multu", IIImul, CPURegsOpnd, [HI, LO]>, + MULT_FM_MM<0x26c>; + + /// Shift Instructions + def SLL_MM : MMRel, shift_rotate_imm<"sll", shamt, CPURegsOpnd>, + SRA_FM_MM<0, 0>; + def SRL_MM : MMRel, shift_rotate_imm<"srl", shamt, CPURegsOpnd>, + SRA_FM_MM<0x40, 0>; + def SRA_MM : MMRel, shift_rotate_imm<"sra", shamt, CPURegsOpnd>, + SRA_FM_MM<0x80, 0>; + def SLLV_MM : MMRel, shift_rotate_reg<"sllv", CPURegsOpnd>, + SRLV_FM_MM<0x10, 0>; + def SRLV_MM : MMRel, shift_rotate_reg<"srlv", CPURegsOpnd>, + SRLV_FM_MM<0x50, 0>; + def SRAV_MM : MMRel, shift_rotate_reg<"srav", CPURegsOpnd>, + SRLV_FM_MM<0x90, 0>; + def ROTR_MM : MMRel, shift_rotate_imm<"rotr", shamt, CPURegsOpnd>, + SRA_FM_MM<0xc0, 0>; + def ROTRV_MM : MMRel, shift_rotate_reg<"rotrv", CPURegsOpnd>, + SRLV_FM_MM<0xd0, 0>; + + /// Load and Store Instructions - aligned + defm LB_MM : LoadM<"lb", CPURegs, sextloadi8>, MMRel, LW_FM_MM<0x7>; + defm LBu_MM : LoadM<"lbu", CPURegs, zextloadi8>, MMRel, LW_FM_MM<0x5>; + defm LH_MM : LoadM<"lh", CPURegs, sextloadi16>, MMRel, LW_FM_MM<0xf>; + defm LHu_MM : LoadM<"lhu", CPURegs, zextloadi16>, MMRel, LW_FM_MM<0xd>; + defm LW_MM : LoadM<"lw", CPURegs>, MMRel, LW_FM_MM<0x3f>; + defm SB_MM : StoreM<"sb", CPURegs, truncstorei8>, MMRel, LW_FM_MM<0x6>; + defm SH_MM : StoreM<"sh", CPURegs, truncstorei16>, MMRel, LW_FM_MM<0xe>; + defm SW_MM : StoreM<"sw", CPURegs>, MMRel, LW_FM_MM<0x3e>; +} diff --git a/lib/Target/Mips/Mips.td b/lib/Target/Mips/Mips.td index 13266233dd..eefb02a494 100644 --- a/lib/Target/Mips/Mips.td +++ b/lib/Target/Mips/Mips.td @@ -44,8 +44,6 @@ def FeatureN64 : SubtargetFeature<"n64", "MipsABI", "N64", "Enable n64 ABI">; def FeatureEABI : SubtargetFeature<"eabi", "MipsABI", "EABI", "Enable eabi ABI">; -def FeatureAndroid : SubtargetFeature<"android", "IsAndroid", "true", - "Target is android">; def FeatureVFPU : SubtargetFeature<"vfpu", "HasVFPU", "true", "Enable vector FPU instructions.">; def FeatureSEInReg : SubtargetFeature<"seinreg", "HasSEInReg", "true", diff --git a/lib/Target/Mips/Mips16FrameLowering.h b/lib/Target/Mips/Mips16FrameLowering.h index 25f4ffb929..54fdb78714 100644 --- a/lib/Target/Mips/Mips16FrameLowering.h +++ b/lib/Target/Mips/Mips16FrameLowering.h @@ -20,7 +20,7 @@ namespace llvm { class Mips16FrameLowering : public MipsFrameLowering { public: explicit Mips16FrameLowering(const MipsSubtarget &STI) - : MipsFrameLowering(STI) {} + : MipsFrameLowering(STI, 8) {} /// emitProlog/emitEpilog - These methods insert prolog and epilog code into /// the function. diff --git a/lib/Target/Mips/Mips16ISelDAGToDAG.cpp b/lib/Target/Mips/Mips16ISelDAGToDAG.cpp new file mode 100644 index 0000000000..c1c635cb9f --- /dev/null +++ b/lib/Target/Mips/Mips16ISelDAGToDAG.cpp @@ -0,0 +1,313 @@ +//===-- Mips16ISelDAGToDAG.cpp - A Dag to Dag Inst Selector for Mips16 ----===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsDAGToDAGISel specialized for mips16. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "mips-isel" +#include "Mips16ISelDAGToDAG.h" +#include "Mips.h" +#include "MCTargetDesc/MipsBaseInfo.h" +#include "MipsAnalyzeImmediate.h" +#include "MipsMachineFunction.h" +#include "MipsRegisterInfo.h" +#include "llvm/CodeGen/MachineConstantPool.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineFunction.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/SelectionDAGNodes.h" +#include "llvm/IR/GlobalValue.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/IR/Type.h" +#include "llvm/Support/CFG.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetMachine.h" +using namespace llvm; + +bool Mips16DAGToDAGISel::runOnMachineFunction(MachineFunction &MF) { + if (!Subtarget.inMips16Mode()) + return false; + return MipsDAGToDAGISel::runOnMachineFunction(MF); +} +/// Select multiply instructions. +std::pair<SDNode*, SDNode*> +Mips16DAGToDAGISel::selectMULT(SDNode *N, unsigned Opc, DebugLoc DL, EVT Ty, + bool HasLo, bool HasHi) { + SDNode *Lo = 0, *Hi = 0; + SDNode *Mul = CurDAG->getMachineNode(Opc, DL, MVT::Glue, N->getOperand(0), + N->getOperand(1)); + SDValue InFlag = SDValue(Mul, 0); + + if (HasLo) { + unsigned Opcode = Mips::Mflo16; + Lo = CurDAG->getMachineNode(Opcode, DL, Ty, MVT::Glue, InFlag); + InFlag = SDValue(Lo, 1); + } + if (HasHi) { + unsigned Opcode = Mips::Mfhi16; + Hi = CurDAG->getMachineNode(Opcode, DL, Ty, InFlag); + } + return std::make_pair(Lo, Hi); +} + +void Mips16DAGToDAGISel::initGlobalBaseReg(MachineFunction &MF) { + MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); + + if (!MipsFI->globalBaseRegSet()) + return; + + MachineBasicBlock &MBB = MF.front(); + MachineBasicBlock::iterator I = MBB.begin(); + MachineRegisterInfo &RegInfo = MF.getRegInfo(); + const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo(); + DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); + unsigned V0, V1, V2, GlobalBaseReg = MipsFI->getGlobalBaseReg(); + const TargetRegisterClass *RC = + (const TargetRegisterClass*)&Mips::CPU16RegsRegClass; + + V0 = RegInfo.createVirtualRegister(RC); + V1 = RegInfo.createVirtualRegister(RC); + V2 = RegInfo.createVirtualRegister(RC); + + BuildMI(MBB, I, DL, TII.get(Mips::LiRxImmX16), V0) + .addExternalSymbol("_gp_disp", MipsII::MO_ABS_HI); + BuildMI(MBB, I, DL, TII.get(Mips::AddiuRxPcImmX16), V1) + .addExternalSymbol("_gp_disp", MipsII::MO_ABS_LO); + BuildMI(MBB, I, DL, TII.get(Mips::SllX16), V2).addReg(V0).addImm(16); + BuildMI(MBB, I, DL, TII.get(Mips::AdduRxRyRz16), GlobalBaseReg) + .addReg(V1).addReg(V2); +} + +// Insert instructions to initialize the Mips16 SP Alias register in the +// first MBB of the function. +// +void Mips16DAGToDAGISel::initMips16SPAliasReg(MachineFunction &MF) { + MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); + + if (!MipsFI->mips16SPAliasRegSet()) + return; + + MachineBasicBlock &MBB = MF.front(); + MachineBasicBlock::iterator I = MBB.begin(); + const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo(); + DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); + unsigned Mips16SPAliasReg = MipsFI->getMips16SPAliasReg(); + + BuildMI(MBB, I, DL, TII.get(Mips::MoveR3216), Mips16SPAliasReg) + .addReg(Mips::SP); +} + +void Mips16DAGToDAGISel::processFunctionAfterISel(MachineFunction &MF) { + initGlobalBaseReg(MF); + initMips16SPAliasReg(MF); +} + +/// getMips16SPAliasReg - Output the instructions required to put the +/// SP into a Mips16 accessible aliased register. +SDValue Mips16DAGToDAGISel::getMips16SPAliasReg() { + unsigned Mips16SPAliasReg = + MF->getInfo<MipsFunctionInfo>()->getMips16SPAliasReg(); + return CurDAG->getRegister(Mips16SPAliasReg, TLI.getPointerTy()); +} + +void Mips16DAGToDAGISel::getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg) { + SDValue AliasFPReg = CurDAG->getRegister(Mips::S0, TLI.getPointerTy()); + if (Parent) { + switch (Parent->getOpcode()) { + case ISD::LOAD: { + LoadSDNode *SD = dyn_cast<LoadSDNode>(Parent); + switch (SD->getMemoryVT().getSizeInBits()) { + case 8: + case 16: + AliasReg = TM.getFrameLowering()->hasFP(*MF)? + AliasFPReg: getMips16SPAliasReg(); + return; + } + break; + } + case ISD::STORE: { + StoreSDNode *SD = dyn_cast<StoreSDNode>(Parent); + switch (SD->getMemoryVT().getSizeInBits()) { + case 8: + case 16: + AliasReg = TM.getFrameLowering()->hasFP(*MF)? + AliasFPReg: getMips16SPAliasReg(); + return; + } + break; + } + } + } + AliasReg = CurDAG->getRegister(Mips::SP, TLI.getPointerTy()); + return; + +} + +bool Mips16DAGToDAGISel::selectAddr16( + SDNode *Parent, SDValue Addr, SDValue &Base, SDValue &Offset, + SDValue &Alias) { + EVT ValTy = Addr.getValueType(); + + Alias = CurDAG->getTargetConstant(0, ValTy); + + // if Address is FI, get the TargetFrameIndex. + if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) { + Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); + Offset = CurDAG->getTargetConstant(0, ValTy); + getMips16SPRefReg(Parent, Alias); + return true; + } + // on PIC code Load GA + if (Addr.getOpcode() == MipsISD::Wrapper) { + Base = Addr.getOperand(0); + Offset = Addr.getOperand(1); + return true; + } + if (TM.getRelocationModel() != Reloc::PIC_) { + if ((Addr.getOpcode() == ISD::TargetExternalSymbol || + Addr.getOpcode() == ISD::TargetGlobalAddress)) + return false; + } + // Addresses of the form FI+const or FI|const + if (CurDAG->isBaseWithConstantOffset(Addr)) { + ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1)); + if (isInt<16>(CN->getSExtValue())) { + + // If the first operand is a FI, get the TargetFI Node + if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode> + (Addr.getOperand(0))) { + Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); + getMips16SPRefReg(Parent, Alias); + } + else + Base = Addr.getOperand(0); + + Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy); + return true; + } + } + // Operand is a result from an ADD. + if (Addr.getOpcode() == ISD::ADD) { + // When loading from constant pools, load the lower address part in + // the instruction itself. Example, instead of: + // lui $2, %hi($CPI1_0) + // addiu $2, $2, %lo($CPI1_0) + // lwc1 $f0, 0($2) + // Generate: + // lui $2, %hi($CPI1_0) + // lwc1 $f0, %lo($CPI1_0)($2) + if (Addr.getOperand(1).getOpcode() == MipsISD::Lo || + Addr.getOperand(1).getOpcode() == MipsISD::GPRel) { + SDValue Opnd0 = Addr.getOperand(1).getOperand(0); + if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) || + isa<JumpTableSDNode>(Opnd0)) { + Base = Addr.getOperand(0); + Offset = Opnd0; + return true; + } + } + + // If an indexed floating point load/store can be emitted, return false. + const LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(Parent); + + if (LS && + (LS->getMemoryVT() == MVT::f32 || LS->getMemoryVT() == MVT::f64) && + Subtarget.hasFPIdx()) + return false; + } + Base = Addr; + Offset = CurDAG->getTargetConstant(0, ValTy); + return true; +} + +/// Select instructions not customized! Used for +/// expanded, promoted and normal instructions +std::pair<bool, SDNode*> Mips16DAGToDAGISel::selectNode(SDNode *Node) { + unsigned Opcode = Node->getOpcode(); + DebugLoc DL = Node->getDebugLoc(); + + /// + // Instruction Selection not handled by the auto-generated + // tablegen selection should be handled here. + /// + EVT NodeTy = Node->getValueType(0); + unsigned MultOpc; + + switch(Opcode) { + default: break; + + case ISD::SUBE: + case ISD::ADDE: { + SDValue InFlag = Node->getOperand(2), CmpLHS; + unsigned Opc = InFlag.getOpcode(); (void)Opc; + assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) || + (Opc == ISD::SUBC || Opc == ISD::SUBE)) && + "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn"); + + unsigned MOp; + if (Opcode == ISD::ADDE) { + CmpLHS = InFlag.getValue(0); + MOp = Mips::AdduRxRyRz16; + } else { + CmpLHS = InFlag.getOperand(0); + MOp = Mips::SubuRxRyRz16; + } + + SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) }; + + SDValue LHS = Node->getOperand(0); + SDValue RHS = Node->getOperand(1); + + EVT VT = LHS.getValueType(); + + unsigned Sltu_op = Mips::SltuRxRyRz16; + SDNode *Carry = CurDAG->getMachineNode(Sltu_op, DL, VT, Ops); + unsigned Addu_op = Mips::AdduRxRyRz16; + SDNode *AddCarry = CurDAG->getMachineNode(Addu_op, DL, VT, + SDValue(Carry,0), RHS); + + SDNode *Result = CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue, LHS, + SDValue(AddCarry,0)); + return std::make_pair(true, Result); + } + + /// Mul with two results + case ISD::SMUL_LOHI: + case ISD::UMUL_LOHI: { + MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MultuRxRy16 : Mips::MultRxRy16); + std::pair<SDNode*, SDNode*> LoHi = selectMULT(Node, MultOpc, DL, NodeTy, + true, true); + if (!SDValue(Node, 0).use_empty()) + ReplaceUses(SDValue(Node, 0), SDValue(LoHi.first, 0)); + + if (!SDValue(Node, 1).use_empty()) + ReplaceUses(SDValue(Node, 1), SDValue(LoHi.second, 0)); + + return std::make_pair(true, (SDNode*)NULL); + } + + case ISD::MULHS: + case ISD::MULHU: { + MultOpc = (Opcode == ISD::MULHU ? Mips::MultuRxRy16 : Mips::MultRxRy16); + SDNode *Result = selectMULT(Node, MultOpc, DL, NodeTy, false, true).second; + return std::make_pair(true, Result); + } + } + + return std::make_pair(false, (SDNode*)NULL); +} + +FunctionPass *llvm::createMips16ISelDag(MipsTargetMachine &TM) { + return new Mips16DAGToDAGISel(TM); +} diff --git a/lib/Target/Mips/Mips16ISelDAGToDAG.h b/lib/Target/Mips/Mips16ISelDAGToDAG.h new file mode 100644 index 0000000000..f05f9b766d --- /dev/null +++ b/lib/Target/Mips/Mips16ISelDAGToDAG.h @@ -0,0 +1,53 @@ +//===---- Mips16ISelDAGToDAG.h - A Dag to Dag Inst Selector for Mips ------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsDAGToDAGISel specialized for mips16. +// +//===----------------------------------------------------------------------===// + +#ifndef MIPS16ISELDAGTODAG_H +#define MIPS16ISELDAGTODAG_H + +#include "MipsISelDAGToDAG.h" + +namespace llvm { + +class Mips16DAGToDAGISel : public MipsDAGToDAGISel { +public: + explicit Mips16DAGToDAGISel(MipsTargetMachine &TM) : MipsDAGToDAGISel(TM) {} + +private: + std::pair<SDNode*, SDNode*> selectMULT(SDNode *N, unsigned Opc, DebugLoc DL, + EVT Ty, bool HasLo, bool HasHi); + + SDValue getMips16SPAliasReg(); + + virtual bool runOnMachineFunction(MachineFunction &MF); + + void getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg); + + virtual bool selectAddr16(SDNode *Parent, SDValue N, SDValue &Base, + SDValue &Offset, SDValue &Alias); + + virtual std::pair<bool, SDNode*> selectNode(SDNode *Node); + + virtual void processFunctionAfterISel(MachineFunction &MF); + + // Insert instructions to initialize the global base register in the + // first MBB of the function. + void initGlobalBaseReg(MachineFunction &MF); + + void initMips16SPAliasReg(MachineFunction &MF); +}; + +FunctionPass *createMips16ISelDag(MipsTargetMachine &TM); + +} + +#endif diff --git a/lib/Target/Mips/Mips16ISelLowering.cpp b/lib/Target/Mips/Mips16ISelLowering.cpp new file mode 100644 index 0000000000..f63318f1e6 --- /dev/null +++ b/lib/Target/Mips/Mips16ISelLowering.cpp @@ -0,0 +1,690 @@ +//===-- Mips16ISelLowering.h - Mips16 DAG Lowering Interface ----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsTargetLowering specialized for mips16. +// +//===----------------------------------------------------------------------===// +#define DEBUG_TYPE "mips-lower" +#include "Mips16ISelLowering.h" +#include "MipsRegisterInfo.h" +#include "MCTargetDesc/MipsBaseInfo.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Target/TargetInstrInfo.h" +#include <set> + +using namespace llvm; + +static cl::opt<bool> +Mips16HardFloat("mips16-hard-float", cl::NotHidden, + cl::desc("MIPS: mips16 hard float enable."), + cl::init(false)); + +static cl::opt<bool> DontExpandCondPseudos16( + "mips16-dont-expand-cond-pseudo", + cl::init(false), + cl::desc("Dont expand conditional move related " + "pseudos for Mips 16"), + cl::Hidden); + +namespace { + std::set<const char*, MipsTargetLowering::LTStr> NoHelperNeeded; +} + +Mips16TargetLowering::Mips16TargetLowering(MipsTargetMachine &TM) + : MipsTargetLowering(TM) { + // + // set up as if mips32 and then revert so we can test the mechanism + // for switching + addRegisterClass(MVT::i32, &Mips::CPURegsRegClass); + addRegisterClass(MVT::f32, &Mips::FGR32RegClass); + computeRegisterProperties(); + clearRegisterClasses(); + + // Set up the register classes + addRegisterClass(MVT::i32, &Mips::CPU16RegsRegClass); + + if (Mips16HardFloat) + setMips16HardFloatLibCalls(); + + setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand); + setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_SWAP, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand); + setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand); + + computeRegisterProperties(); +} + +const MipsTargetLowering * +llvm::createMips16TargetLowering(MipsTargetMachine &TM) { + return new Mips16TargetLowering(TM); +} + +bool +Mips16TargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const { + return false; +} + +MachineBasicBlock * +Mips16TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, + MachineBasicBlock *BB) const { + switch (MI->getOpcode()) { + default: + return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB); + case Mips::SelBeqZ: + return emitSel16(Mips::BeqzRxImm16, MI, BB); + case Mips::SelBneZ: + return emitSel16(Mips::BnezRxImm16, MI, BB); + case Mips::SelTBteqZCmpi: + return emitSeliT16(Mips::BteqzX16, Mips::CmpiRxImmX16, MI, BB); + case Mips::SelTBteqZSlti: + return emitSeliT16(Mips::BteqzX16, Mips::SltiRxImmX16, MI, BB); + case Mips::SelTBteqZSltiu: + return emitSeliT16(Mips::BteqzX16, Mips::SltiuRxImmX16, MI, BB); + case Mips::SelTBtneZCmpi: + return emitSeliT16(Mips::BtnezX16, Mips::CmpiRxImmX16, MI, BB); + case Mips::SelTBtneZSlti: + return emitSeliT16(Mips::BtnezX16, Mips::SltiRxImmX16, MI, BB); + case Mips::SelTBtneZSltiu: + return emitSeliT16(Mips::BtnezX16, Mips::SltiuRxImmX16, MI, BB); + case Mips::SelTBteqZCmp: + return emitSelT16(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB); + case Mips::SelTBteqZSlt: + return emitSelT16(Mips::BteqzX16, Mips::SltRxRy16, MI, BB); + case Mips::SelTBteqZSltu: + return emitSelT16(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB); + case Mips::SelTBtneZCmp: + return emitSelT16(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB); + case Mips::SelTBtneZSlt: + return emitSelT16(Mips::BtnezX16, Mips::SltRxRy16, MI, BB); + case Mips::SelTBtneZSltu: + return emitSelT16(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB); + case Mips::BteqzT8CmpX16: + return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB); + case Mips::BteqzT8SltX16: + return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltRxRy16, MI, BB); + case Mips::BteqzT8SltuX16: + // TBD: figure out a way to get this or remove the instruction + // altogether. + return emitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB); + case Mips::BtnezT8CmpX16: + return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB); + case Mips::BtnezT8SltX16: + return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltRxRy16, MI, BB); + case Mips::BtnezT8SltuX16: + // TBD: figure out a way to get this or remove the instruction + // altogether. + return emitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB); + case Mips::BteqzT8CmpiX16: return emitFEXT_T8I8I16_ins( + Mips::BteqzX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB); + case Mips::BteqzT8SltiX16: return emitFEXT_T8I8I16_ins( + Mips::BteqzX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); + case Mips::BteqzT8SltiuX16: return emitFEXT_T8I8I16_ins( + Mips::BteqzX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); + case Mips::BtnezT8CmpiX16: return emitFEXT_T8I8I16_ins( + Mips::BtnezX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB); + case Mips::BtnezT8SltiX16: return emitFEXT_T8I8I16_ins( + Mips::BtnezX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); + case Mips::BtnezT8SltiuX16: return emitFEXT_T8I8I16_ins( + Mips::BtnezX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); + break; + case Mips::SltCCRxRy16: + return emitFEXT_CCRX16_ins(Mips::SltRxRy16, MI, BB); + break; + case Mips::SltiCCRxImmX16: + return emitFEXT_CCRXI16_ins + (Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); + case Mips::SltiuCCRxImmX16: + return emitFEXT_CCRXI16_ins + (Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); + case Mips::SltuCCRxRy16: + return emitFEXT_CCRX16_ins + (Mips::SltuRxRy16, MI, BB); + } +} + +bool Mips16TargetLowering:: +isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, + unsigned NextStackOffset, + const MipsFunctionInfo& FI) const { + // No tail call optimization for mips16. + return false; +} + +void Mips16TargetLowering::setMips16LibcallName + (RTLIB::Libcall L, const char *Name) { + setLibcallName(L, Name); + NoHelperNeeded.insert(Name); +} + +void Mips16TargetLowering::setMips16HardFloatLibCalls() { + setMips16LibcallName(RTLIB::ADD_F32, "__mips16_addsf3"); + setMips16LibcallName(RTLIB::ADD_F64, "__mips16_adddf3"); + setMips16LibcallName(RTLIB::SUB_F32, "__mips16_subsf3"); + setMips16LibcallName(RTLIB::SUB_F64, "__mips16_subdf3"); + setMips16LibcallName(RTLIB::MUL_F32, "__mips16_mulsf3"); + setMips16LibcallName(RTLIB::MUL_F64, "__mips16_muldf3"); + setMips16LibcallName(RTLIB::DIV_F32, "__mips16_divsf3"); + setMips16LibcallName(RTLIB::DIV_F64, "__mips16_divdf3"); + setMips16LibcallName(RTLIB::FPEXT_F32_F64, "__mips16_extendsfdf2"); + setMips16LibcallName(RTLIB::FPROUND_F64_F32, "__mips16_truncdfsf2"); + setMips16LibcallName(RTLIB::FPTOSINT_F32_I32, "__mips16_fix_truncsfsi"); + setMips16LibcallName(RTLIB::FPTOSINT_F64_I32, "__mips16_fix_truncdfsi"); + setMips16LibcallName(RTLIB::SINTTOFP_I32_F32, "__mips16_floatsisf"); + setMips16LibcallName(RTLIB::SINTTOFP_I32_F64, "__mips16_floatsidf"); + setMips16LibcallName(RTLIB::UINTTOFP_I32_F32, "__mips16_floatunsisf"); + setMips16LibcallName(RTLIB::UINTTOFP_I32_F64, "__mips16_floatunsidf"); + setMips16LibcallName(RTLIB::OEQ_F32, "__mips16_eqsf2"); + setMips16LibcallName(RTLIB::OEQ_F64, "__mips16_eqdf2"); + setMips16LibcallName(RTLIB::UNE_F32, "__mips16_nesf2"); + setMips16LibcallName(RTLIB::UNE_F64, "__mips16_nedf2"); + setMips16LibcallName(RTLIB::OGE_F32, "__mips16_gesf2"); + setMips16LibcallName(RTLIB::OGE_F64, "__mips16_gedf2"); + setMips16LibcallName(RTLIB::OLT_F32, "__mips16_ltsf2"); + setMips16LibcallName(RTLIB::OLT_F64, "__mips16_ltdf2"); + setMips16LibcallName(RTLIB::OLE_F32, "__mips16_lesf2"); + setMips16LibcallName(RTLIB::OLE_F64, "__mips16_ledf2"); + setMips16LibcallName(RTLIB::OGT_F32, "__mips16_gtsf2"); + setMips16LibcallName(RTLIB::OGT_F64, "__mips16_gtdf2"); + setMips16LibcallName(RTLIB::UO_F32, "__mips16_unordsf2"); + setMips16LibcallName(RTLIB::UO_F64, "__mips16_unorddf2"); + setMips16LibcallName(RTLIB::O_F32, "__mips16_unordsf2"); + setMips16LibcallName(RTLIB::O_F64, "__mips16_unorddf2"); +} + + +// +// The Mips16 hard float is a crazy quilt inherited from gcc. I have a much +// cleaner way to do all of this but it will have to wait until the traditional +// gcc mechanism is completed. +// +// For Pic, in order for Mips16 code to call Mips32 code which according the abi +// have either arguments or returned values placed in floating point registers, +// we use a set of helper functions. (This includes functions which return type +// complex which on Mips are returned in a pair of floating point registers). +// +// This is an encoding that we inherited from gcc. +// In Mips traditional O32, N32 ABI, floating point numbers are passed in +// floating point argument registers 1,2 only when the first and optionally +// the second arguments are float (sf) or double (df). +// For Mips16 we are only concerned with the situations where floating point +// arguments are being passed in floating point registers by the ABI, because +// Mips16 mode code cannot execute floating point instructions to load those +// values and hence helper functions are needed. +// The possibilities are (), (sf), (sf, sf), (sf, df), (df), (df, sf), (df, df) +// the helper function suffixs for these are: +// 0, 1, 5, 9, 2, 6, 10 +// this suffix can then be calculated as follows: +// for a given argument Arg: +// Arg1x, Arg2x = 1 : Arg is sf +// 2 : Arg is df +// 0: Arg is neither sf or df +// So this stub is the string for number Arg1x + Arg2x*4. +// However not all numbers between 0 and 10 are possible, we check anyway and +// assert if the impossible exists. +// + +unsigned int Mips16TargetLowering::getMips16HelperFunctionStubNumber + (ArgListTy &Args) const { + unsigned int resultNum = 0; + if (Args.size() >= 1) { + Type *t = Args[0].Ty; + if (t->isFloatTy()) { + resultNum = 1; + } + else if (t->isDoubleTy()) { + resultNum = 2; + } + } + if (resultNum) { + if (Args.size() >=2) { + Type *t = Args[1].Ty; + if (t->isFloatTy()) { + resultNum += 4; + } + else if (t->isDoubleTy()) { + resultNum += 8; + } + } + } + return resultNum; +} + +// +// prefixs are attached to stub numbers depending on the return type . +// return type: float sf_ +// double df_ +// single complex sc_ +// double complext dc_ +// others NO PREFIX +// +// +// The full name of a helper function is__mips16_call_stub + +// return type dependent prefix + stub number +// +// +// This is something that probably should be in a different source file and +// perhaps done differently but my main purpose is to not waste runtime +// on something that we can enumerate in the source. Another possibility is +// to have a python script to generate these mapping tables. This will do +// for now. There are a whole series of helper function mapping arrays, one +// for each return type class as outlined above. There there are 11 possible +// entries. Ones with 0 are ones which should never be selected +// +// All the arrays are similar except for ones which return neither +// sf, df, sc, dc, in which only care about ones which have sf or df as a +// first parameter. +// +#define P_ "__mips16_call_stub_" +#define MAX_STUB_NUMBER 10 +#define T1 P "1", P "2", 0, 0, P "5", P "6", 0, 0, P "9", P "10" +#define T P "0" , T1 +#define P P_ +static char const * vMips16Helper[MAX_STUB_NUMBER+1] = + {0, T1 }; +#undef P +#define P P_ "sf_" +static char const * sfMips16Helper[MAX_STUB_NUMBER+1] = + { T }; +#undef P +#define P P_ "df_" +static char const * dfMips16Helper[MAX_STUB_NUMBER+1] = + { T }; +#undef P +#define P P_ "sc_" +static char const * scMips16Helper[MAX_STUB_NUMBER+1] = + { T }; +#undef P +#define P P_ "dc_" +static char const * dcMips16Helper[MAX_STUB_NUMBER+1] = + { T }; +#undef P +#undef P_ + + +const char* Mips16TargetLowering:: + getMips16HelperFunction + (Type* RetTy, ArgListTy &Args, bool &needHelper) const { + const unsigned int stubNum = getMips16HelperFunctionStubNumber(Args); +#ifndef NDEBUG + const unsigned int maxStubNum = 10; + assert(stubNum <= maxStubNum); + const bool validStubNum[maxStubNum+1] = + {true, true, true, false, false, true, true, false, false, true, true}; + assert(validStubNum[stubNum]); +#endif + const char *result; + if (RetTy->isFloatTy()) { + result = sfMips16Helper[stubNum]; + } + else if (RetTy ->isDoubleTy()) { + result = dfMips16Helper[stubNum]; + } + else if (RetTy->isStructTy()) { + // check if it's complex + if (RetTy->getNumContainedTypes() == 2) { + if ((RetTy->getContainedType(0)->isFloatTy()) && + (RetTy->getContainedType(1)->isFloatTy())) { + result = scMips16Helper[stubNum]; + } + else if ((RetTy->getContainedType(0)->isDoubleTy()) && + (RetTy->getContainedType(1)->isDoubleTy())) { + result = dcMips16Helper[stubNum]; + } + else { + llvm_unreachable("Uncovered condition"); + } + } + else { + llvm_unreachable("Uncovered condition"); + } + } + else { + if (stubNum == 0) { + needHelper = false; + return ""; + } + result = vMips16Helper[stubNum]; + } + needHelper = true; + return result; +} + +void Mips16TargetLowering:: +getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const { + SelectionDAG &DAG = CLI.DAG; + const char* Mips16HelperFunction = 0; + bool NeedMips16Helper = false; + + if (getTargetMachine().Options.UseSoftFloat && Mips16HardFloat) { + // + // currently we don't have symbols tagged with the mips16 or mips32 + // qualifier so we will assume that we don't know what kind it is. + // and generate the helper + // + bool LookupHelper = true; + if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(CLI.Callee)) { + if (NoHelperNeeded.find(S->getSymbol()) != NoHelperNeeded.end()) { + LookupHelper = false; + } + } + if (LookupHelper) Mips16HelperFunction = + getMips16HelperFunction(CLI.RetTy, CLI.Args, NeedMips16Helper); + + } + + SDValue JumpTarget = Callee; + + // T9 should contain the address of the callee function if + // -reloction-model=pic or it is an indirect call. + if (IsPICCall || !GlobalOrExternal) { + unsigned V0Reg = Mips::V0; + if (NeedMips16Helper) { + RegsToPass.push_front(std::make_pair(V0Reg, Callee)); + JumpTarget = DAG.getExternalSymbol(Mips16HelperFunction, getPointerTy()); + JumpTarget = getAddrGlobal(JumpTarget, DAG, MipsII::MO_GOT); + } else + RegsToPass.push_front(std::make_pair((unsigned)Mips::T9, Callee)); + } + + Ops.push_back(JumpTarget); + + MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal, + InternalLinkage, CLI, Callee, Chain); +} + +MachineBasicBlock *Mips16TargetLowering:: +emitSel16(unsigned Opc, MachineInstr *MI, MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + DebugLoc DL = MI->getDebugLoc(); + // To "insert" a SELECT_CC instruction, we actually have to insert the + // diamond control-flow pattern. The incoming instruction knows the + // destination vreg to set, the condition code register to branch on, the + // true/false values to select between, and a branch opcode to use. + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = BB; + ++It; + + // thisMBB: + // ... + // TrueVal = ... + // setcc r1, r2, r3 + // bNE r1, r0, copy1MBB + // fallthrough --> copy0MBB + MachineBasicBlock *thisMBB = BB; + MachineFunction *F = BB->getParent(); + MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); + F->insert(It, copy0MBB); + F->insert(It, sinkMBB); + + // Transfer the remainder of BB and its successor edges to sinkMBB. + sinkMBB->splice(sinkMBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + sinkMBB->transferSuccessorsAndUpdatePHIs(BB); + + // Next, add the true and fallthrough blocks as its successors. + BB->addSuccessor(copy0MBB); + BB->addSuccessor(sinkMBB); + + BuildMI(BB, DL, TII->get(Opc)).addReg(MI->getOperand(3).getReg()) + .addMBB(sinkMBB); + + // copy0MBB: + // %FalseValue = ... + // # fallthrough to sinkMBB + BB = copy0MBB; + + // Update machine-CFG edges + BB->addSuccessor(sinkMBB); + + // sinkMBB: + // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] + // ... + BB = sinkMBB; + + BuildMI(*BB, BB->begin(), DL, + TII->get(Mips::PHI), MI->getOperand(0).getReg()) + .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) + .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); + + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; +} + +MachineBasicBlock *Mips16TargetLowering::emitSelT16 + (unsigned Opc1, unsigned Opc2, + MachineInstr *MI, MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + DebugLoc DL = MI->getDebugLoc(); + // To "insert" a SELECT_CC instruction, we actually have to insert the + // diamond control-flow pattern. The incoming instruction knows the + // destination vreg to set, the condition code register to branch on, the + // true/false values to select between, and a branch opcode to use. + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = BB; + ++It; + + // thisMBB: + // ... + // TrueVal = ... + // setcc r1, r2, r3 + // bNE r1, r0, copy1MBB + // fallthrough --> copy0MBB + MachineBasicBlock *thisMBB = BB; + MachineFunction *F = BB->getParent(); + MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); + F->insert(It, copy0MBB); + F->insert(It, sinkMBB); + + // Transfer the remainder of BB and its successor edges to sinkMBB. + sinkMBB->splice(sinkMBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + sinkMBB->transferSuccessorsAndUpdatePHIs(BB); + + // Next, add the true and fallthrough blocks as its successors. + BB->addSuccessor(copy0MBB); + BB->addSuccessor(sinkMBB); + + BuildMI(BB, DL, TII->get(Opc2)).addReg(MI->getOperand(3).getReg()) + .addReg(MI->getOperand(4).getReg()); + BuildMI(BB, DL, TII->get(Opc1)).addMBB(sinkMBB); + + // copy0MBB: + // %FalseValue = ... + // # fallthrough to sinkMBB + BB = copy0MBB; + + // Update machine-CFG edges + BB->addSuccessor(sinkMBB); + + // sinkMBB: + // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] + // ... + BB = sinkMBB; + + BuildMI(*BB, BB->begin(), DL, + TII->get(Mips::PHI), MI->getOperand(0).getReg()) + .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) + .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); + + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; + +} + +MachineBasicBlock *Mips16TargetLowering::emitSeliT16 + (unsigned Opc1, unsigned Opc2, + MachineInstr *MI, MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + DebugLoc DL = MI->getDebugLoc(); + // To "insert" a SELECT_CC instruction, we actually have to insert the + // diamond control-flow pattern. The incoming instruction knows the + // destination vreg to set, the condition code register to branch on, the + // true/false values to select between, and a branch opcode to use. + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = BB; + ++It; + + // thisMBB: + // ... + // TrueVal = ... + // setcc r1, r2, r3 + // bNE r1, r0, copy1MBB + // fallthrough --> copy0MBB + MachineBasicBlock *thisMBB = BB; + MachineFunction *F = BB->getParent(); + MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); + F->insert(It, copy0MBB); + F->insert(It, sinkMBB); + + // Transfer the remainder of BB and its successor edges to sinkMBB. + sinkMBB->splice(sinkMBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + sinkMBB->transferSuccessorsAndUpdatePHIs(BB); + + // Next, add the true and fallthrough blocks as its successors. + BB->addSuccessor(copy0MBB); + BB->addSuccessor(sinkMBB); + + BuildMI(BB, DL, TII->get(Opc2)).addReg(MI->getOperand(3).getReg()) + .addImm(MI->getOperand(4).getImm()); + BuildMI(BB, DL, TII->get(Opc1)).addMBB(sinkMBB); + + // copy0MBB: + // %FalseValue = ... + // # fallthrough to sinkMBB + BB = copy0MBB; + + // Update machine-CFG edges + BB->addSuccessor(sinkMBB); + + // sinkMBB: + // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] + // ... + BB = sinkMBB; + + BuildMI(*BB, BB->begin(), DL, + TII->get(Mips::PHI), MI->getOperand(0).getReg()) + .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) + .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); + + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; + +} + +MachineBasicBlock + *Mips16TargetLowering::emitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc, + MachineInstr *MI, + MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + unsigned regX = MI->getOperand(0).getReg(); + unsigned regY = MI->getOperand(1).getReg(); + MachineBasicBlock *target = MI->getOperand(2).getMBB(); + BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX) + .addReg(regY); + BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target); + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; +} + +MachineBasicBlock *Mips16TargetLowering::emitFEXT_T8I8I16_ins( + unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc, + MachineInstr *MI, MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + unsigned regX = MI->getOperand(0).getReg(); + int64_t imm = MI->getOperand(1).getImm(); + MachineBasicBlock *target = MI->getOperand(2).getMBB(); + unsigned CmpOpc; + if (isUInt<8>(imm)) + CmpOpc = CmpiOpc; + else if (isUInt<16>(imm)) + CmpOpc = CmpiXOpc; + else + llvm_unreachable("immediate field not usable"); + BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX) + .addImm(imm); + BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target); + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; +} + +static unsigned Mips16WhichOp8uOr16simm + (unsigned shortOp, unsigned longOp, int64_t Imm) { + if (isUInt<8>(Imm)) + return shortOp; + else if (isInt<16>(Imm)) + return longOp; + else + llvm_unreachable("immediate field not usable"); +} + +MachineBasicBlock *Mips16TargetLowering::emitFEXT_CCRX16_ins( + unsigned SltOpc, + MachineInstr *MI, MachineBasicBlock *BB) const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + unsigned CC = MI->getOperand(0).getReg(); + unsigned regX = MI->getOperand(1).getReg(); + unsigned regY = MI->getOperand(2).getReg(); + BuildMI(*BB, MI, MI->getDebugLoc(), + TII->get(SltOpc)).addReg(regX).addReg(regY); + BuildMI(*BB, MI, MI->getDebugLoc(), + TII->get(Mips::MoveR3216), CC).addReg(Mips::T8); + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; +} + +MachineBasicBlock *Mips16TargetLowering::emitFEXT_CCRXI16_ins( + unsigned SltiOpc, unsigned SltiXOpc, + MachineInstr *MI, MachineBasicBlock *BB )const { + if (DontExpandCondPseudos16) + return BB; + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + unsigned CC = MI->getOperand(0).getReg(); + unsigned regX = MI->getOperand(1).getReg(); + int64_t Imm = MI->getOperand(2).getImm(); + unsigned SltOpc = Mips16WhichOp8uOr16simm(SltiOpc, SltiXOpc, Imm); + BuildMI(*BB, MI, MI->getDebugLoc(), + TII->get(SltOpc)).addReg(regX).addImm(Imm); + BuildMI(*BB, MI, MI->getDebugLoc(), + TII->get(Mips::MoveR3216), CC).addReg(Mips::T8); + MI->eraseFromParent(); // The pseudo instruction is gone now. + return BB; + +} diff --git a/lib/Target/Mips/Mips16ISelLowering.h b/lib/Target/Mips/Mips16ISelLowering.h new file mode 100644 index 0000000000..b23e2a1f37 --- /dev/null +++ b/lib/Target/Mips/Mips16ISelLowering.h @@ -0,0 +1,80 @@ +//===-- Mips16ISelLowering.h - Mips16 DAG Lowering Interface ----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsTargetLowering specialized for mips16. +// +//===----------------------------------------------------------------------===// + +#ifndef Mips16ISELLOWERING_H +#define Mips16ISELLOWERING_H + +#include "MipsISelLowering.h" + +namespace llvm { + class Mips16TargetLowering : public MipsTargetLowering { + public: + explicit Mips16TargetLowering(MipsTargetMachine &TM); + + virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const; + + virtual MachineBasicBlock * + EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const; + + private: + virtual bool + isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, + unsigned NextStackOffset, + const MipsFunctionInfo& FI) const; + + void setMips16LibcallName(RTLIB::Libcall, const char *Name); + + void setMips16HardFloatLibCalls(); + + unsigned int + getMips16HelperFunctionStubNumber(ArgListTy &Args) const; + + const char *getMips16HelperFunction + (Type* RetTy, ArgListTy &Args, bool &needHelper) const; + + virtual void + getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const; + + MachineBasicBlock *emitSel16(unsigned Opc, MachineInstr *MI, + MachineBasicBlock *BB) const; + + MachineBasicBlock *emitSeliT16(unsigned Opc1, unsigned Opc2, + MachineInstr *MI, + MachineBasicBlock *BB) const; + + MachineBasicBlock *emitSelT16(unsigned Opc1, unsigned Opc2, + MachineInstr *MI, + MachineBasicBlock *BB) const; + + MachineBasicBlock *emitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc, + MachineInstr *MI, + MachineBasicBlock *BB) const; + + MachineBasicBlock *emitFEXT_T8I8I16_ins( + unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc, + MachineInstr *MI, MachineBasicBlock *BB) const; + + MachineBasicBlock *emitFEXT_CCRX16_ins( + unsigned SltOpc, + MachineInstr *MI, MachineBasicBlock *BB) const; + + MachineBasicBlock *emitFEXT_CCRXI16_ins( + unsigned SltiOpc, unsigned SltiXOpc, + MachineInstr *MI, MachineBasicBlock *BB )const; + }; +} + +#endif // Mips16ISELLOWERING_H diff --git a/lib/Target/Mips/Mips16InstrInfo.cpp b/lib/Target/Mips/Mips16InstrInfo.cpp index fd3cc8f190..17dd2c0796 100644 --- a/lib/Target/Mips/Mips16InstrInfo.cpp +++ b/lib/Target/Mips/Mips16InstrInfo.cpp @@ -98,10 +98,10 @@ void Mips16InstrInfo::copyPhysReg(MachineBasicBlock &MBB, } void Mips16InstrInfo:: -storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, - unsigned SrcReg, bool isKill, int FI, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const { +storeRegToStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, + unsigned SrcReg, bool isKill, int FI, + const TargetRegisterClass *RC, const TargetRegisterInfo *TRI, + int64_t Offset) const { DebugLoc DL; if (I != MBB.end()) DL = I->getDebugLoc(); MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOStore); @@ -110,14 +110,13 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Opc = Mips::SwRxSpImmX16; assert(Opc && "Register class not handled!"); BuildMI(MBB, I, DL, get(Opc)).addReg(SrcReg, getKillRegState(isKill)) - .addFrameIndex(FI).addImm(0).addMemOperand(MMO); + .addFrameIndex(FI).addImm(Offset).addMemOperand(MMO); } void Mips16InstrInfo:: -loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, - unsigned DestReg, int FI, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const { +loadRegFromStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, + unsigned DestReg, int FI, const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, int64_t Offset) const { DebugLoc DL; if (I != MBB.end()) DL = I->getDebugLoc(); MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOLoad); @@ -126,7 +125,7 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, if (Mips::CPU16RegsRegClass.hasSubClassEq(RC)) Opc = Mips::LwRxSpImmX16; assert(Opc && "Register class not handled!"); - BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(0) + BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(Offset) .addMemOperand(MMO); } diff --git a/lib/Target/Mips/Mips16InstrInfo.h b/lib/Target/Mips/Mips16InstrInfo.h index 1cb1dfe196..a77a9043bb 100644 --- a/lib/Target/Mips/Mips16InstrInfo.h +++ b/lib/Target/Mips/Mips16InstrInfo.h @@ -48,17 +48,19 @@ public: unsigned DestReg, unsigned SrcReg, bool KillSrc) const; - virtual void storeRegToStackSlot(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MBBI, - unsigned SrcReg, bool isKill, int FrameIndex, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const; - - virtual void loadRegFromStackSlot(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MBBI, - unsigned DestReg, int FrameIndex, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const; + virtual void storeRegToStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + unsigned SrcReg, bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const; + + virtual void loadRegFromStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + unsigned DestReg, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const; virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const; diff --git a/lib/Target/Mips/Mips16InstrInfo.td b/lib/Target/Mips/Mips16InstrInfo.td index a9e9c52716..aa51aaf465 100644 --- a/lib/Target/Mips/Mips16InstrInfo.td +++ b/lib/Target/Mips/Mips16InstrInfo.td @@ -15,7 +15,7 @@ // Mips Address // def addr16 : - ComplexPattern<iPTR, 3, "SelectAddr16", [frameindex], [SDNPWantParent]>; + ComplexPattern<iPTR, 3, "selectAddr16", [frameindex], [SDNPWantParent]>; // // Address operand @@ -1466,14 +1466,14 @@ def: Mips16Pat<(i32 immZExt16:$in), (LiRxImmX16 immZExt16:$in)>; // MipsDivRem // def: Mips16Pat - <(MipsDivRem CPU16Regs:$rx, CPU16Regs:$ry), + <(MipsDivRem16 CPU16Regs:$rx, CPU16Regs:$ry), (DivRxRy16 CPU16Regs:$rx, CPU16Regs:$ry)>; // // MipsDivRemU // def: Mips16Pat - <(MipsDivRemU CPU16Regs:$rx, CPU16Regs:$ry), + <(MipsDivRemU16 CPU16Regs:$rx, CPU16Regs:$ry), (DivuRxRy16 CPU16Regs:$rx, CPU16Regs:$ry)>; // signed a,b diff --git a/lib/Target/Mips/Mips16RegisterInfo.cpp b/lib/Target/Mips/Mips16RegisterInfo.cpp index 0ea9368949..7ad18f2b4d 100644 --- a/lib/Target/Mips/Mips16RegisterInfo.cpp +++ b/lib/Target/Mips/Mips16RegisterInfo.cpp @@ -1,5 +1,4 @@ - -//===-- Mips16RegisterInfo.cpp - MIPS16 Register Information -== ----------===// +//===-- Mips16RegisterInfo.cpp - MIPS16 Register Information --------------===// // // The LLVM Compiler Infrastructure // @@ -72,6 +71,12 @@ bool Mips16RegisterInfo::saveScavengerRegister return true; } +const TargetRegisterClass * +Mips16RegisterInfo::intRegClass(unsigned Size) const { + assert(Size == 4); + return &Mips::CPU16RegsRegClass; +} + void Mips16RegisterInfo::eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo, int FrameIndex, uint64_t StackSize, diff --git a/lib/Target/Mips/Mips16RegisterInfo.h b/lib/Target/Mips/Mips16RegisterInfo.h index b8f818a478..2b3d2b1a4e 100644 --- a/lib/Target/Mips/Mips16RegisterInfo.h +++ b/lib/Target/Mips/Mips16RegisterInfo.h @@ -37,6 +37,8 @@ public: const TargetRegisterClass *RC, unsigned Reg) const; + virtual const TargetRegisterClass *intRegClass(unsigned Size) const; + private: virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo, int FrameIndex, uint64_t StackSize, diff --git a/lib/Target/Mips/Mips64InstrInfo.td b/lib/Target/Mips/Mips64InstrInfo.td index 494ba87b40..fc533fb03f 100644 --- a/lib/Target/Mips/Mips64InstrInfo.td +++ b/lib/Target/Mips/Mips64InstrInfo.td @@ -66,6 +66,12 @@ let usesCustomInserter = 1, Predicates = [HasStdEnc], defm ATOMIC_CMP_SWAP_I64 : AtomicCmpSwap64<atomic_cmp_swap_64>; } +/// Pseudo instructions for loading and storing accumulator registers. +let isPseudo = 1 in { + defm LOAD_AC128 : LoadM<"load_ac128", ACRegs128>; + defm STORE_AC128 : StoreM<"store_ac128", ACRegs128>; +} + //===----------------------------------------------------------------------===// // Instruction definition //===----------------------------------------------------------------------===// @@ -179,10 +185,16 @@ def DMULT : Mult<"dmult", IIImul, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1c>; def DMULTu : Mult<"dmultu", IIImul, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1d>; -def DSDIV : Div<MipsDivRem, "ddiv", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, - MULT_FM<0, 0x1e>; -def DUDIV : Div<MipsDivRemU, "ddivu", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, - MULT_FM<0, 0x1f>; +def PseudoDMULT : MultDivPseudo<DMULT, ACRegs128, CPU64RegsOpnd, MipsMult, + IIImul>; +def PseudoDMULTu : MultDivPseudo<DMULTu, ACRegs128, CPU64RegsOpnd, MipsMultu, + IIImul>; +def DSDIV : Div<"ddiv", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1e>; +def DUDIV : Div<"ddivu", IIIdiv, CPU64RegsOpnd, [HI64, LO64]>, MULT_FM<0, 0x1f>; +def PseudoDSDIV : MultDivPseudo<DSDIV, ACRegs128, CPU64RegsOpnd, MipsDivRem, + IIIdiv, 0>; +def PseudoDUDIV : MultDivPseudo<DUDIV, ACRegs128, CPU64RegsOpnd, MipsDivRemU, + IIIdiv, 0>; def MTHI64 : MoveToLOHI<"mthi", CPU64Regs, [HI64]>, MTLO_FM<0x11>; def MTLO64 : MoveToLOHI<"mtlo", CPU64Regs, [LO64]>, MTLO_FM<0x13>; @@ -306,6 +318,10 @@ def : MipsPat<(i64 (sext_inreg CPU64Regs:$src, i32)), // bswap MipsPattern def : MipsPat<(bswap CPU64Regs:$rt), (DSHD (DSBH CPU64Regs:$rt))>; +// mflo/hi patterns. +def : MipsPat<(i64 (ExtractLOHI ACRegs128:$ac, imm:$lohi_idx)), + (EXTRACT_SUBREG ACRegs128:$ac, imm:$lohi_idx)>; + //===----------------------------------------------------------------------===// // Instruction aliases //===----------------------------------------------------------------------===// @@ -313,7 +329,7 @@ def : InstAlias<"move $dst, $src", (DADDu CPU64RegsOpnd:$dst, CPU64RegsOpnd:$src, ZERO_64), 1>, Requires<[HasMips64]>; def : InstAlias<"move $dst, $src", - (OR64 CPU64RegsOpnd:$dst, CPU64RegsOpnd:$src, ZERO_64), 0>, + (OR64 CPU64RegsOpnd:$dst, CPU64RegsOpnd:$src, ZERO_64), 1>, Requires<[HasMips64]>; def : InstAlias<"and $rs, $rt, $imm", (DANDi CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, uimm16_64:$imm), @@ -332,13 +348,19 @@ def : InstAlias<"not $rt, $rs", def : InstAlias<"j $rs", (JR64 CPU64Regs:$rs), 0>, Requires<[HasMips64]>; def : InstAlias<"jalr $rs", (JALR64 RA_64, CPU64Regs:$rs)>, Requires<[HasMips64]>; +def : InstAlias<"jal $rs", (JALR64 RA_64, CPU64Regs:$rs), 0>, + Requires<[HasMips64]>; +def : InstAlias<"jal $rd,$rs", (JALR64 CPU64Regs:$rd, CPU64Regs:$rs), 0>, + Requires<[HasMips64]>; def : InstAlias<"daddu $rs, $rt, $imm", (DADDiu CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, simm16_64:$imm), 1>; def : InstAlias<"dadd $rs, $rt, $imm", (DADDi CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, simm16_64:$imm), 1>; - +def : InstAlias<"or $rs, $rt, $imm", + (ORi64 CPU64RegsOpnd:$rs, CPU64RegsOpnd:$rt, uimm16_64:$imm), + 1>, Requires<[HasMips64]>; /// Move between CPU and coprocessor registers let DecoderNamespace = "Mips64" in { diff --git a/lib/Target/Mips/MipsAsmPrinter.cpp b/lib/Target/Mips/MipsAsmPrinter.cpp index fc45fd273f..07181e5f5d 100644 --- a/lib/Target/Mips/MipsAsmPrinter.cpp +++ b/lib/Target/Mips/MipsAsmPrinter.cpp @@ -47,6 +47,10 @@ using namespace llvm; bool MipsAsmPrinter::runOnMachineFunction(MachineFunction &MF) { + // Initialize TargetLoweringObjectFile. + if (Subtarget->allowMixed16_32()) + const_cast<TargetLoweringObjectFile&>(getObjFileLowering()) + .Initialize(OutContext, TM); MipsFI = MF.getInfo<MipsFunctionInfo>(); AsmPrinter::runOnMachineFunction(MF); return true; @@ -249,12 +253,18 @@ void MipsAsmPrinter::EmitFunctionEntryLabel() { void MipsAsmPrinter::EmitFunctionBodyStart() { MCInstLowering.Initialize(Mang, &MF->getContext()); - emitFrameDirective(); + bool IsNakedFunction = + MF->getFunction()-> + getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::Naked); + if (!IsNakedFunction) + emitFrameDirective(); if (OutStreamer.hasRawTextSupport()) { SmallString<128> Str; raw_svector_ostream OS(Str); - printSavedRegsBitmask(OS); + if (!IsNakedFunction) + printSavedRegsBitmask(OS); OutStreamer.EmitRawText(OS.str()); if (!Subtarget->inMips16Mode()) { OutStreamer.EmitRawText(StringRef("\t.set\tnoreorder")); @@ -423,12 +433,18 @@ bool MipsAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNum, unsigned AsmVariant, const char *ExtraCode, raw_ostream &O) { - if (ExtraCode && ExtraCode[0]) - return true; // Unknown modifier. + int Offset = 0; + // Currently we are expecting either no ExtraCode or 'D' + if (ExtraCode) { + if (ExtraCode[0] == 'D') + Offset = 4; + else + return true; // Unknown modifier. + } const MachineOperand &MO = MI->getOperand(OpNum); assert(MO.isReg() && "unexpected inline asm memory operand"); - O << "0($" << MipsInstPrinter::getRegisterName(MO.getReg()) << ")"; + O << Offset << "($" << MipsInstPrinter::getRegisterName(MO.getReg()) << ")"; return false; } diff --git a/lib/Target/Mips/MipsCallingConv.td b/lib/Target/Mips/MipsCallingConv.td index de3a855898..3f2ceabcd5 100644 --- a/lib/Target/Mips/MipsCallingConv.td +++ b/lib/Target/Mips/MipsCallingConv.td @@ -96,6 +96,12 @@ def RetCC_MipsN : CallingConv<[ CCIfType<[f64], CCAssignToReg<[D0_64, D2_64]>> ]>; +// In soft-mode, register A0_64, instead of V1_64, is used to return a long +// double value. +def RetCC_F128Soft : CallingConv<[ + CCIfType<[i64], CCAssignToReg<[V0_64, A0_64]>> +]>; + //===----------------------------------------------------------------------===// // Mips EABI Calling Convention //===----------------------------------------------------------------------===// @@ -139,17 +145,6 @@ def RetCC_MipsEABI : CallingConv<[ ]>; //===----------------------------------------------------------------------===// -// Mips Android Calling Convention -//===----------------------------------------------------------------------===// - -def RetCC_MipsAndroid : CallingConv<[ - // f32 are returned in registers F0, F2, F1, F3 - CCIfType<[f32], CCAssignToReg<[F0, F2, F1, F3]>>, - - CCDelegateTo<RetCC_MipsO32> -]>; - -//===----------------------------------------------------------------------===// // Mips FastCC Calling Convention //===----------------------------------------------------------------------===// def CC_MipsO32_FastCC : CallingConv<[ @@ -214,7 +209,6 @@ def RetCC_Mips : CallingConv<[ CCIfSubtarget<"isABI_EABI()", CCDelegateTo<RetCC_MipsEABI>>, CCIfSubtarget<"isABI_N32()", CCDelegateTo<RetCC_MipsN>>, CCIfSubtarget<"isABI_N64()", CCDelegateTo<RetCC_MipsN>>, - CCIfSubtarget<"isAndroid()", CCDelegateTo<RetCC_MipsAndroid>>, CCDelegateTo<RetCC_MipsO32> ]>; diff --git a/lib/Target/Mips/MipsCodeEmitter.cpp b/lib/Target/Mips/MipsCodeEmitter.cpp index df877b6548..3fc402ba64 100644 --- a/lib/Target/Mips/MipsCodeEmitter.cpp +++ b/lib/Target/Mips/MipsCodeEmitter.cpp @@ -115,6 +115,10 @@ private: void emitGlobalAddressUnaligned(const GlobalValue *GV, unsigned Reloc, int Offset) const; + /// Expand pseudo instructions with accumulator register operands. + void expandACCInstr(MachineBasicBlock::instr_iterator MI, + MachineBasicBlock &MBB, unsigned Opc) const; + /// \brief Expand pseudo instruction. Return true if MI was expanded. bool expandPseudos(MachineBasicBlock::instr_iterator &MI, MachineBasicBlock &MBB) const; @@ -298,6 +302,14 @@ void MipsCodeEmitter::emitWord(unsigned Word) { MCE.emitWordBE(Word); } +void MipsCodeEmitter::expandACCInstr(MachineBasicBlock::instr_iterator MI, + MachineBasicBlock &MBB, + unsigned Opc) const { + // Expand "pseudomult $ac0, $t0, $t1" to "mult $t0, $t1". + BuildMI(MBB, &*MI, MI->getDebugLoc(), II->get(Opc)) + .addReg(MI->getOperand(1).getReg()).addReg(MI->getOperand(2).getReg()); +} + bool MipsCodeEmitter::expandPseudos(MachineBasicBlock::instr_iterator &MI, MachineBasicBlock &MBB) const { switch (MI->getOpcode()) { @@ -309,6 +321,30 @@ bool MipsCodeEmitter::expandPseudos(MachineBasicBlock::instr_iterator &MI, BuildMI(MBB, &*MI, MI->getDebugLoc(), II->get(Mips::JALR), Mips::RA) .addReg(MI->getOperand(0).getReg()); break; + case Mips::PseudoMULT: + expandACCInstr(MI, MBB, Mips::MULT); + break; + case Mips::PseudoMULTu: + expandACCInstr(MI, MBB, Mips::MULTu); + break; + case Mips::PseudoSDIV: + expandACCInstr(MI, MBB, Mips::SDIV); + break; + case Mips::PseudoUDIV: + expandACCInstr(MI, MBB, Mips::UDIV); + break; + case Mips::PseudoMADD: + expandACCInstr(MI, MBB, Mips::MADD); + break; + case Mips::PseudoMADDU: + expandACCInstr(MI, MBB, Mips::MADDU); + break; + case Mips::PseudoMSUB: + expandACCInstr(MI, MBB, Mips::MSUB); + break; + case Mips::PseudoMSUBU: + expandACCInstr(MI, MBB, Mips::MSUBU); + break; default: return false; } diff --git a/lib/Target/Mips/MipsCondMov.td b/lib/Target/Mips/MipsCondMov.td index 559370bc2a..42e4c99f05 100644 --- a/lib/Target/Mips/MipsCondMov.td +++ b/lib/Target/Mips/MipsCondMov.td @@ -58,21 +58,23 @@ multiclass MovzPats0<RegisterClass CRC, RegisterClass DRC, Instruction SLTiuOp> { def : MipsPat<(select (i32 (setge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), (MOVZInst DRC:$T, (SLTOp CRC:$lhs, CRC:$rhs), DRC:$F)>; - def : MipsPat< - (select (i32 (setuge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), - (MOVZInst DRC:$T, (SLTuOp CRC:$lhs, CRC:$rhs), DRC:$F)>; - def : MipsPat< - (select (i32 (setge CRC:$lhs, immSExt16:$rhs)), DRC:$T, DRC:$F), - (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, immSExt16:$rhs), DRC:$F)>; - def : MipsPat< - (select (i32 (setuge CRC:$lh, immSExt16:$rh)), DRC:$T, DRC:$F), - (MOVZInst DRC:$T, (SLTiuOp CRC:$lh, immSExt16:$rh), DRC:$F)>; - def : MipsPat< - (select (i32 (setle CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), - (MOVZInst DRC:$T, (SLTOp CRC:$rhs, CRC:$lhs), DRC:$F)>; - def : MipsPat< - (select (i32 (setule CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), - (MOVZInst DRC:$T, (SLTuOp CRC:$rhs, CRC:$lhs), DRC:$F)>; + def : MipsPat<(select (i32 (setuge CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTuOp CRC:$lhs, CRC:$rhs), DRC:$F)>; + def : MipsPat<(select (i32 (setge CRC:$lhs, immSExt16:$rhs)), DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, immSExt16:$rhs), DRC:$F)>; + def : MipsPat<(select (i32 (setuge CRC:$lh, immSExt16:$rh)), DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTiuOp CRC:$lh, immSExt16:$rh), DRC:$F)>; + def : MipsPat<(select (i32 (setle CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTOp CRC:$rhs, CRC:$lhs), DRC:$F)>; + def : MipsPat<(select (i32 (setule CRC:$lhs, CRC:$rhs)), DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTuOp CRC:$rhs, CRC:$lhs), DRC:$F)>; + def : MipsPat<(select (i32 (setgt CRC:$lhs, immSExt16Plus1:$rhs)), + DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTiOp CRC:$lhs, (Plus1 imm:$rhs)), DRC:$F)>; + def : MipsPat<(select (i32 (setugt CRC:$lhs, immSExt16Plus1:$rhs)), + DRC:$T, DRC:$F), + (MOVZInst DRC:$T, (SLTiuOp CRC:$lhs, (Plus1 imm:$rhs)), + DRC:$F)>; } multiclass MovzPats1<RegisterClass CRC, RegisterClass DRC, diff --git a/lib/Target/Mips/MipsConstantIslandPass.cpp b/lib/Target/Mips/MipsConstantIslandPass.cpp index b5de1ebad2..1951324cf1 100644 --- a/lib/Target/Mips/MipsConstantIslandPass.cpp +++ b/lib/Target/Mips/MipsConstantIslandPass.cpp @@ -80,6 +80,10 @@ FunctionPass *llvm::createMipsConstantIslandPass(MipsTargetMachine &tm) { } bool MipsConstantIslands::runOnMachineFunction(MachineFunction &F) { - return true; + // The intention is for this to be a mips16 only pass for now + // FIXME: + // if (!TM.getSubtarget<MipsSubtarget>().inMips16Mode()) + // return false; + return false; } diff --git a/lib/Target/Mips/MipsDSPInstrFormats.td b/lib/Target/Mips/MipsDSPInstrFormats.td index a72a763fde..cf09113cd8 100644 --- a/lib/Target/Mips/MipsDSPInstrFormats.td +++ b/lib/Target/Mips/MipsDSPInstrFormats.td @@ -219,6 +219,33 @@ class MULT_FMT<bits<6> opcode, bits<6> funct> : DSPInst { let Inst{5-0} = funct; } +// MFHI sub-class format. +class MFHI_FMT<bits<6> funct> : DSPInst { + bits<5> rd; + bits<2> ac; + + let Inst{31-26} = 0; + let Inst{25-23} = 0; + let Inst{22-21} = ac; + let Inst{20-16} = 0; + let Inst{15-11} = rd; + let Inst{10-6} = 0; + let Inst{5-0} = funct; +} + +// MTHI sub-class format. +class MTHI_FMT<bits<6> funct> : DSPInst { + bits<5> rs; + bits<2> ac; + + let Inst{31-26} = 0; + let Inst{25-21} = rs; + let Inst{20-13} = 0; + let Inst{12-11} = ac; + let Inst{10-6} = 0; + let Inst{5-0} = funct; +} + // EXTR.W sub-class format (type 1). class EXTR_W_TY1_FMT<bits<5> op> : DSPInst { bits<5> rt; diff --git a/lib/Target/Mips/MipsDSPInstrInfo.td b/lib/Target/Mips/MipsDSPInstrInfo.td index 9531b91487..c12878a952 100644 --- a/lib/Target/Mips/MipsDSPInstrInfo.td +++ b/lib/Target/Mips/MipsDSPInstrInfo.td @@ -20,17 +20,20 @@ def immZExt10 : ImmLeaf<i32, [{return isUInt<10>(Imm);}]>; def immSExt6 : ImmLeaf<i32, [{return isInt<6>(Imm);}]>; // Mips-specific dsp nodes -def SDT_MipsExtr : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>]>; -def SDT_MipsShilo : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>; -def SDT_MipsDPA : SDTypeProfile<0, 2, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>]>; +def SDT_MipsExtr : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>, + SDTCisVT<2, untyped>]>; +def SDT_MipsShilo : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>, + SDTCisSameAs<0, 2>, SDTCisVT<1, i32>]>; +def SDT_MipsDPA : SDTypeProfile<1, 3, [SDTCisVT<0, untyped>, SDTCisSameAs<0, 3>, + SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>; +def SDT_MipsSHIFT_DSP : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>, + SDTCisVT<2, i32>]>; class MipsDSPBase<string Opc, SDTypeProfile Prof> : - SDNode<!strconcat("MipsISD::", Opc), Prof, - [SDNPHasChain, SDNPInGlue, SDNPOutGlue]>; + SDNode<!strconcat("MipsISD::", Opc), Prof>; class MipsDSPSideEffectBase<string Opc, SDTypeProfile Prof> : - SDNode<!strconcat("MipsISD::", Opc), Prof, - [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPSideEffect]>; + SDNode<!strconcat("MipsISD::", Opc), Prof, [SDNPHasChain, SDNPSideEffect]>; def MipsEXTP : MipsDSPSideEffectBase<"EXTP", SDT_MipsExtr>; def MipsEXTPDP : MipsDSPSideEffectBase<"EXTPDP", SDT_MipsExtr>; @@ -40,7 +43,7 @@ def MipsEXTR_R_W : MipsDSPSideEffectBase<"EXTR_R_W", SDT_MipsExtr>; def MipsEXTR_RS_W : MipsDSPSideEffectBase<"EXTR_RS_W", SDT_MipsExtr>; def MipsSHILO : MipsDSPBase<"SHILO", SDT_MipsShilo>; -def MipsMTHLIP : MipsDSPBase<"MTHLIP", SDT_MipsShilo>; +def MipsMTHLIP : MipsDSPSideEffectBase<"MTHLIP", SDT_MipsShilo>; def MipsMULSAQ_S_W_PH : MipsDSPSideEffectBase<"MULSAQ_S_W_PH", SDT_MipsDPA>; def MipsMAQ_S_W_PHL : MipsDSPSideEffectBase<"MAQ_S_W_PHL", SDT_MipsDPA>; @@ -73,18 +76,19 @@ def MipsMADD_DSP : MipsDSPBase<"MADD_DSP", SDT_MipsDPA>; def MipsMADDU_DSP : MipsDSPBase<"MADDU_DSP", SDT_MipsDPA>; def MipsMSUB_DSP : MipsDSPBase<"MSUB_DSP", SDT_MipsDPA>; def MipsMSUBU_DSP : MipsDSPBase<"MSUBU_DSP", SDT_MipsDPA>; +def MipsSHLL_DSP : MipsDSPBase<"SHLL_DSP", SDT_MipsSHIFT_DSP>; +def MipsSHRA_DSP : MipsDSPBase<"SHRA_DSP", SDT_MipsSHIFT_DSP>; +def MipsSHRL_DSP : MipsDSPBase<"SHRL_DSP", SDT_MipsSHIFT_DSP>; +def MipsSETCC_DSP : MipsDSPBase<"SETCC_DSP", SDTSetCC>; +def MipsSELECT_CC_DSP : MipsDSPBase<"SELECT_CC_DSP", SDTSelectCC>; // Flags. -class UseAC { - list<Register> Uses = [AC0]; +class Uses<list<Register> Regs> { + list<Register> Uses = Regs; } -class UseDSPCtrl { - list<Register> Uses = [DSPCtrl]; -} - -class ClearDefs { - list<Register> Defs = []; +class Defs<list<Register> Regs> { + list<Register> Defs = Regs; } // Instruction encoding. @@ -144,6 +148,10 @@ class MAQ_S_W_PHL_ENC : DPA_W_PH_FMT<0b10100>; class MAQ_S_W_PHR_ENC : DPA_W_PH_FMT<0b10110>; class MAQ_SA_W_PHL_ENC : DPA_W_PH_FMT<0b10000>; class MAQ_SA_W_PHR_ENC : DPA_W_PH_FMT<0b10010>; +class MFHI_ENC : MFHI_FMT<0b010000>; +class MFLO_ENC : MFHI_FMT<0b010010>; +class MTHI_ENC : MTHI_FMT<0b010001>; +class MTLO_ENC : MTHI_FMT<0b010011>; class DPAU_H_QBL_ENC : DPA_W_PH_FMT<0b00011>; class DPAU_H_QBR_ENC : DPA_W_PH_FMT<0b00111>; class DPSU_H_QBL_ENC : DPA_W_PH_FMT<0b01011>; @@ -255,7 +263,6 @@ class ADDU_QB_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rs, $rt"); list<dag> Pattern = [(set RCD:$rd, (OpNode RCS:$rs, RCT:$rt))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class RADDU_W_QB_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -266,7 +273,6 @@ class RADDU_W_QB_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rs"); list<dag> Pattern = [(set RCD:$rd, (OpNode RCS:$rs))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class CMP_EQ_QB_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -277,7 +283,6 @@ class CMP_EQ_QB_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rs, $rt"); list<dag> Pattern = [(OpNode RCS:$rs, RCT:$rt)]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class CMP_EQ_QB_R3_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -288,7 +293,6 @@ class CMP_EQ_QB_R3_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rs, $rt"); list<dag> Pattern = [(set RCD:$rd, (OpNode RCS:$rs, RCT:$rt))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class PRECR_SRA_PH_W_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -299,7 +303,6 @@ class PRECR_SRA_PH_W_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rt, $rs, $sa"); list<dag> Pattern = [(set RCT:$rt, (OpNode RCS:$src, RCS:$rs, immZExt5:$sa))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; string Constraints = "$src = $rt"; } @@ -311,7 +314,6 @@ class ABSQ_S_PH_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rt"); list<dag> Pattern = [(set RCD:$rd, (OpNode RCT:$rt))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class REPL_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -321,7 +323,6 @@ class REPL_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $imm"); list<dag> Pattern = [(set RC:$rd, (OpNode immPat:$imm))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class SHLL_QB_R3_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -331,7 +332,6 @@ class SHLL_QB_R3_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rt, $rs_sa"); list<dag> Pattern = [(set RC:$rd, (OpNode RC:$rt, CPURegs:$rs_sa))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class SHLL_QB_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -342,7 +342,7 @@ class SHLL_QB_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rt, $rs_sa"); list<dag> Pattern = [(set RC:$rd, (OpNode RC:$rt, ImmPat:$rs_sa))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; + bit hasSideEffects = 1; } class LX_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -353,7 +353,6 @@ class LX_DESC_BASE<string instr_asm, SDPatternOperator OpNode, list<dag> Pattern = [(set CPURegs:$rd, (OpNode CPURegs:$base, CPURegs:$index))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; bit mayLoad = 1; } @@ -365,7 +364,6 @@ class ADDUH_QB_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $rs, $rt"); list<dag> Pattern = [(set RCD:$rd, (OpNode RCS:$rs, RCT:$rt))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class APPEND_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -376,62 +374,50 @@ class APPEND_DESC_BASE<string instr_asm, SDPatternOperator OpNode, list<dag> Pattern = [(set CPURegs:$rt, (OpNode CPURegs:$src, CPURegs:$rs, ImmOp:$sa))]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; string Constraints = "$src = $rt"; } class EXTR_W_TY1_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode, InstrItinClass itin> { dag OutOperandList = (outs CPURegs:$rt); - dag InOperandList = (ins ACRegs:$ac, CPURegs:$shift_rs); + dag InOperandList = (ins ACRegsDSP:$ac, CPURegs:$shift_rs); string AsmString = !strconcat(instr_asm, "\t$rt, $ac, $shift_rs"); InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } class EXTR_W_TY1_R1_DESC_BASE<string instr_asm, SDPatternOperator OpNode, InstrItinClass itin> { dag OutOperandList = (outs CPURegs:$rt); - dag InOperandList = (ins ACRegs:$ac, uimm16:$shift_rs); + dag InOperandList = (ins ACRegsDSP:$ac, uimm16:$shift_rs); string AsmString = !strconcat(instr_asm, "\t$rt, $ac, $shift_rs"); InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; -} - -class SHILO_R1_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin, - Instruction realinst> : - PseudoDSP<(outs), (ins simm16:$shift), [(OpNode immSExt6:$shift)]>, - PseudoInstExpansion<(realinst AC0, simm16:$shift)> { - list<Register> Defs = [DSPCtrl, AC0]; - list<Register> Uses = [AC0]; - InstrItinClass Itinerary = itin; } -class SHILO_R1_DESC_BASE<string instr_asm> { - dag OutOperandList = (outs ACRegs:$ac); - dag InOperandList = (ins simm16:$shift); +class SHILO_R1_DESC_BASE<string instr_asm, SDPatternOperator OpNode> { + dag OutOperandList = (outs ACRegsDSP:$ac); + dag InOperandList = (ins simm16:$shift, ACRegsDSP:$acin); string AsmString = !strconcat(instr_asm, "\t$ac, $shift"); + list<dag> Pattern = [(set ACRegsDSP:$ac, + (OpNode immSExt6:$shift, ACRegsDSP:$acin))]; + string Constraints = "$acin = $ac"; } -class SHILO_R2_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin, - Instruction realinst> : - PseudoDSP<(outs), (ins CPURegs:$rs), [(OpNode CPURegs:$rs)]>, - PseudoInstExpansion<(realinst AC0, CPURegs:$rs)> { - list<Register> Defs = [DSPCtrl, AC0]; - list<Register> Uses = [AC0]; - InstrItinClass Itinerary = itin; -} - -class SHILO_R2_DESC_BASE<string instr_asm> { - dag OutOperandList = (outs ACRegs:$ac); - dag InOperandList = (ins CPURegs:$rs); +class SHILO_R2_DESC_BASE<string instr_asm, SDPatternOperator OpNode> { + dag OutOperandList = (outs ACRegsDSP:$ac); + dag InOperandList = (ins CPURegs:$rs, ACRegsDSP:$acin); string AsmString = !strconcat(instr_asm, "\t$ac, $rs"); + list<dag> Pattern = [(set ACRegsDSP:$ac, + (OpNode CPURegs:$rs, ACRegsDSP:$acin))]; + string Constraints = "$acin = $ac"; } -class MTHLIP_DESC_BASE<string instr_asm> { - dag OutOperandList = (outs ACRegs:$ac); - dag InOperandList = (ins CPURegs:$rs); +class MTHLIP_DESC_BASE<string instr_asm, SDPatternOperator OpNode> { + dag OutOperandList = (outs ACRegsDSP:$ac); + dag InOperandList = (ins CPURegs:$rs, ACRegsDSP:$acin); string AsmString = !strconcat(instr_asm, "\t$rs, $ac"); + list<dag> Pattern = [(set ACRegsDSP:$ac, + (OpNode CPURegs:$rs, ACRegsDSP:$acin))]; + string Constraints = "$acin = $ac"; } class RDDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -441,7 +427,6 @@ class RDDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rd, $mask"); list<dag> Pattern = [(set CPURegs:$rd, (OpNode immZExt10:$mask))]; InstrItinClass Itinerary = itin; - list<Register> Uses = [DSPCtrl]; } class WRDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode, @@ -451,43 +436,56 @@ class WRDSP_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rs, $mask"); list<dag> Pattern = [(OpNode CPURegs:$rs, immZExt10:$mask)]; InstrItinClass Itinerary = itin; - list<Register> Defs = [DSPCtrl]; } -class DPA_W_PH_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin, - Instruction realinst> : - PseudoDSP<(outs), (ins CPURegs:$rs, CPURegs:$rt), - [(OpNode CPURegs:$rs, CPURegs:$rt)]>, - PseudoInstExpansion<(realinst AC0, CPURegs:$rs, CPURegs:$rt)> { - list<Register> Defs = [DSPCtrl, AC0]; - list<Register> Uses = [AC0]; - InstrItinClass Itinerary = itin; +class DPA_W_PH_DESC_BASE<string instr_asm, SDPatternOperator OpNode> { + dag OutOperandList = (outs ACRegsDSP:$ac); + dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin); + string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt"); + list<dag> Pattern = [(set ACRegsDSP:$ac, + (OpNode CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin))]; + string Constraints = "$acin = $ac"; } -class DPA_W_PH_DESC_BASE<string instr_asm> { - dag OutOperandList = (outs ACRegs:$ac); +class MULT_DESC_BASE<string instr_asm, SDPatternOperator OpNode, + InstrItinClass itin> { + dag OutOperandList = (outs ACRegsDSP:$ac); dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt); string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt"); + list<dag> Pattern = [(set ACRegsDSP:$ac, (OpNode CPURegs:$rs, CPURegs:$rt))]; + InstrItinClass Itinerary = itin; + int AddedComplexity = 20; + bit isCommutable = 1; } -class MULT_PSEUDO_BASE<SDPatternOperator OpNode, InstrItinClass itin, - Instruction realinst> : - PseudoDSP<(outs), (ins CPURegs:$rs, CPURegs:$rt), - [(OpNode CPURegs:$rs, CPURegs:$rt)]>, - PseudoInstExpansion<(realinst AC0, CPURegs:$rs, CPURegs:$rt)> { - list<Register> Defs = [DSPCtrl, AC0]; +class MADD_DESC_BASE<string instr_asm, SDPatternOperator OpNode, + InstrItinClass itin> { + dag OutOperandList = (outs ACRegsDSP:$ac); + dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin); + string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt"); + list<dag> Pattern = [(set ACRegsDSP:$ac, + (OpNode CPURegs:$rs, CPURegs:$rt, ACRegsDSP:$acin))]; InstrItinClass Itinerary = itin; + int AddedComplexity = 20; + string Constraints = "$acin = $ac"; } -class MULT_DESC_BASE<string instr_asm> { - dag OutOperandList = (outs ACRegs:$ac); - dag InOperandList = (ins CPURegs:$rs, CPURegs:$rt); - string AsmString = !strconcat(instr_asm, "\t$ac, $rs, $rt"); +class MFHI_DESC_BASE<string instr_asm, RegisterClass RC, InstrItinClass itin> { + dag OutOperandList = (outs CPURegs:$rd); + dag InOperandList = (ins RC:$ac); + string AsmString = !strconcat(instr_asm, "\t$rd, $ac"); + InstrItinClass Itinerary = itin; +} + +class MTHI_DESC_BASE<string instr_asm, RegisterClass RC, InstrItinClass itin> { + dag OutOperandList = (outs RC:$ac); + dag InOperandList = (ins CPURegs:$rs); + string AsmString = !strconcat(instr_asm, "\t$rs, $ac"); + InstrItinClass Itinerary = itin; } class BPOSGE32_PSEUDO_DESC_BASE<SDPatternOperator OpNode, InstrItinClass itin> : MipsPseudo<(outs CPURegs:$dst), (ins), [(set CPURegs:$dst, (OpNode))]> { - list<Register> Uses = [DSPCtrl]; bit usesCustomInserter = 1; } @@ -496,7 +494,6 @@ class BPOSGE32_DESC_BASE<string instr_asm, InstrItinClass itin> { dag InOperandList = (ins brtarget:$offset); string AsmString = !strconcat(instr_asm, "\t$offset"); InstrItinClass Itinerary = itin; - list<Register> Uses = [DSPCtrl]; bit isBranch = 1; bit isTerminator = 1; bit hasDelaySlot = 1; @@ -509,7 +506,6 @@ class INSV_DESC_BASE<string instr_asm, SDPatternOperator OpNode, string AsmString = !strconcat(instr_asm, "\t$rt, $rs"); list<dag> Pattern = [(set CPURegs:$rt, (OpNode CPURegs:$src, CPURegs:$rs))]; InstrItinClass Itinerary = itin; - list<Register> Uses = [DSPCtrl]; string Constraints = "$src = $rt"; } @@ -518,178 +514,183 @@ class INSV_DESC_BASE<string instr_asm, SDPatternOperator OpNode, //===----------------------------------------------------------------------===// // Addition/subtraction -class ADDU_QB_DESC : ADDU_QB_DESC_BASE<"addu.qb", int_mips_addu_qb, NoItinerary, - DSPRegs, DSPRegs>, IsCommutable; +class ADDU_QB_DESC : ADDU_QB_DESC_BASE<"addu.qb", null_frag, NoItinerary, + DSPRegs, DSPRegs>, IsCommutable, + Defs<[DSPOutFlag20]>; class ADDU_S_QB_DESC : ADDU_QB_DESC_BASE<"addu_s.qb", int_mips_addu_s_qb, NoItinerary, DSPRegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag20]>; -class SUBU_QB_DESC : ADDU_QB_DESC_BASE<"subu.qb", int_mips_subu_qb, NoItinerary, - DSPRegs, DSPRegs>; +class SUBU_QB_DESC : ADDU_QB_DESC_BASE<"subu.qb", null_frag, NoItinerary, + DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; class SUBU_S_QB_DESC : ADDU_QB_DESC_BASE<"subu_s.qb", int_mips_subu_s_qb, - NoItinerary, DSPRegs, DSPRegs>; + NoItinerary, DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; -class ADDQ_PH_DESC : ADDU_QB_DESC_BASE<"addq.ph", int_mips_addq_ph, NoItinerary, - DSPRegs, DSPRegs>, IsCommutable; +class ADDQ_PH_DESC : ADDU_QB_DESC_BASE<"addq.ph", null_frag, NoItinerary, + DSPRegs, DSPRegs>, IsCommutable, + Defs<[DSPOutFlag20]>; class ADDQ_S_PH_DESC : ADDU_QB_DESC_BASE<"addq_s.ph", int_mips_addq_s_ph, NoItinerary, DSPRegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag20]>; -class SUBQ_PH_DESC : ADDU_QB_DESC_BASE<"subq.ph", int_mips_subq_ph, NoItinerary, - DSPRegs, DSPRegs>; +class SUBQ_PH_DESC : ADDU_QB_DESC_BASE<"subq.ph", null_frag, NoItinerary, + DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; class SUBQ_S_PH_DESC : ADDU_QB_DESC_BASE<"subq_s.ph", int_mips_subq_s_ph, - NoItinerary, DSPRegs, DSPRegs>; + NoItinerary, DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; class ADDQ_S_W_DESC : ADDU_QB_DESC_BASE<"addq_s.w", int_mips_addq_s_w, NoItinerary, CPURegs, CPURegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag20]>; class SUBQ_S_W_DESC : ADDU_QB_DESC_BASE<"subq_s.w", int_mips_subq_s_w, - NoItinerary, CPURegs, CPURegs>; + NoItinerary, CPURegs, CPURegs>, + Defs<[DSPOutFlag20]>; -class ADDSC_DESC : ADDU_QB_DESC_BASE<"addsc", int_mips_addsc, NoItinerary, - CPURegs, CPURegs>, IsCommutable; +class ADDSC_DESC : ADDU_QB_DESC_BASE<"addsc", null_frag, NoItinerary, + CPURegs, CPURegs>, IsCommutable, + Defs<[DSPCarry]>; -class ADDWC_DESC : ADDU_QB_DESC_BASE<"addwc", int_mips_addwc, NoItinerary, +class ADDWC_DESC : ADDU_QB_DESC_BASE<"addwc", null_frag, NoItinerary, CPURegs, CPURegs>, - IsCommutable, UseDSPCtrl; + IsCommutable, Uses<[DSPCarry]>, Defs<[DSPOutFlag20]>; class MODSUB_DESC : ADDU_QB_DESC_BASE<"modsub", int_mips_modsub, NoItinerary, - CPURegs, CPURegs>, ClearDefs; + CPURegs, CPURegs>; class RADDU_W_QB_DESC : RADDU_W_QB_DESC_BASE<"raddu.w.qb", int_mips_raddu_w_qb, - NoItinerary, CPURegs, DSPRegs>, - ClearDefs; + NoItinerary, CPURegs, DSPRegs>; // Absolute value class ABSQ_S_PH_DESC : ABSQ_S_PH_R2_DESC_BASE<"absq_s.ph", int_mips_absq_s_ph, - NoItinerary, DSPRegs>; + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag20]>; class ABSQ_S_W_DESC : ABSQ_S_PH_R2_DESC_BASE<"absq_s.w", int_mips_absq_s_w, - NoItinerary, CPURegs>; + NoItinerary, CPURegs>, + Defs<[DSPOutFlag20]>; // Precision reduce/expand class PRECRQ_QB_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"precrq.qb.ph", int_mips_precrq_qb_ph, - NoItinerary, DSPRegs, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs, DSPRegs>; class PRECRQ_PH_W_DESC : CMP_EQ_QB_R3_DESC_BASE<"precrq.ph.w", int_mips_precrq_ph_w, - NoItinerary, DSPRegs, CPURegs>, - ClearDefs; + NoItinerary, DSPRegs, CPURegs>; class PRECRQ_RS_PH_W_DESC : CMP_EQ_QB_R3_DESC_BASE<"precrq_rs.ph.w", int_mips_precrq_rs_ph_w, NoItinerary, DSPRegs, - CPURegs>; + CPURegs>, + Defs<[DSPOutFlag22]>; class PRECRQU_S_QB_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"precrqu_s.qb.ph", int_mips_precrqu_s_qb_ph, NoItinerary, DSPRegs, - DSPRegs>; + DSPRegs>, + Defs<[DSPOutFlag22]>; class PRECEQ_W_PHL_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceq.w.phl", int_mips_preceq_w_phl, - NoItinerary, CPURegs, DSPRegs>, - ClearDefs; + NoItinerary, CPURegs, DSPRegs>; class PRECEQ_W_PHR_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceq.w.phr", int_mips_preceq_w_phr, - NoItinerary, CPURegs, DSPRegs>, - ClearDefs; + NoItinerary, CPURegs, DSPRegs>; class PRECEQU_PH_QBL_DESC : ABSQ_S_PH_R2_DESC_BASE<"precequ.ph.qbl", int_mips_precequ_ph_qbl, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEQU_PH_QBR_DESC : ABSQ_S_PH_R2_DESC_BASE<"precequ.ph.qbr", int_mips_precequ_ph_qbr, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEQU_PH_QBLA_DESC : ABSQ_S_PH_R2_DESC_BASE<"precequ.ph.qbla", int_mips_precequ_ph_qbla, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEQU_PH_QBRA_DESC : ABSQ_S_PH_R2_DESC_BASE<"precequ.ph.qbra", int_mips_precequ_ph_qbra, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEU_PH_QBL_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceu.ph.qbl", int_mips_preceu_ph_qbl, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEU_PH_QBR_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceu.ph.qbr", int_mips_preceu_ph_qbr, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEU_PH_QBLA_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceu.ph.qbla", int_mips_preceu_ph_qbla, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; class PRECEU_PH_QBRA_DESC : ABSQ_S_PH_R2_DESC_BASE<"preceu.ph.qbra", int_mips_preceu_ph_qbra, - NoItinerary, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs>; // Shift -class SHLL_QB_DESC : SHLL_QB_R2_DESC_BASE<"shll.qb", int_mips_shll_qb, immZExt3, - NoItinerary, DSPRegs>; +class SHLL_QB_DESC : SHLL_QB_R2_DESC_BASE<"shll.qb", null_frag, immZExt3, + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; class SHLLV_QB_DESC : SHLL_QB_R3_DESC_BASE<"shllv.qb", int_mips_shll_qb, - NoItinerary, DSPRegs>; + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; -class SHRL_QB_DESC : SHLL_QB_R2_DESC_BASE<"shrl.qb", int_mips_shrl_qb, immZExt3, - NoItinerary, DSPRegs>, ClearDefs; +class SHRL_QB_DESC : SHLL_QB_R2_DESC_BASE<"shrl.qb", null_frag, immZExt3, + NoItinerary, DSPRegs>; class SHRLV_QB_DESC : SHLL_QB_R3_DESC_BASE<"shrlv.qb", int_mips_shrl_qb, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; -class SHLL_PH_DESC : SHLL_QB_R2_DESC_BASE<"shll.ph", int_mips_shll_ph, immZExt4, - NoItinerary, DSPRegs>; +class SHLL_PH_DESC : SHLL_QB_R2_DESC_BASE<"shll.ph", null_frag, immZExt4, + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; class SHLLV_PH_DESC : SHLL_QB_R3_DESC_BASE<"shllv.ph", int_mips_shll_ph, - NoItinerary, DSPRegs>; + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; class SHLL_S_PH_DESC : SHLL_QB_R2_DESC_BASE<"shll_s.ph", int_mips_shll_s_ph, - immZExt4, NoItinerary, DSPRegs>; + immZExt4, NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; class SHLLV_S_PH_DESC : SHLL_QB_R3_DESC_BASE<"shllv_s.ph", int_mips_shll_s_ph, - NoItinerary, DSPRegs>; + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag22]>; -class SHRA_PH_DESC : SHLL_QB_R2_DESC_BASE<"shra.ph", int_mips_shra_ph, immZExt4, - NoItinerary, DSPRegs>, ClearDefs; +class SHRA_PH_DESC : SHLL_QB_R2_DESC_BASE<"shra.ph", null_frag, immZExt4, + NoItinerary, DSPRegs>; class SHRAV_PH_DESC : SHLL_QB_R3_DESC_BASE<"shrav.ph", int_mips_shra_ph, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class SHRA_R_PH_DESC : SHLL_QB_R2_DESC_BASE<"shra_r.ph", int_mips_shra_r_ph, - immZExt4, NoItinerary, DSPRegs>, - ClearDefs; + immZExt4, NoItinerary, DSPRegs>; class SHRAV_R_PH_DESC : SHLL_QB_R3_DESC_BASE<"shrav_r.ph", int_mips_shra_r_ph, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class SHLL_S_W_DESC : SHLL_QB_R2_DESC_BASE<"shll_s.w", int_mips_shll_s_w, - immZExt5, NoItinerary, CPURegs>; + immZExt5, NoItinerary, CPURegs>, + Defs<[DSPOutFlag22]>; class SHLLV_S_W_DESC : SHLL_QB_R3_DESC_BASE<"shllv_s.w", int_mips_shll_s_w, - NoItinerary, CPURegs>; + NoItinerary, CPURegs>, + Defs<[DSPOutFlag22]>; class SHRA_R_W_DESC : SHLL_QB_R2_DESC_BASE<"shra_r.w", int_mips_shra_r_w, - immZExt5, NoItinerary, CPURegs>, - ClearDefs; + immZExt5, NoItinerary, CPURegs>; class SHRAV_R_W_DESC : SHLL_QB_R3_DESC_BASE<"shrav_r.w", int_mips_shra_r_w, NoItinerary, CPURegs>; @@ -697,77 +698,91 @@ class SHRAV_R_W_DESC : SHLL_QB_R3_DESC_BASE<"shrav_r.w", int_mips_shra_r_w, // Multiplication class MULEU_S_PH_QBL_DESC : ADDU_QB_DESC_BASE<"muleu_s.ph.qbl", int_mips_muleu_s_ph_qbl, - NoItinerary, DSPRegs, DSPRegs>; + NoItinerary, DSPRegs, DSPRegs>, + Defs<[DSPOutFlag21]>; class MULEU_S_PH_QBR_DESC : ADDU_QB_DESC_BASE<"muleu_s.ph.qbr", int_mips_muleu_s_ph_qbr, - NoItinerary, DSPRegs, DSPRegs>; + NoItinerary, DSPRegs, DSPRegs>, + Defs<[DSPOutFlag21]>; class MULEQ_S_W_PHL_DESC : ADDU_QB_DESC_BASE<"muleq_s.w.phl", int_mips_muleq_s_w_phl, NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag21]>; class MULEQ_S_W_PHR_DESC : ADDU_QB_DESC_BASE<"muleq_s.w.phr", int_mips_muleq_s_w_phr, NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag21]>; class MULQ_RS_PH_DESC : ADDU_QB_DESC_BASE<"mulq_rs.ph", int_mips_mulq_rs_ph, NoItinerary, DSPRegs, DSPRegs>, - IsCommutable; - -class MULSAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsaq_s.w.ph">; - -class MAQ_S_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phl">; + IsCommutable, Defs<[DSPOutFlag21]>; -class MAQ_S_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phr">; +class MULSAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsaq_s.w.ph", + MipsMULSAQ_S_W_PH>, + Defs<[DSPOutFlag16_19]>; -class MAQ_SA_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phl">; +class MAQ_S_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phl", MipsMAQ_S_W_PHL>, + Defs<[DSPOutFlag16_19]>; -class MAQ_SA_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phr">; +class MAQ_S_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_s.w.phr", MipsMAQ_S_W_PHR>, + Defs<[DSPOutFlag16_19]>; -// Dot product with accumulate/subtract -class DPAU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbl">; - -class DPAU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbr">; - -class DPSU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbl">; +class MAQ_SA_W_PHL_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phl", MipsMAQ_SA_W_PHL>, + Defs<[DSPOutFlag16_19]>; -class DPSU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbr">; +class MAQ_SA_W_PHR_DESC : DPA_W_PH_DESC_BASE<"maq_sa.w.phr", MipsMAQ_SA_W_PHR>, + Defs<[DSPOutFlag16_19]>; -class DPAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaq_s.w.ph">; +// Move from/to hi/lo. +class MFHI_DESC : MFHI_DESC_BASE<"mfhi", HIRegsDSP, NoItinerary>; +class MFLO_DESC : MFHI_DESC_BASE<"mflo", LORegsDSP, NoItinerary>; +class MTHI_DESC : MTHI_DESC_BASE<"mthi", HIRegsDSP, NoItinerary>; +class MTLO_DESC : MTHI_DESC_BASE<"mtlo", LORegsDSP, NoItinerary>; -class DPSQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsq_s.w.ph">; +// Dot product with accumulate/subtract +class DPAU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbl", MipsDPAU_H_QBL>; -class DPAQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpaq_sa.l.w">; +class DPAU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpau.h.qbr", MipsDPAU_H_QBR>; -class DPSQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpsq_sa.l.w">; +class DPSU_H_QBL_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbl", MipsDPSU_H_QBL>; -class MULT_DSP_DESC : MULT_DESC_BASE<"mult">; +class DPSU_H_QBR_DESC : DPA_W_PH_DESC_BASE<"dpsu.h.qbr", MipsDPSU_H_QBR>; -class MULTU_DSP_DESC : MULT_DESC_BASE<"multu">; +class DPAQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaq_s.w.ph", MipsDPAQ_S_W_PH>, + Defs<[DSPOutFlag16_19]>; -class MADD_DSP_DESC : MULT_DESC_BASE<"madd">; +class DPSQ_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsq_s.w.ph", MipsDPSQ_S_W_PH>, + Defs<[DSPOutFlag16_19]>; -class MADDU_DSP_DESC : MULT_DESC_BASE<"maddu">; +class DPAQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpaq_sa.l.w", MipsDPAQ_SA_L_W>, + Defs<[DSPOutFlag16_19]>; -class MSUB_DSP_DESC : MULT_DESC_BASE<"msub">; +class DPSQ_SA_L_W_DESC : DPA_W_PH_DESC_BASE<"dpsq_sa.l.w", MipsDPSQ_SA_L_W>, + Defs<[DSPOutFlag16_19]>; -class MSUBU_DSP_DESC : MULT_DESC_BASE<"msubu">; +class MULT_DSP_DESC : MULT_DESC_BASE<"mult", MipsMult, NoItinerary>; +class MULTU_DSP_DESC : MULT_DESC_BASE<"multu", MipsMultu, NoItinerary>; +class MADD_DSP_DESC : MADD_DESC_BASE<"madd", MipsMAdd, NoItinerary>; +class MADDU_DSP_DESC : MADD_DESC_BASE<"maddu", MipsMAddu, NoItinerary>; +class MSUB_DSP_DESC : MADD_DESC_BASE<"msub", MipsMSub, NoItinerary>; +class MSUBU_DSP_DESC : MADD_DESC_BASE<"msubu", MipsMSubu, NoItinerary>; // Comparison class CMPU_EQ_QB_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmpu.eq.qb", int_mips_cmpu_eq_qb, NoItinerary, - DSPRegs>, IsCommutable; + DSPRegs>, + IsCommutable, Defs<[DSPCCond]>; class CMPU_LT_QB_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmpu.lt.qb", int_mips_cmpu_lt_qb, NoItinerary, - DSPRegs>, IsCommutable; + DSPRegs>, Defs<[DSPCCond]>; class CMPU_LE_QB_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmpu.le.qb", int_mips_cmpu_le_qb, NoItinerary, - DSPRegs>, IsCommutable; + DSPRegs>, Defs<[DSPCCond]>; class CMPGU_EQ_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgu.eq.qb", int_mips_cmpgu_eq_qb, @@ -776,222 +791,237 @@ class CMPGU_EQ_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgu.eq.qb", class CMPGU_LT_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgu.lt.qb", int_mips_cmpgu_lt_qb, - NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + NoItinerary, CPURegs, DSPRegs>; class CMPGU_LE_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgu.le.qb", int_mips_cmpgu_le_qb, - NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + NoItinerary, CPURegs, DSPRegs>; class CMP_EQ_PH_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmp.eq.ph", int_mips_cmp_eq_ph, NoItinerary, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPCCond]>; class CMP_LT_PH_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmp.lt.ph", int_mips_cmp_lt_ph, NoItinerary, DSPRegs>, - IsCommutable; + Defs<[DSPCCond]>; class CMP_LE_PH_DESC : CMP_EQ_QB_R2_DESC_BASE<"cmp.le.ph", int_mips_cmp_le_ph, NoItinerary, DSPRegs>, - IsCommutable; + Defs<[DSPCCond]>; // Misc class BITREV_DESC : ABSQ_S_PH_R2_DESC_BASE<"bitrev", int_mips_bitrev, - NoItinerary, CPURegs>, ClearDefs; + NoItinerary, CPURegs>; class PACKRL_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"packrl.ph", int_mips_packrl_ph, - NoItinerary, DSPRegs, DSPRegs>, - ClearDefs; + NoItinerary, DSPRegs, DSPRegs>; class REPL_QB_DESC : REPL_DESC_BASE<"repl.qb", int_mips_repl_qb, immZExt8, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class REPL_PH_DESC : REPL_DESC_BASE<"repl.ph", int_mips_repl_ph, immZExt10, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class REPLV_QB_DESC : ABSQ_S_PH_R2_DESC_BASE<"replv.qb", int_mips_repl_qb, - NoItinerary, DSPRegs, CPURegs>, - ClearDefs; + NoItinerary, DSPRegs, CPURegs>; class REPLV_PH_DESC : ABSQ_S_PH_R2_DESC_BASE<"replv.ph", int_mips_repl_ph, - NoItinerary, DSPRegs, CPURegs>, - ClearDefs; + NoItinerary, DSPRegs, CPURegs>; class PICK_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"pick.qb", int_mips_pick_qb, NoItinerary, DSPRegs, DSPRegs>, - ClearDefs, UseDSPCtrl; + Uses<[DSPCCond]>; class PICK_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"pick.ph", int_mips_pick_ph, NoItinerary, DSPRegs, DSPRegs>, - ClearDefs, UseDSPCtrl; + Uses<[DSPCCond]>; -class LWX_DESC : LX_DESC_BASE<"lwx", int_mips_lwx, NoItinerary>, ClearDefs; +class LWX_DESC : LX_DESC_BASE<"lwx", int_mips_lwx, NoItinerary>; -class LHX_DESC : LX_DESC_BASE<"lhx", int_mips_lhx, NoItinerary>, ClearDefs; +class LHX_DESC : LX_DESC_BASE<"lhx", int_mips_lhx, NoItinerary>; -class LBUX_DESC : LX_DESC_BASE<"lbux", int_mips_lbux, NoItinerary>, ClearDefs; +class LBUX_DESC : LX_DESC_BASE<"lbux", int_mips_lbux, NoItinerary>; class BPOSGE32_DESC : BPOSGE32_DESC_BASE<"bposge32", NoItinerary>; // Extr -class EXTP_DESC : EXTR_W_TY1_R1_DESC_BASE<"extp", MipsEXTP, NoItinerary>; +class EXTP_DESC : EXTR_W_TY1_R1_DESC_BASE<"extp", MipsEXTP, NoItinerary>, + Uses<[DSPPos]>, Defs<[DSPEFI]>; -class EXTPV_DESC : EXTR_W_TY1_R2_DESC_BASE<"extpv", MipsEXTP, NoItinerary>; +class EXTPV_DESC : EXTR_W_TY1_R2_DESC_BASE<"extpv", MipsEXTP, NoItinerary>, + Uses<[DSPPos]>, Defs<[DSPEFI]>; -class EXTPDP_DESC : EXTR_W_TY1_R1_DESC_BASE<"extpdp", MipsEXTPDP, NoItinerary>; +class EXTPDP_DESC : EXTR_W_TY1_R1_DESC_BASE<"extpdp", MipsEXTPDP, NoItinerary>, + Uses<[DSPPos]>, Defs<[DSPPos, DSPEFI]>; class EXTPDPV_DESC : EXTR_W_TY1_R2_DESC_BASE<"extpdpv", MipsEXTPDP, - NoItinerary>; + NoItinerary>, + Uses<[DSPPos]>, Defs<[DSPPos, DSPEFI]>; -class EXTR_W_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr.w", MipsEXTR_W, NoItinerary>; +class EXTR_W_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr.w", MipsEXTR_W, NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTRV_W_DESC : EXTR_W_TY1_R2_DESC_BASE<"extrv.w", MipsEXTR_W, - NoItinerary>; + NoItinerary>, Defs<[DSPOutFlag23]>; class EXTR_R_W_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr_r.w", MipsEXTR_R_W, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTRV_R_W_DESC : EXTR_W_TY1_R2_DESC_BASE<"extrv_r.w", MipsEXTR_R_W, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTR_RS_W_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr_rs.w", MipsEXTR_RS_W, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTRV_RS_W_DESC : EXTR_W_TY1_R2_DESC_BASE<"extrv_rs.w", MipsEXTR_RS_W, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTR_S_H_DESC : EXTR_W_TY1_R1_DESC_BASE<"extr_s.h", MipsEXTR_S_H, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; class EXTRV_S_H_DESC : EXTR_W_TY1_R2_DESC_BASE<"extrv_s.h", MipsEXTR_S_H, - NoItinerary>; + NoItinerary>, + Defs<[DSPOutFlag23]>; -class SHILO_DESC : SHILO_R1_DESC_BASE<"shilo">; +class SHILO_DESC : SHILO_R1_DESC_BASE<"shilo", MipsSHILO>; -class SHILOV_DESC : SHILO_R2_DESC_BASE<"shilov">; +class SHILOV_DESC : SHILO_R2_DESC_BASE<"shilov", MipsSHILO>; -class MTHLIP_DESC : MTHLIP_DESC_BASE<"mthlip">; +class MTHLIP_DESC : MTHLIP_DESC_BASE<"mthlip", MipsMTHLIP>, Defs<[DSPPos]>; class RDDSP_DESC : RDDSP_DESC_BASE<"rddsp", int_mips_rddsp, NoItinerary>; class WRDSP_DESC : WRDSP_DESC_BASE<"wrdsp", int_mips_wrdsp, NoItinerary>; -class INSV_DESC : INSV_DESC_BASE<"insv", int_mips_insv, NoItinerary>; +class INSV_DESC : INSV_DESC_BASE<"insv", int_mips_insv, NoItinerary>, + Uses<[DSPPos, DSPSCount]>; //===----------------------------------------------------------------------===// // MIPS DSP Rev 2 // Addition/subtraction class ADDU_PH_DESC : ADDU_QB_DESC_BASE<"addu.ph", int_mips_addu_ph, NoItinerary, - DSPRegs, DSPRegs>, IsCommutable; + DSPRegs, DSPRegs>, IsCommutable, + Defs<[DSPOutFlag20]>; class ADDU_S_PH_DESC : ADDU_QB_DESC_BASE<"addu_s.ph", int_mips_addu_s_ph, NoItinerary, DSPRegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag20]>; class SUBU_PH_DESC : ADDU_QB_DESC_BASE<"subu.ph", int_mips_subu_ph, NoItinerary, - DSPRegs, DSPRegs>; + DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; class SUBU_S_PH_DESC : ADDU_QB_DESC_BASE<"subu_s.ph", int_mips_subu_s_ph, - NoItinerary, DSPRegs, DSPRegs>; + NoItinerary, DSPRegs, DSPRegs>, + Defs<[DSPOutFlag20]>; class ADDUH_QB_DESC : ADDUH_QB_DESC_BASE<"adduh.qb", int_mips_adduh_qb, - NoItinerary, DSPRegs>, - ClearDefs, IsCommutable; + NoItinerary, DSPRegs>, IsCommutable; class ADDUH_R_QB_DESC : ADDUH_QB_DESC_BASE<"adduh_r.qb", int_mips_adduh_r_qb, - NoItinerary, DSPRegs>, - ClearDefs, IsCommutable; + NoItinerary, DSPRegs>, IsCommutable; class SUBUH_QB_DESC : ADDUH_QB_DESC_BASE<"subuh.qb", int_mips_subuh_qb, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class SUBUH_R_QB_DESC : ADDUH_QB_DESC_BASE<"subuh_r.qb", int_mips_subuh_r_qb, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class ADDQH_PH_DESC : ADDUH_QB_DESC_BASE<"addqh.ph", int_mips_addqh_ph, - NoItinerary, DSPRegs>, - ClearDefs, IsCommutable; + NoItinerary, DSPRegs>, IsCommutable; class ADDQH_R_PH_DESC : ADDUH_QB_DESC_BASE<"addqh_r.ph", int_mips_addqh_r_ph, - NoItinerary, DSPRegs>, - ClearDefs, IsCommutable; + NoItinerary, DSPRegs>, IsCommutable; class SUBQH_PH_DESC : ADDUH_QB_DESC_BASE<"subqh.ph", int_mips_subqh_ph, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class SUBQH_R_PH_DESC : ADDUH_QB_DESC_BASE<"subqh_r.ph", int_mips_subqh_r_ph, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class ADDQH_W_DESC : ADDUH_QB_DESC_BASE<"addqh.w", int_mips_addqh_w, - NoItinerary, CPURegs>, - ClearDefs, IsCommutable; + NoItinerary, CPURegs>, IsCommutable; class ADDQH_R_W_DESC : ADDUH_QB_DESC_BASE<"addqh_r.w", int_mips_addqh_r_w, - NoItinerary, CPURegs>, - ClearDefs, IsCommutable; + NoItinerary, CPURegs>, IsCommutable; class SUBQH_W_DESC : ADDUH_QB_DESC_BASE<"subqh.w", int_mips_subqh_w, - NoItinerary, CPURegs>, ClearDefs; + NoItinerary, CPURegs>; class SUBQH_R_W_DESC : ADDUH_QB_DESC_BASE<"subqh_r.w", int_mips_subqh_r_w, - NoItinerary, CPURegs>, ClearDefs; + NoItinerary, CPURegs>; // Comparison class CMPGDU_EQ_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgdu.eq.qb", int_mips_cmpgdu_eq_qb, NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPCCond]>; class CMPGDU_LT_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgdu.lt.qb", int_mips_cmpgdu_lt_qb, NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + Defs<[DSPCCond]>; class CMPGDU_LE_QB_DESC : CMP_EQ_QB_R3_DESC_BASE<"cmpgdu.le.qb", int_mips_cmpgdu_le_qb, NoItinerary, CPURegs, DSPRegs>, - IsCommutable; + Defs<[DSPCCond]>; // Absolute class ABSQ_S_QB_DESC : ABSQ_S_PH_R2_DESC_BASE<"absq_s.qb", int_mips_absq_s_qb, - NoItinerary, DSPRegs>; + NoItinerary, DSPRegs>, + Defs<[DSPOutFlag20]>; // Multiplication -class MUL_PH_DESC : ADDUH_QB_DESC_BASE<"mul.ph", int_mips_mul_ph, NoItinerary, - DSPRegs>, IsCommutable; +class MUL_PH_DESC : ADDUH_QB_DESC_BASE<"mul.ph", null_frag, NoItinerary, + DSPRegs>, IsCommutable, + Defs<[DSPOutFlag21]>; class MUL_S_PH_DESC : ADDUH_QB_DESC_BASE<"mul_s.ph", int_mips_mul_s_ph, - NoItinerary, DSPRegs>, IsCommutable; + NoItinerary, DSPRegs>, IsCommutable, + Defs<[DSPOutFlag21]>; class MULQ_S_W_DESC : ADDUH_QB_DESC_BASE<"mulq_s.w", int_mips_mulq_s_w, - NoItinerary, CPURegs>, IsCommutable; + NoItinerary, CPURegs>, IsCommutable, + Defs<[DSPOutFlag21]>; class MULQ_RS_W_DESC : ADDUH_QB_DESC_BASE<"mulq_rs.w", int_mips_mulq_rs_w, - NoItinerary, CPURegs>, IsCommutable; + NoItinerary, CPURegs>, IsCommutable, + Defs<[DSPOutFlag21]>; class MULQ_S_PH_DESC : ADDU_QB_DESC_BASE<"mulq_s.ph", int_mips_mulq_s_ph, NoItinerary, DSPRegs, DSPRegs>, - IsCommutable; + IsCommutable, Defs<[DSPOutFlag21]>; // Dot product with accumulate/subtract -class DPA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpa.w.ph">; +class DPA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpa.w.ph", MipsDPA_W_PH>; -class DPS_W_PH_DESC : DPA_W_PH_DESC_BASE<"dps.w.ph">; +class DPS_W_PH_DESC : DPA_W_PH_DESC_BASE<"dps.w.ph", MipsDPS_W_PH>; -class DPAQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_s.w.ph">; +class DPAQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_s.w.ph", MipsDPAQX_S_W_PH>, + Defs<[DSPOutFlag16_19]>; -class DPAQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_sa.w.ph">; +class DPAQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpaqx_sa.w.ph", + MipsDPAQX_SA_W_PH>, + Defs<[DSPOutFlag16_19]>; -class DPAX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpax.w.ph">; +class DPAX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpax.w.ph", MipsDPAX_W_PH>; -class DPSX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsx.w.ph">; +class DPSX_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsx.w.ph", MipsDPSX_W_PH>; -class DPSQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_s.w.ph">; +class DPSQX_S_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_s.w.ph", MipsDPSQX_S_W_PH>, + Defs<[DSPOutFlag16_19]>; -class DPSQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_sa.w.ph">; +class DPSQX_SA_W_PH_DESC : DPA_W_PH_DESC_BASE<"dpsqx_sa.w.ph", + MipsDPSQX_SA_W_PH>, + Defs<[DSPOutFlag16_19]>; -class MULSA_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsa.w.ph">; +class MULSA_W_PH_DESC : DPA_W_PH_DESC_BASE<"mulsa.w.ph", MipsMULSA_W_PH>; // Precision reduce/expand class PRECR_QB_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"precr.qb.ph", @@ -1001,45 +1031,45 @@ class PRECR_QB_PH_DESC : CMP_EQ_QB_R3_DESC_BASE<"precr.qb.ph", class PRECR_SRA_PH_W_DESC : PRECR_SRA_PH_W_DESC_BASE<"precr_sra.ph.w", int_mips_precr_sra_ph_w, NoItinerary, DSPRegs, - CPURegs>, ClearDefs; + CPURegs>; class PRECR_SRA_R_PH_W_DESC : PRECR_SRA_PH_W_DESC_BASE<"precr_sra_r.ph.w", int_mips_precr_sra_r_ph_w, NoItinerary, DSPRegs, - CPURegs>, ClearDefs; + CPURegs>; // Shift -class SHRA_QB_DESC : SHLL_QB_R2_DESC_BASE<"shra.qb", int_mips_shra_qb, immZExt3, - NoItinerary, DSPRegs>, ClearDefs; +class SHRA_QB_DESC : SHLL_QB_R2_DESC_BASE<"shra.qb", null_frag, immZExt3, + NoItinerary, DSPRegs>; class SHRAV_QB_DESC : SHLL_QB_R3_DESC_BASE<"shrav.qb", int_mips_shra_qb, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; class SHRA_R_QB_DESC : SHLL_QB_R2_DESC_BASE<"shra_r.qb", int_mips_shra_r_qb, - immZExt3, NoItinerary, DSPRegs>, - ClearDefs; + immZExt3, NoItinerary, DSPRegs>; class SHRAV_R_QB_DESC : SHLL_QB_R3_DESC_BASE<"shrav_r.qb", int_mips_shra_r_qb, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; -class SHRL_PH_DESC : SHLL_QB_R2_DESC_BASE<"shrl.ph", int_mips_shrl_ph, immZExt4, - NoItinerary, DSPRegs>, ClearDefs; +class SHRL_PH_DESC : SHLL_QB_R2_DESC_BASE<"shrl.ph", null_frag, immZExt4, + NoItinerary, DSPRegs>; class SHRLV_PH_DESC : SHLL_QB_R3_DESC_BASE<"shrlv.ph", int_mips_shrl_ph, - NoItinerary, DSPRegs>, ClearDefs; + NoItinerary, DSPRegs>; // Misc class APPEND_DESC : APPEND_DESC_BASE<"append", int_mips_append, immZExt5, - NoItinerary>, ClearDefs; + NoItinerary>; class BALIGN_DESC : APPEND_DESC_BASE<"balign", int_mips_balign, immZExt2, - NoItinerary>, ClearDefs; + NoItinerary>; class PREPEND_DESC : APPEND_DESC_BASE<"prepend", int_mips_prepend, immZExt5, - NoItinerary>, ClearDefs; + NoItinerary>; // Pseudos. -def BPOSGE32_PSEUDO : BPOSGE32_PSEUDO_DESC_BASE<int_mips_bposge32, NoItinerary>; +def BPOSGE32_PSEUDO : BPOSGE32_PSEUDO_DESC_BASE<int_mips_bposge32, + NoItinerary>, Uses<[DSPPos]>; // Instruction defs. // MIPS DSP Rev 1 @@ -1099,6 +1129,10 @@ def MAQ_S_W_PHL : MAQ_S_W_PHL_ENC, MAQ_S_W_PHL_DESC; def MAQ_S_W_PHR : MAQ_S_W_PHR_ENC, MAQ_S_W_PHR_DESC; def MAQ_SA_W_PHL : MAQ_SA_W_PHL_ENC, MAQ_SA_W_PHL_DESC; def MAQ_SA_W_PHR : MAQ_SA_W_PHR_ENC, MAQ_SA_W_PHR_DESC; +def MFHI_DSP : MFHI_ENC, MFHI_DESC; +def MFLO_DSP : MFLO_ENC, MFLO_DESC; +def MTHI_DSP : MTHI_ENC, MTHI_DESC; +def MTLO_DSP : MTLO_ENC, MTLO_DESC; def DPAU_H_QBL : DPAU_H_QBL_ENC, DPAU_H_QBL_DESC; def DPAU_H_QBR : DPAU_H_QBR_ENC, DPAU_H_QBR_DESC; def DPSU_H_QBL : DPSU_H_QBL_ENC, DPSU_H_QBL_DESC; @@ -1206,70 +1240,35 @@ def PREPEND : PREPEND_ENC, PREPEND_DESC; } // Pseudos. -def MULSAQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMULSAQ_S_W_PH, NoItinerary, - MULSAQ_S_W_PH>; -def MAQ_S_W_PHL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_S_W_PHL, NoItinerary, - MAQ_S_W_PHL>; -def MAQ_S_W_PHR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_S_W_PHR, NoItinerary, - MAQ_S_W_PHR>; -def MAQ_SA_W_PHL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_SA_W_PHL, NoItinerary, - MAQ_SA_W_PHL>; -def MAQ_SA_W_PHR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMAQ_SA_W_PHR, NoItinerary, - MAQ_SA_W_PHR>; -def DPAU_H_QBL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAU_H_QBL, NoItinerary, - DPAU_H_QBL>; -def DPAU_H_QBR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAU_H_QBR, NoItinerary, - DPAU_H_QBR>; -def DPSU_H_QBL_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSU_H_QBL, NoItinerary, - DPSU_H_QBL>; -def DPSU_H_QBR_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSU_H_QBR, NoItinerary, - DPSU_H_QBR>; -def DPAQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQ_S_W_PH, NoItinerary, - DPAQ_S_W_PH>; -def DPSQ_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQ_S_W_PH, NoItinerary, - DPSQ_S_W_PH>; -def DPAQ_SA_L_W_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQ_SA_L_W, NoItinerary, - DPAQ_SA_L_W>; -def DPSQ_SA_L_W_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQ_SA_L_W, NoItinerary, - DPSQ_SA_L_W>; - -def MULT_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMULT, NoItinerary, MULT_DSP>, - IsCommutable; -def MULTU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMULTU, NoItinerary, MULTU_DSP>, - IsCommutable; -def MADD_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMADD_DSP, NoItinerary, MADD_DSP>, - IsCommutable, UseAC; -def MADDU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMADDU_DSP, NoItinerary, MADDU_DSP>, - IsCommutable, UseAC; -def MSUB_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMSUB_DSP, NoItinerary, MSUB_DSP>, - UseAC; -def MSUBU_DSP_PSEUDO : MULT_PSEUDO_BASE<MipsMSUBU_DSP, NoItinerary, MSUBU_DSP>, - UseAC; - -def SHILO_PSEUDO : SHILO_R1_PSEUDO_BASE<MipsSHILO, NoItinerary, SHILO>; -def SHILOV_PSEUDO : SHILO_R2_PSEUDO_BASE<MipsSHILO, NoItinerary, SHILOV>; -def MTHLIP_PSEUDO : SHILO_R2_PSEUDO_BASE<MipsMTHLIP, NoItinerary, MTHLIP>; +let isPseudo = 1 in { + // Pseudo instructions for loading and storing accumulator registers. + defm LOAD_AC_DSP : LoadM<"load_ac_dsp", ACRegsDSP>; + defm STORE_AC_DSP : StoreM<"store_ac_dsp", ACRegsDSP>; + + // Pseudos for loading and storing ccond field of DSP control register. + defm LOAD_CCOND_DSP : LoadM<"load_ccond_dsp", DSPCC>; + defm STORE_CCOND_DSP : StoreM<"store_ccond_dsp", DSPCC>; +} -let Predicates = [HasDSPR2] in { +// Pseudo CMP and PICK instructions. +class PseudoCMP<Instruction RealInst> : + PseudoDSP<(outs DSPCC:$cmp), (ins DSPRegs:$rs, DSPRegs:$rt), []>, + PseudoInstExpansion<(RealInst DSPRegs:$rs, DSPRegs:$rt)>, NeverHasSideEffects; -def DPA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPA_W_PH, NoItinerary, DPA_W_PH>; -def DPS_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPS_W_PH, NoItinerary, DPS_W_PH>; -def DPAQX_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQX_S_W_PH, NoItinerary, - DPAQX_S_W_PH>; -def DPAQX_SA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAQX_SA_W_PH, NoItinerary, - DPAQX_SA_W_PH>; -def DPAX_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPAX_W_PH, NoItinerary, - DPAX_W_PH>; -def DPSX_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSX_W_PH, NoItinerary, - DPSX_W_PH>; -def DPSQX_S_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQX_S_W_PH, NoItinerary, - DPSQX_S_W_PH>; -def DPSQX_SA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsDPSQX_SA_W_PH, NoItinerary, - DPSQX_SA_W_PH>; -def MULSA_W_PH_PSEUDO : DPA_W_PH_PSEUDO_BASE<MipsMULSA_W_PH, NoItinerary, - MULSA_W_PH>; +class PseudoPICK<Instruction RealInst> : + PseudoDSP<(outs DSPRegs:$rd), (ins DSPCC:$cmp, DSPRegs:$rs, DSPRegs:$rt), []>, + PseudoInstExpansion<(RealInst DSPRegs:$rd, DSPRegs:$rs, DSPRegs:$rt)>, + NeverHasSideEffects; -} +def PseudoCMP_EQ_PH : PseudoCMP<CMP_EQ_PH>; +def PseudoCMP_LT_PH : PseudoCMP<CMP_LT_PH>; +def PseudoCMP_LE_PH : PseudoCMP<CMP_LE_PH>; +def PseudoCMPU_EQ_QB : PseudoCMP<CMPU_EQ_QB>; +def PseudoCMPU_LT_QB : PseudoCMP<CMPU_LT_QB>; +def PseudoCMPU_LE_QB : PseudoCMP<CMPU_LE_QB>; + +def PseudoPICK_PH : PseudoPICK<PICK_PH>; +def PseudoPICK_QB : PseudoPICK<PICK_QB>; // Patterns. class DSPPat<dag pattern, dag result, Predicate pred = HasDSP> : @@ -1294,12 +1293,103 @@ def : DSPPat<(store (v2i16 DSPRegs:$val), addr:$a), def : DSPPat<(store (v4i8 DSPRegs:$val), addr:$a), (SW (COPY_TO_REGCLASS DSPRegs:$val, CPURegs), addr:$a)>; +// Binary operations. +class DSPBinPat<Instruction Inst, ValueType ValTy, SDPatternOperator Node, + Predicate Pred = HasDSP> : + DSPPat<(Node ValTy:$a, ValTy:$b), (Inst ValTy:$a, ValTy:$b), Pred>; + +def : DSPBinPat<ADDQ_PH, v2i16, int_mips_addq_ph>; +def : DSPBinPat<ADDQ_PH, v2i16, add>; +def : DSPBinPat<SUBQ_PH, v2i16, int_mips_subq_ph>; +def : DSPBinPat<SUBQ_PH, v2i16, sub>; +def : DSPBinPat<MUL_PH, v2i16, int_mips_mul_ph, HasDSPR2>; +def : DSPBinPat<MUL_PH, v2i16, mul, HasDSPR2>; +def : DSPBinPat<ADDU_QB, v4i8, int_mips_addu_qb>; +def : DSPBinPat<ADDU_QB, v4i8, add>; +def : DSPBinPat<SUBU_QB, v4i8, int_mips_subu_qb>; +def : DSPBinPat<SUBU_QB, v4i8, sub>; +def : DSPBinPat<ADDSC, i32, int_mips_addsc>; +def : DSPBinPat<ADDSC, i32, addc>; +def : DSPBinPat<ADDWC, i32, int_mips_addwc>; +def : DSPBinPat<ADDWC, i32, adde>; + +// Shift immediate patterns. +class DSPShiftPat<Instruction Inst, ValueType ValTy, SDPatternOperator Node, + SDPatternOperator Imm, Predicate Pred = HasDSP> : + DSPPat<(Node ValTy:$a, Imm:$shamt), (Inst ValTy:$a, Imm:$shamt), Pred>; + +def : DSPShiftPat<SHLL_PH, v2i16, MipsSHLL_DSP, imm>; +def : DSPShiftPat<SHRA_PH, v2i16, MipsSHRA_DSP, imm>; +def : DSPShiftPat<SHRL_PH, v2i16, MipsSHRL_DSP, imm, HasDSPR2>; +def : DSPShiftPat<SHLL_PH, v2i16, int_mips_shll_ph, immZExt4>; +def : DSPShiftPat<SHRA_PH, v2i16, int_mips_shra_ph, immZExt4>; +def : DSPShiftPat<SHRL_PH, v2i16, int_mips_shrl_ph, immZExt4, HasDSPR2>; +def : DSPShiftPat<SHLL_QB, v4i8, MipsSHLL_DSP, imm>; +def : DSPShiftPat<SHRA_QB, v4i8, MipsSHRA_DSP, imm, HasDSPR2>; +def : DSPShiftPat<SHRL_QB, v4i8, MipsSHRL_DSP, imm>; +def : DSPShiftPat<SHLL_QB, v4i8, int_mips_shll_qb, immZExt3>; +def : DSPShiftPat<SHRA_QB, v4i8, int_mips_shra_qb, immZExt3, HasDSPR2>; +def : DSPShiftPat<SHRL_QB, v4i8, int_mips_shrl_qb, immZExt3>; + +// SETCC/SELECT_CC patterns. +class DSPSetCCPat<Instruction Cmp, Instruction Pick, ValueType ValTy, + CondCode CC> : + DSPPat<(ValTy (MipsSETCC_DSP ValTy:$a, ValTy:$b, CC)), + (ValTy (Pick (ValTy (Cmp ValTy:$a, ValTy:$b)), + (ValTy (COPY_TO_REGCLASS (ADDiu ZERO, -1), DSPRegs)), + (ValTy ZERO)))>; + +class DSPSetCCPatInv<Instruction Cmp, Instruction Pick, ValueType ValTy, + CondCode CC> : + DSPPat<(ValTy (MipsSETCC_DSP ValTy:$a, ValTy:$b, CC)), + (ValTy (Pick (ValTy (Cmp ValTy:$a, ValTy:$b)), + (ValTy ZERO), + (ValTy (COPY_TO_REGCLASS (ADDiu ZERO, -1), DSPRegs))))>; + +class DSPSelectCCPat<Instruction Cmp, Instruction Pick, ValueType ValTy, + CondCode CC> : + DSPPat<(ValTy (MipsSELECT_CC_DSP ValTy:$a, ValTy:$b, ValTy:$c, ValTy:$d, CC)), + (ValTy (Pick (ValTy (Cmp ValTy:$a, ValTy:$b)), $c, $d))>; + +class DSPSelectCCPatInv<Instruction Cmp, Instruction Pick, ValueType ValTy, + CondCode CC> : + DSPPat<(ValTy (MipsSELECT_CC_DSP ValTy:$a, ValTy:$b, ValTy:$c, ValTy:$d, CC)), + (ValTy (Pick (ValTy (Cmp ValTy:$a, ValTy:$b)), $d, $c))>; + +def : DSPSetCCPat<PseudoCMP_EQ_PH, PseudoPICK_PH, v2i16, SETEQ>; +def : DSPSetCCPat<PseudoCMP_LT_PH, PseudoPICK_PH, v2i16, SETLT>; +def : DSPSetCCPat<PseudoCMP_LE_PH, PseudoPICK_PH, v2i16, SETLE>; +def : DSPSetCCPatInv<PseudoCMP_EQ_PH, PseudoPICK_PH, v2i16, SETNE>; +def : DSPSetCCPatInv<PseudoCMP_LT_PH, PseudoPICK_PH, v2i16, SETGE>; +def : DSPSetCCPatInv<PseudoCMP_LE_PH, PseudoPICK_PH, v2i16, SETGT>; +def : DSPSetCCPat<PseudoCMPU_EQ_QB, PseudoPICK_QB, v4i8, SETEQ>; +def : DSPSetCCPat<PseudoCMPU_LT_QB, PseudoPICK_QB, v4i8, SETULT>; +def : DSPSetCCPat<PseudoCMPU_LE_QB, PseudoPICK_QB, v4i8, SETULE>; +def : DSPSetCCPatInv<PseudoCMPU_EQ_QB, PseudoPICK_QB, v4i8, SETNE>; +def : DSPSetCCPatInv<PseudoCMPU_LT_QB, PseudoPICK_QB, v4i8, SETUGE>; +def : DSPSetCCPatInv<PseudoCMPU_LE_QB, PseudoPICK_QB, v4i8, SETUGT>; + +def : DSPSelectCCPat<PseudoCMP_EQ_PH, PseudoPICK_PH, v2i16, SETEQ>; +def : DSPSelectCCPat<PseudoCMP_LT_PH, PseudoPICK_PH, v2i16, SETLT>; +def : DSPSelectCCPat<PseudoCMP_LE_PH, PseudoPICK_PH, v2i16, SETLE>; +def : DSPSelectCCPatInv<PseudoCMP_EQ_PH, PseudoPICK_PH, v2i16, SETNE>; +def : DSPSelectCCPatInv<PseudoCMP_LT_PH, PseudoPICK_PH, v2i16, SETGE>; +def : DSPSelectCCPatInv<PseudoCMP_LE_PH, PseudoPICK_PH, v2i16, SETGT>; +def : DSPSelectCCPat<PseudoCMPU_EQ_QB, PseudoPICK_QB, v4i8, SETEQ>; +def : DSPSelectCCPat<PseudoCMPU_LT_QB, PseudoPICK_QB, v4i8, SETULT>; +def : DSPSelectCCPat<PseudoCMPU_LE_QB, PseudoPICK_QB, v4i8, SETULE>; +def : DSPSelectCCPatInv<PseudoCMPU_EQ_QB, PseudoPICK_QB, v4i8, SETNE>; +def : DSPSelectCCPatInv<PseudoCMPU_LT_QB, PseudoPICK_QB, v4i8, SETUGE>; +def : DSPSelectCCPatInv<PseudoCMPU_LE_QB, PseudoPICK_QB, v4i8, SETUGT>; + // Extr patterns. class EXTR_W_TY1_R2_Pat<SDPatternOperator OpNode, Instruction Instr> : - DSPPat<(i32 (OpNode CPURegs:$rs)), (Instr AC0, CPURegs:$rs)>; + DSPPat<(i32 (OpNode CPURegs:$rs, ACRegsDSP:$ac)), + (Instr ACRegsDSP:$ac, CPURegs:$rs)>; class EXTR_W_TY1_R1_Pat<SDPatternOperator OpNode, Instruction Instr> : - DSPPat<(i32 (OpNode immZExt5:$shift)), (Instr AC0, immZExt5:$shift)>; + DSPPat<(i32 (OpNode immZExt5:$shift, ACRegsDSP:$ac)), + (Instr ACRegsDSP:$ac, immZExt5:$shift)>; def : EXTR_W_TY1_R1_Pat<MipsEXTP, EXTP>; def : EXTR_W_TY1_R2_Pat<MipsEXTP, EXTPV>; @@ -1313,3 +1403,19 @@ def : EXTR_W_TY1_R1_Pat<MipsEXTR_RS_W, EXTR_RS_W>; def : EXTR_W_TY1_R2_Pat<MipsEXTR_RS_W, EXTRV_RS_W>; def : EXTR_W_TY1_R1_Pat<MipsEXTR_S_H, EXTR_S_H>; def : EXTR_W_TY1_R2_Pat<MipsEXTR_S_H, EXTRV_S_H>; + +// mflo/hi patterns. +let AddedComplexity = 20 in +def : DSPPat<(i32 (ExtractLOHI ACRegsDSP:$ac, imm:$lohi_idx)), + (EXTRACT_SUBREG ACRegsDSP:$ac, imm:$lohi_idx)>; + +// Indexed load patterns. +class IndexedLoadPat<SDPatternOperator LoadNode, Instruction Instr> : + DSPPat<(i32 (LoadNode (add i32:$base, i32:$index))), + (Instr i32:$base, i32:$index)>; + +let AddedComplexity = 20 in { + def : IndexedLoadPat<zextloadi8, LBUX>; + def : IndexedLoadPat<sextloadi16, LHX>; + def : IndexedLoadPat<load, LWX>; +} diff --git a/lib/Target/Mips/MipsDelaySlotFiller.cpp b/lib/Target/Mips/MipsDelaySlotFiller.cpp index 49475d1740..cf350b5b97 100644 --- a/lib/Target/Mips/MipsDelaySlotFiller.cpp +++ b/lib/Target/Mips/MipsDelaySlotFiller.cpp @@ -14,11 +14,17 @@ #define DEBUG_TYPE "delay-slot-filler" #include "Mips.h" +#include "MipsInstrInfo.h" #include "MipsTargetMachine.h" #include "llvm/ADT/BitVector.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/ValueTracking.h" +#include "llvm/CodeGen/MachineBranchProbabilityInfo.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/PseudoSourceValue.h" #include "llvm/Support/CommandLine.h" #include "llvm/Target/TargetInstrInfo.h" #include "llvm/Target/TargetMachine.h" @@ -36,18 +42,59 @@ static cl::opt<bool> DisableDelaySlotFiller( cl::desc("Fill all delay slots with NOPs."), cl::Hidden); -// This option can be used to silence complaints by machine verifier passes. -static cl::opt<bool> SkipDelaySlotFiller( - "skip-mips-delay-filler", +static cl::opt<bool> DisableForwardSearch( + "disable-mips-df-forward-search", + cl::init(true), + cl::desc("Disallow MIPS delay filler to search forward."), + cl::Hidden); + +static cl::opt<bool> DisableSuccBBSearch( + "disable-mips-df-succbb-search", + cl::init(true), + cl::desc("Disallow MIPS delay filler to search successor basic blocks."), + cl::Hidden); + +static cl::opt<bool> DisableBackwardSearch( + "disable-mips-df-backward-search", cl::init(false), - cl::desc("Skip MIPS' delay slot filling pass."), + cl::desc("Disallow MIPS delay filler to search backward."), cl::Hidden); namespace { + typedef MachineBasicBlock::iterator Iter; + typedef MachineBasicBlock::reverse_iterator ReverseIter; + typedef SmallDenseMap<MachineBasicBlock*, MachineInstr*, 2> BB2BrMap; + + /// \brief A functor comparing edge weight of two blocks. + struct CmpWeight { + CmpWeight(const MachineBasicBlock &S, + const MachineBranchProbabilityInfo &P) : Src(S), Prob(P) {} + + bool operator()(const MachineBasicBlock *Dst0, + const MachineBasicBlock *Dst1) const { + return Prob.getEdgeWeight(&Src, Dst0) < Prob.getEdgeWeight(&Src, Dst1); + } + + const MachineBasicBlock &Src; + const MachineBranchProbabilityInfo &Prob; + }; + class RegDefsUses { public: RegDefsUses(TargetMachine &TM); void init(const MachineInstr &MI); + + /// This function sets all caller-saved registers in Defs. + void setCallerSaved(const MachineInstr &MI); + + /// This function sets all unallocatable registers in Defs. + void setUnallocatableRegs(const MachineFunction &MF); + + /// Set bits in Uses corresponding to MBB's live-out registers except for + /// the registers that are live-in to SuccBB. + void addLiveOut(const MachineBasicBlock &MBB, + const MachineBasicBlock &SuccBB); + bool update(const MachineInstr &MI, unsigned Begin, unsigned End); private: @@ -61,6 +108,72 @@ namespace { BitVector Defs, Uses; }; + /// Base class for inspecting loads and stores. + class InspectMemInstr { + public: + InspectMemInstr(bool ForbidMemInstr_) + : OrigSeenLoad(false), OrigSeenStore(false), SeenLoad(false), + SeenStore(false), ForbidMemInstr(ForbidMemInstr_) {} + + /// Return true if MI cannot be moved to delay slot. + bool hasHazard(const MachineInstr &MI); + + virtual ~InspectMemInstr() {} + + protected: + /// Flags indicating whether loads or stores have been seen. + bool OrigSeenLoad, OrigSeenStore, SeenLoad, SeenStore; + + /// Memory instructions are not allowed to move to delay slot if this flag + /// is true. + bool ForbidMemInstr; + + private: + virtual bool hasHazard_(const MachineInstr &MI) = 0; + }; + + /// This subclass rejects any memory instructions. + class NoMemInstr : public InspectMemInstr { + public: + NoMemInstr() : InspectMemInstr(true) {} + private: + virtual bool hasHazard_(const MachineInstr &MI) { return true; } + }; + + /// This subclass accepts loads from stacks and constant loads. + class LoadFromStackOrConst : public InspectMemInstr { + public: + LoadFromStackOrConst() : InspectMemInstr(false) {} + private: + virtual bool hasHazard_(const MachineInstr &MI); + }; + + /// This subclass uses memory dependence information to determine whether a + /// memory instruction can be moved to a delay slot. + class MemDefsUses : public InspectMemInstr { + public: + MemDefsUses(const MachineFrameInfo *MFI); + + private: + virtual bool hasHazard_(const MachineInstr &MI); + + /// Update Defs and Uses. Return true if there exist dependences that + /// disqualify the delay slot candidate between V and values in Uses and + /// Defs. + bool updateDefsUses(const Value *V, bool MayStore); + + /// Get the list of underlying objects of MI's memory operand. + bool getUnderlyingObjects(const MachineInstr &MI, + SmallVectorImpl<const Value *> &Objects) const; + + const MachineFrameInfo *MFI; + SmallPtrSet<const Value*, 4> Uses, Defs; + + /// Flags indicating whether loads or stores with no underlying objects have + /// been seen. + bool SeenNoObjLoad, SeenNoObjStore; + }; + class Filler : public MachineFunctionPass { public: Filler(TargetMachine &tm) @@ -81,19 +194,54 @@ namespace { return Changed; } - private: - typedef MachineBasicBlock::iterator Iter; - typedef MachineBasicBlock::reverse_iterator ReverseIter; + void getAnalysisUsage(AnalysisUsage &AU) const { + AU.addRequired<MachineBranchProbabilityInfo>(); + MachineFunctionPass::getAnalysisUsage(AU); + } + private: bool runOnMachineBasicBlock(MachineBasicBlock &MBB); /// This function checks if it is valid to move Candidate to the delay slot - /// and returns true if it isn't. It also updates load and store flags and - /// register defs and uses. - bool delayHasHazard(const MachineInstr &Candidate, bool &SawLoad, - bool &SawStore, RegDefsUses &RegDU) const; - - bool findDelayInstr(MachineBasicBlock &MBB, Iter slot, Iter &Filler) const; + /// and returns true if it isn't. It also updates memory and register + /// dependence information. + bool delayHasHazard(const MachineInstr &Candidate, RegDefsUses &RegDU, + InspectMemInstr &IM) const; + + /// This function searches range [Begin, End) for an instruction that can be + /// moved to the delay slot. Returns true on success. + template<typename IterTy> + bool searchRange(MachineBasicBlock &MBB, IterTy Begin, IterTy End, + RegDefsUses &RegDU, InspectMemInstr &IM, + IterTy &Filler) const; + + /// This function searches in the backward direction for an instruction that + /// can be moved to the delay slot. Returns true on success. + bool searchBackward(MachineBasicBlock &MBB, Iter Slot) const; + + /// This function searches MBB in the forward direction for an instruction + /// that can be moved to the delay slot. Returns true on success. + bool searchForward(MachineBasicBlock &MBB, Iter Slot) const; + + /// This function searches one of MBB's successor blocks for an instruction + /// that can be moved to the delay slot and inserts clones of the + /// instruction into the successor's predecessor blocks. + bool searchSuccBBs(MachineBasicBlock &MBB, Iter Slot) const; + + /// Pick a successor block of MBB. Return NULL if MBB doesn't have a + /// successor block that is not a landing pad. + MachineBasicBlock *selectSuccBB(MachineBasicBlock &B) const; + + /// This function analyzes MBB and returns an instruction with an unoccupied + /// slot that branches to Dst. + std::pair<MipsInstrInfo::BranchType, MachineInstr *> + getBranch(MachineBasicBlock &MBB, const MachineBasicBlock &Dst) const; + + /// Examine Pred and see if it is possible to insert an instruction into + /// one of its branches delay slot or its end. + bool examinePred(MachineBasicBlock &Pred, const MachineBasicBlock &Succ, + RegDefsUses &RegDU, bool &HasMultipleSuccs, + BB2BrMap &BrMap) const; bool terminateSearch(const MachineInstr &Candidate) const; @@ -105,6 +253,45 @@ namespace { char Filler::ID = 0; } // end of anonymous namespace +static bool hasUnoccupiedSlot(const MachineInstr *MI) { + return MI->hasDelaySlot() && !MI->isBundledWithSucc(); +} + +/// This function inserts clones of Filler into predecessor blocks. +static void insertDelayFiller(Iter Filler, const BB2BrMap &BrMap) { + MachineFunction *MF = Filler->getParent()->getParent(); + + for (BB2BrMap::const_iterator I = BrMap.begin(); I != BrMap.end(); ++I) { + if (I->second) { + MIBundleBuilder(I->second).append(MF->CloneMachineInstr(&*Filler)); + ++UsefulSlots; + } else { + I->first->insert(I->first->end(), MF->CloneMachineInstr(&*Filler)); + } + } +} + +/// This function adds registers Filler defines to MBB's live-in register list. +static void addLiveInRegs(Iter Filler, MachineBasicBlock &MBB) { + for (unsigned I = 0, E = Filler->getNumOperands(); I != E; ++I) { + const MachineOperand &MO = Filler->getOperand(I); + unsigned R; + + if (!MO.isReg() || !MO.isDef() || !(R = MO.getReg())) + continue; + +#ifndef NDEBUG + const MachineFunction &MF = *MBB.getParent(); + assert(MF.getTarget().getRegisterInfo()->getAllocatableSet(MF).test(R) && + "Shouldn't move an instruction with unallocatable registers across " + "basic block boundaries."); +#endif + + if (!MBB.isLiveIn(R)) + MBB.addLiveIn(R); + } +} + RegDefsUses::RegDefsUses(TargetMachine &TM) : TRI(*TM.getRegisterInfo()), Defs(TRI.getNumRegs(), false), Uses(TRI.getNumRegs(), false) {} @@ -126,6 +313,45 @@ void RegDefsUses::init(const MachineInstr &MI) { } } +void RegDefsUses::setCallerSaved(const MachineInstr &MI) { + assert(MI.isCall()); + + // If MI is a call, add all caller-saved registers to Defs. + BitVector CallerSavedRegs(TRI.getNumRegs(), true); + + CallerSavedRegs.reset(Mips::ZERO); + CallerSavedRegs.reset(Mips::ZERO_64); + + for (const MCPhysReg *R = TRI.getCalleeSavedRegs(); *R; ++R) + for (MCRegAliasIterator AI(*R, &TRI, true); AI.isValid(); ++AI) + CallerSavedRegs.reset(*AI); + + Defs |= CallerSavedRegs; +} + +void RegDefsUses::setUnallocatableRegs(const MachineFunction &MF) { + BitVector AllocSet = TRI.getAllocatableSet(MF); + + for (int R = AllocSet.find_first(); R != -1; R = AllocSet.find_next(R)) + for (MCRegAliasIterator AI(R, &TRI, false); AI.isValid(); ++AI) + AllocSet.set(*AI); + + AllocSet.set(Mips::ZERO); + AllocSet.set(Mips::ZERO_64); + + Defs |= AllocSet.flip(); +} + +void RegDefsUses::addLiveOut(const MachineBasicBlock &MBB, + const MachineBasicBlock &SuccBB) { + for (MachineBasicBlock::const_succ_iterator SI = MBB.succ_begin(), + SE = MBB.succ_end(); SI != SE; ++SI) + if (*SI != &SuccBB) + for (MachineBasicBlock::livein_iterator LI = (*SI)->livein_begin(), + LE = (*SI)->livein_end(); LI != LE; ++LI) + Uses.set(*LI); +} + bool RegDefsUses::update(const MachineInstr &MI, unsigned Begin, unsigned End) { BitVector NewDefs(TRI.getNumRegs()), NewUses(TRI.getNumRegs()); bool HasHazard = false; @@ -164,28 +390,134 @@ bool RegDefsUses::isRegInSet(const BitVector &RegSet, unsigned Reg) const { return false; } +bool InspectMemInstr::hasHazard(const MachineInstr &MI) { + if (!MI.mayStore() && !MI.mayLoad()) + return false; + + if (ForbidMemInstr) + return true; + + OrigSeenLoad = SeenLoad; + OrigSeenStore = SeenStore; + SeenLoad |= MI.mayLoad(); + SeenStore |= MI.mayStore(); + + // If MI is an ordered or volatile memory reference, disallow moving + // subsequent loads and stores to delay slot. + if (MI.hasOrderedMemoryRef() && (OrigSeenLoad || OrigSeenStore)) { + ForbidMemInstr = true; + return true; + } + + return hasHazard_(MI); +} + +bool LoadFromStackOrConst::hasHazard_(const MachineInstr &MI) { + if (MI.mayStore()) + return true; + + if (!MI.hasOneMemOperand() || !(*MI.memoperands_begin())->getValue()) + return true; + + const Value *V = (*MI.memoperands_begin())->getValue(); + + if (isa<FixedStackPseudoSourceValue>(V)) + return false; + + if (const PseudoSourceValue *PSV = dyn_cast<const PseudoSourceValue>(V)) + return !PSV->PseudoSourceValue::isConstant(0) && + (V != PseudoSourceValue::getStack()); + + return true; +} + +MemDefsUses::MemDefsUses(const MachineFrameInfo *MFI_) + : InspectMemInstr(false), MFI(MFI_), SeenNoObjLoad(false), + SeenNoObjStore(false) {} + +bool MemDefsUses::hasHazard_(const MachineInstr &MI) { + bool HasHazard = false; + SmallVector<const Value *, 4> Objs; + + // Check underlying object list. + if (getUnderlyingObjects(MI, Objs)) { + for (SmallVector<const Value *, 4>::const_iterator I = Objs.begin(); + I != Objs.end(); ++I) + HasHazard |= updateDefsUses(*I, MI.mayStore()); + + return HasHazard; + } + + // No underlying objects found. + HasHazard = MI.mayStore() && (OrigSeenLoad || OrigSeenStore); + HasHazard |= MI.mayLoad() || OrigSeenStore; + + SeenNoObjLoad |= MI.mayLoad(); + SeenNoObjStore |= MI.mayStore(); + + return HasHazard; +} + +bool MemDefsUses::updateDefsUses(const Value *V, bool MayStore) { + if (MayStore) + return !Defs.insert(V) || Uses.count(V) || SeenNoObjStore || SeenNoObjLoad; + + Uses.insert(V); + return Defs.count(V) || SeenNoObjStore; +} + +bool MemDefsUses:: +getUnderlyingObjects(const MachineInstr &MI, + SmallVectorImpl<const Value *> &Objects) const { + if (!MI.hasOneMemOperand() || !(*MI.memoperands_begin())->getValue()) + return false; + + const Value *V = (*MI.memoperands_begin())->getValue(); + + SmallVector<Value *, 4> Objs; + GetUnderlyingObjects(const_cast<Value *>(V), Objs); + + for (SmallVector<Value*, 4>::iterator I = Objs.begin(), E = Objs.end(); + I != E; ++I) { + if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(*I)) { + if (PSV->isAliased(MFI)) + return false; + } else if (!isIdentifiedObject(V)) + return false; + + Objects.push_back(*I); + } + + return true; +} + /// runOnMachineBasicBlock - Fill in delay slots for the given basic block. /// We assume there is only one delay slot per delayed instruction. bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) { bool Changed = false; for (Iter I = MBB.begin(); I != MBB.end(); ++I) { - if (!I->hasDelaySlot()) + if (!hasUnoccupiedSlot(&*I)) continue; ++FilledSlots; Changed = true; - Iter D; // Delay slot filling is disabled at -O0. - if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None) && - findDelayInstr(MBB, I, D)) { - MBB.splice(llvm::next(I), &MBB, D); - ++UsefulSlots; - } else - BuildMI(MBB, llvm::next(I), I->getDebugLoc(), TII->get(Mips::NOP)); + if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None)) { + if (searchBackward(MBB, I)) + continue; + + if (I->isTerminator()) { + if (searchSuccBBs(MBB, I)) + continue; + } else if (searchForward(MBB, I)) { + continue; + } + } - // Bundle the delay slot filler to the instruction with the delay slot. + // Bundle the NOP to the instruction with the delay slot. + BuildMI(MBB, llvm::next(I), I->getDebugLoc(), TII->get(Mips::NOP)); MIBundleBuilder(MBB, I, llvm::next(llvm::next(I))); } @@ -198,70 +530,188 @@ FunctionPass *llvm::createMipsDelaySlotFillerPass(MipsTargetMachine &tm) { return new Filler(tm); } -// @LOCALMOD-START -extern bool IsDangerousLoad(const MachineInstr &MI, int *AddrIdx); -extern bool IsDangerousStore(const MachineInstr &MI, int *AddrIdx); -// @LOCALMOD-END +template<typename IterTy> +bool Filler::searchRange(MachineBasicBlock &MBB, IterTy Begin, IterTy End, + RegDefsUses &RegDU, InspectMemInstr& IM, + IterTy &Filler) const { + for (IterTy I = Begin; I != End; ++I) { + // skip debug value + if (I->isDebugValue()) + continue; + + if (terminateSearch(*I)) + break; + + assert((!I->isCall() && !I->isReturn() && !I->isBranch()) && + "Cannot put calls, returns or branches in delay slot."); + + if (delayHasHazard(*I, RegDU, IM)) + continue; + + Filler = I; + return true; + } + + return false; +} + +bool Filler::searchBackward(MachineBasicBlock &MBB, Iter Slot) const { + if (DisableBackwardSearch) + return false; -bool Filler::findDelayInstr(MachineBasicBlock &MBB, Iter Slot, - Iter &Filler) const { RegDefsUses RegDU(TM); + MemDefsUses MemDU(MBB.getParent()->getFrameInfo()); + ReverseIter Filler; RegDU.init(*Slot); - bool SawLoad = false; - bool SawStore = false; + if (searchRange(MBB, ReverseIter(Slot), MBB.rend(), RegDU, MemDU, Filler)) { + MBB.splice(llvm::next(Slot), &MBB, llvm::next(Filler).base()); + MIBundleBuilder(MBB, Slot, llvm::next(llvm::next(Slot))); + ++UsefulSlots; + return true; + } - for (ReverseIter I(Slot); I != MBB.rend(); ++I) { - // skip debug value - if (I->isDebugValue()) - continue; + return false; +} - // @LOCALMOD-START - Don't put in delay slot instructions that could be masked. - // - // Should not allow: - // ERET, DERET or WAIT, PAUSE. Need to add these to instruction - // list. TBD. - if (Triple(TM.getTargetTriple()).isOSNaCl()) { - int Dummy; - Iter FI(llvm::next(I).base()); - if (terminateSearch(*I) || (IsDangerousLoad(*FI, &Dummy) - || IsDangerousStore(*FI, &Dummy) - || FI->modifiesRegister(Mips::SP, TM.getRegisterInfo()))) - break; - } else { - if (terminateSearch(*I)) - break; - } - // @LOCALMOD-END +bool Filler::searchForward(MachineBasicBlock &MBB, Iter Slot) const { + // Can handle only calls. + if (DisableForwardSearch || !Slot->isCall()) + return false; - if (delayHasHazard(*I, SawLoad, SawStore, RegDU)) - continue; + RegDefsUses RegDU(TM); + NoMemInstr NM; + Iter Filler; + + RegDU.setCallerSaved(*Slot); - Filler = llvm::next(I).base(); + if (searchRange(MBB, llvm::next(Slot), MBB.end(), RegDU, NM, Filler)) { + MBB.splice(llvm::next(Slot), &MBB, Filler); + MIBundleBuilder(MBB, Slot, llvm::next(llvm::next(Slot))); + ++UsefulSlots; return true; } return false; } -bool Filler::delayHasHazard(const MachineInstr &Candidate, bool &SawLoad, - bool &SawStore, RegDefsUses &RegDU) const { - bool HasHazard = (Candidate.isImplicitDef() || Candidate.isKill()); +bool Filler::searchSuccBBs(MachineBasicBlock &MBB, Iter Slot) const { + if (DisableSuccBBSearch) + return false; + + MachineBasicBlock *SuccBB = selectSuccBB(MBB); + + if (!SuccBB) + return false; + + RegDefsUses RegDU(TM); + bool HasMultipleSuccs = false; + BB2BrMap BrMap; + OwningPtr<InspectMemInstr> IM; + Iter Filler; + + // Iterate over SuccBB's predecessor list. + for (MachineBasicBlock::pred_iterator PI = SuccBB->pred_begin(), + PE = SuccBB->pred_end(); PI != PE; ++PI) + if (!examinePred(**PI, *SuccBB, RegDU, HasMultipleSuccs, BrMap)) + return false; + + // Do not allow moving instructions which have unallocatable register operands + // across basic block boundaries. + RegDU.setUnallocatableRegs(*MBB.getParent()); + + // Only allow moving loads from stack or constants if any of the SuccBB's + // predecessors have multiple successors. + if (HasMultipleSuccs) { + IM.reset(new LoadFromStackOrConst()); + } else { + const MachineFrameInfo *MFI = MBB.getParent()->getFrameInfo(); + IM.reset(new MemDefsUses(MFI)); + } + + if (!searchRange(MBB, SuccBB->begin(), SuccBB->end(), RegDU, *IM, Filler)) + return false; - // Loads or stores cannot be moved past a store to the delay slot - // and stores cannot be moved past a load. - if (Candidate.mayStore() || Candidate.hasOrderedMemoryRef()) { - HasHazard |= SawStore | SawLoad; - SawStore = true; - } else if (Candidate.mayLoad()) { - HasHazard |= SawStore; - SawLoad = true; + insertDelayFiller(Filler, BrMap); + addLiveInRegs(Filler, *SuccBB); + Filler->eraseFromParent(); + + return true; +} + +MachineBasicBlock *Filler::selectSuccBB(MachineBasicBlock &B) const { + if (B.succ_empty()) + return NULL; + + // Select the successor with the larget edge weight. + CmpWeight Cmp(B, getAnalysis<MachineBranchProbabilityInfo>()); + MachineBasicBlock *S = *std::max_element(B.succ_begin(), B.succ_end(), Cmp); + return S->isLandingPad() ? NULL : S; +} + +std::pair<MipsInstrInfo::BranchType, MachineInstr *> +Filler::getBranch(MachineBasicBlock &MBB, const MachineBasicBlock &Dst) const { + const MipsInstrInfo *TII = + static_cast<const MipsInstrInfo*>(TM.getInstrInfo()); + MachineBasicBlock *TrueBB = 0, *FalseBB = 0; + SmallVector<MachineInstr*, 2> BranchInstrs; + SmallVector<MachineOperand, 2> Cond; + + MipsInstrInfo::BranchType R = + TII->AnalyzeBranch(MBB, TrueBB, FalseBB, Cond, false, BranchInstrs); + + if ((R == MipsInstrInfo::BT_None) || (R == MipsInstrInfo::BT_NoBranch)) + return std::make_pair(R, (MachineInstr*)NULL); + + if (R != MipsInstrInfo::BT_CondUncond) { + if (!hasUnoccupiedSlot(BranchInstrs[0])) + return std::make_pair(MipsInstrInfo::BT_None, (MachineInstr*)NULL); + + assert(((R != MipsInstrInfo::BT_Uncond) || (TrueBB == &Dst))); + + return std::make_pair(R, BranchInstrs[0]); } - assert((!Candidate.isCall() && !Candidate.isReturn()) && - "Cannot put calls or returns in delay slot."); + assert((TrueBB == &Dst) || (FalseBB == &Dst)); + + // Examine the conditional branch. See if its slot is occupied. + if (hasUnoccupiedSlot(BranchInstrs[0])) + return std::make_pair(MipsInstrInfo::BT_Cond, BranchInstrs[0]); + + // If that fails, try the unconditional branch. + if (hasUnoccupiedSlot(BranchInstrs[1]) && (FalseBB == &Dst)) + return std::make_pair(MipsInstrInfo::BT_Uncond, BranchInstrs[1]); + + return std::make_pair(MipsInstrInfo::BT_None, (MachineInstr*)NULL); +} + +bool Filler::examinePred(MachineBasicBlock &Pred, const MachineBasicBlock &Succ, + RegDefsUses &RegDU, bool &HasMultipleSuccs, + BB2BrMap &BrMap) const { + std::pair<MipsInstrInfo::BranchType, MachineInstr *> P = + getBranch(Pred, Succ); + + // Return if either getBranch wasn't able to analyze the branches or there + // were no branches with unoccupied slots. + if (P.first == MipsInstrInfo::BT_None) + return false; + + if ((P.first != MipsInstrInfo::BT_Uncond) && + (P.first != MipsInstrInfo::BT_NoBranch)) { + HasMultipleSuccs = true; + RegDU.addLiveOut(Pred, Succ); + } + + BrMap[&Pred] = P.second; + return true; +} + +bool Filler::delayHasHazard(const MachineInstr &Candidate, RegDefsUses &RegDU, + InspectMemInstr &IM) const { + bool HasHazard = (Candidate.isImplicitDef() || Candidate.isKill()); + HasHazard |= IM.hasHazard(Candidate); HasHazard |= RegDU.update(Candidate, 0, Candidate.getNumOperands()); return HasHazard; diff --git a/lib/Target/Mips/MipsFrameLowering.h b/lib/Target/Mips/MipsFrameLowering.h index df52d92da8..6a5f79d0df 100644 --- a/lib/Target/Mips/MipsFrameLowering.h +++ b/lib/Target/Mips/MipsFrameLowering.h @@ -26,9 +26,8 @@ protected: const MipsSubtarget &STI; public: - explicit MipsFrameLowering(const MipsSubtarget &sti) - : TargetFrameLowering(StackGrowsDown, sti.hasMips64() ? 16 : 8, 0, - sti.hasMips64() ? 16 : 8), STI(sti) {} + explicit MipsFrameLowering(const MipsSubtarget &sti, unsigned Alignment) + : TargetFrameLowering(StackGrowsDown, Alignment, 0, Alignment), STI(sti) {} static const MipsFrameLowering *create(MipsTargetMachine &TM, const MipsSubtarget &ST); @@ -39,7 +38,7 @@ protected: uint64_t estimateStackSize(const MachineFunction &MF) const; }; -/// Create MipsInstrInfo objects. +/// Create MipsFrameLowering objects. const MipsFrameLowering *createMips16FrameLowering(const MipsSubtarget &ST); const MipsFrameLowering *createMipsSEFrameLowering(const MipsSubtarget &ST); diff --git a/lib/Target/Mips/MipsISelDAGToDAG.cpp b/lib/Target/Mips/MipsISelDAGToDAG.cpp index 78c74ef879..968e536484 100644 --- a/lib/Target/Mips/MipsISelDAGToDAG.cpp +++ b/lib/Target/Mips/MipsISelDAGToDAG.cpp @@ -12,19 +12,18 @@ //===----------------------------------------------------------------------===// #define DEBUG_TYPE "mips-isel" +#include "MipsISelDAGToDAG.h" +#include "Mips16ISelDAGToDAG.h" +#include "MipsSEISelDAGToDAG.h" #include "Mips.h" #include "MCTargetDesc/MipsBaseInfo.h" -#include "MipsAnalyzeImmediate.h" #include "MipsMachineFunction.h" #include "MipsRegisterInfo.h" -#include "MipsSubtarget.h" -#include "MipsTargetMachine.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineRegisterInfo.h" -#include "llvm/CodeGen/SelectionDAGISel.h" #include "llvm/CodeGen/SelectionDAGNodes.h" #include "llvm/IR/GlobalValue.h" #include "llvm/IR/Instructions.h" @@ -45,270 +44,11 @@ using namespace llvm; // MipsDAGToDAGISel - MIPS specific code to select MIPS machine // instructions for SelectionDAG operations. //===----------------------------------------------------------------------===// -namespace { - -class MipsDAGToDAGISel : public SelectionDAGISel { - - /// TM - Keep a reference to MipsTargetMachine. - MipsTargetMachine &TM; - - /// Subtarget - Keep a pointer to the MipsSubtarget around so that we can - /// make the right decision when generating code for different targets. - const MipsSubtarget &Subtarget; - -public: - explicit MipsDAGToDAGISel(MipsTargetMachine &tm) : - SelectionDAGISel(tm), - TM(tm), Subtarget(tm.getSubtarget<MipsSubtarget>()) {} - - // Pass Name - virtual const char *getPassName() const { - return "MIPS DAG->DAG Pattern Instruction Selection"; - } - - virtual bool runOnMachineFunction(MachineFunction &MF); - -private: - // Include the pieces autogenerated from the target description. - #include "MipsGenDAGISel.inc" - - /// getTargetMachine - Return a reference to the TargetMachine, casted - /// to the target-specific type. - const MipsTargetMachine &getTargetMachine() { - return static_cast<const MipsTargetMachine &>(TM); - } - - /// getInstrInfo - Return a reference to the TargetInstrInfo, casted - /// to the target-specific type. - const MipsInstrInfo *getInstrInfo() { - return getTargetMachine().getInstrInfo(); - } - - SDNode *getGlobalBaseReg(); - - SDValue getMips16SPAliasReg(); - - void getMips16SPRefReg(SDNode *parent, SDValue &AliasReg); - - std::pair<SDNode*, SDNode*> SelectMULT(SDNode *N, unsigned Opc, DebugLoc dl, - EVT Ty, bool HasLo, bool HasHi); - - SDNode *Select(SDNode *N); - - // Complex Pattern. - /// (reg + imm). - bool selectAddrRegImm(SDValue Addr, SDValue &Base, SDValue &Offset) const; - - /// Fall back on this function if all else fails. - bool selectAddrDefault(SDValue Addr, SDValue &Base, SDValue &Offset) const; - - /// Match integer address pattern. - bool selectIntAddr(SDValue Addr, SDValue &Base, SDValue &Offset) const; - - bool SelectAddr16(SDNode *Parent, SDValue N, SDValue &Base, SDValue &Offset, - SDValue &Alias); - - // getImm - Return a target constant with the specified value. - inline SDValue getImm(const SDNode *Node, unsigned Imm) { - return CurDAG->getTargetConstant(Imm, Node->getValueType(0)); - } - - void ProcessFunctionAfterISel(MachineFunction &MF); - bool ReplaceUsesWithZeroReg(MachineRegisterInfo *MRI, const MachineInstr&); - void InitGlobalBaseReg(MachineFunction &MF); - void InitMips16SPAliasReg(MachineFunction &MF); - - virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op, - char ConstraintCode, - std::vector<SDValue> &OutOps); -}; - -} - -// Insert instructions to initialize the global base register in the -// first MBB of the function. When the ABI is O32 and the relocation model is -// PIC, the necessary instructions are emitted later to prevent optimization -// passes from moving them. -void MipsDAGToDAGISel::InitGlobalBaseReg(MachineFunction &MF) { - MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); - - if (!MipsFI->globalBaseRegSet()) - return; - - MachineBasicBlock &MBB = MF.front(); - MachineBasicBlock::iterator I = MBB.begin(); - MachineRegisterInfo &RegInfo = MF.getRegInfo(); - const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo(); - DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); - unsigned V0, V1, V2, GlobalBaseReg = MipsFI->getGlobalBaseReg(); - const TargetRegisterClass *RC; - - if (Subtarget.isABI_N64()) - RC = (const TargetRegisterClass*)&Mips::CPU64RegsRegClass; - else if (Subtarget.inMips16Mode()) - RC = (const TargetRegisterClass*)&Mips::CPU16RegsRegClass; - else - RC = (const TargetRegisterClass*)&Mips::CPURegsRegClass; - - V0 = RegInfo.createVirtualRegister(RC); - V1 = RegInfo.createVirtualRegister(RC); - V2 = RegInfo.createVirtualRegister(RC); - - if (Subtarget.isABI_N64()) { - MF.getRegInfo().addLiveIn(Mips::T9_64); - MBB.addLiveIn(Mips::T9_64); - - // lui $v0, %hi(%neg(%gp_rel(fname))) - // daddu $v1, $v0, $t9 - // daddiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname))) - const GlobalValue *FName = MF.getFunction(); - BuildMI(MBB, I, DL, TII.get(Mips::LUi64), V0) - .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI); - BuildMI(MBB, I, DL, TII.get(Mips::DADDu), V1).addReg(V0) - .addReg(Mips::T9_64); - BuildMI(MBB, I, DL, TII.get(Mips::DADDiu), GlobalBaseReg).addReg(V1) - .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO); - return; - } - - if (Subtarget.inMips16Mode()) { - BuildMI(MBB, I, DL, TII.get(Mips::LiRxImmX16), V0) - .addExternalSymbol("_gp_disp", MipsII::MO_ABS_HI); - BuildMI(MBB, I, DL, TII.get(Mips::AddiuRxPcImmX16), V1) - .addExternalSymbol("_gp_disp", MipsII::MO_ABS_LO); - BuildMI(MBB, I, DL, TII.get(Mips::SllX16), V2).addReg(V0).addImm(16); - BuildMI(MBB, I, DL, TII.get(Mips::AdduRxRyRz16), GlobalBaseReg) - .addReg(V1).addReg(V2); - return; - } - - if (MF.getTarget().getRelocationModel() == Reloc::Static) { - // Set global register to __gnu_local_gp. - // - // lui $v0, %hi(__gnu_local_gp) - // addiu $globalbasereg, $v0, %lo(__gnu_local_gp) - BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0) - .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_HI); - BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V0) - .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_LO); - return; - } - - MF.getRegInfo().addLiveIn(Mips::T9); - MBB.addLiveIn(Mips::T9); - - if (Subtarget.isABI_N32()) { - // lui $v0, %hi(%neg(%gp_rel(fname))) - // addu $v1, $v0, $t9 - // addiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname))) - const GlobalValue *FName = MF.getFunction(); - BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0) - .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI); - BuildMI(MBB, I, DL, TII.get(Mips::ADDu), V1).addReg(V0).addReg(Mips::T9); - BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V1) - .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO); - return; - } - - assert(Subtarget.isABI_O32()); - - // For O32 ABI, the following instruction sequence is emitted to initialize - // the global base register: - // - // 0. lui $2, %hi(_gp_disp) - // 1. addiu $2, $2, %lo(_gp_disp) - // 2. addu $globalbasereg, $2, $t9 - // - // We emit only the last instruction here. - // - // GNU linker requires that the first two instructions appear at the beginning - // of a function and no instructions be inserted before or between them. - // The two instructions are emitted during lowering to MC layer in order to - // avoid any reordering. - // - // Register $2 (Mips::V0) is added to the list of live-in registers to ensure - // the value instruction 1 (addiu) defines is valid when instruction 2 (addu) - // reads it. - MF.getRegInfo().addLiveIn(Mips::V0); - MBB.addLiveIn(Mips::V0); - BuildMI(MBB, I, DL, TII.get(Mips::ADDu), GlobalBaseReg) - .addReg(Mips::V0).addReg(Mips::T9); -} - -// Insert instructions to initialize the Mips16 SP Alias register in the -// first MBB of the function. -// -void MipsDAGToDAGISel::InitMips16SPAliasReg(MachineFunction &MF) { - MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); - - if (!MipsFI->mips16SPAliasRegSet()) - return; - - MachineBasicBlock &MBB = MF.front(); - MachineBasicBlock::iterator I = MBB.begin(); - const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo(); - DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); - unsigned Mips16SPAliasReg = MipsFI->getMips16SPAliasReg(); - - BuildMI(MBB, I, DL, TII.get(Mips::MoveR3216), Mips16SPAliasReg) - .addReg(Mips::SP); -} - - -bool MipsDAGToDAGISel::ReplaceUsesWithZeroReg(MachineRegisterInfo *MRI, - const MachineInstr& MI) { - unsigned DstReg = 0, ZeroReg = 0; - - // Check if MI is "addiu $dst, $zero, 0" or "daddiu $dst, $zero, 0". - if ((MI.getOpcode() == Mips::ADDiu) && - (MI.getOperand(1).getReg() == Mips::ZERO) && - (MI.getOperand(2).getImm() == 0)) { - DstReg = MI.getOperand(0).getReg(); - ZeroReg = Mips::ZERO; - } else if ((MI.getOpcode() == Mips::DADDiu) && - (MI.getOperand(1).getReg() == Mips::ZERO_64) && - (MI.getOperand(2).getImm() == 0)) { - DstReg = MI.getOperand(0).getReg(); - ZeroReg = Mips::ZERO_64; - } - - if (!DstReg) - return false; - - // Replace uses with ZeroReg. - for (MachineRegisterInfo::use_iterator U = MRI->use_begin(DstReg), - E = MRI->use_end(); U != E;) { - MachineOperand &MO = U.getOperand(); - unsigned OpNo = U.getOperandNo(); - MachineInstr *MI = MO.getParent(); - ++U; - - // Do not replace if it is a phi's operand or is tied to def operand. - if (MI->isPHI() || MI->isRegTiedToDefOperand(OpNo) || MI->isPseudo()) - continue; - - MO.setReg(ZeroReg); - } - - return true; -} - -void MipsDAGToDAGISel::ProcessFunctionAfterISel(MachineFunction &MF) { - InitGlobalBaseReg(MF); - InitMips16SPAliasReg(MF); - - MachineRegisterInfo *MRI = &MF.getRegInfo(); - - for (MachineFunction::iterator MFI = MF.begin(), MFE = MF.end(); MFI != MFE; - ++MFI) - for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I) - ReplaceUsesWithZeroReg(MRI, *I); -} bool MipsDAGToDAGISel::runOnMachineFunction(MachineFunction &MF) { bool Ret = SelectionDAGISel::runOnMachineFunction(MF); - ProcessFunctionAfterISel(MF); + processFunctionAfterISel(MF); return Ret; } @@ -320,233 +60,36 @@ SDNode *MipsDAGToDAGISel::getGlobalBaseReg() { return CurDAG->getRegister(GlobalBaseReg, TLI.getPointerTy()).getNode(); } -/// getMips16SPAliasReg - Output the instructions required to put the -/// SP into a Mips16 accessible aliased register. -SDValue MipsDAGToDAGISel::getMips16SPAliasReg() { - unsigned Mips16SPAliasReg = - MF->getInfo<MipsFunctionInfo>()->getMips16SPAliasReg(); - return CurDAG->getRegister(Mips16SPAliasReg, TLI.getPointerTy()); -} - /// ComplexPattern used on MipsInstrInfo /// Used on Mips Load/Store instructions bool MipsDAGToDAGISel::selectAddrRegImm(SDValue Addr, SDValue &Base, SDValue &Offset) const { - EVT ValTy = Addr.getValueType(); - - // if Address is FI, get the TargetFrameIndex. - if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) { - Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); - Offset = CurDAG->getTargetConstant(0, ValTy); - return true; - } - - // on PIC code Load GA - if (Addr.getOpcode() == MipsISD::Wrapper) { - Base = Addr.getOperand(0); - Offset = Addr.getOperand(1); - return true; - } - - if (TM.getRelocationModel() != Reloc::PIC_) { - if ((Addr.getOpcode() == ISD::TargetExternalSymbol || - Addr.getOpcode() == ISD::TargetGlobalAddress)) - return false; - } - - // Addresses of the form FI+const or FI|const - if (CurDAG->isBaseWithConstantOffset(Addr)) { - ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1)); - if (isInt<16>(CN->getSExtValue())) { - - // If the first operand is a FI, get the TargetFI Node - if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode> - (Addr.getOperand(0))) - Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); - else - Base = Addr.getOperand(0); - - Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy); - return true; - } - } - - // Operand is a result from an ADD. - if (Addr.getOpcode() == ISD::ADD) { - // When loading from constant pools, load the lower address part in - // the instruction itself. Example, instead of: - // lui $2, %hi($CPI1_0) - // addiu $2, $2, %lo($CPI1_0) - // lwc1 $f0, 0($2) - // Generate: - // lui $2, %hi($CPI1_0) - // lwc1 $f0, %lo($CPI1_0)($2) - if (Addr.getOperand(1).getOpcode() == MipsISD::Lo || - Addr.getOperand(1).getOpcode() == MipsISD::GPRel) { - SDValue Opnd0 = Addr.getOperand(1).getOperand(0); - if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) || - isa<JumpTableSDNode>(Opnd0)) { - Base = Addr.getOperand(0); - Offset = Opnd0; - return true; - } - } - } - + llvm_unreachable("Unimplemented function."); return false; } bool MipsDAGToDAGISel::selectAddrDefault(SDValue Addr, SDValue &Base, SDValue &Offset) const { - Base = Addr; - Offset = CurDAG->getTargetConstant(0, Addr.getValueType()); - return true; + llvm_unreachable("Unimplemented function."); + return false; } bool MipsDAGToDAGISel::selectIntAddr(SDValue Addr, SDValue &Base, SDValue &Offset) const { - return selectAddrRegImm(Addr, Base, Offset) || - selectAddrDefault(Addr, Base, Offset); -} - -void MipsDAGToDAGISel::getMips16SPRefReg(SDNode *Parent, SDValue &AliasReg) { - SDValue AliasFPReg = CurDAG->getRegister(Mips::S0, TLI.getPointerTy()); - if (Parent) { - switch (Parent->getOpcode()) { - case ISD::LOAD: { - LoadSDNode *SD = dyn_cast<LoadSDNode>(Parent); - switch (SD->getMemoryVT().getSizeInBits()) { - case 8: - case 16: - AliasReg = TM.getFrameLowering()->hasFP(*MF)? - AliasFPReg: getMips16SPAliasReg(); - return; - } - break; - } - case ISD::STORE: { - StoreSDNode *SD = dyn_cast<StoreSDNode>(Parent); - switch (SD->getMemoryVT().getSizeInBits()) { - case 8: - case 16: - AliasReg = TM.getFrameLowering()->hasFP(*MF)? - AliasFPReg: getMips16SPAliasReg(); - return; - } - break; - } - } - } - AliasReg = CurDAG->getRegister(Mips::SP, TLI.getPointerTy()); - return; - -} -bool MipsDAGToDAGISel::SelectAddr16( - SDNode *Parent, SDValue Addr, SDValue &Base, SDValue &Offset, - SDValue &Alias) { - EVT ValTy = Addr.getValueType(); - - Alias = CurDAG->getTargetConstant(0, ValTy); - - // if Address is FI, get the TargetFrameIndex. - if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) { - Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); - Offset = CurDAG->getTargetConstant(0, ValTy); - getMips16SPRefReg(Parent, Alias); - return true; - } - // on PIC code Load GA - if (Addr.getOpcode() == MipsISD::Wrapper) { - Base = Addr.getOperand(0); - Offset = Addr.getOperand(1); - return true; - } - if (TM.getRelocationModel() != Reloc::PIC_) { - if ((Addr.getOpcode() == ISD::TargetExternalSymbol || - Addr.getOpcode() == ISD::TargetGlobalAddress)) - return false; - } - // Addresses of the form FI+const or FI|const - if (CurDAG->isBaseWithConstantOffset(Addr)) { - ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1)); - if (isInt<16>(CN->getSExtValue())) { - - // If the first operand is a FI, get the TargetFI Node - if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode> - (Addr.getOperand(0))) { - Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); - getMips16SPRefReg(Parent, Alias); - } - else - Base = Addr.getOperand(0); - - Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy); - return true; - } - } - // Operand is a result from an ADD. - if (Addr.getOpcode() == ISD::ADD) { - // When loading from constant pools, load the lower address part in - // the instruction itself. Example, instead of: - // lui $2, %hi($CPI1_0) - // addiu $2, $2, %lo($CPI1_0) - // lwc1 $f0, 0($2) - // Generate: - // lui $2, %hi($CPI1_0) - // lwc1 $f0, %lo($CPI1_0)($2) - if (Addr.getOperand(1).getOpcode() == MipsISD::Lo || - Addr.getOperand(1).getOpcode() == MipsISD::GPRel) { - SDValue Opnd0 = Addr.getOperand(1).getOperand(0); - if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) || - isa<JumpTableSDNode>(Opnd0)) { - Base = Addr.getOperand(0); - Offset = Opnd0; - return true; - } - } - - // If an indexed floating point load/store can be emitted, return false. - const LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(Parent); - - if (LS && - (LS->getMemoryVT() == MVT::f32 || LS->getMemoryVT() == MVT::f64) && - Subtarget.hasFPIdx()) - return false; - } - Base = Addr; - Offset = CurDAG->getTargetConstant(0, ValTy); - return true; + llvm_unreachable("Unimplemented function."); + return false; } -/// Select multiply instructions. -std::pair<SDNode*, SDNode*> -MipsDAGToDAGISel::SelectMULT(SDNode *N, unsigned Opc, DebugLoc dl, EVT Ty, - bool HasLo, bool HasHi) { - SDNode *Lo = 0, *Hi = 0; - SDNode *Mul = CurDAG->getMachineNode(Opc, dl, MVT::Glue, N->getOperand(0), - N->getOperand(1)); - SDValue InFlag = SDValue(Mul, 0); - - if (HasLo) { - unsigned Opcode = Subtarget.inMips16Mode() ? Mips::Mflo16 : - (Ty == MVT::i32 ? Mips::MFLO : Mips::MFLO64); - Lo = CurDAG->getMachineNode(Opcode, dl, Ty, MVT::Glue, InFlag); - InFlag = SDValue(Lo, 1); - } - if (HasHi) { - unsigned Opcode = Subtarget.inMips16Mode() ? Mips::Mfhi16 : - (Ty == MVT::i32 ? Mips::MFHI : Mips::MFHI64); - Hi = CurDAG->getMachineNode(Opcode, dl, Ty, InFlag); - } - return std::make_pair(Lo, Hi); +bool MipsDAGToDAGISel::selectAddr16(SDNode *Parent, SDValue N, SDValue &Base, + SDValue &Offset, SDValue &Alias) { + llvm_unreachable("Unimplemented function."); + return false; } - /// Select instructions not customized! Used for /// expanded, promoted and normal instructions SDNode* MipsDAGToDAGISel::Select(SDNode *Node) { unsigned Opcode = Node->getOpcode(); - DebugLoc dl = Node->getDebugLoc(); // Dump information about the Node being selected DEBUG(errs() << "Selecting: "; Node->dump(CurDAG); errs() << "\n"); @@ -557,167 +100,19 @@ SDNode* MipsDAGToDAGISel::Select(SDNode *Node) { return NULL; } - /// - // Instruction Selection not handled by the auto-generated - // tablegen selection should be handled here. - /// - EVT NodeTy = Node->getValueType(0); - unsigned MultOpc; + // See if subclasses can handle this node. + std::pair<bool, SDNode*> Ret = selectNode(Node); + + if (Ret.first) + return Ret.second; switch(Opcode) { default: break; - case ISD::SUBE: - case ISD::ADDE: { - bool inMips16Mode = Subtarget.inMips16Mode(); - SDValue InFlag = Node->getOperand(2), CmpLHS; - unsigned Opc = InFlag.getOpcode(); (void)Opc; - assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) || - (Opc == ISD::SUBC || Opc == ISD::SUBE)) && - "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn"); - - unsigned MOp; - if (Opcode == ISD::ADDE) { - CmpLHS = InFlag.getValue(0); - if (inMips16Mode) - MOp = Mips::AdduRxRyRz16; - else - MOp = Mips::ADDu; - } else { - CmpLHS = InFlag.getOperand(0); - if (inMips16Mode) - MOp = Mips::SubuRxRyRz16; - else - MOp = Mips::SUBu; - } - - SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) }; - - SDValue LHS = Node->getOperand(0); - SDValue RHS = Node->getOperand(1); - - EVT VT = LHS.getValueType(); - - unsigned Sltu_op = inMips16Mode? Mips::SltuRxRyRz16: Mips::SLTu; - SDNode *Carry = CurDAG->getMachineNode(Sltu_op, dl, VT, Ops, 2); - unsigned Addu_op = inMips16Mode? Mips::AdduRxRyRz16 : Mips::ADDu; - SDNode *AddCarry = CurDAG->getMachineNode(Addu_op, dl, VT, - SDValue(Carry,0), RHS); - - return CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue, - LHS, SDValue(AddCarry,0)); - } - - /// Mul with two results - case ISD::SMUL_LOHI: - case ISD::UMUL_LOHI: { - if (NodeTy == MVT::i32) { - if (Subtarget.inMips16Mode()) - MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MultuRxRy16 : - Mips::MultRxRy16); - else - MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::MULTu : Mips::MULT); - } - else - MultOpc = (Opcode == ISD::UMUL_LOHI ? Mips::DMULTu : Mips::DMULT); - - std::pair<SDNode*, SDNode*> LoHi = SelectMULT(Node, MultOpc, dl, NodeTy, - true, true); - - if (!SDValue(Node, 0).use_empty()) - ReplaceUses(SDValue(Node, 0), SDValue(LoHi.first, 0)); - - if (!SDValue(Node, 1).use_empty()) - ReplaceUses(SDValue(Node, 1), SDValue(LoHi.second, 0)); - - return NULL; - } - - /// Special Muls - case ISD::MUL: { - // Mips32 has a 32-bit three operand mul instruction. - if (Subtarget.hasMips32() && NodeTy == MVT::i32) - break; - return SelectMULT(Node, NodeTy == MVT::i32 ? Mips::MULT : Mips::DMULT, - dl, NodeTy, true, false).first; - } - case ISD::MULHS: - case ISD::MULHU: { - if (NodeTy == MVT::i32) { - if (Subtarget.inMips16Mode()) - MultOpc = (Opcode == ISD::MULHU ? - Mips::MultuRxRy16 : Mips::MultRxRy16); - else - MultOpc = (Opcode == ISD::MULHU ? Mips::MULTu : Mips::MULT); - } - else - MultOpc = (Opcode == ISD::MULHU ? Mips::DMULTu : Mips::DMULT); - - return SelectMULT(Node, MultOpc, dl, NodeTy, false, true).second; - } - // Get target GOT address. case ISD::GLOBAL_OFFSET_TABLE: return getGlobalBaseReg(); - case ISD::ConstantFP: { - ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(Node); - if (Node->getValueType(0) == MVT::f64 && CN->isExactlyValue(+0.0)) { - if (Subtarget.hasMips64()) { - SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl, - Mips::ZERO_64, MVT::i64); - return CurDAG->getMachineNode(Mips::DMTC1, dl, MVT::f64, Zero); - } - - SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl, - Mips::ZERO, MVT::i32); - return CurDAG->getMachineNode(Mips::BuildPairF64, dl, MVT::f64, Zero, - Zero); - } - break; - } - - case ISD::Constant: { - const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Node); - unsigned Size = CN->getValueSizeInBits(0); - - if (Size == 32) - break; - - MipsAnalyzeImmediate AnalyzeImm; - int64_t Imm = CN->getSExtValue(); - - const MipsAnalyzeImmediate::InstSeq &Seq = - AnalyzeImm.Analyze(Imm, Size, false); - - MipsAnalyzeImmediate::InstSeq::const_iterator Inst = Seq.begin(); - DebugLoc DL = CN->getDebugLoc(); - SDNode *RegOpnd; - SDValue ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd), - MVT::i64); - - // The first instruction can be a LUi which is different from other - // instructions (ADDiu, ORI and SLL) in that it does not have a register - // operand. - if (Inst->Opc == Mips::LUi64) - RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, ImmOpnd); - else - RegOpnd = - CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, - CurDAG->getRegister(Mips::ZERO_64, MVT::i64), - ImmOpnd); - - // The remaining instructions in the sequence are handled here. - for (++Inst; Inst != Seq.end(); ++Inst) { - ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd), - MVT::i64); - RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, - SDValue(RegOpnd, 0), ImmOpnd); - } - - return RegOpnd; - } - #ifndef NDEBUG case ISD::LOAD: case ISD::STORE: @@ -726,31 +121,6 @@ SDNode* MipsDAGToDAGISel::Select(SDNode *Node) { "Unexpected unaligned loads/stores."); break; #endif - - case MipsISD::ThreadPointer: { - EVT PtrVT = TLI.getPointerTy(); - unsigned RdhwrOpc, SrcReg, DestReg; - - if (PtrVT == MVT::i32) { - RdhwrOpc = Mips::RDHWR; - SrcReg = Mips::HWR29; - DestReg = Mips::V1; - } else { - RdhwrOpc = Mips::RDHWR64; - SrcReg = Mips::HWR29_64; - DestReg = Mips::V1_64; - } - - SDNode *Rdhwr = - CurDAG->getMachineNode(RdhwrOpc, Node->getDebugLoc(), - Node->getValueType(0), - CurDAG->getRegister(SrcReg, PtrVT)); - SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, DestReg, - SDValue(Rdhwr, 0)); - SDValue ResNode = CurDAG->getCopyFromReg(Chain, dl, DestReg, PtrVT); - ReplaceUses(SDValue(Node, 0), ResNode); - return ResNode.getNode(); - } } // Select the default instruction @@ -776,5 +146,8 @@ SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode, /// createMipsISelDag - This pass converts a legalized DAG into a /// MIPS-specific DAG, ready for instruction scheduling. FunctionPass *llvm::createMipsISelDag(MipsTargetMachine &TM) { - return new MipsDAGToDAGISel(TM); + if (TM.getSubtargetImpl()->inMips16Mode()) + return llvm::createMips16ISelDag(TM); + + return llvm::createMipsSEISelDag(TM); } diff --git a/lib/Target/Mips/MipsISelDAGToDAG.h b/lib/Target/Mips/MipsISelDAGToDAG.h new file mode 100644 index 0000000000..cf0f9c58aa --- /dev/null +++ b/lib/Target/Mips/MipsISelDAGToDAG.h @@ -0,0 +1,93 @@ +//===---- MipsISelDAGToDAG.h - A Dag to Dag Inst Selector for Mips --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines an instruction selector for the MIPS target. +// +//===----------------------------------------------------------------------===// + +#ifndef MIPSISELDAGTODAG_H +#define MIPSISELDAGTODAG_H + +#include "Mips.h" +#include "MipsSubtarget.h" +#include "MipsTargetMachine.h" +#include "llvm/CodeGen/SelectionDAGISel.h" + +//===----------------------------------------------------------------------===// +// Instruction Selector Implementation +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// MipsDAGToDAGISel - MIPS specific code to select MIPS machine +// instructions for SelectionDAG operations. +//===----------------------------------------------------------------------===// +namespace llvm { + +class MipsDAGToDAGISel : public SelectionDAGISel { +public: + explicit MipsDAGToDAGISel(MipsTargetMachine &TM) + : SelectionDAGISel(TM), Subtarget(TM.getSubtarget<MipsSubtarget>()) {} + + // Pass Name + virtual const char *getPassName() const { + return "MIPS DAG->DAG Pattern Instruction Selection"; + } + + virtual bool runOnMachineFunction(MachineFunction &MF); + +protected: + SDNode *getGlobalBaseReg(); + + /// Keep a pointer to the MipsSubtarget around so that we can make the right + /// decision when generating code for different targets. + const MipsSubtarget &Subtarget; + +private: + // Include the pieces autogenerated from the target description. + #include "MipsGenDAGISel.inc" + + // Complex Pattern. + /// (reg + imm). + virtual bool selectAddrRegImm(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + /// Fall back on this function if all else fails. + virtual bool selectAddrDefault(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + /// Match integer address pattern. + virtual bool selectIntAddr(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectAddr16(SDNode *Parent, SDValue N, SDValue &Base, + SDValue &Offset, SDValue &Alias); + + virtual SDNode *Select(SDNode *N); + + virtual std::pair<bool, SDNode*> selectNode(SDNode *Node) = 0; + + // getImm - Return a target constant with the specified value. + inline SDValue getImm(const SDNode *Node, uint64_t Imm) { + return CurDAG->getTargetConstant(Imm, Node->getValueType(0)); + } + + virtual void processFunctionAfterISel(MachineFunction &MF) = 0; + + virtual bool SelectInlineAsmMemoryOperand(const SDValue &Op, + char ConstraintCode, + std::vector<SDValue> &OutOps); +}; + +/// createMipsISelDag - This pass converts a legalized DAG into a +/// MIPS-specific DAG, ready for instruction scheduling. +FunctionPass *createMipsISelDag(MipsTargetMachine &TM); + +} + +#endif diff --git a/lib/Target/Mips/MipsISelLowering.cpp b/lib/Target/Mips/MipsISelLowering.cpp index 30409a2b3d..1d9ca6e1d1 100644 --- a/lib/Target/Mips/MipsISelLowering.cpp +++ b/lib/Target/Mips/MipsISelLowering.cpp @@ -12,7 +12,6 @@ // //===----------------------------------------------------------------------===// #define DEBUG_TYPE "mips-lower" -#include <set> #include "MipsISelLowering.h" #include "InstPrinter/MipsInstPrinter.h" #include "MCTargetDesc/MipsBaseInfo.h" @@ -30,9 +29,7 @@ #include "llvm/CodeGen/ValueTypes.h" #include "llvm/IR/CallingConv.h" #include "llvm/IR/DerivedTypes.h" -#include "llvm/IR/Function.h" #include "llvm/IR/GlobalVariable.h" -#include "llvm/IR/Intrinsics.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" @@ -43,26 +40,9 @@ using namespace llvm; STATISTIC(NumTailCalls, "Number of tail calls"); static cl::opt<bool> -EnableMipsTailCalls("enable-mips-tail-calls", cl::Hidden, - cl::desc("MIPS: Enable tail calls."), cl::init(false)); - -static cl::opt<bool> LargeGOT("mxgot", cl::Hidden, cl::desc("MIPS: Enable GOT larger than 64k."), cl::init(false)); -static cl::opt<bool> -Mips16HardFloat("mips16-hard-float", cl::NotHidden, - cl::desc("MIPS: mips16 hard float enable."), - cl::init(false)); - -static cl::opt<bool> DontExpandCondPseudos16( - "mips16-dont-expand-cond-pseudo", - cl::init(false), - cl::desc("Dont expand conditional move related " - "pseudos for Mips 16"), - cl::Hidden); - - static const uint16_t O32IntRegs[4] = { Mips::A0, Mips::A1, Mips::A2, Mips::A3 }; @@ -80,7 +60,7 @@ static const uint16_t Mips64DPRegs[8] = { // If I is a shifted mask, set the size (Size) and the first bit of the // mask (Pos), and return true. // For example, if I is 0x003ff800, (Pos, Size) = (11, 11). -static bool IsShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) { +static bool isShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) { if (!isShiftedMask_64(I)) return false; @@ -89,7 +69,7 @@ static bool IsShiftedMask(uint64_t I, uint64_t &Pos, uint64_t &Size) { return true; } -static SDValue GetGlobalReg(SelectionDAG &DAG, EVT Ty) { +SDValue MipsTargetLowering::getGlobalReg(SelectionDAG &DAG, EVT Ty) const { MipsFunctionInfo *FI = DAG.getMachineFunction().getInfo<MipsFunctionInfo>(); return DAG.getRegister(FI->getGlobalBaseReg(), Ty); } @@ -124,11 +104,12 @@ static SDValue getAddrNonPIC(SDValue Op, SelectionDAG &DAG) { DAG.getNode(MipsISD::Lo, DL, Ty, Lo)); } -static SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) { +SDValue MipsTargetLowering::getAddrLocal(SDValue Op, SelectionDAG &DAG, + bool HasMips64) const { DebugLoc DL = Op.getDebugLoc(); EVT Ty = Op.getValueType(); unsigned GOTFlag = HasMips64 ? MipsII::MO_GOT_PAGE : MipsII::MO_GOT; - SDValue GOT = DAG.getNode(MipsISD::Wrapper, DL, Ty, GetGlobalReg(DAG, Ty), + SDValue GOT = DAG.getNode(MipsISD::Wrapper, DL, Ty, getGlobalReg(DAG, Ty), getTargetNode(Op, DAG, GOTFlag)); SDValue Load = DAG.getLoad(Ty, DL, DAG.getEntryNode(), GOT, MachinePointerInfo::getGOT(), false, false, false, @@ -138,21 +119,23 @@ static SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) { return DAG.getNode(ISD::ADD, DL, Ty, Load, Lo); } -static SDValue getAddrGlobal(SDValue Op, SelectionDAG &DAG, unsigned Flag) { +SDValue MipsTargetLowering::getAddrGlobal(SDValue Op, SelectionDAG &DAG, + unsigned Flag) const { DebugLoc DL = Op.getDebugLoc(); EVT Ty = Op.getValueType(); - SDValue Tgt = DAG.getNode(MipsISD::Wrapper, DL, Ty, GetGlobalReg(DAG, Ty), + SDValue Tgt = DAG.getNode(MipsISD::Wrapper, DL, Ty, getGlobalReg(DAG, Ty), getTargetNode(Op, DAG, Flag)); return DAG.getLoad(Ty, DL, DAG.getEntryNode(), Tgt, MachinePointerInfo::getGOT(), false, false, false, 0); } -static SDValue getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG, - unsigned HiFlag, unsigned LoFlag) { +SDValue MipsTargetLowering::getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG, + unsigned HiFlag, + unsigned LoFlag) const { DebugLoc DL = Op.getDebugLoc(); EVT Ty = Op.getValueType(); SDValue Hi = DAG.getNode(MipsISD::Hi, DL, Ty, getTargetNode(Op, DAG, HiFlag)); - Hi = DAG.getNode(ISD::ADD, DL, Ty, Hi, GetGlobalReg(DAG, Ty)); + Hi = DAG.getNode(ISD::ADD, DL, Ty, Hi, getGlobalReg(DAG, Ty)); SDValue Wrapper = DAG.getNode(MipsISD::Wrapper, DL, Ty, Hi, getTargetNode(Op, DAG, LoFlag)); return DAG.getLoad(Ty, DL, DAG.getEntryNode(), Wrapper, @@ -174,12 +157,18 @@ const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const { case MipsISD::CMovFP_T: return "MipsISD::CMovFP_T"; case MipsISD::CMovFP_F: return "MipsISD::CMovFP_F"; case MipsISD::FPRound: return "MipsISD::FPRound"; + case MipsISD::ExtractLOHI: return "MipsISD::ExtractLOHI"; + case MipsISD::InsertLOHI: return "MipsISD::InsertLOHI"; + case MipsISD::Mult: return "MipsISD::Mult"; + case MipsISD::Multu: return "MipsISD::Multu"; case MipsISD::MAdd: return "MipsISD::MAdd"; case MipsISD::MAddu: return "MipsISD::MAddu"; case MipsISD::MSub: return "MipsISD::MSub"; case MipsISD::MSubu: return "MipsISD::MSubu"; case MipsISD::DivRem: return "MipsISD::DivRem"; case MipsISD::DivRemU: return "MipsISD::DivRemU"; + case MipsISD::DivRem16: return "MipsISD::DivRem16"; + case MipsISD::DivRemU16: return "MipsISD::DivRemU16"; case MipsISD::BuildPairF64: return "MipsISD::BuildPairF64"; case MipsISD::ExtractElementF64: return "MipsISD::ExtractElementF64"; case MipsISD::Wrapper: return "MipsISD::Wrapper"; @@ -204,117 +193,29 @@ const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const { case MipsISD::MTHLIP: return "MipsISD::MTHLIP"; case MipsISD::MULT: return "MipsISD::MULT"; case MipsISD::MULTU: return "MipsISD::MULTU"; - case MipsISD::MADD_DSP: return "MipsISD::MADD_DSPDSP"; + case MipsISD::MADD_DSP: return "MipsISD::MADD_DSP"; case MipsISD::MADDU_DSP: return "MipsISD::MADDU_DSP"; case MipsISD::MSUB_DSP: return "MipsISD::MSUB_DSP"; case MipsISD::MSUBU_DSP: return "MipsISD::MSUBU_DSP"; + case MipsISD::SHLL_DSP: return "MipsISD::SHLL_DSP"; + case MipsISD::SHRA_DSP: return "MipsISD::SHRA_DSP"; + case MipsISD::SHRL_DSP: return "MipsISD::SHRL_DSP"; + case MipsISD::SETCC_DSP: return "MipsISD::SETCC_DSP"; + case MipsISD::SELECT_CC_DSP: return "MipsISD::SELECT_CC_DSP"; default: return NULL; } } -namespace { - struct ltstr { - bool operator()(const char *s1, const char *s2) const - { - return strcmp(s1, s2) < 0; - } - }; - - std::set<const char*, ltstr> noHelperNeeded; -} - -void MipsTargetLowering::SetMips16LibcallName - (RTLIB::Libcall l, const char *Name) { - setLibcallName(l, Name); - noHelperNeeded.insert(Name); -} - -void MipsTargetLowering::setMips16HardFloatLibCalls() { - SetMips16LibcallName(RTLIB::ADD_F32, "__mips16_addsf3"); - SetMips16LibcallName(RTLIB::ADD_F64, "__mips16_adddf3"); - SetMips16LibcallName(RTLIB::SUB_F32, "__mips16_subsf3"); - SetMips16LibcallName(RTLIB::SUB_F64, "__mips16_subdf3"); - SetMips16LibcallName(RTLIB::MUL_F32, "__mips16_mulsf3"); - SetMips16LibcallName(RTLIB::MUL_F64, "__mips16_muldf3"); - SetMips16LibcallName(RTLIB::DIV_F32, "__mips16_divsf3"); - SetMips16LibcallName(RTLIB::DIV_F64, "__mips16_divdf3"); - SetMips16LibcallName(RTLIB::FPEXT_F32_F64, "__mips16_extendsfdf2"); - SetMips16LibcallName(RTLIB::FPROUND_F64_F32, "__mips16_truncdfsf2"); - SetMips16LibcallName(RTLIB::FPTOSINT_F32_I32, "__mips16_fix_truncsfsi"); - SetMips16LibcallName(RTLIB::FPTOSINT_F64_I32, "__mips16_fix_truncdfsi"); - SetMips16LibcallName(RTLIB::SINTTOFP_I32_F32, "__mips16_floatsisf"); - SetMips16LibcallName(RTLIB::SINTTOFP_I32_F64, "__mips16_floatsidf"); - SetMips16LibcallName(RTLIB::UINTTOFP_I32_F32, "__mips16_floatunsisf"); - SetMips16LibcallName(RTLIB::UINTTOFP_I32_F64, "__mips16_floatunsidf"); - SetMips16LibcallName(RTLIB::OEQ_F32, "__mips16_eqsf2"); - SetMips16LibcallName(RTLIB::OEQ_F64, "__mips16_eqdf2"); - SetMips16LibcallName(RTLIB::UNE_F32, "__mips16_nesf2"); - SetMips16LibcallName(RTLIB::UNE_F64, "__mips16_nedf2"); - SetMips16LibcallName(RTLIB::OGE_F32, "__mips16_gesf2"); - SetMips16LibcallName(RTLIB::OGE_F64, "__mips16_gedf2"); - SetMips16LibcallName(RTLIB::OLT_F32, "__mips16_ltsf2"); - SetMips16LibcallName(RTLIB::OLT_F64, "__mips16_ltdf2"); - SetMips16LibcallName(RTLIB::OLE_F32, "__mips16_lesf2"); - SetMips16LibcallName(RTLIB::OLE_F64, "__mips16_ledf2"); - SetMips16LibcallName(RTLIB::OGT_F32, "__mips16_gtsf2"); - SetMips16LibcallName(RTLIB::OGT_F64, "__mips16_gtdf2"); - SetMips16LibcallName(RTLIB::UO_F32, "__mips16_unordsf2"); - SetMips16LibcallName(RTLIB::UO_F64, "__mips16_unorddf2"); - SetMips16LibcallName(RTLIB::O_F32, "__mips16_unordsf2"); - SetMips16LibcallName(RTLIB::O_F64, "__mips16_unorddf2"); -} - MipsTargetLowering:: MipsTargetLowering(MipsTargetMachine &TM) : TargetLowering(TM, new MipsTargetObjectFile()), Subtarget(&TM.getSubtarget<MipsSubtarget>()), HasMips64(Subtarget->hasMips64()), IsN64(Subtarget->isABI_N64()), IsO32(Subtarget->isABI_O32()) { - // Mips does not have i1 type, so use i32 for // setcc operations results (slt, sgt, ...). setBooleanContents(ZeroOrOneBooleanContent); - setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct? - - // Set up the register classes - addRegisterClass(MVT::i32, &Mips::CPURegsRegClass); - - if (HasMips64) - addRegisterClass(MVT::i64, &Mips::CPU64RegsRegClass); - - if (Subtarget->inMips16Mode()) { - addRegisterClass(MVT::i32, &Mips::CPU16RegsRegClass); - if (Mips16HardFloat) - setMips16HardFloatLibCalls(); - } - - if (Subtarget->hasDSP()) { - MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8}; - - for (unsigned i = 0; i < array_lengthof(VecTys); ++i) { - addRegisterClass(VecTys[i], &Mips::DSPRegsRegClass); - - // Expand all builtin opcodes. - for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc) - setOperationAction(Opc, VecTys[i], Expand); - - setOperationAction(ISD::LOAD, VecTys[i], Legal); - setOperationAction(ISD::STORE, VecTys[i], Legal); - setOperationAction(ISD::BITCAST, VecTys[i], Legal); - } - } - - if (!TM.Options.UseSoftFloat) { - addRegisterClass(MVT::f32, &Mips::FGR32RegClass); - - // When dealing with single precision only, use libcalls - if (!Subtarget->isSingleFloat()) { - if (HasMips64) - addRegisterClass(MVT::f64, &Mips::FGR64RegClass); - else - addRegisterClass(MVT::f64, &Mips::AFGR64RegClass); - } - } + setBooleanVectorContents(ZeroOrNegativeOneBooleanContent); // Load extented operations for i1 types must be promoted setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote); @@ -332,6 +233,7 @@ MipsTargetLowering(MipsTargetMachine &TM) AddPromotedToType(ISD::SETCC, MVT::i1, MVT::i32); // Mips Custom Operations + setOperationAction(ISD::BR_JT, MVT::Other, Custom); setOperationAction(ISD::GlobalAddress, MVT::i32, Custom); setOperationAction(ISD::BlockAddress, MVT::i32, Custom); setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom); @@ -348,18 +250,6 @@ MipsTargetLowering(MipsTargetMachine &TM) setOperationAction(ISD::VASTART, MVT::Other, Custom); setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom); setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom); - if (Subtarget->inMips16Mode()) { - setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand); - setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand); - } - else { - setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom); - setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom); - } - if (!Subtarget->inMips16Mode()) { - setOperationAction(ISD::LOAD, MVT::i32, Custom); - setOperationAction(ISD::STORE, MVT::i32, Custom); - } if (!TM.Options.NoNaNsFPMath) { setOperationAction(ISD::FABS, MVT::f32, Custom); @@ -397,8 +287,10 @@ MipsTargetLowering(MipsTargetMachine &TM) setOperationAction(ISD::UREM, MVT::i64, Expand); // Operations not directly supported by Mips. - setOperationAction(ISD::BR_JT, MVT::Other, Expand); - setOperationAction(ISD::BR_CC, MVT::Other, Expand); + setOperationAction(ISD::BR_CC, MVT::f32, Expand); + setOperationAction(ISD::BR_CC, MVT::f64, Expand); + setOperationAction(ISD::BR_CC, MVT::i32, Expand); + setOperationAction(ISD::BR_CC, MVT::i64, Expand); setOperationAction(ISD::SELECT_CC, MVT::Other, Expand); setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand); setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand); @@ -475,21 +367,6 @@ MipsTargetLowering(MipsTargetMachine &TM) setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Expand); setOperationAction(ISD::ATOMIC_STORE, MVT::i64, Expand); - if (Subtarget->inMips16Mode()) { - setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_SWAP, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand); - setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand); - } - setInsertFencesForAtomic(true); if (!Subtarget->hasSEInReg()) { @@ -521,8 +398,7 @@ MipsTargetLowering(MipsTargetMachine &TM) } // @LOCALMOD-END - setTargetDAGCombine(ISD::ADDE); - setTargetDAGCombine(ISD::SUBE); + setTargetDAGCombine(ISD::SDIVREM); setTargetDAGCombine(ISD::UDIVREM); setTargetDAGCombine(ISD::SELECT); @@ -533,7 +409,6 @@ MipsTargetLowering(MipsTargetMachine &TM) setMinFunctionAlignment(HasMips64 ? 3 : 2); setStackPointerRegisterToSaveRestore(IsN64 ? Mips::SP_64 : Mips::SP); - computeRegisterProperties(); setExceptionPointerRegister(IsN64 ? Mips::A0_64 : Mips::A0); setExceptionSelectorRegister(IsN64 ? Mips::A1_64 : Mips::A1); @@ -541,22 +416,11 @@ MipsTargetLowering(MipsTargetMachine &TM) MaxStoresPerMemcpy = 16; } -bool -MipsTargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const { - MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy; - - if (Subtarget->inMips16Mode()) - return false; +const MipsTargetLowering *MipsTargetLowering::create(MipsTargetMachine &TM) { + if (TM.getSubtargetImpl()->inMips16Mode()) + return llvm::createMips16TargetLowering(TM); - switch (SVT) { - case MVT::i64: - case MVT::i32: - if (Fast) - *Fast = true; - return true; - default: - return false; - } + return llvm::createMipsSETargetLowering(TM); } EVT MipsTargetLowering::getSetCCResultType(EVT VT) const { @@ -565,179 +429,7 @@ EVT MipsTargetLowering::getSetCCResultType(EVT VT) const { return VT.changeVectorElementTypeToInteger(); } -// SelectMadd - -// Transforms a subgraph in CurDAG if the following pattern is found: -// (addc multLo, Lo0), (adde multHi, Hi0), -// where, -// multHi/Lo: product of multiplication -// Lo0: initial value of Lo register -// Hi0: initial value of Hi register -// Return true if pattern matching was successful. -static bool SelectMadd(SDNode *ADDENode, SelectionDAG *CurDAG) { - // ADDENode's second operand must be a flag output of an ADDC node in order - // for the matching to be successful. - SDNode *ADDCNode = ADDENode->getOperand(2).getNode(); - - if (ADDCNode->getOpcode() != ISD::ADDC) - return false; - - SDValue MultHi = ADDENode->getOperand(0); - SDValue MultLo = ADDCNode->getOperand(0); - SDNode *MultNode = MultHi.getNode(); - unsigned MultOpc = MultHi.getOpcode(); - - // MultHi and MultLo must be generated by the same node, - if (MultLo.getNode() != MultNode) - return false; - - // and it must be a multiplication. - if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI) - return false; - - // MultLo amd MultHi must be the first and second output of MultNode - // respectively. - if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0) - return false; - - // Transform this to a MADD only if ADDENode and ADDCNode are the only users - // of the values of MultNode, in which case MultNode will be removed in later - // phases. - // If there exist users other than ADDENode or ADDCNode, this function returns - // here, which will result in MultNode being mapped to a single MULT - // instruction node rather than a pair of MULT and MADD instructions being - // produced. - if (!MultHi.hasOneUse() || !MultLo.hasOneUse()) - return false; - - SDValue Chain = CurDAG->getEntryNode(); - DebugLoc dl = ADDENode->getDebugLoc(); - - // create MipsMAdd(u) node - MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MAddu : MipsISD::MAdd; - - SDValue MAdd = CurDAG->getNode(MultOpc, dl, MVT::Glue, - MultNode->getOperand(0),// Factor 0 - MultNode->getOperand(1),// Factor 1 - ADDCNode->getOperand(1),// Lo0 - ADDENode->getOperand(1));// Hi0 - - // create CopyFromReg nodes - SDValue CopyFromLo = CurDAG->getCopyFromReg(Chain, dl, Mips::LO, MVT::i32, - MAdd); - SDValue CopyFromHi = CurDAG->getCopyFromReg(CopyFromLo.getValue(1), dl, - Mips::HI, MVT::i32, - CopyFromLo.getValue(2)); - - // replace uses of adde and addc here - if (!SDValue(ADDCNode, 0).use_empty()) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDCNode, 0), CopyFromLo); - - if (!SDValue(ADDENode, 0).use_empty()) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDENode, 0), CopyFromHi); - - return true; -} - -// SelectMsub - -// Transforms a subgraph in CurDAG if the following pattern is found: -// (addc Lo0, multLo), (sube Hi0, multHi), -// where, -// multHi/Lo: product of multiplication -// Lo0: initial value of Lo register -// Hi0: initial value of Hi register -// Return true if pattern matching was successful. -static bool SelectMsub(SDNode *SUBENode, SelectionDAG *CurDAG) { - // SUBENode's second operand must be a flag output of an SUBC node in order - // for the matching to be successful. - SDNode *SUBCNode = SUBENode->getOperand(2).getNode(); - - if (SUBCNode->getOpcode() != ISD::SUBC) - return false; - - SDValue MultHi = SUBENode->getOperand(1); - SDValue MultLo = SUBCNode->getOperand(1); - SDNode *MultNode = MultHi.getNode(); - unsigned MultOpc = MultHi.getOpcode(); - - // MultHi and MultLo must be generated by the same node, - if (MultLo.getNode() != MultNode) - return false; - - // and it must be a multiplication. - if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI) - return false; - - // MultLo amd MultHi must be the first and second output of MultNode - // respectively. - if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0) - return false; - - // Transform this to a MSUB only if SUBENode and SUBCNode are the only users - // of the values of MultNode, in which case MultNode will be removed in later - // phases. - // If there exist users other than SUBENode or SUBCNode, this function returns - // here, which will result in MultNode being mapped to a single MULT - // instruction node rather than a pair of MULT and MSUB instructions being - // produced. - if (!MultHi.hasOneUse() || !MultLo.hasOneUse()) - return false; - - SDValue Chain = CurDAG->getEntryNode(); - DebugLoc dl = SUBENode->getDebugLoc(); - - // create MipsSub(u) node - MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MSubu : MipsISD::MSub; - - SDValue MSub = CurDAG->getNode(MultOpc, dl, MVT::Glue, - MultNode->getOperand(0),// Factor 0 - MultNode->getOperand(1),// Factor 1 - SUBCNode->getOperand(0),// Lo0 - SUBENode->getOperand(0));// Hi0 - - // create CopyFromReg nodes - SDValue CopyFromLo = CurDAG->getCopyFromReg(Chain, dl, Mips::LO, MVT::i32, - MSub); - SDValue CopyFromHi = CurDAG->getCopyFromReg(CopyFromLo.getValue(1), dl, - Mips::HI, MVT::i32, - CopyFromLo.getValue(2)); - - // replace uses of sube and subc here - if (!SDValue(SUBCNode, 0).use_empty()) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBCNode, 0), CopyFromLo); - - if (!SDValue(SUBENode, 0).use_empty()) - CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBENode, 0), CopyFromHi); - - return true; -} - -static SDValue PerformADDECombine(SDNode *N, SelectionDAG &DAG, - TargetLowering::DAGCombinerInfo &DCI, - const MipsSubtarget *Subtarget) { - if (DCI.isBeforeLegalize()) - return SDValue(); - - if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 && - SelectMadd(N, &DAG)) - return SDValue(N, 0); - - return SDValue(); -} - -static SDValue PerformSUBECombine(SDNode *N, SelectionDAG &DAG, - TargetLowering::DAGCombinerInfo &DCI, - const MipsSubtarget *Subtarget) { - if (DCI.isBeforeLegalize()) - return SDValue(); - - if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 && - SelectMsub(N, &DAG)) - return SDValue(N, 0); - - return SDValue(); -} - -static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG, +static SDValue performDivRemCombine(SDNode *N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const MipsSubtarget *Subtarget) { if (DCI.isBeforeLegalizeOps()) @@ -746,18 +438,18 @@ static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG, EVT Ty = N->getValueType(0); unsigned LO = (Ty == MVT::i32) ? Mips::LO : Mips::LO64; unsigned HI = (Ty == MVT::i32) ? Mips::HI : Mips::HI64; - unsigned opc = N->getOpcode() == ISD::SDIVREM ? MipsISD::DivRem : - MipsISD::DivRemU; - DebugLoc dl = N->getDebugLoc(); + unsigned Opc = N->getOpcode() == ISD::SDIVREM ? MipsISD::DivRem16 : + MipsISD::DivRemU16; + DebugLoc DL = N->getDebugLoc(); - SDValue DivRem = DAG.getNode(opc, dl, MVT::Glue, + SDValue DivRem = DAG.getNode(Opc, DL, MVT::Glue, N->getOperand(0), N->getOperand(1)); SDValue InChain = DAG.getEntryNode(); SDValue InGlue = DivRem; // insert MFLO if (N->hasAnyUseOfValue(0)) { - SDValue CopyFromLo = DAG.getCopyFromReg(InChain, dl, LO, Ty, + SDValue CopyFromLo = DAG.getCopyFromReg(InChain, DL, LO, Ty, InGlue); DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), CopyFromLo); InChain = CopyFromLo.getValue(1); @@ -766,7 +458,7 @@ static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG, // insert MFHI if (N->hasAnyUseOfValue(1)) { - SDValue CopyFromHi = DAG.getCopyFromReg(InChain, dl, + SDValue CopyFromHi = DAG.getCopyFromReg(InChain, DL, HI, Ty, InGlue); DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), CopyFromHi); } @@ -774,7 +466,7 @@ static SDValue PerformDivRemCombine(SDNode *N, SelectionDAG &DAG, return SDValue(); } -static Mips::CondCode FPCondCCodeToFCC(ISD::CondCode CC) { +static Mips::CondCode condCodeToFCC(ISD::CondCode CC) { switch (CC) { default: llvm_unreachable("Unknown fp condition code!"); case ISD::SETEQ: @@ -801,8 +493,9 @@ static Mips::CondCode FPCondCCodeToFCC(ISD::CondCode CC) { } -// Returns true if condition code has to be inverted. -static bool InvertFPCondCode(Mips::CondCode CC) { +/// This function returns true if the floating point conditional branches and +/// conditional moves which use condition code CC should be inverted. +static bool invertFPCondCodeUser(Mips::CondCode CC) { if (CC >= Mips::FCOND_F && CC <= Mips::FCOND_NGT) return false; @@ -814,7 +507,7 @@ static bool InvertFPCondCode(Mips::CondCode CC) { // Creates and returns an FPCmp node from a setcc node. // Returns Op if setcc is not a floating point comparison. -static SDValue CreateFPCmp(SelectionDAG &DAG, const SDValue &Op) { +static SDValue createFPCmp(SelectionDAG &DAG, const SDValue &Op) { // must be a SETCC node if (Op.getOpcode() != ISD::SETCC) return Op; @@ -825,28 +518,27 @@ static SDValue CreateFPCmp(SelectionDAG &DAG, const SDValue &Op) { return Op; SDValue RHS = Op.getOperand(1); - DebugLoc dl = Op.getDebugLoc(); + DebugLoc DL = Op.getDebugLoc(); // Assume the 3rd operand is a CondCodeSDNode. Add code to check the type of // node if necessary. ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get(); - return DAG.getNode(MipsISD::FPCmp, dl, MVT::Glue, LHS, RHS, - DAG.getConstant(FPCondCCodeToFCC(CC), MVT::i32)); + return DAG.getNode(MipsISD::FPCmp, DL, MVT::Glue, LHS, RHS, + DAG.getConstant(condCodeToFCC(CC), MVT::i32)); } // Creates and returns a CMovFPT/F node. -static SDValue CreateCMovFP(SelectionDAG &DAG, SDValue Cond, SDValue True, +static SDValue createCMovFP(SelectionDAG &DAG, SDValue Cond, SDValue True, SDValue False, DebugLoc DL) { - bool invert = InvertFPCondCode((Mips::CondCode) - cast<ConstantSDNode>(Cond.getOperand(2)) - ->getSExtValue()); + ConstantSDNode *CC = cast<ConstantSDNode>(Cond.getOperand(2)); + bool invert = invertFPCondCodeUser((Mips::CondCode)CC->getSExtValue()); return DAG.getNode((invert ? MipsISD::CMovFP_F : MipsISD::CMovFP_T), DL, True.getValueType(), True, False, Cond); } -static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG, +static SDValue performSELECTCombine(SDNode *N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const MipsSubtarget *Subtarget) { if (DCI.isBeforeLegalizeOps()) @@ -879,7 +571,7 @@ static SDValue PerformSELECTCombine(SDNode *N, SelectionDAG &DAG, return DAG.getNode(ISD::SELECT, DL, FalseTy, SetCC, False, True); } -static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG, +static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const MipsSubtarget *Subtarget) { // Pattern match EXT. @@ -905,7 +597,7 @@ static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG, // Op's second operand must be a shifted mask. if (!(CN = dyn_cast<ConstantSDNode>(Mask)) || - !IsShiftedMask(CN->getZExtValue(), SMPos, SMSize)) + !isShiftedMask(CN->getZExtValue(), SMPos, SMSize)) return SDValue(); // Return if the shifted mask does not start at bit 0 or the sum of its size @@ -919,7 +611,7 @@ static SDValue PerformANDCombine(SDNode *N, SelectionDAG &DAG, DAG.getConstant(SMSize, MVT::i32)); } -static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG, +static SDValue performORCombine(SDNode *N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const MipsSubtarget *Subtarget) { // Pattern match INS. @@ -938,7 +630,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG, return SDValue(); if (!(CN = dyn_cast<ConstantSDNode>(And0.getOperand(1))) || - !IsShiftedMask(~CN->getSExtValue(), SMPos0, SMSize0)) + !isShiftedMask(~CN->getSExtValue(), SMPos0, SMSize0)) return SDValue(); // See if Op's second operand matches (and (shl $src, pos), mask1). @@ -946,7 +638,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG, return SDValue(); if (!(CN = dyn_cast<ConstantSDNode>(And1.getOperand(1))) || - !IsShiftedMask(CN->getZExtValue(), SMPos1, SMSize1)) + !isShiftedMask(CN->getZExtValue(), SMPos1, SMSize1)) return SDValue(); // The shift masks must have the same position and size. @@ -973,7 +665,7 @@ static SDValue PerformORCombine(SDNode *N, SelectionDAG &DAG, DAG.getConstant(SMSize0, MVT::i32), And0.getOperand(0)); } -static SDValue PerformADDCombine(SDNode *N, SelectionDAG &DAG, +static SDValue performADDCombine(SDNode *N, SelectionDAG &DAG, TargetLowering::DAGCombinerInfo &DCI, const MipsSubtarget *Subtarget) { // (add v0, (add v1, abs_lo(tjt))) => (add (add v0, v1), abs_lo(tjt)) @@ -1003,25 +695,21 @@ static SDValue PerformADDCombine(SDNode *N, SelectionDAG &DAG, SDValue MipsTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { SelectionDAG &DAG = DCI.DAG; - unsigned opc = N->getOpcode(); + unsigned Opc = N->getOpcode(); - switch (opc) { + switch (Opc) { default: break; - case ISD::ADDE: - return PerformADDECombine(N, DAG, DCI, Subtarget); - case ISD::SUBE: - return PerformSUBECombine(N, DAG, DCI, Subtarget); case ISD::SDIVREM: case ISD::UDIVREM: - return PerformDivRemCombine(N, DAG, DCI, Subtarget); + return performDivRemCombine(N, DAG, DCI, Subtarget); case ISD::SELECT: - return PerformSELECTCombine(N, DAG, DCI, Subtarget); + return performSELECTCombine(N, DAG, DCI, Subtarget); case ISD::AND: - return PerformANDCombine(N, DAG, DCI, Subtarget); + return performANDCombine(N, DAG, DCI, Subtarget); case ISD::OR: - return PerformORCombine(N, DAG, DCI, Subtarget); + return performORCombine(N, DAG, DCI, Subtarget); case ISD::ADD: - return PerformADDCombine(N, DAG, DCI, Subtarget); + return performADDCombine(N, DAG, DCI, Subtarget); } return SDValue(); @@ -1041,10 +729,7 @@ void MipsTargetLowering::ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results, SelectionDAG &DAG) const { - SDValue Res = LowerOperation(SDValue(N, 0), DAG); - - for (unsigned I = 0, E = Res->getNumValues(); I != E; ++I) - Results.push_back(Res.getValue(I)); + return LowerOperationWrapper(N, Results, DAG); } SDValue MipsTargetLowering:: @@ -1052,35 +737,29 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const { switch (Op.getOpcode()) { - case ISD::BRCOND: return LowerBRCOND(Op, DAG); - case ISD::ConstantPool: return LowerConstantPool(Op, DAG); - case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG); - case ISD::BlockAddress: return LowerBlockAddress(Op, DAG); - case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG); - case ISD::JumpTable: return LowerJumpTable(Op, DAG); - case ISD::SELECT: return LowerSELECT(Op, DAG); - case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG); - case ISD::SETCC: return LowerSETCC(Op, DAG); - case ISD::VASTART: return LowerVASTART(Op, DAG); - case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG); - case ISD::FABS: return LowerFABS(Op, DAG); - case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG); - case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG); - case ISD::EH_RETURN: return LowerEH_RETURN(Op, DAG); - case ISD::MEMBARRIER: return LowerMEMBARRIER(Op, DAG); - case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, DAG); - case ISD::SHL_PARTS: return LowerShiftLeftParts(Op, DAG); - case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG, true); - case ISD::SRL_PARTS: return LowerShiftRightParts(Op, DAG, false); - case ISD::LOAD: return LowerLOAD(Op, DAG); - case ISD::STORE: return LowerSTORE(Op, DAG); - case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG); - case ISD::INTRINSIC_W_CHAIN: return LowerINTRINSIC_W_CHAIN(Op, DAG); - case ISD::ADD: return LowerADD(Op, DAG); - // @LOCALMOD-BEGIN - case ISD::NACL_TP_TLS_OFFSET: return LowerNaClTpTlsOffset(Op, DAG); - case ISD::NACL_TP_TDB_OFFSET: return LowerNaClTpTdbOffset(Op, DAG); - // @LOCALMOD-END + case ISD::BR_JT: return lowerBR_JT(Op, DAG); + case ISD::BRCOND: return lowerBRCOND(Op, DAG); + case ISD::ConstantPool: return lowerConstantPool(Op, DAG); + case ISD::GlobalAddress: return lowerGlobalAddress(Op, DAG); + case ISD::BlockAddress: return lowerBlockAddress(Op, DAG); + case ISD::GlobalTLSAddress: return lowerGlobalTLSAddress(Op, DAG); + case ISD::JumpTable: return lowerJumpTable(Op, DAG); + case ISD::SELECT: return lowerSELECT(Op, DAG); + case ISD::SELECT_CC: return lowerSELECT_CC(Op, DAG); + case ISD::SETCC: return lowerSETCC(Op, DAG); + case ISD::VASTART: return lowerVASTART(Op, DAG); + case ISD::FCOPYSIGN: return lowerFCOPYSIGN(Op, DAG); + case ISD::FABS: return lowerFABS(Op, DAG); + case ISD::FRAMEADDR: return lowerFRAMEADDR(Op, DAG); + case ISD::RETURNADDR: return lowerRETURNADDR(Op, DAG); + case ISD::EH_RETURN: return lowerEH_RETURN(Op, DAG); + case ISD::ATOMIC_FENCE: return lowerATOMIC_FENCE(Op, DAG); + case ISD::SHL_PARTS: return lowerShiftLeftParts(Op, DAG); + case ISD::SRA_PARTS: return lowerShiftRightParts(Op, DAG, true); + case ISD::SRL_PARTS: return lowerShiftRightParts(Op, DAG, false); + case ISD::LOAD: return lowerLOAD(Op, DAG); + case ISD::STORE: return lowerSTORE(Op, DAG); + case ISD::ADD: return lowerADD(Op, DAG); } return SDValue(); } @@ -1089,447 +768,17 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const // Lower helper functions //===----------------------------------------------------------------------===// -// AddLiveIn - This helper function adds the specified physical register to the +// addLiveIn - This helper function adds the specified physical register to the // MachineFunction as a live in value. It also creates a corresponding // virtual register for it. static unsigned -AddLiveIn(MachineFunction &MF, unsigned PReg, const TargetRegisterClass *RC) +addLiveIn(MachineFunction &MF, unsigned PReg, const TargetRegisterClass *RC) { unsigned VReg = MF.getRegInfo().createVirtualRegister(RC); MF.getRegInfo().addLiveIn(PReg, VReg); return VReg; } -// Get fp branch code (not opcode) from condition code. -static Mips::FPBranchCode GetFPBranchCodeFromCond(Mips::CondCode CC) { - if (CC >= Mips::FCOND_F && CC <= Mips::FCOND_NGT) - return Mips::BRANCH_T; - - assert((CC >= Mips::FCOND_T && CC <= Mips::FCOND_GT) && - "Invalid CondCode."); - - return Mips::BRANCH_F; -} - -/* -static MachineBasicBlock* ExpandCondMov(MachineInstr *MI, MachineBasicBlock *BB, - DebugLoc dl, - const MipsSubtarget *Subtarget, - const TargetInstrInfo *TII, - bool isFPCmp, unsigned Opc) { - // There is no need to expand CMov instructions if target has - // conditional moves. - if (Subtarget->hasCondMov()) - return BB; - - // To "insert" a SELECT_CC instruction, we actually have to insert the - // diamond control-flow pattern. The incoming instruction knows the - // destination vreg to set, the condition code register to branch on, the - // true/false values to select between, and a branch opcode to use. - const BasicBlock *LLVM_BB = BB->getBasicBlock(); - MachineFunction::iterator It = BB; - ++It; - - // thisMBB: - // ... - // TrueVal = ... - // setcc r1, r2, r3 - // bNE r1, r0, copy1MBB - // fallthrough --> copy0MBB - MachineBasicBlock *thisMBB = BB; - MachineFunction *F = BB->getParent(); - MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); - F->insert(It, copy0MBB); - F->insert(It, sinkMBB); - - // Transfer the remainder of BB and its successor edges to sinkMBB. - sinkMBB->splice(sinkMBB->begin(), BB, - llvm::next(MachineBasicBlock::iterator(MI)), - BB->end()); - sinkMBB->transferSuccessorsAndUpdatePHIs(BB); - - // Next, add the true and fallthrough blocks as its successors. - BB->addSuccessor(copy0MBB); - BB->addSuccessor(sinkMBB); - - // Emit the right instruction according to the type of the operands compared - if (isFPCmp) - BuildMI(BB, dl, TII->get(Opc)).addMBB(sinkMBB); - else - BuildMI(BB, dl, TII->get(Opc)).addReg(MI->getOperand(2).getReg()) - .addReg(Mips::ZERO).addMBB(sinkMBB); - - // copy0MBB: - // %FalseValue = ... - // # fallthrough to sinkMBB - BB = copy0MBB; - - // Update machine-CFG edges - BB->addSuccessor(sinkMBB); - - // sinkMBB: - // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] - // ... - BB = sinkMBB; - - if (isFPCmp) - BuildMI(*BB, BB->begin(), dl, - TII->get(Mips::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB) - .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB); - else - BuildMI(*BB, BB->begin(), dl, - TII->get(Mips::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(3).getReg()).addMBB(thisMBB) - .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB); - - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; -} -*/ - -MachineBasicBlock * -MipsTargetLowering::EmitBPOSGE32(MachineInstr *MI, MachineBasicBlock *BB) const{ - // $bb: - // bposge32_pseudo $vr0 - // => - // $bb: - // bposge32 $tbb - // $fbb: - // li $vr2, 0 - // b $sink - // $tbb: - // li $vr1, 1 - // $sink: - // $vr0 = phi($vr2, $fbb, $vr1, $tbb) - - MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo(); - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - const TargetRegisterClass *RC = &Mips::CPURegsRegClass; - DebugLoc DL = MI->getDebugLoc(); - const BasicBlock *LLVM_BB = BB->getBasicBlock(); - MachineFunction::iterator It = llvm::next(MachineFunction::iterator(BB)); - MachineFunction *F = BB->getParent(); - MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *Sink = F->CreateMachineBasicBlock(LLVM_BB); - F->insert(It, FBB); - F->insert(It, TBB); - F->insert(It, Sink); - - // Transfer the remainder of BB and its successor edges to Sink. - Sink->splice(Sink->begin(), BB, llvm::next(MachineBasicBlock::iterator(MI)), - BB->end()); - Sink->transferSuccessorsAndUpdatePHIs(BB); - - // Add successors. - BB->addSuccessor(FBB); - BB->addSuccessor(TBB); - FBB->addSuccessor(Sink); - TBB->addSuccessor(Sink); - - // Insert the real bposge32 instruction to $BB. - BuildMI(BB, DL, TII->get(Mips::BPOSGE32)).addMBB(TBB); - - // Fill $FBB. - unsigned VR2 = RegInfo.createVirtualRegister(RC); - BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), VR2) - .addReg(Mips::ZERO).addImm(0); - BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink); - - // Fill $TBB. - unsigned VR1 = RegInfo.createVirtualRegister(RC); - BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), VR1) - .addReg(Mips::ZERO).addImm(1); - - // Insert phi function to $Sink. - BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI), - MI->getOperand(0).getReg()) - .addReg(VR2).addMBB(FBB).addReg(VR1).addMBB(TBB); - - MI->eraseFromParent(); // The pseudo instruction is gone now. - return Sink; -} - -MachineBasicBlock *MipsTargetLowering::EmitSel16(unsigned Opc, MachineInstr *MI, - MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); - // To "insert" a SELECT_CC instruction, we actually have to insert the - // diamond control-flow pattern. The incoming instruction knows the - // destination vreg to set, the condition code register to branch on, the - // true/false values to select between, and a branch opcode to use. - const BasicBlock *LLVM_BB = BB->getBasicBlock(); - MachineFunction::iterator It = BB; - ++It; - - // thisMBB: - // ... - // TrueVal = ... - // setcc r1, r2, r3 - // bNE r1, r0, copy1MBB - // fallthrough --> copy0MBB - MachineBasicBlock *thisMBB = BB; - MachineFunction *F = BB->getParent(); - MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); - F->insert(It, copy0MBB); - F->insert(It, sinkMBB); - - // Transfer the remainder of BB and its successor edges to sinkMBB. - sinkMBB->splice(sinkMBB->begin(), BB, - llvm::next(MachineBasicBlock::iterator(MI)), - BB->end()); - sinkMBB->transferSuccessorsAndUpdatePHIs(BB); - - // Next, add the true and fallthrough blocks as its successors. - BB->addSuccessor(copy0MBB); - BB->addSuccessor(sinkMBB); - - BuildMI(BB, dl, TII->get(Opc)).addReg(MI->getOperand(3).getReg()) - .addMBB(sinkMBB); - - // copy0MBB: - // %FalseValue = ... - // # fallthrough to sinkMBB - BB = copy0MBB; - - // Update machine-CFG edges - BB->addSuccessor(sinkMBB); - - // sinkMBB: - // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] - // ... - BB = sinkMBB; - - BuildMI(*BB, BB->begin(), dl, - TII->get(Mips::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) - .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); - - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; -} - -MachineBasicBlock *MipsTargetLowering::EmitSelT16 - (unsigned Opc1, unsigned Opc2, - MachineInstr *MI, MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); - // To "insert" a SELECT_CC instruction, we actually have to insert the - // diamond control-flow pattern. The incoming instruction knows the - // destination vreg to set, the condition code register to branch on, the - // true/false values to select between, and a branch opcode to use. - const BasicBlock *LLVM_BB = BB->getBasicBlock(); - MachineFunction::iterator It = BB; - ++It; - - // thisMBB: - // ... - // TrueVal = ... - // setcc r1, r2, r3 - // bNE r1, r0, copy1MBB - // fallthrough --> copy0MBB - MachineBasicBlock *thisMBB = BB; - MachineFunction *F = BB->getParent(); - MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); - F->insert(It, copy0MBB); - F->insert(It, sinkMBB); - - // Transfer the remainder of BB and its successor edges to sinkMBB. - sinkMBB->splice(sinkMBB->begin(), BB, - llvm::next(MachineBasicBlock::iterator(MI)), - BB->end()); - sinkMBB->transferSuccessorsAndUpdatePHIs(BB); - - // Next, add the true and fallthrough blocks as its successors. - BB->addSuccessor(copy0MBB); - BB->addSuccessor(sinkMBB); - - BuildMI(BB, dl, TII->get(Opc2)).addReg(MI->getOperand(3).getReg()) - .addReg(MI->getOperand(4).getReg()); - BuildMI(BB, dl, TII->get(Opc1)).addMBB(sinkMBB); - - // copy0MBB: - // %FalseValue = ... - // # fallthrough to sinkMBB - BB = copy0MBB; - - // Update machine-CFG edges - BB->addSuccessor(sinkMBB); - - // sinkMBB: - // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] - // ... - BB = sinkMBB; - - BuildMI(*BB, BB->begin(), dl, - TII->get(Mips::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) - .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); - - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; - -} - - -MachineBasicBlock *MipsTargetLowering::EmitSeliT16 - (unsigned Opc1, unsigned Opc2, - MachineInstr *MI, MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); - // To "insert" a SELECT_CC instruction, we actually have to insert the - // diamond control-flow pattern. The incoming instruction knows the - // destination vreg to set, the condition code register to branch on, the - // true/false values to select between, and a branch opcode to use. - const BasicBlock *LLVM_BB = BB->getBasicBlock(); - MachineFunction::iterator It = BB; - ++It; - - // thisMBB: - // ... - // TrueVal = ... - // setcc r1, r2, r3 - // bNE r1, r0, copy1MBB - // fallthrough --> copy0MBB - MachineBasicBlock *thisMBB = BB; - MachineFunction *F = BB->getParent(); - MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB); - MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB); - F->insert(It, copy0MBB); - F->insert(It, sinkMBB); - - // Transfer the remainder of BB and its successor edges to sinkMBB. - sinkMBB->splice(sinkMBB->begin(), BB, - llvm::next(MachineBasicBlock::iterator(MI)), - BB->end()); - sinkMBB->transferSuccessorsAndUpdatePHIs(BB); - - // Next, add the true and fallthrough blocks as its successors. - BB->addSuccessor(copy0MBB); - BB->addSuccessor(sinkMBB); - - BuildMI(BB, dl, TII->get(Opc2)).addReg(MI->getOperand(3).getReg()) - .addImm(MI->getOperand(4).getImm()); - BuildMI(BB, dl, TII->get(Opc1)).addMBB(sinkMBB); - - // copy0MBB: - // %FalseValue = ... - // # fallthrough to sinkMBB - BB = copy0MBB; - - // Update machine-CFG edges - BB->addSuccessor(sinkMBB); - - // sinkMBB: - // %Result = phi [ %TrueValue, thisMBB ], [ %FalseValue, copy0MBB ] - // ... - BB = sinkMBB; - - BuildMI(*BB, BB->begin(), dl, - TII->get(Mips::PHI), MI->getOperand(0).getReg()) - .addReg(MI->getOperand(1).getReg()).addMBB(thisMBB) - .addReg(MI->getOperand(2).getReg()).addMBB(copy0MBB); - - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; - -} - - -MachineBasicBlock - *MipsTargetLowering::EmitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc, - MachineInstr *MI, - MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - unsigned regX = MI->getOperand(0).getReg(); - unsigned regY = MI->getOperand(1).getReg(); - MachineBasicBlock *target = MI->getOperand(2).getMBB(); - BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX).addReg(regY); - BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target); - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; -} - - -MachineBasicBlock *MipsTargetLowering::EmitFEXT_T8I8I16_ins( - unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc, - MachineInstr *MI, MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - unsigned regX = MI->getOperand(0).getReg(); - int64_t imm = MI->getOperand(1).getImm(); - MachineBasicBlock *target = MI->getOperand(2).getMBB(); - unsigned CmpOpc; - if (isUInt<8>(imm)) - CmpOpc = CmpiOpc; - else if (isUInt<16>(imm)) - CmpOpc = CmpiXOpc; - else - llvm_unreachable("immediate field not usable"); - BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(CmpOpc)).addReg(regX).addImm(imm); - BuildMI(*BB, MI, MI->getDebugLoc(), TII->get(BtOpc)).addMBB(target); - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; -} - - -static unsigned Mips16WhichOp8uOr16simm - (unsigned shortOp, unsigned longOp, int64_t Imm) { - if (isUInt<8>(Imm)) - return shortOp; - else if (isInt<16>(Imm)) - return longOp; - else - llvm_unreachable("immediate field not usable"); -} - -MachineBasicBlock *MipsTargetLowering::EmitFEXT_CCRX16_ins( - unsigned SltOpc, - MachineInstr *MI, MachineBasicBlock *BB) const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - unsigned CC = MI->getOperand(0).getReg(); - unsigned regX = MI->getOperand(1).getReg(); - unsigned regY = MI->getOperand(2).getReg(); - BuildMI(*BB, MI, MI->getDebugLoc(), - TII->get(SltOpc)).addReg(regX).addReg(regY); - BuildMI(*BB, MI, MI->getDebugLoc(), - TII->get(Mips::MoveR3216), CC).addReg(Mips::T8); - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; -} -MachineBasicBlock *MipsTargetLowering::EmitFEXT_CCRXI16_ins( - unsigned SltiOpc, unsigned SltiXOpc, - MachineInstr *MI, MachineBasicBlock *BB )const { - if (DontExpandCondPseudos16) - return BB; - const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - unsigned CC = MI->getOperand(0).getReg(); - unsigned regX = MI->getOperand(1).getReg(); - int64_t Imm = MI->getOperand(2).getImm(); - unsigned SltOpc = Mips16WhichOp8uOr16simm(SltiOpc, SltiXOpc, Imm); - BuildMI(*BB, MI, MI->getDebugLoc(), - TII->get(SltOpc)).addReg(regX).addImm(Imm); - BuildMI(*BB, MI, MI->getDebugLoc(), - TII->get(Mips::MoveR3216), CC).addReg(Mips::T8); - MI->eraseFromParent(); // The pseudo instruction is gone now. - return BB; - -} MachineBasicBlock * MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *BB) const { @@ -1538,185 +787,114 @@ MipsTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, llvm_unreachable("Unexpected instr type to insert"); case Mips::ATOMIC_LOAD_ADD_I8: case Mips::ATOMIC_LOAD_ADD_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, Mips::ADDu); + return emitAtomicBinaryPartword(MI, BB, 1, Mips::ADDu); case Mips::ATOMIC_LOAD_ADD_I16: case Mips::ATOMIC_LOAD_ADD_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, Mips::ADDu); + return emitAtomicBinaryPartword(MI, BB, 2, Mips::ADDu); case Mips::ATOMIC_LOAD_ADD_I32: case Mips::ATOMIC_LOAD_ADD_I32_P8: - return EmitAtomicBinary(MI, BB, 4, Mips::ADDu); + return emitAtomicBinary(MI, BB, 4, Mips::ADDu); case Mips::ATOMIC_LOAD_ADD_I64: case Mips::ATOMIC_LOAD_ADD_I64_P8: - return EmitAtomicBinary(MI, BB, 8, Mips::DADDu); + return emitAtomicBinary(MI, BB, 8, Mips::DADDu); case Mips::ATOMIC_LOAD_AND_I8: case Mips::ATOMIC_LOAD_AND_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, Mips::AND); + return emitAtomicBinaryPartword(MI, BB, 1, Mips::AND); case Mips::ATOMIC_LOAD_AND_I16: case Mips::ATOMIC_LOAD_AND_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, Mips::AND); + return emitAtomicBinaryPartword(MI, BB, 2, Mips::AND); case Mips::ATOMIC_LOAD_AND_I32: case Mips::ATOMIC_LOAD_AND_I32_P8: - return EmitAtomicBinary(MI, BB, 4, Mips::AND); + return emitAtomicBinary(MI, BB, 4, Mips::AND); case Mips::ATOMIC_LOAD_AND_I64: case Mips::ATOMIC_LOAD_AND_I64_P8: - return EmitAtomicBinary(MI, BB, 8, Mips::AND64); + return emitAtomicBinary(MI, BB, 8, Mips::AND64); case Mips::ATOMIC_LOAD_OR_I8: case Mips::ATOMIC_LOAD_OR_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, Mips::OR); + return emitAtomicBinaryPartword(MI, BB, 1, Mips::OR); case Mips::ATOMIC_LOAD_OR_I16: case Mips::ATOMIC_LOAD_OR_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, Mips::OR); + return emitAtomicBinaryPartword(MI, BB, 2, Mips::OR); case Mips::ATOMIC_LOAD_OR_I32: case Mips::ATOMIC_LOAD_OR_I32_P8: - return EmitAtomicBinary(MI, BB, 4, Mips::OR); + return emitAtomicBinary(MI, BB, 4, Mips::OR); case Mips::ATOMIC_LOAD_OR_I64: case Mips::ATOMIC_LOAD_OR_I64_P8: - return EmitAtomicBinary(MI, BB, 8, Mips::OR64); + return emitAtomicBinary(MI, BB, 8, Mips::OR64); case Mips::ATOMIC_LOAD_XOR_I8: case Mips::ATOMIC_LOAD_XOR_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, Mips::XOR); + return emitAtomicBinaryPartword(MI, BB, 1, Mips::XOR); case Mips::ATOMIC_LOAD_XOR_I16: case Mips::ATOMIC_LOAD_XOR_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, Mips::XOR); + return emitAtomicBinaryPartword(MI, BB, 2, Mips::XOR); case Mips::ATOMIC_LOAD_XOR_I32: case Mips::ATOMIC_LOAD_XOR_I32_P8: - return EmitAtomicBinary(MI, BB, 4, Mips::XOR); + return emitAtomicBinary(MI, BB, 4, Mips::XOR); case Mips::ATOMIC_LOAD_XOR_I64: case Mips::ATOMIC_LOAD_XOR_I64_P8: - return EmitAtomicBinary(MI, BB, 8, Mips::XOR64); + return emitAtomicBinary(MI, BB, 8, Mips::XOR64); case Mips::ATOMIC_LOAD_NAND_I8: case Mips::ATOMIC_LOAD_NAND_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, 0, true); + return emitAtomicBinaryPartword(MI, BB, 1, 0, true); case Mips::ATOMIC_LOAD_NAND_I16: case Mips::ATOMIC_LOAD_NAND_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, 0, true); + return emitAtomicBinaryPartword(MI, BB, 2, 0, true); case Mips::ATOMIC_LOAD_NAND_I32: case Mips::ATOMIC_LOAD_NAND_I32_P8: - return EmitAtomicBinary(MI, BB, 4, 0, true); + return emitAtomicBinary(MI, BB, 4, 0, true); case Mips::ATOMIC_LOAD_NAND_I64: case Mips::ATOMIC_LOAD_NAND_I64_P8: - return EmitAtomicBinary(MI, BB, 8, 0, true); + return emitAtomicBinary(MI, BB, 8, 0, true); case Mips::ATOMIC_LOAD_SUB_I8: case Mips::ATOMIC_LOAD_SUB_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, Mips::SUBu); + return emitAtomicBinaryPartword(MI, BB, 1, Mips::SUBu); case Mips::ATOMIC_LOAD_SUB_I16: case Mips::ATOMIC_LOAD_SUB_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, Mips::SUBu); + return emitAtomicBinaryPartword(MI, BB, 2, Mips::SUBu); case Mips::ATOMIC_LOAD_SUB_I32: case Mips::ATOMIC_LOAD_SUB_I32_P8: - return EmitAtomicBinary(MI, BB, 4, Mips::SUBu); + return emitAtomicBinary(MI, BB, 4, Mips::SUBu); case Mips::ATOMIC_LOAD_SUB_I64: case Mips::ATOMIC_LOAD_SUB_I64_P8: - return EmitAtomicBinary(MI, BB, 8, Mips::DSUBu); + return emitAtomicBinary(MI, BB, 8, Mips::DSUBu); case Mips::ATOMIC_SWAP_I8: case Mips::ATOMIC_SWAP_I8_P8: - return EmitAtomicBinaryPartword(MI, BB, 1, 0); + return emitAtomicBinaryPartword(MI, BB, 1, 0); case Mips::ATOMIC_SWAP_I16: case Mips::ATOMIC_SWAP_I16_P8: - return EmitAtomicBinaryPartword(MI, BB, 2, 0); + return emitAtomicBinaryPartword(MI, BB, 2, 0); case Mips::ATOMIC_SWAP_I32: case Mips::ATOMIC_SWAP_I32_P8: - return EmitAtomicBinary(MI, BB, 4, 0); + return emitAtomicBinary(MI, BB, 4, 0); case Mips::ATOMIC_SWAP_I64: case Mips::ATOMIC_SWAP_I64_P8: - return EmitAtomicBinary(MI, BB, 8, 0); + return emitAtomicBinary(MI, BB, 8, 0); case Mips::ATOMIC_CMP_SWAP_I8: case Mips::ATOMIC_CMP_SWAP_I8_P8: - return EmitAtomicCmpSwapPartword(MI, BB, 1); + return emitAtomicCmpSwapPartword(MI, BB, 1); case Mips::ATOMIC_CMP_SWAP_I16: case Mips::ATOMIC_CMP_SWAP_I16_P8: - return EmitAtomicCmpSwapPartword(MI, BB, 2); + return emitAtomicCmpSwapPartword(MI, BB, 2); case Mips::ATOMIC_CMP_SWAP_I32: case Mips::ATOMIC_CMP_SWAP_I32_P8: - return EmitAtomicCmpSwap(MI, BB, 4); + return emitAtomicCmpSwap(MI, BB, 4); case Mips::ATOMIC_CMP_SWAP_I64: case Mips::ATOMIC_CMP_SWAP_I64_P8: - return EmitAtomicCmpSwap(MI, BB, 8); - case Mips::BPOSGE32_PSEUDO: - return EmitBPOSGE32(MI, BB); - case Mips::SelBeqZ: - return EmitSel16(Mips::BeqzRxImm16, MI, BB); - case Mips::SelBneZ: - return EmitSel16(Mips::BnezRxImm16, MI, BB); - case Mips::SelTBteqZCmpi: - return EmitSeliT16(Mips::BteqzX16, Mips::CmpiRxImmX16, MI, BB); - case Mips::SelTBteqZSlti: - return EmitSeliT16(Mips::BteqzX16, Mips::SltiRxImmX16, MI, BB); - case Mips::SelTBteqZSltiu: - return EmitSeliT16(Mips::BteqzX16, Mips::SltiuRxImmX16, MI, BB); - case Mips::SelTBtneZCmpi: - return EmitSeliT16(Mips::BtnezX16, Mips::CmpiRxImmX16, MI, BB); - case Mips::SelTBtneZSlti: - return EmitSeliT16(Mips::BtnezX16, Mips::SltiRxImmX16, MI, BB); - case Mips::SelTBtneZSltiu: - return EmitSeliT16(Mips::BtnezX16, Mips::SltiuRxImmX16, MI, BB); - case Mips::SelTBteqZCmp: - return EmitSelT16(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB); - case Mips::SelTBteqZSlt: - return EmitSelT16(Mips::BteqzX16, Mips::SltRxRy16, MI, BB); - case Mips::SelTBteqZSltu: - return EmitSelT16(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB); - case Mips::SelTBtneZCmp: - return EmitSelT16(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB); - case Mips::SelTBtneZSlt: - return EmitSelT16(Mips::BtnezX16, Mips::SltRxRy16, MI, BB); - case Mips::SelTBtneZSltu: - return EmitSelT16(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB); - case Mips::BteqzT8CmpX16: - return EmitFEXT_T8I816_ins(Mips::BteqzX16, Mips::CmpRxRy16, MI, BB); - case Mips::BteqzT8SltX16: - return EmitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltRxRy16, MI, BB); - case Mips::BteqzT8SltuX16: - // TBD: figure out a way to get this or remove the instruction - // altogether. - return EmitFEXT_T8I816_ins(Mips::BteqzX16, Mips::SltuRxRy16, MI, BB); - case Mips::BtnezT8CmpX16: - return EmitFEXT_T8I816_ins(Mips::BtnezX16, Mips::CmpRxRy16, MI, BB); - case Mips::BtnezT8SltX16: - return EmitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltRxRy16, MI, BB); - case Mips::BtnezT8SltuX16: - // TBD: figure out a way to get this or remove the instruction - // altogether. - return EmitFEXT_T8I816_ins(Mips::BtnezX16, Mips::SltuRxRy16, MI, BB); - case Mips::BteqzT8CmpiX16: return EmitFEXT_T8I8I16_ins( - Mips::BteqzX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB); - case Mips::BteqzT8SltiX16: return EmitFEXT_T8I8I16_ins( - Mips::BteqzX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); - case Mips::BteqzT8SltiuX16: return EmitFEXT_T8I8I16_ins( - Mips::BteqzX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); - case Mips::BtnezT8CmpiX16: return EmitFEXT_T8I8I16_ins( - Mips::BtnezX16, Mips::CmpiRxImm16, Mips::CmpiRxImmX16, MI, BB); - case Mips::BtnezT8SltiX16: return EmitFEXT_T8I8I16_ins( - Mips::BtnezX16, Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); - case Mips::BtnezT8SltiuX16: return EmitFEXT_T8I8I16_ins( - Mips::BtnezX16, Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); - break; - case Mips::SltCCRxRy16: - return EmitFEXT_CCRX16_ins(Mips::SltRxRy16, MI, BB); - break; - case Mips::SltiCCRxImmX16: - return EmitFEXT_CCRXI16_ins - (Mips::SltiRxImm16, Mips::SltiRxImmX16, MI, BB); - case Mips::SltiuCCRxImmX16: - return EmitFEXT_CCRXI16_ins - (Mips::SltiuRxImm16, Mips::SltiuRxImmX16, MI, BB); - case Mips::SltuCCRxRy16: - return EmitFEXT_CCRX16_ins - (Mips::SltuRxRy16, MI, BB); + return emitAtomicCmpSwap(MI, BB, 8); } } // This function also handles Mips::ATOMIC_SWAP_I32 (when BinOpcode == 0), and // Mips::ATOMIC_LOAD_NAND_I32 (when Nand == true) MachineBasicBlock * -MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, +MipsTargetLowering::emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode, bool Nand) const { assert((Size == 4 || Size == 8) && "Unsupported size for EmitAtomicBinary."); @@ -1725,7 +903,7 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, MachineRegisterInfo &RegInfo = MF->getRegInfo(); const TargetRegisterClass *RC = getRegClassFor(MVT::getIntegerVT(Size * 8)); const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); + DebugLoc DL = MI->getDebugLoc(); unsigned LL, SC, AND, NOR, ZERO, BEQ; if (Size == 4) { @@ -1781,20 +959,20 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, // sc success, storeval, 0(ptr) // beq success, $0, loopMBB BB = loopMBB; - BuildMI(BB, dl, TII->get(LL), OldVal).addReg(Ptr).addImm(0); + BuildMI(BB, DL, TII->get(LL), OldVal).addReg(Ptr).addImm(0); if (Nand) { // and andres, oldval, incr // nor storeval, $0, andres - BuildMI(BB, dl, TII->get(AND), AndRes).addReg(OldVal).addReg(Incr); - BuildMI(BB, dl, TII->get(NOR), StoreVal).addReg(ZERO).addReg(AndRes); + BuildMI(BB, DL, TII->get(AND), AndRes).addReg(OldVal).addReg(Incr); + BuildMI(BB, DL, TII->get(NOR), StoreVal).addReg(ZERO).addReg(AndRes); } else if (BinOpcode) { // <binop> storeval, oldval, incr - BuildMI(BB, dl, TII->get(BinOpcode), StoreVal).addReg(OldVal).addReg(Incr); + BuildMI(BB, DL, TII->get(BinOpcode), StoreVal).addReg(OldVal).addReg(Incr); } else { StoreVal = Incr; } - BuildMI(BB, dl, TII->get(SC), Success).addReg(StoreVal).addReg(Ptr).addImm(0); - BuildMI(BB, dl, TII->get(BEQ)).addReg(Success).addReg(ZERO).addMBB(loopMBB); + BuildMI(BB, DL, TII->get(SC), Success).addReg(StoreVal).addReg(Ptr).addImm(0); + BuildMI(BB, DL, TII->get(BEQ)).addReg(Success).addReg(ZERO).addMBB(loopMBB); MI->eraseFromParent(); // The instruction is gone now. @@ -1802,7 +980,7 @@ MipsTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, } MachineBasicBlock * -MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, +MipsTargetLowering::emitAtomicBinaryPartword(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode, bool Nand) const { @@ -1813,7 +991,7 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, MachineRegisterInfo &RegInfo = MF->getRegInfo(); const TargetRegisterClass *RC = getRegClassFor(MVT::i32); const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); + DebugLoc DL = MI->getDebugLoc(); unsigned LL = IsN64 ? Mips::LL_P8 : Mips::LL; unsigned SC = IsN64 ? Mips::SC_P8 : Mips::SC; @@ -1872,18 +1050,18 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, // sll incr2,incr,shiftamt int64_t MaskImm = (Size == 1) ? 255 : 65535; - BuildMI(BB, dl, TII->get(Mips::ADDiu), MaskLSB2) + BuildMI(BB, DL, TII->get(Mips::ADDiu), MaskLSB2) .addReg(Mips::ZERO).addImm(-4); - BuildMI(BB, dl, TII->get(Mips::AND), AlignedAddr) + BuildMI(BB, DL, TII->get(Mips::AND), AlignedAddr) .addReg(Ptr).addReg(MaskLSB2); - BuildMI(BB, dl, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3); - BuildMI(BB, dl, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3); - BuildMI(BB, dl, TII->get(Mips::ORi), MaskUpper) + BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3); + BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3); + BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper) .addReg(Mips::ZERO).addImm(MaskImm); - BuildMI(BB, dl, TII->get(Mips::SLLV), Mask) + BuildMI(BB, DL, TII->get(Mips::SLLV), Mask) .addReg(ShiftAmt).addReg(MaskUpper); - BuildMI(BB, dl, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask); - BuildMI(BB, dl, TII->get(Mips::SLLV), Incr2).addReg(ShiftAmt).addReg(Incr); + BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask); + BuildMI(BB, DL, TII->get(Mips::SLLV), Incr2).addReg(ShiftAmt).addReg(Incr); // atomic.load.binop // loopMBB: @@ -1905,32 +1083,32 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, // beq success,$0,loopMBB BB = loopMBB; - BuildMI(BB, dl, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0); + BuildMI(BB, DL, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0); if (Nand) { // and andres, oldval, incr2 // nor binopres, $0, andres // and newval, binopres, mask - BuildMI(BB, dl, TII->get(Mips::AND), AndRes).addReg(OldVal).addReg(Incr2); - BuildMI(BB, dl, TII->get(Mips::NOR), BinOpRes) + BuildMI(BB, DL, TII->get(Mips::AND), AndRes).addReg(OldVal).addReg(Incr2); + BuildMI(BB, DL, TII->get(Mips::NOR), BinOpRes) .addReg(Mips::ZERO).addReg(AndRes); - BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask); + BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask); } else if (BinOpcode) { // <binop> binopres, oldval, incr2 // and newval, binopres, mask - BuildMI(BB, dl, TII->get(BinOpcode), BinOpRes).addReg(OldVal).addReg(Incr2); - BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask); + BuildMI(BB, DL, TII->get(BinOpcode), BinOpRes).addReg(OldVal).addReg(Incr2); + BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(BinOpRes).addReg(Mask); } else {// atomic.swap // and newval, incr2, mask - BuildMI(BB, dl, TII->get(Mips::AND), NewVal).addReg(Incr2).addReg(Mask); + BuildMI(BB, DL, TII->get(Mips::AND), NewVal).addReg(Incr2).addReg(Mask); } - BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal0) + BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal0) .addReg(OldVal).addReg(Mask2); - BuildMI(BB, dl, TII->get(Mips::OR), StoreVal) + BuildMI(BB, DL, TII->get(Mips::OR), StoreVal) .addReg(MaskedOldVal0).addReg(NewVal); - BuildMI(BB, dl, TII->get(SC), Success) + BuildMI(BB, DL, TII->get(SC), Success) .addReg(StoreVal).addReg(AlignedAddr).addImm(0); - BuildMI(BB, dl, TII->get(Mips::BEQ)) + BuildMI(BB, DL, TII->get(Mips::BEQ)) .addReg(Success).addReg(Mips::ZERO).addMBB(loopMBB); // sinkMBB: @@ -1941,13 +1119,13 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, BB = sinkMBB; int64_t ShiftImm = (Size == 1) ? 24 : 16; - BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal1) + BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal1) .addReg(OldVal).addReg(Mask); - BuildMI(BB, dl, TII->get(Mips::SRLV), SrlRes) + BuildMI(BB, DL, TII->get(Mips::SRLV), SrlRes) .addReg(ShiftAmt).addReg(MaskedOldVal1); - BuildMI(BB, dl, TII->get(Mips::SLL), SllRes) + BuildMI(BB, DL, TII->get(Mips::SLL), SllRes) .addReg(SrlRes).addImm(ShiftImm); - BuildMI(BB, dl, TII->get(Mips::SRA), Dest) + BuildMI(BB, DL, TII->get(Mips::SRA), Dest) .addReg(SllRes).addImm(ShiftImm); MI->eraseFromParent(); // The instruction is gone now. @@ -1956,7 +1134,7 @@ MipsTargetLowering::EmitAtomicBinaryPartword(MachineInstr *MI, } MachineBasicBlock * -MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, +MipsTargetLowering::emitAtomicCmpSwap(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size) const { assert((Size == 4 || Size == 8) && "Unsupported size for EmitAtomicCmpSwap."); @@ -1965,7 +1143,7 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, MachineRegisterInfo &RegInfo = MF->getRegInfo(); const TargetRegisterClass *RC = getRegClassFor(MVT::getIntegerVT(Size * 8)); const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); + DebugLoc DL = MI->getDebugLoc(); unsigned LL, SC, ZERO, BNE, BEQ; if (Size == 4) { @@ -2019,17 +1197,17 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, // ll dest, 0(ptr) // bne dest, oldval, exitMBB BB = loop1MBB; - BuildMI(BB, dl, TII->get(LL), Dest).addReg(Ptr).addImm(0); - BuildMI(BB, dl, TII->get(BNE)) + BuildMI(BB, DL, TII->get(LL), Dest).addReg(Ptr).addImm(0); + BuildMI(BB, DL, TII->get(BNE)) .addReg(Dest).addReg(OldVal).addMBB(exitMBB); // loop2MBB: // sc success, newval, 0(ptr) // beq success, $0, loop1MBB BB = loop2MBB; - BuildMI(BB, dl, TII->get(SC), Success) + BuildMI(BB, DL, TII->get(SC), Success) .addReg(NewVal).addReg(Ptr).addImm(0); - BuildMI(BB, dl, TII->get(BEQ)) + BuildMI(BB, DL, TII->get(BEQ)) .addReg(Success).addReg(ZERO).addMBB(loop1MBB); MI->eraseFromParent(); // The instruction is gone now. @@ -2038,7 +1216,7 @@ MipsTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, } MachineBasicBlock * -MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, +MipsTargetLowering::emitAtomicCmpSwapPartword(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size) const { assert((Size == 1 || Size == 2) && @@ -2048,7 +1226,7 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, MachineRegisterInfo &RegInfo = MF->getRegInfo(); const TargetRegisterClass *RC = getRegClassFor(MVT::i32); const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); - DebugLoc dl = MI->getDebugLoc(); + DebugLoc DL = MI->getDebugLoc(); unsigned LL = IsN64 ? Mips::LL_P8 : Mips::LL; unsigned SC = IsN64 ? Mips::SC_P8 : Mips::SC; @@ -2115,24 +1293,24 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, // andi maskednewval,newval,255 // sll shiftednewval,maskednewval,shiftamt int64_t MaskImm = (Size == 1) ? 255 : 65535; - BuildMI(BB, dl, TII->get(Mips::ADDiu), MaskLSB2) + BuildMI(BB, DL, TII->get(Mips::ADDiu), MaskLSB2) .addReg(Mips::ZERO).addImm(-4); - BuildMI(BB, dl, TII->get(Mips::AND), AlignedAddr) + BuildMI(BB, DL, TII->get(Mips::AND), AlignedAddr) .addReg(Ptr).addReg(MaskLSB2); - BuildMI(BB, dl, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3); - BuildMI(BB, dl, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3); - BuildMI(BB, dl, TII->get(Mips::ORi), MaskUpper) + BuildMI(BB, DL, TII->get(Mips::ANDi), PtrLSB2).addReg(Ptr).addImm(3); + BuildMI(BB, DL, TII->get(Mips::SLL), ShiftAmt).addReg(PtrLSB2).addImm(3); + BuildMI(BB, DL, TII->get(Mips::ORi), MaskUpper) .addReg(Mips::ZERO).addImm(MaskImm); - BuildMI(BB, dl, TII->get(Mips::SLLV), Mask) + BuildMI(BB, DL, TII->get(Mips::SLLV), Mask) .addReg(ShiftAmt).addReg(MaskUpper); - BuildMI(BB, dl, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask); - BuildMI(BB, dl, TII->get(Mips::ANDi), MaskedCmpVal) + BuildMI(BB, DL, TII->get(Mips::NOR), Mask2).addReg(Mips::ZERO).addReg(Mask); + BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedCmpVal) .addReg(CmpVal).addImm(MaskImm); - BuildMI(BB, dl, TII->get(Mips::SLLV), ShiftedCmpVal) + BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedCmpVal) .addReg(ShiftAmt).addReg(MaskedCmpVal); - BuildMI(BB, dl, TII->get(Mips::ANDi), MaskedNewVal) + BuildMI(BB, DL, TII->get(Mips::ANDi), MaskedNewVal) .addReg(NewVal).addImm(MaskImm); - BuildMI(BB, dl, TII->get(Mips::SLLV), ShiftedNewVal) + BuildMI(BB, DL, TII->get(Mips::SLLV), ShiftedNewVal) .addReg(ShiftAmt).addReg(MaskedNewVal); // loop1MBB: @@ -2140,10 +1318,10 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, // and maskedoldval0,oldval,mask // bne maskedoldval0,shiftedcmpval,sinkMBB BB = loop1MBB; - BuildMI(BB, dl, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0); - BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal0) + BuildMI(BB, DL, TII->get(LL), OldVal).addReg(AlignedAddr).addImm(0); + BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal0) .addReg(OldVal).addReg(Mask); - BuildMI(BB, dl, TII->get(Mips::BNE)) + BuildMI(BB, DL, TII->get(Mips::BNE)) .addReg(MaskedOldVal0).addReg(ShiftedCmpVal).addMBB(sinkMBB); // loop2MBB: @@ -2152,13 +1330,13 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, // sc success,storeval,0(alignedaddr) // beq success,$0,loop1MBB BB = loop2MBB; - BuildMI(BB, dl, TII->get(Mips::AND), MaskedOldVal1) + BuildMI(BB, DL, TII->get(Mips::AND), MaskedOldVal1) .addReg(OldVal).addReg(Mask2); - BuildMI(BB, dl, TII->get(Mips::OR), StoreVal) + BuildMI(BB, DL, TII->get(Mips::OR), StoreVal) .addReg(MaskedOldVal1).addReg(ShiftedNewVal); - BuildMI(BB, dl, TII->get(SC), Success) + BuildMI(BB, DL, TII->get(SC), Success) .addReg(StoreVal).addReg(AlignedAddr).addImm(0); - BuildMI(BB, dl, TII->get(Mips::BEQ)) + BuildMI(BB, DL, TII->get(Mips::BEQ)) .addReg(Success).addReg(Mips::ZERO).addMBB(loop1MBB); // sinkMBB: @@ -2168,11 +1346,11 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, BB = sinkMBB; int64_t ShiftImm = (Size == 1) ? 24 : 16; - BuildMI(BB, dl, TII->get(Mips::SRLV), SrlRes) + BuildMI(BB, DL, TII->get(Mips::SRLV), SrlRes) .addReg(ShiftAmt).addReg(MaskedOldVal0); - BuildMI(BB, dl, TII->get(Mips::SLL), SllRes) + BuildMI(BB, DL, TII->get(Mips::SLL), SllRes) .addReg(SrlRes).addImm(ShiftImm); - BuildMI(BB, dl, TII->get(Mips::SRA), Dest) + BuildMI(BB, DL, TII->get(Mips::SRA), Dest) .addReg(SllRes).addImm(ShiftImm); MI->eraseFromParent(); // The instruction is gone now. @@ -2183,16 +1361,46 @@ MipsTargetLowering::EmitAtomicCmpSwapPartword(MachineInstr *MI, //===----------------------------------------------------------------------===// // Misc Lower Operation implementation //===----------------------------------------------------------------------===// +SDValue MipsTargetLowering::lowerBR_JT(SDValue Op, SelectionDAG &DAG) const { + SDValue Chain = Op.getOperand(0); + SDValue Table = Op.getOperand(1); + SDValue Index = Op.getOperand(2); + DebugLoc DL = Op.getDebugLoc(); + EVT PTy = getPointerTy(); + unsigned EntrySize = + DAG.getMachineFunction().getJumpTableInfo()->getEntrySize(*getDataLayout()); + + Index = DAG.getNode(ISD::MUL, DL, PTy, Index, + DAG.getConstant(EntrySize, PTy)); + SDValue Addr = DAG.getNode(ISD::ADD, DL, PTy, Index, Table); + + EVT MemVT = EVT::getIntegerVT(*DAG.getContext(), EntrySize * 8); + Addr = DAG.getExtLoad(ISD::SEXTLOAD, DL, PTy, Chain, Addr, + MachinePointerInfo::getJumpTable(), MemVT, false, false, + 0); + Chain = Addr.getValue(1); + + if ((getTargetMachine().getRelocationModel() == Reloc::PIC_) || IsN64) { + // For PIC, the sequence is: + // BRIND(load(Jumptable + index) + RelocBase) + // RelocBase can be JumpTable, GOT or some sort of global base. + Addr = DAG.getNode(ISD::ADD, DL, PTy, Addr, + getPICJumpTableRelocBase(Table, DAG)); + } + + return DAG.getNode(ISD::BRIND, DL, MVT::Other, Chain, Addr); +} + SDValue MipsTargetLowering:: -LowerBRCOND(SDValue Op, SelectionDAG &DAG) const +lowerBRCOND(SDValue Op, SelectionDAG &DAG) const { // The first operand is the chain, the second is the condition, the third is // the block to branch to if the condition is true. SDValue Chain = Op.getOperand(0); SDValue Dest = Op.getOperand(2); - DebugLoc dl = Op.getDebugLoc(); + DebugLoc DL = Op.getDebugLoc(); - SDValue CondRes = CreateFPCmp(DAG, Op.getOperand(1)); + SDValue CondRes = createFPCmp(DAG, Op.getOperand(1)); // Return if flag is not set by a floating point comparison. if (CondRes.getOpcode() != MipsISD::FPCmp) @@ -2201,27 +1409,27 @@ LowerBRCOND(SDValue Op, SelectionDAG &DAG) const SDValue CCNode = CondRes.getOperand(2); Mips::CondCode CC = (Mips::CondCode)cast<ConstantSDNode>(CCNode)->getZExtValue(); - SDValue BrCode = DAG.getConstant(GetFPBranchCodeFromCond(CC), MVT::i32); - - return DAG.getNode(MipsISD::FPBrcond, dl, Op.getValueType(), Chain, BrCode, + unsigned Opc = invertFPCondCodeUser(CC) ? Mips::BRANCH_F : Mips::BRANCH_T; + SDValue BrCode = DAG.getConstant(Opc, MVT::i32); + return DAG.getNode(MipsISD::FPBrcond, DL, Op.getValueType(), Chain, BrCode, Dest, CondRes); } SDValue MipsTargetLowering:: -LowerSELECT(SDValue Op, SelectionDAG &DAG) const +lowerSELECT(SDValue Op, SelectionDAG &DAG) const { - SDValue Cond = CreateFPCmp(DAG, Op.getOperand(0)); + SDValue Cond = createFPCmp(DAG, Op.getOperand(0)); // Return if flag is not set by a floating point comparison. if (Cond.getOpcode() != MipsISD::FPCmp) return Op; - return CreateCMovFP(DAG, Cond, Op.getOperand(1), Op.getOperand(2), + return createCMovFP(DAG, Cond, Op.getOperand(1), Op.getOperand(2), Op.getDebugLoc()); } SDValue MipsTargetLowering:: -LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const +lowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const { DebugLoc DL = Op.getDebugLoc(); EVT Ty = Op.getOperand(0).getValueType(); @@ -2233,8 +1441,8 @@ LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const Op.getOperand(3)); } -SDValue MipsTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const { - SDValue Cond = CreateFPCmp(DAG, Op); +SDValue MipsTargetLowering::lowerSETCC(SDValue Op, SelectionDAG &DAG) const { + SDValue Cond = createFPCmp(DAG, Op); assert(Cond.getOpcode() == MipsISD::FPCmp && "Floating point operand expected."); @@ -2242,13 +1450,13 @@ SDValue MipsTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const { SDValue True = DAG.getConstant(1, MVT::i32); SDValue False = DAG.getConstant(0, MVT::i32); - return CreateCMovFP(DAG, Cond, True, False, Op.getDebugLoc()); + return createCMovFP(DAG, Cond, True, False, Op.getDebugLoc()); } -SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op, +SDValue MipsTargetLowering::lowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const { // FIXME there isn't actually debug info here - DebugLoc dl = Op.getDebugLoc(); + DebugLoc DL = Op.getDebugLoc(); const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal(); if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64) { @@ -2257,12 +1465,12 @@ SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op, // %gp_rel relocation if (TLOF.IsGlobalInSmallSection(GV, getTargetMachine())) { - SDValue GA = DAG.getTargetGlobalAddress(GV, dl, MVT::i32, 0, + SDValue GA = DAG.getTargetGlobalAddress(GV, DL, MVT::i32, 0, MipsII::MO_GPREL); - SDValue GPRelNode = DAG.getNode(MipsISD::GPRel, dl, + SDValue GPRelNode = DAG.getNode(MipsISD::GPRel, DL, DAG.getVTList(MVT::i32), &GA, 1); SDValue GPReg = DAG.getRegister(Mips::GP, MVT::i32); - return DAG.getNode(ISD::ADD, dl, MVT::i32, GPReg, GPRelNode); + return DAG.getNode(ISD::ADD, DL, MVT::i32, GPReg, GPRelNode); } // %hi/%lo relocation @@ -2280,7 +1488,7 @@ SDValue MipsTargetLowering::LowerGlobalAddress(SDValue Op, HasMips64 ? MipsII::MO_GOT_DISP : MipsII::MO_GOT16); } -SDValue MipsTargetLowering::LowerBlockAddress(SDValue Op, +SDValue MipsTargetLowering::lowerBlockAddress(SDValue Op, SelectionDAG &DAG) const { if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64) return getAddrNonPIC(Op, DAG); @@ -2356,7 +1564,7 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const // Local Exec TLS Model. GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op); - DebugLoc dl = GA->getDebugLoc(); + DebugLoc DL = GA->getDebugLoc(); const GlobalValue *GV = GA->getGlobal(); EVT PtrVT = getPointerTy(); @@ -2388,9 +1596,9 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const unsigned Flag = (model == TLSModel::LocalDynamic) ? MipsII::MO_TLSLDM : MipsII::MO_TLSGD; - SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, Flag); - SDValue Argument = DAG.getNode(MipsISD::Wrapper, dl, PtrVT, - GetGlobalReg(DAG, PtrVT), TGA); + SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, Flag); + SDValue Argument = DAG.getNode(MipsISD::Wrapper, DL, PtrVT, + getGlobalReg(DAG, PtrVT), TGA); unsigned PtrSize = PtrVT.getSizeInBits(); IntegerType *PtrTy = Type::getIntNTy(*DAG.getContext(), PtrSize); @@ -2404,9 +1612,9 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const TargetLowering::CallLoweringInfo CLI(DAG.getEntryNode(), PtrTy, false, false, false, false, 0, CallingConv::C, - /*isTailCall=*/false, /*doesNotRet=*/false, + /*IsTailCall=*/false, /*doesNotRet=*/false, /*isReturnValueUsed=*/true, - TlsGetAddr, Args, DAG, dl); + TlsGetAddr, Args, DAG, DL); std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI); SDValue Ret = CallResult.first; @@ -2414,44 +1622,44 @@ LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const if (model != TLSModel::LocalDynamic) return Ret; - SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, + SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, MipsII::MO_DTPREL_HI); - SDValue Hi = DAG.getNode(MipsISD::Hi, dl, PtrVT, TGAHi); - SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, + SDValue Hi = DAG.getNode(MipsISD::Hi, DL, PtrVT, TGAHi); + SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, MipsII::MO_DTPREL_LO); - SDValue Lo = DAG.getNode(MipsISD::Lo, dl, PtrVT, TGALo); - SDValue Add = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Ret); - return DAG.getNode(ISD::ADD, dl, PtrVT, Add, Lo); + SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo); + SDValue Add = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Ret); + return DAG.getNode(ISD::ADD, DL, PtrVT, Add, Lo); } SDValue Offset; if (model == TLSModel::InitialExec) { // Initial Exec TLS Model - SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, + SDValue TGA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, MipsII::MO_GOTTPREL); - TGA = DAG.getNode(MipsISD::Wrapper, dl, PtrVT, GetGlobalReg(DAG, PtrVT), + TGA = DAG.getNode(MipsISD::Wrapper, DL, PtrVT, getGlobalReg(DAG, PtrVT), TGA); - Offset = DAG.getLoad(PtrVT, dl, + Offset = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), TGA, MachinePointerInfo(), false, false, false, 0); } else { // Local Exec TLS Model assert(model == TLSModel::LocalExec); - SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, + SDValue TGAHi = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, MipsII::MO_TPREL_HI); - SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, + SDValue TGALo = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, MipsII::MO_TPREL_LO); - SDValue Hi = DAG.getNode(MipsISD::Hi, dl, PtrVT, TGAHi); - SDValue Lo = DAG.getNode(MipsISD::Lo, dl, PtrVT, TGALo); - Offset = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo); + SDValue Hi = DAG.getNode(MipsISD::Hi, DL, PtrVT, TGAHi); + SDValue Lo = DAG.getNode(MipsISD::Lo, DL, PtrVT, TGALo); + Offset = DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo); } - SDValue ThreadPointer = DAG.getNode(MipsISD::ThreadPointer, dl, PtrVT); - return DAG.getNode(ISD::ADD, dl, PtrVT, ThreadPointer, Offset); + SDValue ThreadPointer = DAG.getNode(MipsISD::ThreadPointer, DL, PtrVT); + return DAG.getNode(ISD::ADD, DL, PtrVT, ThreadPointer, Offset); } SDValue MipsTargetLowering:: -LowerJumpTable(SDValue Op, SelectionDAG &DAG) const +lowerJumpTable(SDValue Op, SelectionDAG &DAG) const { if (getTargetMachine().getRelocationModel() != Reloc::PIC_ && !IsN64) return getAddrNonPIC(Op, DAG); @@ -2460,7 +1668,7 @@ LowerJumpTable(SDValue Op, SelectionDAG &DAG) const } SDValue MipsTargetLowering:: -LowerConstantPool(SDValue Op, SelectionDAG &DAG) const +lowerConstantPool(SDValue Op, SelectionDAG &DAG) const { // gp_rel relocation // FIXME: we should reference the constant pool using small data sections, @@ -2478,22 +1686,22 @@ LowerConstantPool(SDValue Op, SelectionDAG &DAG) const return getAddrLocal(Op, DAG, HasMips64); } -SDValue MipsTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const { +SDValue MipsTargetLowering::lowerVASTART(SDValue Op, SelectionDAG &DAG) const { MachineFunction &MF = DAG.getMachineFunction(); MipsFunctionInfo *FuncInfo = MF.getInfo<MipsFunctionInfo>(); - DebugLoc dl = Op.getDebugLoc(); + DebugLoc DL = Op.getDebugLoc(); SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), getPointerTy()); // vastart just stores the address of the VarArgsFrameIndex slot into the // memory location argument. const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue(); - return DAG.getStore(Op.getOperand(0), dl, FI, Op.getOperand(1), + return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1), MachinePointerInfo(SV), false, false, 0); } -static SDValue LowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) { +static SDValue lowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) { EVT TyX = Op.getOperand(0).getValueType(); EVT TyY = Op.getOperand(1).getValueType(); SDValue Const1 = DAG.getConstant(1, MVT::i32); @@ -2538,7 +1746,7 @@ static SDValue LowerFCOPYSIGN32(SDValue Op, SelectionDAG &DAG, bool HasR2) { return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, LowX, Res); } -static SDValue LowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) { +static SDValue lowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) { unsigned WidthX = Op.getOperand(0).getValueSizeInBits(); unsigned WidthY = Op.getOperand(1).getValueSizeInBits(); EVT TyX = MVT::getIntegerVT(WidthX), TyY = MVT::getIntegerVT(WidthY); @@ -2587,14 +1795,14 @@ static SDValue LowerFCOPYSIGN64(SDValue Op, SelectionDAG &DAG, bool HasR2) { } SDValue -MipsTargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const { +MipsTargetLowering::lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const { if (Subtarget->hasMips64()) - return LowerFCOPYSIGN64(Op, DAG, Subtarget->hasMips32r2()); + return lowerFCOPYSIGN64(Op, DAG, Subtarget->hasMips32r2()); - return LowerFCOPYSIGN32(Op, DAG, Subtarget->hasMips32r2()); + return lowerFCOPYSIGN32(Op, DAG, Subtarget->hasMips32r2()); } -static SDValue LowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) { +static SDValue lowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) { SDValue Res, Const1 = DAG.getConstant(1, MVT::i32); DebugLoc DL = Op.getDebugLoc(); @@ -2623,7 +1831,7 @@ static SDValue LowerFABS32(SDValue Op, SelectionDAG &DAG, bool HasR2) { return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, LowX, Res); } -static SDValue LowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) { +static SDValue lowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) { SDValue Res, Const1 = DAG.getConstant(1, MVT::i32); DebugLoc DL = Op.getDebugLoc(); @@ -2644,15 +1852,15 @@ static SDValue LowerFABS64(SDValue Op, SelectionDAG &DAG, bool HasR2) { } SDValue -MipsTargetLowering::LowerFABS(SDValue Op, SelectionDAG &DAG) const { +MipsTargetLowering::lowerFABS(SDValue Op, SelectionDAG &DAG) const { if (Subtarget->hasMips64() && (Op.getValueType() == MVT::f64)) - return LowerFABS64(Op, DAG, Subtarget->hasMips32r2()); + return lowerFABS64(Op, DAG, Subtarget->hasMips32r2()); - return LowerFABS32(Op, DAG, Subtarget->hasMips32r2()); + return lowerFABS32(Op, DAG, Subtarget->hasMips32r2()); } SDValue MipsTargetLowering:: -LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const { +lowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const { // check the depth assert((cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() == 0) && "Frame address can only be determined for current frame."); @@ -2660,13 +1868,13 @@ LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const { MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo(); MFI->setFrameAddressIsTaken(true); EVT VT = Op.getValueType(); - DebugLoc dl = Op.getDebugLoc(); - SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, + DebugLoc DL = Op.getDebugLoc(); + SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), DL, IsN64 ? Mips::FP_64 : Mips::FP, VT); return FrameAddr; } -SDValue MipsTargetLowering::LowerRETURNADDR(SDValue Op, +SDValue MipsTargetLowering::lowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const { // check the depth assert((cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue() == 0) && @@ -2687,7 +1895,7 @@ SDValue MipsTargetLowering::LowerRETURNADDR(SDValue Op, // generated from __builtin_eh_return (offset, handler) // The effect of this is to adjust the stack pointer by "offset" // and then branch to "handler". -SDValue MipsTargetLowering::LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) +SDValue MipsTargetLowering::lowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const { MachineFunction &MF = DAG.getMachineFunction(); MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); @@ -2711,26 +1919,17 @@ SDValue MipsTargetLowering::LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) Chain.getValue(1)); } -// TODO: set SType according to the desired memory barrier behavior. -SDValue -MipsTargetLowering::LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const { - unsigned SType = 0; - DebugLoc dl = Op.getDebugLoc(); - return DAG.getNode(MipsISD::Sync, dl, MVT::Other, Op.getOperand(0), - DAG.getConstant(SType, MVT::i32)); -} - -SDValue MipsTargetLowering::LowerATOMIC_FENCE(SDValue Op, +SDValue MipsTargetLowering::lowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG) const { // FIXME: Need pseudo-fence for 'singlethread' fences // FIXME: Set SType for weaker fences where supported/appropriate. unsigned SType = 0; - DebugLoc dl = Op.getDebugLoc(); - return DAG.getNode(MipsISD::Sync, dl, MVT::Other, Op.getOperand(0), + DebugLoc DL = Op.getDebugLoc(); + return DAG.getNode(MipsISD::Sync, DL, MVT::Other, Op.getOperand(0), DAG.getConstant(SType, MVT::i32)); } -SDValue MipsTargetLowering::LowerShiftLeftParts(SDValue Op, +SDValue MipsTargetLowering::lowerShiftLeftParts(SDValue Op, SelectionDAG &DAG) const { DebugLoc DL = Op.getDebugLoc(); SDValue Lo = Op.getOperand(0), Hi = Op.getOperand(1); @@ -2761,7 +1960,7 @@ SDValue MipsTargetLowering::LowerShiftLeftParts(SDValue Op, return DAG.getMergeValues(Ops, 2, DL); } -SDValue MipsTargetLowering::LowerShiftRightParts(SDValue Op, SelectionDAG &DAG, +SDValue MipsTargetLowering::lowerShiftRightParts(SDValue Op, SelectionDAG &DAG, bool IsSRA) const { DebugLoc DL = Op.getDebugLoc(); SDValue Lo = Op.getOperand(0), Hi = Op.getOperand(1); @@ -2802,7 +2001,7 @@ SDValue MipsTargetLowering::LowerShiftRightParts(SDValue Op, SelectionDAG &DAG, return DAG.getMergeValues(Ops, 2, DL); } -static SDValue CreateLoadLR(unsigned Opc, SelectionDAG &DAG, LoadSDNode *LD, +static SDValue createLoadLR(unsigned Opc, SelectionDAG &DAG, LoadSDNode *LD, SDValue Chain, SDValue Src, unsigned Offset) { SDValue Ptr = LD->getBasePtr(); EVT VT = LD->getValueType(0), MemVT = LD->getMemoryVT(); @@ -2820,7 +2019,7 @@ static SDValue CreateLoadLR(unsigned Opc, SelectionDAG &DAG, LoadSDNode *LD, } // Expand an unaligned 32 or 64-bit integer load node. -SDValue MipsTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const { +SDValue MipsTargetLowering::lowerLOAD(SDValue Op, SelectionDAG &DAG) const { LoadSDNode *LD = cast<LoadSDNode>(Op); EVT MemVT = LD->getMemoryVT(); @@ -2842,15 +2041,15 @@ SDValue MipsTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const { // (set tmp, (ldl (add baseptr, 7), undef)) // (set dst, (ldr baseptr, tmp)) if ((VT == MVT::i64) && (ExtType == ISD::NON_EXTLOAD)) { - SDValue LDL = CreateLoadLR(MipsISD::LDL, DAG, LD, Chain, Undef, + SDValue LDL = createLoadLR(MipsISD::LDL, DAG, LD, Chain, Undef, IsLittle ? 7 : 0); - return CreateLoadLR(MipsISD::LDR, DAG, LD, LDL.getValue(1), LDL, + return createLoadLR(MipsISD::LDR, DAG, LD, LDL.getValue(1), LDL, IsLittle ? 0 : 7); } - SDValue LWL = CreateLoadLR(MipsISD::LWL, DAG, LD, Chain, Undef, + SDValue LWL = createLoadLR(MipsISD::LWL, DAG, LD, Chain, Undef, IsLittle ? 3 : 0); - SDValue LWR = CreateLoadLR(MipsISD::LWR, DAG, LD, LWL.getValue(1), LWL, + SDValue LWR = createLoadLR(MipsISD::LWR, DAG, LD, LWL.getValue(1), LWL, IsLittle ? 0 : 3); // Expand @@ -2881,7 +2080,7 @@ SDValue MipsTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const { return DAG.getMergeValues(Ops, 2, DL); } -static SDValue CreateStoreLR(unsigned Opc, SelectionDAG &DAG, StoreSDNode *SD, +static SDValue createStoreLR(unsigned Opc, SelectionDAG &DAG, StoreSDNode *SD, SDValue Chain, unsigned Offset) { SDValue Ptr = SD->getBasePtr(), Value = SD->getValue(); EVT MemVT = SD->getMemoryVT(), BasePtrVT = Ptr.getValueType(); @@ -2898,7 +2097,7 @@ static SDValue CreateStoreLR(unsigned Opc, SelectionDAG &DAG, StoreSDNode *SD, } // Expand an unaligned 32 or 64-bit integer store node. -SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const { +SDValue MipsTargetLowering::lowerSTORE(SDValue Op, SelectionDAG &DAG) const { StoreSDNode *SD = cast<StoreSDNode>(Op); EVT MemVT = SD->getMemoryVT(); @@ -2918,9 +2117,9 @@ SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const { // (swl val, (add baseptr, 3)) // (swr val, baseptr) if ((VT == MVT::i32) || SD->isTruncatingStore()) { - SDValue SWL = CreateStoreLR(MipsISD::SWL, DAG, SD, Chain, + SDValue SWL = createStoreLR(MipsISD::SWL, DAG, SD, Chain, IsLittle ? 3 : 0); - return CreateStoreLR(MipsISD::SWR, DAG, SD, SWL, IsLittle ? 0 : 3); + return createStoreLR(MipsISD::SWR, DAG, SD, SWL, IsLittle ? 0 : 3); } assert(VT == MVT::i64); @@ -2930,160 +2129,11 @@ SDValue MipsTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const { // to // (sdl val, (add baseptr, 7)) // (sdr val, baseptr) - SDValue SDL = CreateStoreLR(MipsISD::SDL, DAG, SD, Chain, IsLittle ? 7 : 0); - return CreateStoreLR(MipsISD::SDR, DAG, SD, SDL, IsLittle ? 0 : 7); + SDValue SDL = createStoreLR(MipsISD::SDL, DAG, SD, Chain, IsLittle ? 7 : 0); + return createStoreLR(MipsISD::SDR, DAG, SD, SDL, IsLittle ? 0 : 7); } -// This function expands mips intrinsic nodes which have 64-bit input operands -// or output values. -// -// out64 = intrinsic-node in64 -// => -// lo = copy (extract-element (in64, 0)) -// hi = copy (extract-element (in64, 1)) -// mips-specific-node -// v0 = copy lo -// v1 = copy hi -// out64 = merge-values (v0, v1) -// -static SDValue LowerDSPIntr(SDValue Op, SelectionDAG &DAG, - unsigned Opc, bool HasI64In, bool HasI64Out) { - DebugLoc DL = Op.getDebugLoc(); - bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other; - SDValue Chain = HasChainIn ? Op->getOperand(0) : DAG.getEntryNode(); - SmallVector<SDValue, 3> Ops; - - if (HasI64In) { - SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, - Op->getOperand(1 + HasChainIn), - DAG.getConstant(0, MVT::i32)); - SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, - Op->getOperand(1 + HasChainIn), - DAG.getConstant(1, MVT::i32)); - - Chain = DAG.getCopyToReg(Chain, DL, Mips::LO, InLo, SDValue()); - Chain = DAG.getCopyToReg(Chain, DL, Mips::HI, InHi, Chain.getValue(1)); - - Ops.push_back(Chain); - Ops.append(Op->op_begin() + HasChainIn + 2, Op->op_end()); - Ops.push_back(Chain.getValue(1)); - } else { - Ops.push_back(Chain); - Ops.append(Op->op_begin() + HasChainIn + 1, Op->op_end()); - } - - if (!HasI64Out) - return DAG.getNode(Opc, DL, Op->value_begin(), Op->getNumValues(), - Ops.begin(), Ops.size()); - - SDValue Intr = DAG.getNode(Opc, DL, DAG.getVTList(MVT::Other, MVT::Glue), - Ops.begin(), Ops.size()); - SDValue OutLo = DAG.getCopyFromReg(Intr.getValue(0), DL, Mips::LO, MVT::i32, - Intr.getValue(1)); - SDValue OutHi = DAG.getCopyFromReg(OutLo.getValue(1), DL, Mips::HI, MVT::i32, - OutLo.getValue(2)); - SDValue Out = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, OutLo, OutHi); - - if (!HasChainIn) - return Out; - - SDValue Vals[] = { Out, OutHi.getValue(1) }; - return DAG.getMergeValues(Vals, 2, DL); -} - -SDValue MipsTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, - SelectionDAG &DAG) const { - switch (cast<ConstantSDNode>(Op->getOperand(0))->getZExtValue()) { - default: - return SDValue(); - // @LOCALMOD-BEGIN - case Intrinsic::nacl_read_tp: - return GetNaClThreadPointer(DAG, Op->getDebugLoc()); - // @LOCALMOD-END - case Intrinsic::mips_shilo: - return LowerDSPIntr(Op, DAG, MipsISD::SHILO, true, true); - case Intrinsic::mips_dpau_h_qbl: - return LowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL, true, true); - case Intrinsic::mips_dpau_h_qbr: - return LowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR, true, true); - case Intrinsic::mips_dpsu_h_qbl: - return LowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL, true, true); - case Intrinsic::mips_dpsu_h_qbr: - return LowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR, true, true); - case Intrinsic::mips_dpa_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH, true, true); - case Intrinsic::mips_dps_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH, true, true); - case Intrinsic::mips_dpax_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH, true, true); - case Intrinsic::mips_dpsx_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH, true, true); - case Intrinsic::mips_mulsa_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH, true, true); - case Intrinsic::mips_mult: - return LowerDSPIntr(Op, DAG, MipsISD::MULT, false, true); - case Intrinsic::mips_multu: - return LowerDSPIntr(Op, DAG, MipsISD::MULTU, false, true); - case Intrinsic::mips_madd: - return LowerDSPIntr(Op, DAG, MipsISD::MADD_DSP, true, true); - case Intrinsic::mips_maddu: - return LowerDSPIntr(Op, DAG, MipsISD::MADDU_DSP, true, true); - case Intrinsic::mips_msub: - return LowerDSPIntr(Op, DAG, MipsISD::MSUB_DSP, true, true); - case Intrinsic::mips_msubu: - return LowerDSPIntr(Op, DAG, MipsISD::MSUBU_DSP, true, true); - } -} - -SDValue MipsTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op, - SelectionDAG &DAG) const { - switch (cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue()) { - default: - return SDValue(); - case Intrinsic::mips_extp: - return LowerDSPIntr(Op, DAG, MipsISD::EXTP, true, false); - case Intrinsic::mips_extpdp: - return LowerDSPIntr(Op, DAG, MipsISD::EXTPDP, true, false); - case Intrinsic::mips_extr_w: - return LowerDSPIntr(Op, DAG, MipsISD::EXTR_W, true, false); - case Intrinsic::mips_extr_r_w: - return LowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W, true, false); - case Intrinsic::mips_extr_rs_w: - return LowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W, true, false); - case Intrinsic::mips_extr_s_h: - return LowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H, true, false); - case Intrinsic::mips_mthlip: - return LowerDSPIntr(Op, DAG, MipsISD::MTHLIP, true, true); - case Intrinsic::mips_mulsaq_s_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH, true, true); - case Intrinsic::mips_maq_s_w_phl: - return LowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL, true, true); - case Intrinsic::mips_maq_s_w_phr: - return LowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR, true, true); - case Intrinsic::mips_maq_sa_w_phl: - return LowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL, true, true); - case Intrinsic::mips_maq_sa_w_phr: - return LowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR, true, true); - case Intrinsic::mips_dpaq_s_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH, true, true); - case Intrinsic::mips_dpsq_s_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH, true, true); - case Intrinsic::mips_dpaq_sa_l_w: - return LowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W, true, true); - case Intrinsic::mips_dpsq_sa_l_w: - return LowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W, true, true); - case Intrinsic::mips_dpaqx_s_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH, true, true); - case Intrinsic::mips_dpaqx_sa_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH, true, true); - case Intrinsic::mips_dpsqx_s_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH, true, true); - case Intrinsic::mips_dpsqx_sa_w_ph: - return LowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH, true, true); - } -} - -SDValue MipsTargetLowering::LowerADD(SDValue Op, SelectionDAG &DAG) const { +SDValue MipsTargetLowering::lowerADD(SDValue Op, SelectionDAG &DAG) const { if (Op->getOperand(0).getOpcode() != ISD::FRAMEADDR || cast<ConstantSDNode> (Op->getOperand(0).getOperand(0))->getZExtValue() != 0 @@ -3220,28 +2270,6 @@ static unsigned getNextIntArgReg(unsigned Reg) { return (Reg == Mips::A0) ? Mips::A1 : Mips::A3; } -/// IsEligibleForTailCallOptimization - Check whether the call is eligible -/// for tail call optimization. -bool MipsTargetLowering:: -IsEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, - unsigned NextStackOffset, - const MipsFunctionInfo& FI) const { - if (!EnableMipsTailCalls) - return false; - - // No tail call optimization for mips16. - if (Subtarget->inMips16Mode()) - return false; - - // Return false if either the callee or caller has a byval argument. - if (MipsCCInfo.hasByValArg() || FI.hasByvalArg()) - return false; - - // Return true if the callee's argument area is no larger than the - // caller's. - return NextStackOffset <= FI.getIncomingArgSize(); -} - SDValue MipsTargetLowering::passArgOnStack(SDValue StackPtr, unsigned Offset, SDValue Chain, SDValue Arg, DebugLoc DL, @@ -3260,161 +2288,48 @@ MipsTargetLowering::passArgOnStack(SDValue StackPtr, unsigned Offset, /*isVolatile=*/ true, false, 0); } -// -// The Mips16 hard float is a crazy quilt inherited from gcc. I have a much -// cleaner way to do all of this but it will have to wait until the traditional -// gcc mechanism is completed. -// -// For Pic, in order for Mips16 code to call Mips32 code which according the abi -// have either arguments or returned values placed in floating point registers, -// we use a set of helper functions. (This includes functions which return type -// complex which on Mips are returned in a pair of floating point registers). -// -// This is an encoding that we inherited from gcc. -// In Mips traditional O32, N32 ABI, floating point numbers are passed in -// floating point argument registers 1,2 only when the first and optionally -// the second arguments are float (sf) or double (df). -// For Mips16 we are only concerned with the situations where floating point -// arguments are being passed in floating point registers by the ABI, because -// Mips16 mode code cannot execute floating point instructions to load those -// values and hence helper functions are needed. -// The possibilities are (), (sf), (sf, sf), (sf, df), (df), (df, sf), (df, df) -// the helper function suffixs for these are: -// 0, 1, 5, 9, 2, 6, 10 -// this suffix can then be calculated as follows: -// for a given argument Arg: -// Arg1x, Arg2x = 1 : Arg is sf -// 2 : Arg is df -// 0: Arg is neither sf or df -// So this stub is the string for number Arg1x + Arg2x*4. -// However not all numbers between 0 and 10 are possible, we check anyway and -// assert if the impossible exists. -// - -unsigned int MipsTargetLowering::getMips16HelperFunctionStubNumber - (ArgListTy &Args) const { - unsigned int resultNum = 0; - if (Args.size() >= 1) { - Type *t = Args[0].Ty; - if (t->isFloatTy()) { - resultNum = 1; - } - else if (t->isDoubleTy()) { - resultNum = 2; - } - } - if (resultNum) { - if (Args.size() >=2) { - Type *t = Args[1].Ty; - if (t->isFloatTy()) { - resultNum += 4; - } - else if (t->isDoubleTy()) { - resultNum += 8; - } - } +void MipsTargetLowering:: +getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const { + // Insert node "GP copy globalreg" before call to function. + // + // R_MIPS_CALL* operators (emitted when non-internal functions are called + // in PIC mode) allow symbols to be resolved via lazy binding. + // The lazy binding stub requires GP to point to the GOT. + if (IsPICCall && !InternalLinkage) { + unsigned GPReg = IsN64 ? Mips::GP_64 : Mips::GP; + EVT Ty = IsN64 ? MVT::i64 : MVT::i32; + RegsToPass.push_back(std::make_pair(GPReg, getGlobalReg(CLI.DAG, Ty))); } - return resultNum; -} -// -// prefixs are attached to stub numbers depending on the return type . -// return type: float sf_ -// double df_ -// single complex sc_ -// double complext dc_ -// others NO PREFIX -// -// -// The full name of a helper function is__mips16_call_stub + -// return type dependent prefix + stub number -// -// -// This is something that probably should be in a different source file and -// perhaps done differently but my main purpose is to not waste runtime -// on something that we can enumerate in the source. Another possibility is -// to have a python script to generate these mapping tables. This will do -// for now. There are a whole series of helper function mapping arrays, one -// for each return type class as outlined above. There there are 11 possible -// entries. Ones with 0 are ones which should never be selected -// -// All the arrays are similar except for ones which return neither -// sf, df, sc, dc, in which only care about ones which have sf or df as a -// first parameter. -// -#define P_ "__mips16_call_stub_" -#define MAX_STUB_NUMBER 10 -#define T1 P "1", P "2", 0, 0, P "5", P "6", 0, 0, P "9", P "10" -#define T P "0" , T1 -#define P P_ -static char const * vMips16Helper[MAX_STUB_NUMBER+1] = - {0, T1 }; -#undef P -#define P P_ "sf_" -static char const * sfMips16Helper[MAX_STUB_NUMBER+1] = - { T }; -#undef P -#define P P_ "df_" -static char const * dfMips16Helper[MAX_STUB_NUMBER+1] = - { T }; -#undef P -#define P P_ "sc_" -static char const * scMips16Helper[MAX_STUB_NUMBER+1] = - { T }; -#undef P -#define P P_ "dc_" -static char const * dcMips16Helper[MAX_STUB_NUMBER+1] = - { T }; -#undef P -#undef P_ - - -const char* MipsTargetLowering:: - getMips16HelperFunction - (Type* RetTy, ArgListTy &Args, bool &needHelper) const { - const unsigned int stubNum = getMips16HelperFunctionStubNumber(Args); -#ifndef NDEBUG - const unsigned int maxStubNum = 10; - assert(stubNum <= maxStubNum); - const bool validStubNum[maxStubNum+1] = - {true, true, true, false, false, true, true, false, false, true, true}; - assert(validStubNum[stubNum]); -#endif - const char *result; - if (RetTy->isFloatTy()) { - result = sfMips16Helper[stubNum]; - } - else if (RetTy ->isDoubleTy()) { - result = dfMips16Helper[stubNum]; - } - else if (RetTy->isStructTy()) { - // check if it's complex - if (RetTy->getNumContainedTypes() == 2) { - if ((RetTy->getContainedType(0)->isFloatTy()) && - (RetTy->getContainedType(1)->isFloatTy())) { - result = scMips16Helper[stubNum]; - } - else if ((RetTy->getContainedType(0)->isDoubleTy()) && - (RetTy->getContainedType(1)->isDoubleTy())) { - result = dcMips16Helper[stubNum]; - } - else { - llvm_unreachable("Uncovered condition"); - } - } - else { - llvm_unreachable("Uncovered condition"); - } - } - else { - if (stubNum == 0) { - needHelper = false; - return ""; - } - result = vMips16Helper[stubNum]; + // Build a sequence of copy-to-reg nodes chained together with token + // chain and flag operands which copy the outgoing args into registers. + // The InFlag in necessary since all emitted instructions must be + // stuck together. + SDValue InFlag; + + for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) { + Chain = CLI.DAG.getCopyToReg(Chain, CLI.DL, RegsToPass[i].first, + RegsToPass[i].second, InFlag); + InFlag = Chain.getValue(1); } - needHelper = true; - return result; + + // Add argument registers to the end of the list so that they are + // known live into the call. + for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) + Ops.push_back(CLI.DAG.getRegister(RegsToPass[i].first, + RegsToPass[i].second.getValueType())); + + // Add a register mask operand representing the call-preserved registers. + const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo(); + const uint32_t *Mask = TRI->getCallPreservedMask(CLI.CallConv); + assert(Mask && "Missing call preserved mask for calling convention"); + Ops.push_back(CLI.DAG.getRegisterMask(Mask)); + + if (InFlag.getNode()) + Ops.push_back(InFlag); } /// LowerCall - functions arguments are copied from virtual regs to @@ -3423,36 +2338,16 @@ SDValue MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, SmallVectorImpl<SDValue> &InVals) const { SelectionDAG &DAG = CLI.DAG; - DebugLoc &dl = CLI.DL; + DebugLoc &DL = CLI.DL; SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs; SmallVector<SDValue, 32> &OutVals = CLI.OutVals; SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins; SDValue Chain = CLI.Chain; SDValue Callee = CLI.Callee; - bool &isTailCall = CLI.IsTailCall; + bool &IsTailCall = CLI.IsTailCall; CallingConv::ID CallConv = CLI.CallConv; - bool isVarArg = CLI.IsVarArg; - - const char* mips16HelperFunction = 0; - bool needMips16Helper = false; - - if (Subtarget->inMips16Mode() && getTargetMachine().Options.UseSoftFloat && - Mips16HardFloat) { - // - // currently we don't have symbols tagged with the mips16 or mips32 - // qualifier so we will assume that we don't know what kind it is. - // and generate the helper - // - bool lookupHelper = true; - if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) { - if (noHelperNeeded.find(S->getSymbol()) != noHelperNeeded.end()) { - lookupHelper = false; - } - } - if (lookupHelper) mips16HelperFunction = - getMips16HelperFunction(CLI.RetTy, CLI.Args, needMips16Helper); + bool IsVarArg = CLI.IsVarArg; - } MachineFunction &MF = DAG.getMachineFunction(); MachineFrameInfo *MFI = MF.getFrameInfo(); const TargetFrameLowering *TFL = MF.getTarget().getFrameLowering(); @@ -3460,22 +2355,24 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, // Analyze operands of the call, assigning locations to each operand. SmallVector<CCValAssign, 16> ArgLocs; - CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), + CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), getTargetMachine(), ArgLocs, *DAG.getContext()); MipsCC MipsCCInfo(CallConv, IsO32, CCInfo); - MipsCCInfo.analyzeCallOperands(Outs, isVarArg); + MipsCCInfo.analyzeCallOperands(Outs, IsVarArg, + getTargetMachine().Options.UseSoftFloat, + Callee.getNode(), CLI.Args); // Get a count of how many bytes are to be pushed on the stack. unsigned NextStackOffset = CCInfo.getNextStackOffset(); // Check if it's really possible to do a tail call. - if (isTailCall) - isTailCall = - IsEligibleForTailCallOptimization(MipsCCInfo, NextStackOffset, + if (IsTailCall) + IsTailCall = + isEligibleForTailCallOptimization(MipsCCInfo, NextStackOffset, *MF.getInfo<MipsFunctionInfo>()); - if (isTailCall) + if (IsTailCall) ++NumTailCalls; // Chain is the output chain of the last Load/Store or CopyToReg node. @@ -3485,10 +2382,10 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, NextStackOffset = RoundUpToAlignment(NextStackOffset, StackAlignment); SDValue NextStackOffsetVal = DAG.getIntPtrConstant(NextStackOffset, true); - if (!isTailCall) + if (!IsTailCall) Chain = DAG.getCALLSEQ_START(Chain, NextStackOffsetVal); - SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, + SDValue StackPtr = DAG.getCopyFromReg(Chain, DL, IsN64 ? Mips::SP_64 : Mips::SP, getPointerTy()); @@ -3509,9 +2406,9 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, assert(Flags.getByValSize() && "ByVal args of size 0 should have been ignored by front-end."); assert(ByValArg != MipsCCInfo.byval_end()); - assert(!isTailCall && + assert(!IsTailCall && "Do not tail-call optimize if there is a byval argument."); - passByValArg(Chain, dl, RegsToPass, MemOpChains, StackPtr, MFI, DAG, Arg, + passByValArg(Chain, DL, RegsToPass, MemOpChains, StackPtr, MFI, DAG, Arg, MipsCCInfo, *ByValArg, Flags, Subtarget->isLittle()); ++ByValArg; continue; @@ -3523,12 +2420,13 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, case CCValAssign::Full: if (VA.isRegLoc()) { if ((ValVT == MVT::f32 && LocVT == MVT::i32) || - (ValVT == MVT::f64 && LocVT == MVT::i64)) - Arg = DAG.getNode(ISD::BITCAST, dl, LocVT, Arg); + (ValVT == MVT::f64 && LocVT == MVT::i64) || + (ValVT == MVT::i64 && LocVT == MVT::f64)) + Arg = DAG.getNode(ISD::BITCAST, DL, LocVT, Arg); else if (ValVT == MVT::f64 && LocVT == MVT::i32) { - SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, dl, MVT::i32, + SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Arg, DAG.getConstant(0, MVT::i32)); - SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, dl, MVT::i32, + SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Arg, DAG.getConstant(1, MVT::i32)); if (!Subtarget->isLittle()) std::swap(Lo, Hi); @@ -3541,13 +2439,13 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, } break; case CCValAssign::SExt: - Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, LocVT, Arg); + Arg = DAG.getNode(ISD::SIGN_EXTEND, DL, LocVT, Arg); break; case CCValAssign::ZExt: - Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, LocVT, Arg); + Arg = DAG.getNode(ISD::ZERO_EXTEND, DL, LocVT, Arg); break; case CCValAssign::AExt: - Arg = DAG.getNode(ISD::ANY_EXTEND, dl, LocVT, Arg); + Arg = DAG.getNode(ISD::ANY_EXTEND, DL, LocVT, Arg); break; } @@ -3564,13 +2462,13 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, // emit ISD::STORE whichs stores the // parameter value to a stack Location MemOpChains.push_back(passArgOnStack(StackPtr, VA.getLocMemOffset(), - Chain, Arg, dl, isTailCall, DAG)); + Chain, Arg, DL, IsTailCall, DAG)); } // Transform all store nodes into one single node because all store // nodes are independent of each other. if (!MemOpChains.empty()) - Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, + Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &MemOpChains[0], MemOpChains.size()); // If the callee is a GlobalAddress/ExternalSymbol node (quite common, every @@ -3592,7 +2490,7 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, else Callee = getAddrGlobal(Callee, DAG, MipsII::MO_GOT_CALL); } else - Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, getPointerTy(), 0, + Callee = DAG.getTargetGlobalAddress(G->getGlobal(), DL, getPointerTy(), 0, MipsII::MO_NO_FLAG); GlobalOrExternal = true; } @@ -3609,80 +2507,17 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, GlobalOrExternal = true; } - SDValue JumpTarget = Callee; - - // T9 should contain the address of the callee function if - // -reloction-model=pic or it is an indirect call. - if (IsPICCall || !GlobalOrExternal) { - unsigned T9Reg = IsN64 ? Mips::T9_64 : Mips::T9; - unsigned V0Reg = Mips::V0; - if (needMips16Helper) { - RegsToPass.push_front(std::make_pair(V0Reg, Callee)); - JumpTarget = DAG.getExternalSymbol( - mips16HelperFunction, getPointerTy()); - JumpTarget = getAddrGlobal(JumpTarget, DAG, MipsII::MO_GOT); - } - else { - RegsToPass.push_front(std::make_pair(T9Reg, Callee)); - - if (!Subtarget->inMips16Mode()) - JumpTarget = SDValue(); - } - } - - // Insert node "GP copy globalreg" before call to function. - // - // R_MIPS_CALL* operators (emitted when non-internal functions are called - // in PIC mode) allow symbols to be resolved via lazy binding. - // The lazy binding stub requires GP to point to the GOT. - if (IsPICCall && !InternalLinkage) { - unsigned GPReg = IsN64 ? Mips::GP_64 : Mips::GP; - EVT Ty = IsN64 ? MVT::i64 : MVT::i32; - RegsToPass.push_back(std::make_pair(GPReg, GetGlobalReg(DAG, Ty))); - } - - // Build a sequence of copy-to-reg nodes chained together with token - // chain and flag operands which copy the outgoing args into registers. - // The InFlag in necessary since all emitted instructions must be - // stuck together. - SDValue InFlag; - - for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) { - Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first, - RegsToPass[i].second, InFlag); - InFlag = Chain.getValue(1); - } - - // MipsJmpLink = #chain, #target_address, #opt_in_flags... - // = Chain, Callee, Reg#1, Reg#2, ... - // - // Returns a chain & a flag for retval copy to use. - SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue); SmallVector<SDValue, 8> Ops(1, Chain); + SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue); - if (JumpTarget.getNode()) - Ops.push_back(JumpTarget); - - // Add argument registers to the end of the list so that they are - // known live into the call. - for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) - Ops.push_back(DAG.getRegister(RegsToPass[i].first, - RegsToPass[i].second.getValueType())); - - // Add a register mask operand representing the call-preserved registers. - const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo(); - const uint32_t *Mask = TRI->getCallPreservedMask(CallConv); - assert(Mask && "Missing call preserved mask for calling convention"); - Ops.push_back(DAG.getRegisterMask(Mask)); - - if (InFlag.getNode()) - Ops.push_back(InFlag); + getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal, InternalLinkage, + CLI, Callee, Chain); - if (isTailCall) - return DAG.getNode(MipsISD::TailCall, dl, MVT::Other, &Ops[0], Ops.size()); + if (IsTailCall) + return DAG.getNode(MipsISD::TailCall, DL, MVT::Other, &Ops[0], Ops.size()); - Chain = DAG.getNode(MipsISD::JmpLink, dl, NodeTys, &Ops[0], Ops.size()); - InFlag = Chain.getValue(1); + Chain = DAG.getNode(MipsISD::JmpLink, DL, NodeTys, &Ops[0], Ops.size()); + SDValue InFlag = Chain.getValue(1); // Create the CALLSEQ_END node. Chain = DAG.getCALLSEQ_END(Chain, NextStackOffsetVal, @@ -3691,31 +2526,40 @@ MipsTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, // Handle result values, copying them out of physregs into vregs that we // return. - return LowerCallResult(Chain, InFlag, CallConv, isVarArg, - Ins, dl, DAG, InVals); + return LowerCallResult(Chain, InFlag, CallConv, IsVarArg, + Ins, DL, DAG, InVals, CLI.Callee.getNode(), CLI.RetTy); } /// LowerCallResult - Lower the result values of a call into the /// appropriate copies out of appropriate physical registers. SDValue MipsTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag, - CallingConv::ID CallConv, bool isVarArg, + CallingConv::ID CallConv, bool IsVarArg, const SmallVectorImpl<ISD::InputArg> &Ins, - DebugLoc dl, SelectionDAG &DAG, - SmallVectorImpl<SDValue> &InVals) const { + DebugLoc DL, SelectionDAG &DAG, + SmallVectorImpl<SDValue> &InVals, + const SDNode *CallNode, + const Type *RetTy) const { // Assign locations to each value returned by this call. SmallVector<CCValAssign, 16> RVLocs; - CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), + CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), getTargetMachine(), RVLocs, *DAG.getContext()); + MipsCC MipsCCInfo(CallConv, IsO32, CCInfo); - CCInfo.AnalyzeCallResult(Ins, RetCC_Mips); + MipsCCInfo.analyzeCallResult(Ins, getTargetMachine().Options.UseSoftFloat, + CallNode, RetTy); // Copy all of the result registers out of their specified physreg. for (unsigned i = 0; i != RVLocs.size(); ++i) { - Chain = DAG.getCopyFromReg(Chain, dl, RVLocs[i].getLocReg(), - RVLocs[i].getValVT(), InFlag).getValue(1); - InFlag = Chain.getValue(2); - InVals.push_back(Chain.getValue(0)); + SDValue Val = DAG.getCopyFromReg(Chain, DL, RVLocs[i].getLocReg(), + RVLocs[i].getLocVT(), InFlag); + Chain = Val.getValue(1); + InFlag = Val.getValue(2); + + if (RVLocs[i].getValVT() != RVLocs[i].getLocVT()) + Val = DAG.getNode(ISD::BITCAST, DL, RVLocs[i].getValVT(), Val); + + InVals.push_back(Val); } return Chain; @@ -3729,9 +2573,9 @@ MipsTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag, SDValue MipsTargetLowering::LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv, - bool isVarArg, + bool IsVarArg, const SmallVectorImpl<ISD::InputArg> &Ins, - DebugLoc dl, SelectionDAG &DAG, + DebugLoc DL, SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const { MachineFunction &MF = DAG.getMachineFunction(); @@ -3745,16 +2589,17 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, // Assign locations to all of the incoming arguments. SmallVector<CCValAssign, 16> ArgLocs; - CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), + CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), getTargetMachine(), ArgLocs, *DAG.getContext()); MipsCC MipsCCInfo(CallConv, IsO32, CCInfo); + Function::const_arg_iterator FuncArg = + DAG.getMachineFunction().getFunction()->arg_begin(); + bool UseSoftFloat = getTargetMachine().Options.UseSoftFloat; - MipsCCInfo.analyzeFormalArguments(Ins); + MipsCCInfo.analyzeFormalArguments(Ins, UseSoftFloat, FuncArg); MipsFI->setFormalArgInfo(CCInfo.getNextStackOffset(), MipsCCInfo.hasByValArg()); - Function::const_arg_iterator FuncArg = - DAG.getMachineFunction().getFunction()->arg_begin(); unsigned CurArgIdx = 0; MipsCC::byval_iterator ByValArg = MipsCCInfo.byval_begin(); @@ -3770,7 +2615,7 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, assert(Flags.getByValSize() && "ByVal args of size 0 should have been ignored by front-end."); assert(ByValArg != MipsCCInfo.byval_end()); - copyByValRegs(Chain, dl, OutChains, DAG, Flags, InVals, &*FuncArg, + copyByValRegs(Chain, DL, OutChains, DAG, Flags, InVals, &*FuncArg, MipsCCInfo, *ByValArg); ++ByValArg; continue; @@ -3796,8 +2641,8 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, // Transform the arguments stored on // physical registers into virtual ones - unsigned Reg = AddLiveIn(DAG.getMachineFunction(), ArgReg, RC); - SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT); + unsigned Reg = addLiveIn(DAG.getMachineFunction(), ArgReg, RC); + SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, Reg, RegVT); // If this is an 8 or 16-bit value, it has been passed promoted // to 32 bits. Insert an assert[sz]ext to capture this, then @@ -3809,22 +2654,24 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, else if (VA.getLocInfo() == CCValAssign::ZExt) Opcode = ISD::AssertZext; if (Opcode) - ArgValue = DAG.getNode(Opcode, dl, RegVT, ArgValue, + ArgValue = DAG.getNode(Opcode, DL, RegVT, ArgValue, DAG.getValueType(ValVT)); - ArgValue = DAG.getNode(ISD::TRUNCATE, dl, ValVT, ArgValue); + ArgValue = DAG.getNode(ISD::TRUNCATE, DL, ValVT, ArgValue); } - // Handle floating point arguments passed in integer registers. + // Handle floating point arguments passed in integer registers and + // long double arguments passed in floating point registers. if ((RegVT == MVT::i32 && ValVT == MVT::f32) || - (RegVT == MVT::i64 && ValVT == MVT::f64)) - ArgValue = DAG.getNode(ISD::BITCAST, dl, ValVT, ArgValue); + (RegVT == MVT::i64 && ValVT == MVT::f64) || + (RegVT == MVT::f64 && ValVT == MVT::i64)) + ArgValue = DAG.getNode(ISD::BITCAST, DL, ValVT, ArgValue); else if (IsO32 && RegVT == MVT::i32 && ValVT == MVT::f64) { - unsigned Reg2 = AddLiveIn(DAG.getMachineFunction(), + unsigned Reg2 = addLiveIn(DAG.getMachineFunction(), getNextIntArgReg(ArgReg), RC); - SDValue ArgValue2 = DAG.getCopyFromReg(Chain, dl, Reg2, RegVT); + SDValue ArgValue2 = DAG.getCopyFromReg(Chain, DL, Reg2, RegVT); if (!Subtarget->isLittle()) std::swap(ArgValue, ArgValue2); - ArgValue = DAG.getNode(MipsISD::BuildPairF64, dl, MVT::f64, + ArgValue = DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, ArgValue, ArgValue2); } @@ -3840,7 +2687,7 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, // Create load nodes to retrieve arguments from the stack SDValue FIN = DAG.getFrameIndex(FI, getPointerTy()); - InVals.push_back(DAG.getLoad(ValVT, dl, Chain, FIN, + InVals.push_back(DAG.getLoad(ValVT, DL, Chain, FIN, MachinePointerInfo::getFixedStack(FI), false, false, false, 0)); } @@ -3856,18 +2703,18 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, createVirtualRegister(getRegClassFor(IsN64 ? MVT::i64 : MVT::i32)); MipsFI->setSRetReturnReg(Reg); } - SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]); - Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Chain); + SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), DL, Reg, InVals[0]); + Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Copy, Chain); } - if (isVarArg) - writeVarArgRegs(OutChains, MipsCCInfo, Chain, dl, DAG); + if (IsVarArg) + writeVarArgRegs(OutChains, MipsCCInfo, Chain, DL, DAG); // All stores are grouped in one node to allow the matching between // the size of Ins and InVals. This only happens when on varg functions if (!OutChains.empty()) { OutChains.push_back(Chain); - Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, + Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &OutChains[0], OutChains.size()); } @@ -3880,42 +2727,48 @@ MipsTargetLowering::LowerFormalArguments(SDValue Chain, bool MipsTargetLowering::CanLowerReturn(CallingConv::ID CallConv, - MachineFunction &MF, bool isVarArg, + MachineFunction &MF, bool IsVarArg, const SmallVectorImpl<ISD::OutputArg> &Outs, LLVMContext &Context) const { SmallVector<CCValAssign, 16> RVLocs; - CCState CCInfo(CallConv, isVarArg, MF, getTargetMachine(), + CCState CCInfo(CallConv, IsVarArg, MF, getTargetMachine(), RVLocs, Context); return CCInfo.CheckReturn(Outs, RetCC_Mips); } SDValue MipsTargetLowering::LowerReturn(SDValue Chain, - CallingConv::ID CallConv, bool isVarArg, + CallingConv::ID CallConv, bool IsVarArg, const SmallVectorImpl<ISD::OutputArg> &Outs, const SmallVectorImpl<SDValue> &OutVals, - DebugLoc dl, SelectionDAG &DAG) const { - + DebugLoc DL, SelectionDAG &DAG) const { // CCValAssign - represent the assignment of // the return value to a location SmallVector<CCValAssign, 16> RVLocs; + MachineFunction &MF = DAG.getMachineFunction(); // CCState - Info about the registers and stack slot. - CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), - getTargetMachine(), RVLocs, *DAG.getContext()); + CCState CCInfo(CallConv, IsVarArg, MF, getTargetMachine(), RVLocs, + *DAG.getContext()); + MipsCC MipsCCInfo(CallConv, IsO32, CCInfo); - // Analize return values. - CCInfo.AnalyzeReturn(Outs, RetCC_Mips); + // Analyze return values. + MipsCCInfo.analyzeReturn(Outs, getTargetMachine().Options.UseSoftFloat, + MF.getFunction()->getReturnType()); SDValue Flag; SmallVector<SDValue, 4> RetOps(1, Chain); // Copy the result values into the output registers. for (unsigned i = 0; i != RVLocs.size(); ++i) { + SDValue Val = OutVals[i]; CCValAssign &VA = RVLocs[i]; assert(VA.isRegLoc() && "Can only return in registers!"); - Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), OutVals[i], Flag); + if (RVLocs[i].getValVT() != RVLocs[i].getLocVT()) + Val = DAG.getNode(ISD::BITCAST, DL, RVLocs[i].getLocVT(), Val); + + Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), Val, Flag); // Guarantee that all emitted copies are stuck together with flags. Flag = Chain.getValue(1); @@ -3926,17 +2779,16 @@ MipsTargetLowering::LowerReturn(SDValue Chain, // the sret argument into $v0 for the return. We saved the argument into // a virtual register in the entry block, so now we copy the value out // and into $v0. - if (DAG.getMachineFunction().getFunction()->hasStructRetAttr()) { - MachineFunction &MF = DAG.getMachineFunction(); + if (MF.getFunction()->hasStructRetAttr()) { MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); unsigned Reg = MipsFI->getSRetReturnReg(); if (!Reg) llvm_unreachable("sret virtual register not created in the entry block"); - SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy()); + SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, getPointerTy()); unsigned V0 = IsN64 ? Mips::V0_64 : Mips::V0; - Chain = DAG.getCopyToReg(Chain, dl, V0, Val, Flag); + Chain = DAG.getCopyToReg(Chain, DL, V0, Val, Flag); Flag = Chain.getValue(1); RetOps.push_back(DAG.getRegister(V0, getPointerTy())); } @@ -3948,7 +2800,7 @@ MipsTargetLowering::LowerReturn(SDValue Chain, RetOps.push_back(Flag); // Return on Mips is always a "jr $ra" - return DAG.getNode(MipsISD::Ret, dl, MVT::Other, &RetOps[0], RetOps.size()); + return DAG.getNode(MipsISD::Ret, DL, MVT::Other, &RetOps[0], RetOps.size()); } //===----------------------------------------------------------------------===// @@ -3981,6 +2833,8 @@ getConstraintType(const std::string &Constraint) const case 'l': case 'x': return C_RegisterClass; + case 'R': + return C_Memory; } } return TargetLowering::getConstraintType(Constraint); @@ -4029,6 +2883,9 @@ MipsTargetLowering::getSingleConstraintMatchWeight( if (isa<ConstantInt>(CallOperandVal)) weight = CW_Constant; break; + case 'R': + weight = CW_Memory; + break; } return weight; } @@ -4071,8 +2928,8 @@ getRegForInlineAsmConstraint(const std::string &Constraint, EVT VT) const return std::make_pair((unsigned)Mips::T9_64, &Mips::CPU64RegsRegClass); case 'l': // register suitable for indirect jump if (VT == MVT::i32) - return std::make_pair((unsigned)Mips::LO, &Mips::HILORegClass); - return std::make_pair((unsigned)Mips::LO64, &Mips::HILO64RegClass); + return std::make_pair((unsigned)Mips::LO, &Mips::LORegsRegClass); + return std::make_pair((unsigned)Mips::LO64, &Mips::LORegs64RegClass); case 'x': // register suitable for indirect jump // Fixme: Not triggering the use of both hi and low // This will generate an error message @@ -4229,6 +3086,46 @@ unsigned MipsTargetLowering::getJumpTableEncoding() const { return TargetLowering::getJumpTableEncoding(); } +/// This function returns true if CallSym is a long double emulation routine. +static bool isF128SoftLibCall(const char *CallSym) { + const char *const LibCalls[] = + {"__addtf3", "__divtf3", "__eqtf2", "__extenddftf2", "__extendsftf2", + "__fixtfdi", "__fixtfsi", "__fixtfti", "__fixunstfdi", "__fixunstfsi", + "__fixunstfti", "__floatditf", "__floatsitf", "__floattitf", + "__floatunditf", "__floatunsitf", "__floatuntitf", "__getf2", "__gttf2", + "__letf2", "__lttf2", "__multf3", "__netf2", "__powitf2", "__subtf3", + "__trunctfdf2", "__trunctfsf2", "__unordtf2", + "ceill", "copysignl", "cosl", "exp2l", "expl", "floorl", "fmal", "fmodl", + "log10l", "log2l", "logl", "nearbyintl", "powl", "rintl", "sinl", "sqrtl", + "truncl"}; + + const char * const *End = LibCalls + array_lengthof(LibCalls); + + // Check that LibCalls is sorted alphabetically. + MipsTargetLowering::LTStr Comp; + +#ifndef NDEBUG + for (const char * const *I = LibCalls; I < End - 1; ++I) + assert(Comp(*I, *(I + 1))); +#endif + + return std::binary_search(LibCalls, End, CallSym, Comp); +} + +/// This function returns true if Ty is fp128 or i128 which was originally a +/// fp128. +static bool originalTypeIsF128(const Type *Ty, const SDNode *CallNode) { + if (Ty->isFP128Ty()) + return true; + + const ExternalSymbolSDNode *ES = + dyn_cast_or_null<const ExternalSymbolSDNode>(CallNode); + + // If the Ty is i128 and the function being called is a long double emulation + // routine, then the original type is f128. + return (ES && Ty->isIntegerTy(128) && isF128SoftLibCall(ES->getSymbol())); +} + MipsTargetLowering::MipsCC::MipsCC(CallingConv::ID CC, bool IsO32_, CCState &Info) : CCInfo(Info), CallConv(CC), IsO32(IsO32_) { @@ -4238,7 +3135,8 @@ MipsTargetLowering::MipsCC::MipsCC(CallingConv::ID CC, bool IsO32_, void MipsTargetLowering::MipsCC:: analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args, - bool IsVarArg) { + bool IsVarArg, bool IsSoftFloat, const SDNode *CallNode, + std::vector<ArgListEntry> &FuncArgs) { assert((CallConv != CallingConv::Fast || !IsVarArg) && "CallingConv::Fast shouldn't be used for vararg functions."); @@ -4257,8 +3155,11 @@ analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args, if (IsVarArg && !Args[I].IsFixed) R = VarFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo); - else - R = FixedFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo); + else { + MVT RegVT = getRegVT(ArgVT, FuncArgs[Args[I].OrigArgIndex].Ty, CallNode, + IsSoftFloat); + R = FixedFn(I, ArgVT, RegVT, CCValAssign::Full, ArgFlags, CCInfo); + } if (R) { #ifndef NDEBUG @@ -4271,20 +3172,26 @@ analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Args, } void MipsTargetLowering::MipsCC:: -analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args) { +analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args, + bool IsSoftFloat, Function::const_arg_iterator FuncArg) { unsigned NumArgs = Args.size(); llvm::CCAssignFn *FixedFn = fixedArgFn(); + unsigned CurArgIdx = 0; for (unsigned I = 0; I != NumArgs; ++I) { MVT ArgVT = Args[I].VT; ISD::ArgFlagsTy ArgFlags = Args[I].Flags; + std::advance(FuncArg, Args[I].OrigArgIndex - CurArgIdx); + CurArgIdx = Args[I].OrigArgIndex; if (ArgFlags.isByVal()) { handleByValArg(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags); continue; } - if (!FixedFn(I, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo)) + MVT RegVT = getRegVT(ArgVT, FuncArg->getType(), 0, IsSoftFloat); + + if (!FixedFn(I, ArgVT, RegVT, CCValAssign::Full, ArgFlags, CCInfo)) continue; #ifndef NDEBUG @@ -4295,6 +3202,44 @@ analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Args) { } } +template<typename Ty> +void MipsTargetLowering::MipsCC:: +analyzeReturn(const SmallVectorImpl<Ty> &RetVals, bool IsSoftFloat, + const SDNode *CallNode, const Type *RetTy) const { + CCAssignFn *Fn; + + if (IsSoftFloat && originalTypeIsF128(RetTy, CallNode)) + Fn = RetCC_F128Soft; + else + Fn = RetCC_Mips; + + for (unsigned I = 0, E = RetVals.size(); I < E; ++I) { + MVT VT = RetVals[I].VT; + ISD::ArgFlagsTy Flags = RetVals[I].Flags; + MVT RegVT = this->getRegVT(VT, RetTy, CallNode, IsSoftFloat); + + if (Fn(I, VT, RegVT, CCValAssign::Full, Flags, this->CCInfo)) { +#ifndef NDEBUG + dbgs() << "Call result #" << I << " has unhandled type " + << EVT(VT).getEVTString() << '\n'; +#endif + llvm_unreachable(0); + } + } +} + +void MipsTargetLowering::MipsCC:: +analyzeCallResult(const SmallVectorImpl<ISD::InputArg> &Ins, bool IsSoftFloat, + const SDNode *CallNode, const Type *RetTy) const { + analyzeReturn(Ins, IsSoftFloat, CallNode, RetTy); +} + +void MipsTargetLowering::MipsCC:: +analyzeReturn(const SmallVectorImpl<ISD::OutputArg> &Outs, bool IsSoftFloat, + const Type *RetTy) const { + analyzeReturn(Outs, IsSoftFloat, 0, RetTy); +} + void MipsTargetLowering::MipsCC::handleByValArg(unsigned ValNo, MVT ValVT, MVT LocVT, @@ -4369,6 +3314,21 @@ void MipsTargetLowering::MipsCC::allocateRegs(ByValArgInfo &ByVal, CCInfo.AllocateReg(IntArgRegs[I], ShadowRegs[I]); } +MVT MipsTargetLowering::MipsCC::getRegVT(MVT VT, const Type *OrigTy, + const SDNode *CallNode, + bool IsSoftFloat) const { + if (IsSoftFloat || IsO32) + return VT; + + // Check if the original type was fp128. + if (originalTypeIsF128(OrigTy, CallNode)) { + assert(VT == MVT::i64); + return MVT::f64; + } + + return VT; +} + void MipsTargetLowering:: copyByValRegs(SDValue Chain, DebugLoc DL, std::vector<SDValue> &OutChains, SelectionDAG &DAG, const ISD::ArgFlagsTy &Flags, @@ -4401,7 +3361,7 @@ copyByValRegs(SDValue Chain, DebugLoc DL, std::vector<SDValue> &OutChains, for (unsigned I = 0; I < ByVal.NumRegs; ++I) { unsigned ArgReg = CC.intArgRegs()[ByVal.FirstIdx + I]; - unsigned VReg = AddLiveIn(MF, ArgReg, RC); + unsigned VReg = addLiveIn(MF, ArgReg, RC); unsigned Offset = I * CC.regSize(); SDValue StorePtr = DAG.getNode(ISD::ADD, DL, PtrTy, FIN, DAG.getConstant(Offset, PtrTy)); @@ -4543,7 +3503,7 @@ MipsTargetLowering::writeVarArgRegs(std::vector<SDValue> &OutChains, // in the caller's stack frame, while for N32/64, it is allocated in the // callee's stack frame. for (unsigned I = Idx; I < NumRegs; ++I, VaArgOffset += RegSize) { - unsigned Reg = AddLiveIn(MF, ArgRegs[I], RC); + unsigned Reg = addLiveIn(MF, ArgRegs[I], RC); SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, Reg, RegTy); FI = MFI->CreateFixedObject(RegSize, VaArgOffset, true); SDValue PtrOff = DAG.getFrameIndex(FI, getPointerTy()); diff --git a/lib/Target/Mips/MipsISelLowering.h b/lib/Target/Mips/MipsISelLowering.h index 47be1a013a..6c1e7ae3ec 100644 --- a/lib/Target/Mips/MipsISelLowering.h +++ b/lib/Target/Mips/MipsISelLowering.h @@ -19,6 +19,7 @@ #include "MipsSubtarget.h" #include "llvm/CodeGen/CallingConvLower.h" #include "llvm/CodeGen/SelectionDAG.h" +#include "llvm/IR/Function.h" #include "llvm/Target/TargetLowering.h" #include <deque> #include <string> @@ -67,6 +68,16 @@ namespace llvm { EH_RETURN, + // Node used to extract integer from accumulator. + ExtractLOHI, + + // Node used to insert integers to accumulator. + InsertLOHI, + + // Mult nodes. + Mult, + Multu, + // MAdd/Sub nodes MAdd, MAddu, @@ -76,6 +87,8 @@ namespace llvm { // DivRem(u) DivRem, DivRemU, + DivRem16, + DivRemU16, BuildPairF64, ExtractElementF64, @@ -130,6 +143,15 @@ namespace llvm { MSUB_DSP, MSUBU_DSP, + // DSP shift nodes. + SHLL_DSP, + SHRA_DSP, + SHRL_DSP, + + // DSP setcc and select_cc nodes. + SETCC_DSP, + SELECT_CC_DSP, + // Load/Store Left/Right nodes. LWL = ISD::FIRST_TARGET_MEMORY_OPCODE, LWR, @@ -151,9 +173,9 @@ namespace llvm { public: explicit MipsTargetLowering(MipsTargetMachine &TM); - virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i32; } + static const MipsTargetLowering *create(MipsTargetMachine &TM); - virtual bool allowsUnalignedMemoryAccesses (EVT VT, bool *Fast) const; + virtual MVT getScalarShiftAmountTy(EVT LHSTy) const { return MVT::i32; } virtual void LowerOperationWrapper(SDNode *N, SmallVectorImpl<SDValue> &Results, @@ -176,17 +198,34 @@ namespace llvm { EVT getSetCCResultType(EVT VT) const; virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const; - private: - void SetMips16LibcallName(RTLIB::Libcall, const char *Name); + virtual MachineBasicBlock * + EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const; + + struct LTStr { + bool operator()(const char *S1, const char *S2) const { + return strcmp(S1, S2) < 0; + } + }; + + protected: + SDValue getGlobalReg(SelectionDAG &DAG, EVT Ty) const; + + SDValue getAddrLocal(SDValue Op, SelectionDAG &DAG, bool HasMips64) const; - void setMips16HardFloatLibCalls(); + SDValue getAddrGlobal(SDValue Op, SelectionDAG &DAG, unsigned Flag) const; - unsigned int - getMips16HelperFunctionStubNumber(ArgListTy &Args) const; + SDValue getAddrGlobalLargeGOT(SDValue Op, SelectionDAG &DAG, + unsigned HiFlag, unsigned LoFlag) const; - const char *getMips16HelperFunction - (Type* RetTy, ArgListTy &Args, bool &needHelper) const; + /// This function fills Ops, which is the list of operands that will later + /// be used when a function call node is created. It also generates + /// copyToReg nodes to set up argument registers. + virtual void + getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const; /// ByValArgInfo - Byval argument information. struct ByValArgInfo { @@ -204,8 +243,20 @@ namespace llvm { MipsCC(CallingConv::ID CallConv, bool IsO32, CCState &Info); void analyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Outs, - bool IsVarArg); - void analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Ins); + bool IsVarArg, bool IsSoftFloat, + const SDNode *CallNode, + std::vector<ArgListEntry> &FuncArgs); + void analyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Ins, + bool IsSoftFloat, + Function::const_arg_iterator FuncArg); + + void analyzeCallResult(const SmallVectorImpl<ISD::InputArg> &Ins, + bool IsSoftFloat, const SDNode *CallNode, + const Type *RetTy) const; + + void analyzeReturn(const SmallVectorImpl<ISD::OutputArg> &Outs, + bool IsSoftFloat, const Type *RetTy) const; + const CCState &getCCInfo() const { return CCInfo; } /// hasByValArg - Returns true if function has byval arguments. @@ -248,6 +299,17 @@ namespace llvm { void allocateRegs(ByValArgInfo &ByVal, unsigned ByValSize, unsigned Align); + /// Return the type of the register which is used to pass an argument or + /// return a value. This function returns f64 if the argument is an i64 + /// value which has been generated as a result of softening an f128 value. + /// Otherwise, it just returns VT. + MVT getRegVT(MVT VT, const Type *OrigTy, const SDNode *CallNode, + bool IsSoftFloat) const; + + template<typename Ty> + void analyzeReturn(const SmallVectorImpl<Ty> &RetVals, bool IsSoftFloat, + const SDNode *CallNode, const Type *RetTy) const; + CCState &CCInfo; CallingConv::ID CallConv; bool IsO32; @@ -259,45 +321,46 @@ namespace llvm { bool HasMips64, IsN64, IsO32; + private: // Lower Operand helpers SDValue LowerCallResult(SDValue Chain, SDValue InFlag, CallingConv::ID CallConv, bool isVarArg, const SmallVectorImpl<ISD::InputArg> &Ins, DebugLoc dl, SelectionDAG &DAG, - SmallVectorImpl<SDValue> &InVals) const; + SmallVectorImpl<SDValue> &InVals, + const SDNode *CallNode, const Type *RetTy) const; // Lower Operand specifics - SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG& DAG) const; - SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG& DAG) const; - SDValue LowerShiftLeftParts(SDValue Op, SelectionDAG& DAG) const; - SDValue LowerShiftRightParts(SDValue Op, SelectionDAG& DAG, + SDValue lowerBR_JT(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerBRCOND(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerConstantPool(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerBlockAddress(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerJumpTable(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerSELECT(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerSETCC(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerVASTART(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerFABS(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerATOMIC_FENCE(SDValue Op, SelectionDAG& DAG) const; + SDValue lowerShiftLeftParts(SDValue Op, SelectionDAG& DAG) const; + SDValue lowerShiftRightParts(SDValue Op, SelectionDAG& DAG, bool IsSRA) const; - SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerADD(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerLOAD(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerSTORE(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerADD(SDValue Op, SelectionDAG &DAG) const; - /// IsEligibleForTailCallOptimization - Check whether the call is eligible + /// isEligibleForTailCallOptimization - Check whether the call is eligible /// for tail call optimization. - bool IsEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, - unsigned NextStackOffset, - const MipsFunctionInfo& FI) const; + virtual bool + isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, + unsigned NextStackOffset, + const MipsFunctionInfo& FI) const = 0; /// copyByValArg - Copy argument registers which were used to pass a byval /// argument to the stack. Create a stack frame object for the byval @@ -357,10 +420,6 @@ namespace llvm { const SmallVectorImpl<SDValue> &OutVals, DebugLoc dl, SelectionDAG &DAG) const; - virtual MachineBasicBlock * - EmitInstrWithCustomInserter(MachineInstr *MI, - MachineBasicBlock *MBB) const; - // Inline asm support ConstraintType getConstraintType(const std::string &Constraint) const; @@ -399,40 +458,20 @@ namespace llvm { virtual unsigned getJumpTableEncoding() const; - MachineBasicBlock *EmitBPOSGE32(MachineInstr *MI, - MachineBasicBlock *BB) const; - MachineBasicBlock *EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, + MachineBasicBlock *emitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode, bool Nand = false) const; - MachineBasicBlock *EmitAtomicBinaryPartword(MachineInstr *MI, + MachineBasicBlock *emitAtomicBinaryPartword(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size, unsigned BinOpcode, bool Nand = false) const; - MachineBasicBlock *EmitAtomicCmpSwap(MachineInstr *MI, + MachineBasicBlock *emitAtomicCmpSwap(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size) const; - MachineBasicBlock *EmitAtomicCmpSwapPartword(MachineInstr *MI, + MachineBasicBlock *emitAtomicCmpSwapPartword(MachineInstr *MI, MachineBasicBlock *BB, unsigned Size) const; - MachineBasicBlock *EmitSel16(unsigned Opc, MachineInstr *MI, - MachineBasicBlock *BB) const; - MachineBasicBlock *EmitSeliT16(unsigned Opc1, unsigned Opc2, - MachineInstr *MI, - MachineBasicBlock *BB) const; - - MachineBasicBlock *EmitSelT16(unsigned Opc1, unsigned Opc2, - MachineInstr *MI, - MachineBasicBlock *BB) const; - MachineBasicBlock *EmitFEXT_T8I816_ins(unsigned BtOpc, unsigned CmpOpc, - MachineInstr *MI, - MachineBasicBlock *BB) const; - MachineBasicBlock *EmitFEXT_T8I8I16_ins( - unsigned BtOpc, unsigned CmpiOpc, unsigned CmpiXOpc, - MachineInstr *MI, MachineBasicBlock *BB) const; - MachineBasicBlock *EmitFEXT_CCRX16_ins( - unsigned SltOpc, - MachineInstr *MI, MachineBasicBlock *BB) const; - MachineBasicBlock *EmitFEXT_CCRXI16_ins( - unsigned SltiOpc, unsigned SltiXOpc, - MachineInstr *MI, MachineBasicBlock *BB )const; - }; + + /// Create MipsTargetLowering objects. + const MipsTargetLowering *createMips16TargetLowering(MipsTargetMachine &TM); + const MipsTargetLowering *createMipsSETargetLowering(MipsTargetMachine &TM); } #endif // MipsISELLOWERING_H diff --git a/lib/Target/Mips/MipsInstrFPU.td b/lib/Target/Mips/MipsInstrFPU.td index d22400d211..af75392d11 100644 --- a/lib/Target/Mips/MipsInstrFPU.td +++ b/lib/Target/Mips/MipsInstrFPU.td @@ -503,32 +503,27 @@ let Predicates = [IsFP64bit, HasStdEnc] in { def : MipsPat<(f64 (fextend FGR32:$src)), (CVT_D64_S FGR32:$src)>; } -// Load/Store patterns. +// Patterns for loads/stores with a reg+imm operand. let AddedComplexity = 40 in { let Predicates = [IsN64, HasStdEnc] in { - def : MipsPat<(f32 (load addrRegImm:$a)), (LWC1_P8 addrRegImm:$a)>; - def : MipsPat<(store FGR32:$v, addrRegImm:$a), - (SWC1_P8 FGR32:$v, addrRegImm:$a)>; - def : MipsPat<(f64 (load addrRegImm:$a)), (LDC164_P8 addrRegImm:$a)>; - def : MipsPat<(store FGR64:$v, addrRegImm:$a), - (SDC164_P8 FGR64:$v, addrRegImm:$a)>; + def : LoadRegImmPat<LWC1_P8, f32, load>; + def : StoreRegImmPat<SWC1_P8, f32>; + def : LoadRegImmPat<LDC164_P8, f64, load>; + def : StoreRegImmPat<SDC164_P8, f64>; } let Predicates = [NotN64, HasStdEnc] in { - def : MipsPat<(f32 (load addrRegImm:$a)), (LWC1 addrRegImm:$a)>; - def : MipsPat<(store FGR32:$v, addrRegImm:$a), - (SWC1 FGR32:$v, addrRegImm:$a)>; + def : LoadRegImmPat<LWC1, f32, load>; + def : StoreRegImmPat<SWC1, f32>; } let Predicates = [NotN64, HasMips64, HasStdEnc] in { - def : MipsPat<(f64 (load addrRegImm:$a)), (LDC164 addrRegImm:$a)>; - def : MipsPat<(store FGR64:$v, addrRegImm:$a), - (SDC164 FGR64:$v, addrRegImm:$a)>; + def : LoadRegImmPat<LDC164, f64, load>; + def : StoreRegImmPat<SDC164, f64>; } let Predicates = [NotN64, NotMips64, HasStdEnc] in { - def : MipsPat<(f64 (load addrRegImm:$a)), (LDC1 addrRegImm:$a)>; - def : MipsPat<(store AFGR64:$v, addrRegImm:$a), - (SDC1 AFGR64:$v, addrRegImm:$a)>; + def : LoadRegImmPat<LDC1, f64, load>; + def : StoreRegImmPat<SDC1, f64>; } } diff --git a/lib/Target/Mips/MipsInstrFormats.td b/lib/Target/Mips/MipsInstrFormats.td index ee432c8753..ea07372221 100644 --- a/lib/Target/Mips/MipsInstrFormats.td +++ b/lib/Target/Mips/MipsInstrFormats.td @@ -36,6 +36,24 @@ def FrmFR : Format<4>; def FrmFI : Format<5>; def FrmOther : Format<6>; // Instruction w/ a custom format +class MMRel; + +def Std2MicroMips : InstrMapping { + let FilterClass = "MMRel"; + // Instructions with the same BaseOpcode and isNVStore values form a row. + let RowFields = ["BaseOpcode"]; + // Instructions with the same predicate sense form a column. + let ColFields = ["Arch"]; + // The key column is the unpredicated instructions. + let KeyCol = ["se"]; + // Value columns are PredSense=true and PredSense=false + let ValueCols = [["se"], ["micromips"]]; +} + +class StdArch { + string Arch = "se"; +} + // Generic Mips Format class MipsInst<dag outs, dag ins, string asmstr, list<dag> pattern, InstrItinClass itin, Format f>: Instruction @@ -74,9 +92,11 @@ class MipsInst<dag outs, dag ins, string asmstr, list<dag> pattern, // Mips32/64 Instruction Format class InstSE<dag outs, dag ins, string asmstr, list<dag> pattern, - InstrItinClass itin, Format f>: + InstrItinClass itin, Format f, string opstr = ""> : MipsInst<outs, ins, asmstr, pattern, itin, f> { let Predicates = [HasStdEnc]; + string BaseOpcode = opstr; + string Arch; } // Mips Pseudo Instructions Format @@ -192,7 +212,7 @@ class MFC3OP_FM<bits<6> op, bits<5> mfmt> let Inst{2-0} = sel; } -class ADD_FM<bits<6> op, bits<6> funct> { +class ADD_FM<bits<6> op, bits<6> funct> : StdArch { bits<5> rd; bits<5> rs; bits<5> rt; @@ -207,7 +227,7 @@ class ADD_FM<bits<6> op, bits<6> funct> { let Inst{5-0} = funct; } -class ADDI_FM<bits<6> op> { +class ADDI_FM<bits<6> op> : StdArch { bits<5> rs; bits<5> rt; bits<16> imm16; @@ -220,7 +240,7 @@ class ADDI_FM<bits<6> op> { let Inst{15-0} = imm16; } -class SRA_FM<bits<6> funct, bit rotate> { +class SRA_FM<bits<6> funct, bit rotate> : StdArch { bits<5> rd; bits<5> rt; bits<5> shamt; @@ -236,7 +256,7 @@ class SRA_FM<bits<6> funct, bit rotate> { let Inst{5-0} = funct; } -class SRLV_FM<bits<6> funct, bit rotate> { +class SRLV_FM<bits<6> funct, bit rotate> : StdArch { bits<5> rd; bits<5> rt; bits<5> rs; @@ -288,7 +308,7 @@ class B_FM { let Inst{15-0} = offset; } -class SLTI_FM<bits<6> op> { +class SLTI_FM<bits<6> op> : StdArch { bits<5> rt; bits<5> rs; bits<16> imm16; @@ -413,7 +433,7 @@ class SYNC_FM { let Inst{5-0} = 0xf; } -class MULT_FM<bits<6> op, bits<6> funct> { +class MULT_FM<bits<6> op, bits<6> funct> : StdArch { bits<5> rs; bits<5> rt; @@ -529,7 +549,7 @@ class MFC1_FM<bits<5> funct> { let Inst{10-0} = 0; } -class LW_FM<bits<6> op> { +class LW_FM<bits<6> op> : StdArch { bits<5> rt; bits<21> addr; diff --git a/lib/Target/Mips/MipsInstrInfo.cpp b/lib/Target/Mips/MipsInstrInfo.cpp index 76644c1584..ad92d41209 100644 --- a/lib/Target/Mips/MipsInstrInfo.cpp +++ b/lib/Target/Mips/MipsInstrInfo.cpp @@ -93,81 +93,11 @@ bool MipsInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, MachineBasicBlock *&FBB, SmallVectorImpl<MachineOperand> &Cond, - bool AllowModify) const -{ - - MachineBasicBlock::reverse_iterator I = MBB.rbegin(), REnd = MBB.rend(); - - // Skip all the debug instructions. - while (I != REnd && I->isDebugValue()) - ++I; - - if (I == REnd || !isUnpredicatedTerminator(&*I)) { - // If this block ends with no branches (it just falls through to its succ) - // just return false, leaving TBB/FBB null. - TBB = FBB = NULL; - return false; - } - - MachineInstr *LastInst = &*I; - unsigned LastOpc = LastInst->getOpcode(); - - // Not an analyzable branch (must be an indirect jump). - if (!GetAnalyzableBrOpc(LastOpc)) - return true; - - // Get the second to last instruction in the block. - unsigned SecondLastOpc = 0; - MachineInstr *SecondLastInst = NULL; - - if (++I != REnd) { - SecondLastInst = &*I; - SecondLastOpc = GetAnalyzableBrOpc(SecondLastInst->getOpcode()); - - // Not an analyzable branch (must be an indirect jump). - if (isUnpredicatedTerminator(SecondLastInst) && !SecondLastOpc) - return true; - } - - // If there is only one terminator instruction, process it. - if (!SecondLastOpc) { - // Unconditional branch - if (LastOpc == UncondBrOpc) { - TBB = LastInst->getOperand(0).getMBB(); - return false; - } - - // Conditional branch - AnalyzeCondBr(LastInst, LastOpc, TBB, Cond); - return false; - } + bool AllowModify) const { + SmallVector<MachineInstr*, 2> BranchInstrs; + BranchType BT = AnalyzeBranch(MBB, TBB, FBB, Cond, AllowModify, BranchInstrs); - // If we reached here, there are two branches. - // If there are three terminators, we don't know what sort of block this is. - if (++I != REnd && isUnpredicatedTerminator(&*I)) - return true; - - // If second to last instruction is an unconditional branch, - // analyze it and remove the last instruction. - if (SecondLastOpc == UncondBrOpc) { - // Return if the last instruction cannot be removed. - if (!AllowModify) - return true; - - TBB = SecondLastInst->getOperand(0).getMBB(); - LastInst->eraseFromParent(); - return false; - } - - // Conditional branch followed by an unconditional branch. - // The last one must be unconditional. - if (LastOpc != UncondBrOpc) - return true; - - AnalyzeCondBr(SecondLastInst, SecondLastOpc, TBB, Cond); - FBB = LastInst->getOperand(0).getMBB(); - - return false; + return (BT == BT_None) || (BT == BT_Indirect); } void MipsInstrInfo::BuildCondBr(MachineBasicBlock &MBB, @@ -256,6 +186,90 @@ ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const return false; } +MipsInstrInfo::BranchType MipsInstrInfo:: +AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, + MachineBasicBlock *&FBB, SmallVectorImpl<MachineOperand> &Cond, + bool AllowModify, + SmallVectorImpl<MachineInstr*> &BranchInstrs) const { + + MachineBasicBlock::reverse_iterator I = MBB.rbegin(), REnd = MBB.rend(); + + // Skip all the debug instructions. + while (I != REnd && I->isDebugValue()) + ++I; + + if (I == REnd || !isUnpredicatedTerminator(&*I)) { + // This block ends with no branches (it just falls through to its succ). + // Leave TBB/FBB null. + TBB = FBB = NULL; + return BT_NoBranch; + } + + MachineInstr *LastInst = &*I; + unsigned LastOpc = LastInst->getOpcode(); + BranchInstrs.push_back(LastInst); + + // Not an analyzable branch (e.g., indirect jump). + if (!GetAnalyzableBrOpc(LastOpc)) + return LastInst->isIndirectBranch() ? BT_Indirect : BT_None; + + // Get the second to last instruction in the block. + unsigned SecondLastOpc = 0; + MachineInstr *SecondLastInst = NULL; + + if (++I != REnd) { + SecondLastInst = &*I; + SecondLastOpc = GetAnalyzableBrOpc(SecondLastInst->getOpcode()); + + // Not an analyzable branch (must be an indirect jump). + if (isUnpredicatedTerminator(SecondLastInst) && !SecondLastOpc) + return BT_None; + } + + // If there is only one terminator instruction, process it. + if (!SecondLastOpc) { + // Unconditional branch + if (LastOpc == UncondBrOpc) { + TBB = LastInst->getOperand(0).getMBB(); + return BT_Uncond; + } + + // Conditional branch + AnalyzeCondBr(LastInst, LastOpc, TBB, Cond); + return BT_Cond; + } + + // If we reached here, there are two branches. + // If there are three terminators, we don't know what sort of block this is. + if (++I != REnd && isUnpredicatedTerminator(&*I)) + return BT_None; + + BranchInstrs.insert(BranchInstrs.begin(), SecondLastInst); + + // If second to last instruction is an unconditional branch, + // analyze it and remove the last instruction. + if (SecondLastOpc == UncondBrOpc) { + // Return if the last instruction cannot be removed. + if (!AllowModify) + return BT_None; + + TBB = SecondLastInst->getOperand(0).getMBB(); + LastInst->eraseFromParent(); + BranchInstrs.pop_back(); + return BT_Uncond; + } + + // Conditional branch followed by an unconditional branch. + // The last one must be unconditional. + if (LastOpc != UncondBrOpc) + return BT_None; + + AnalyzeCondBr(SecondLastInst, SecondLastOpc, TBB, Cond); + FBB = LastInst->getOperand(0).getMBB(); + + return BT_CondUncond; +} + /// Return the number of bytes of code the specified instruction may be. unsigned MipsInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const { switch (MI->getOpcode()) { diff --git a/lib/Target/Mips/MipsInstrInfo.h b/lib/Target/Mips/MipsInstrInfo.h index aca2bc7ae9..8c05d97bea 100644 --- a/lib/Target/Mips/MipsInstrInfo.h +++ b/lib/Target/Mips/MipsInstrInfo.h @@ -31,6 +31,15 @@ protected: unsigned UncondBrOpc; public: + enum BranchType { + BT_None, // Couldn't analyze branch. + BT_NoBranch, // No branches found. + BT_Uncond, // One unconditional branch. + BT_Cond, // One conditional branch. + BT_CondUncond, // A conditional branch followed by an unconditional branch. + BT_Indirect // One indirct branch. + }; + explicit MipsInstrInfo(MipsTargetMachine &TM, unsigned UncondBrOpc); static const MipsInstrInfo *create(MipsTargetMachine &TM); @@ -51,6 +60,12 @@ public: virtual bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const; + BranchType AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, + MachineBasicBlock *&FBB, + SmallVectorImpl<MachineOperand> &Cond, + bool AllowModify, + SmallVectorImpl<MachineInstr*> &BranchInstrs) const; + virtual MachineInstr* emitFrameIndexDebugValue(MachineFunction &MF, int FrameIx, uint64_t Offset, const MDNode *MDPtr, @@ -71,6 +86,36 @@ public: /// Return the number of bytes of code the specified instruction may be. unsigned GetInstSizeInBytes(const MachineInstr *MI) const; + virtual void storeRegToStackSlot(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + unsigned SrcReg, bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const { + storeRegToStack(MBB, MBBI, SrcReg, isKill, FrameIndex, RC, TRI, 0); + } + + virtual void loadRegFromStackSlot(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MBBI, + unsigned DestReg, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const { + loadRegFromStack(MBB, MBBI, DestReg, FrameIndex, RC, TRI, 0); + } + + virtual void storeRegToStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + unsigned SrcReg, bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const = 0; + + virtual void loadRegFromStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + unsigned DestReg, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const = 0; + protected: bool isZeroImm(const MachineOperand &op) const; diff --git a/lib/Target/Mips/MipsInstrInfo.td b/lib/Target/Mips/MipsInstrInfo.td index 64db6246a1..10d11dd835 100644 --- a/lib/Target/Mips/MipsInstrInfo.td +++ b/lib/Target/Mips/MipsInstrInfo.td @@ -23,13 +23,16 @@ def SDT_MipsCMov : SDTypeProfile<1, 4, [SDTCisSameAs<0, 1>, SDTCisInt<4>]>; def SDT_MipsCallSeqStart : SDCallSeqStart<[SDTCisVT<0, i32>]>; def SDT_MipsCallSeqEnd : SDCallSeqEnd<[SDTCisVT<0, i32>, SDTCisVT<1, i32>]>; -def SDT_MipsMAddMSub : SDTypeProfile<0, 4, - [SDTCisVT<0, i32>, SDTCisSameAs<0, 1>, - SDTCisSameAs<1, 2>, - SDTCisSameAs<2, 3>]>; -def SDT_MipsDivRem : SDTypeProfile<0, 2, - [SDTCisInt<0>, - SDTCisSameAs<0, 1>]>; +def SDT_ExtractLOHI : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisVT<1, untyped>, + SDTCisVT<2, i32>]>; +def SDT_InsertLOHI : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>, + SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>; +def SDT_MipsMultDiv : SDTypeProfile<1, 2, [SDTCisVT<0, untyped>, SDTCisInt<1>, + SDTCisSameAs<1, 2>]>; +def SDT_MipsMAddMSub : SDTypeProfile<1, 3, + [SDTCisVT<0, untyped>, SDTCisSameAs<0, 3>, + SDTCisVT<1, i32>, SDTCisSameAs<1, 2>]>; +def SDT_MipsDivRem16 : SDTypeProfile<0, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>]>; def SDT_MipsThreadPointer : SDTypeProfile<1, 0, [SDTCisPtrTy<0>]>; @@ -82,20 +85,27 @@ def callseq_end : SDNode<"ISD::CALLSEQ_END", SDT_MipsCallSeqEnd, [SDNPHasChain, SDNPSideEffect, SDNPOptInGlue, SDNPOutGlue]>; +// Node used to extract integer from LO/HI register. +def ExtractLOHI : SDNode<"MipsISD::ExtractLOHI", SDT_ExtractLOHI>; + +// Node used to insert 32-bit integers to LOHI register pair. +def InsertLOHI : SDNode<"MipsISD::InsertLOHI", SDT_InsertLOHI>; + +// Mult nodes. +def MipsMult : SDNode<"MipsISD::Mult", SDT_MipsMultDiv>; +def MipsMultu : SDNode<"MipsISD::Multu", SDT_MipsMultDiv>; + // MAdd*/MSub* nodes -def MipsMAdd : SDNode<"MipsISD::MAdd", SDT_MipsMAddMSub, - [SDNPOptInGlue, SDNPOutGlue]>; -def MipsMAddu : SDNode<"MipsISD::MAddu", SDT_MipsMAddMSub, - [SDNPOptInGlue, SDNPOutGlue]>; -def MipsMSub : SDNode<"MipsISD::MSub", SDT_MipsMAddMSub, - [SDNPOptInGlue, SDNPOutGlue]>; -def MipsMSubu : SDNode<"MipsISD::MSubu", SDT_MipsMAddMSub, - [SDNPOptInGlue, SDNPOutGlue]>; +def MipsMAdd : SDNode<"MipsISD::MAdd", SDT_MipsMAddMSub>; +def MipsMAddu : SDNode<"MipsISD::MAddu", SDT_MipsMAddMSub>; +def MipsMSub : SDNode<"MipsISD::MSub", SDT_MipsMAddMSub>; +def MipsMSubu : SDNode<"MipsISD::MSubu", SDT_MipsMAddMSub>; // DivRem(u) nodes -def MipsDivRem : SDNode<"MipsISD::DivRem", SDT_MipsDivRem, - [SDNPOutGlue]>; -def MipsDivRemU : SDNode<"MipsISD::DivRemU", SDT_MipsDivRem, +def MipsDivRem : SDNode<"MipsISD::DivRem", SDT_MipsMultDiv>; +def MipsDivRemU : SDNode<"MipsISD::DivRemU", SDT_MipsMultDiv>; +def MipsDivRem16 : SDNode<"MipsISD::DivRem16", SDT_MipsDivRem16, [SDNPOutGlue]>; +def MipsDivRemU16 : SDNode<"MipsISD::DivRemU16", SDT_MipsDivRem16, [SDNPOutGlue]>; // Target constant nodes that are not part of any isel patterns and remain @@ -169,6 +179,8 @@ def NoNaNsFPMath : Predicate<"TM.Options.NoNaNsFPMath">, AssemblerPredicate<"FeatureMips32">; def HasStdEnc : Predicate<"Subtarget.hasStandardEncoding()">, AssemblerPredicate<"!FeatureMips16">; +def NotDSP : Predicate<"!Subtarget.hasDSP()">; + def IsNaCl : Predicate<"Subtarget.isTargetNaCl()">; def IsNotNaCl : Predicate<"!Subtarget.isTargetNaCl()">; @@ -258,6 +270,7 @@ def mem : Operand<i32> { let MIOperandInfo = (ops CPURegs, simm16); let EncoderMethod = "getMemEncoding"; let ParserMatchClass = MipsMemAsmOperand; + let OperandType = "OPERAND_MEMORY"; } def mem64 : Operand<i64> { @@ -265,18 +278,21 @@ def mem64 : Operand<i64> { let MIOperandInfo = (ops CPU64Regs, simm16_64); let EncoderMethod = "getMemEncoding"; let ParserMatchClass = MipsMemAsmOperand; + let OperandType = "OPERAND_MEMORY"; } def mem_ea : Operand<i32> { let PrintMethod = "printMemOperandEA"; let MIOperandInfo = (ops CPURegs, simm16); let EncoderMethod = "getMemEncoding"; + let OperandType = "OPERAND_MEMORY"; } def mem_ea_64 : Operand<i64> { let PrintMethod = "printMemOperandEA"; let MIOperandInfo = (ops CPU64Regs, simm16_64); let EncoderMethod = "getMemEncoding"; + let OperandType = "OPERAND_MEMORY"; } // size operand of ext instruction @@ -301,6 +317,9 @@ def HI16 : SDNodeXForm<imm, [{ return getImm(N, (N->getZExtValue() >> 16) & 0xFFFF); }]>; +// Plus 1. +def Plus1 : SDNodeXForm<imm, [{ return getImm(N, N->getSExtValue() + 1); }]>; + // Node immediate fits as 16-bit sign extended on target immediate. // e.g. addi, andi def immSExt8 : PatLeaf<(imm), [{ return isInt<8>(N->getSExtValue()); }]>; @@ -333,6 +352,11 @@ def immLow16Zero : PatLeaf<(imm), [{ // shamt field must fit in 5 bits. def immZExt5 : ImmLeaf<i32, [{return Imm == (Imm & 0x1f);}]>; +// True if (N + 1) fits in 16-bit field. +def immSExt16Plus1 : PatLeaf<(imm), [{ + return isInt<17>(N->getSExtValue()) && isInt<16>(N->getSExtValue() + 1); +}]>; + // Mips Address Mode! SDNode frameindex could possibily be a match // since load and store instructions from stack used it. def addr : @@ -354,11 +378,9 @@ class ArithLogicR<string opstr, RegisterOperand RO, bit isComm = 0, SDPatternOperator OpNode = null_frag>: InstSE<(outs RO:$rd), (ins RO:$rs, RO:$rt), !strconcat(opstr, "\t$rd, $rs, $rt"), - [(set RO:$rd, (OpNode RO:$rs, RO:$rt))], Itin, FrmR> { + [(set RO:$rd, (OpNode RO:$rs, RO:$rt))], Itin, FrmR, opstr> { let isCommutable = isComm; let isReMaterializable = 1; - string BaseOpcode; - string Arch; } // Arithmetic and logical instructions with 2 register operands. @@ -367,15 +389,15 @@ class ArithLogicI<string opstr, Operand Od, RegisterOperand RO, SDPatternOperator OpNode = null_frag> : InstSE<(outs RO:$rt), (ins RO:$rs, Od:$imm16), !strconcat(opstr, "\t$rt, $rs, $imm16"), - [(set RO:$rt, (OpNode RO:$rs, imm_type:$imm16))], IIAlu, FrmI> { + [(set RO:$rt, (OpNode RO:$rs, imm_type:$imm16))], + IIAlu, FrmI, opstr> { let isReMaterializable = 1; } // Arithmetic Multiply ADD/SUB -class MArithR<string opstr, SDPatternOperator op = null_frag, bit isComm = 0> : +class MArithR<string opstr, bit isComm = 0> : InstSE<(outs), (ins CPURegsOpnd:$rs, CPURegsOpnd:$rt), - !strconcat(opstr, "\t$rs, $rt"), - [(op CPURegsOpnd:$rs, CPURegsOpnd:$rt, LO, HI)], IIImul, FrmR> { + !strconcat(opstr, "\t$rs, $rt"), [], IIImul, FrmR> { let Defs = [HI, LO]; let Uses = [HI, LO]; let isCommutable = isComm; @@ -385,7 +407,7 @@ class MArithR<string opstr, SDPatternOperator op = null_frag, bit isComm = 0> : class LogicNOR<string opstr, RegisterOperand RC>: InstSE<(outs RC:$rd), (ins RC:$rs, RC:$rt), !strconcat(opstr, "\t$rd, $rs, $rt"), - [(set RC:$rd, (not (or RC:$rs, RC:$rt)))], IIAlu, FrmR> { + [(set RC:$rd, (not (or RC:$rs, RC:$rt)))], IIAlu, FrmR, opstr> { let isCommutable = 1; } @@ -395,13 +417,13 @@ class shift_rotate_imm<string opstr, Operand ImmOpnd, SDPatternOperator PF = null_frag> : InstSE<(outs RC:$rd), (ins RC:$rt, ImmOpnd:$shamt), !strconcat(opstr, "\t$rd, $rt, $shamt"), - [(set RC:$rd, (OpNode RC:$rt, PF:$shamt))], IIAlu, FrmR>; + [(set RC:$rd, (OpNode RC:$rt, PF:$shamt))], IIAlu, FrmR, opstr>; class shift_rotate_reg<string opstr, RegisterOperand RC, SDPatternOperator OpNode = null_frag>: InstSE<(outs RC:$rd), (ins CPURegsOpnd:$rs, RC:$rt), !strconcat(opstr, "\t$rd, $rt, $rs"), - [(set RC:$rd, (OpNode RC:$rt, CPURegsOpnd:$rs))], IIAlu, FrmR>; + [(set RC:$rd, (OpNode RC:$rt, CPURegsOpnd:$rs))], IIAlu, FrmR, opstr>; // Load Upper Imediate class LoadUpper<string opstr, RegisterClass RC, Operand Imm>: @@ -421,33 +443,43 @@ class FMem<bits<6> op, dag outs, dag ins, string asmstr, list<dag> pattern, // Memory Load/Store class Load<string opstr, SDPatternOperator OpNode, RegisterClass RC, - Operand MemOpnd> : + Operand MemOpnd, ComplexPattern Addr, string ofsuffix> : InstSE<(outs RC:$rt), (ins MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"), - [(set RC:$rt, (OpNode addr:$addr))], NoItinerary, FrmI> { + [(set RC:$rt, (OpNode Addr:$addr))], NoItinerary, FrmI, + !strconcat(opstr, ofsuffix)> { let DecoderMethod = "DecodeMem"; let canFoldAsLoad = 1; + let mayLoad = 1; } class Store<string opstr, SDPatternOperator OpNode, RegisterClass RC, - Operand MemOpnd> : + Operand MemOpnd, ComplexPattern Addr, string ofsuffix> : InstSE<(outs), (ins RC:$rt, MemOpnd:$addr), !strconcat(opstr, "\t$rt, $addr"), - [(OpNode RC:$rt, addr:$addr)], NoItinerary, FrmI> { + [(OpNode RC:$rt, Addr:$addr)], NoItinerary, FrmI, + !strconcat(opstr, ofsuffix)> { let DecoderMethod = "DecodeMem"; + let mayStore = 1; } multiclass LoadM<string opstr, RegisterClass RC, - SDPatternOperator OpNode = null_frag> { - def NAME : Load<opstr, OpNode, RC, mem>, Requires<[NotN64, HasStdEnc]>; - def _P8 : Load<opstr, OpNode, RC, mem64>, Requires<[IsN64, HasStdEnc]> { + SDPatternOperator OpNode = null_frag, + ComplexPattern Addr = addr> { + def NAME : Load<opstr, OpNode, RC, mem, Addr, "">, + Requires<[NotN64, HasStdEnc]>; + def _P8 : Load<opstr, OpNode, RC, mem64, Addr, "_p8">, + Requires<[IsN64, HasStdEnc]> { let DecoderNamespace = "Mips64"; let isCodeGenOnly = 1; } } multiclass StoreM<string opstr, RegisterClass RC, - SDPatternOperator OpNode = null_frag> { - def NAME : Store<opstr, OpNode, RC, mem>, Requires<[NotN64, HasStdEnc]>; - def _P8 : Store<opstr, OpNode, RC, mem64>, Requires<[IsN64, HasStdEnc]> { + SDPatternOperator OpNode = null_frag, + ComplexPattern Addr = addr> { + def NAME : Store<opstr, OpNode, RC, mem, Addr, "">, + Requires<[NotN64, HasStdEnc]>; + def _P8 : Store<opstr, OpNode, RC, mem64, Addr, "_p8">, + Requires<[IsN64, HasStdEnc]> { let DecoderNamespace = "Mips64"; let isCodeGenOnly = 1; } @@ -517,14 +549,15 @@ class CBranchZero<string opstr, PatFrag cond_op, RegisterClass RC> : class SetCC_R<string opstr, PatFrag cond_op, RegisterClass RC> : InstSE<(outs CPURegsOpnd:$rd), (ins RC:$rs, RC:$rt), !strconcat(opstr, "\t$rd, $rs, $rt"), - [(set CPURegsOpnd:$rd, (cond_op RC:$rs, RC:$rt))], IIAlu, FrmR>; + [(set CPURegsOpnd:$rd, (cond_op RC:$rs, RC:$rt))], + IIAlu, FrmR, opstr>; class SetCC_I<string opstr, PatFrag cond_op, Operand Od, PatLeaf imm_type, RegisterClass RC>: InstSE<(outs CPURegsOpnd:$rt), (ins RC:$rs, Od:$imm16), !strconcat(opstr, "\t$rt, $rs, $imm16"), [(set CPURegsOpnd:$rt, (cond_op RC:$rs, imm_type:$imm16))], - IIAlu, FrmI>; + IIAlu, FrmI, opstr>; // Jump class JumpFJ<DAGOperand opnd, string opstr, SDPatternOperator operator, @@ -611,17 +644,40 @@ class SYNC_FT : class Mult<string opstr, InstrItinClass itin, RegisterOperand RO, list<Register> DefRegs> : InstSE<(outs), (ins RO:$rs, RO:$rt), !strconcat(opstr, "\t$rs, $rt"), [], - itin, FrmR> { + itin, FrmR, opstr> { let isCommutable = 1; let Defs = DefRegs; let neverHasSideEffects = 1; } -class Div<SDNode op, string opstr, InstrItinClass itin, RegisterOperand RO, +// Pseudo multiply/divide instruction with explicit accumulator register +// operands. +class MultDivPseudo<Instruction RealInst, RegisterClass R0, RegisterOperand R1, + SDPatternOperator OpNode, InstrItinClass Itin, + bit IsComm = 1, bit HasSideEffects = 0> : + PseudoSE<(outs R0:$ac), (ins R1:$rs, R1:$rt), + [(set R0:$ac, (OpNode R1:$rs, R1:$rt))], Itin>, + PseudoInstExpansion<(RealInst R1:$rs, R1:$rt)> { + let isCommutable = IsComm; + let hasSideEffects = HasSideEffects; +} + +// Pseudo multiply add/sub instruction with explicit accumulator register +// operands. +class MAddSubPseudo<Instruction RealInst, SDPatternOperator OpNode> + : PseudoSE<(outs ACRegs:$ac), + (ins CPURegsOpnd:$rs, CPURegsOpnd:$rt, ACRegs:$acin), + [(set ACRegs:$ac, + (OpNode CPURegsOpnd:$rs, CPURegsOpnd:$rt, ACRegs:$acin))], + IIImul>, + PseudoInstExpansion<(RealInst CPURegsOpnd:$rs, CPURegsOpnd:$rt)> { + string Constraints = "$acin = $ac"; +} + +class Div<string opstr, InstrItinClass itin, RegisterOperand RO, list<Register> DefRegs> : - InstSE<(outs), (ins RO:$rs, RO:$rt), - !strconcat(opstr, "\t$$zero, $rs, $rt"), [(op RO:$rs, RO:$rt)], itin, - FrmR> { + InstSE<(outs), (ins RO:$rs, RO:$rt), !strconcat(opstr, "\t$$zero, $rs, $rt"), + [], itin, FrmR> { let Defs = DefRegs; } @@ -815,6 +871,12 @@ let usesCustomInserter = 1 in { defm ATOMIC_CMP_SWAP_I32 : AtomicCmpSwap32<atomic_cmp_swap_32>; } +/// Pseudo instructions for loading and storing accumulator registers. +let isPseudo = 1 in { + defm LOAD_AC64 : LoadM<"load_ac64", ACRegs>; + defm STORE_AC64 : StoreM<"store_ac64", ACRegs>; +} + //===----------------------------------------------------------------------===// // Instruction definition //===----------------------------------------------------------------------===// @@ -823,60 +885,70 @@ let usesCustomInserter = 1 in { //===----------------------------------------------------------------------===// /// Arithmetic Instructions (ALU Immediate) -def ADDiu : ArithLogicI<"addiu", simm16, CPURegsOpnd, immSExt16, add>, +def ADDiu : MMRel, ArithLogicI<"addiu", simm16, CPURegsOpnd, immSExt16, add>, ADDI_FM<0x9>, IsAsCheapAsAMove; -def ADDi : ArithLogicI<"addi", simm16, CPURegsOpnd>, ADDI_FM<0x8>; -def SLTi : SetCC_I<"slti", setlt, simm16, immSExt16, CPURegs>, SLTI_FM<0xa>; -def SLTiu : SetCC_I<"sltiu", setult, simm16, immSExt16, CPURegs>, SLTI_FM<0xb>; -def ANDi : ArithLogicI<"andi", uimm16, CPURegsOpnd, immZExt16, and>, +def ADDi : MMRel, ArithLogicI<"addi", simm16, CPURegsOpnd>, ADDI_FM<0x8>; +def SLTi : MMRel, SetCC_I<"slti", setlt, simm16, immSExt16, CPURegs>, + SLTI_FM<0xa>; +def SLTiu : MMRel, SetCC_I<"sltiu", setult, simm16, immSExt16, CPURegs>, + SLTI_FM<0xb>; +def ANDi : MMRel, ArithLogicI<"andi", uimm16, CPURegsOpnd, immZExt16, and>, ADDI_FM<0xc>; -def ORi : ArithLogicI<"ori", uimm16, CPURegsOpnd, immZExt16, or>, +def ORi : MMRel, ArithLogicI<"ori", uimm16, CPURegsOpnd, immZExt16, or>, ADDI_FM<0xd>; -def XORi : ArithLogicI<"xori", uimm16, CPURegsOpnd, immZExt16, xor>, +def XORi : MMRel, ArithLogicI<"xori", uimm16, CPURegsOpnd, immZExt16, xor>, ADDI_FM<0xe>; -def LUi : LoadUpper<"lui", CPURegs, uimm16>, LUI_FM; +def LUi : MMRel, LoadUpper<"lui", CPURegs, uimm16>, LUI_FM; /// Arithmetic Instructions (3-Operand, R-Type) -def ADDu : ArithLogicR<"addu", CPURegsOpnd, 1, IIAlu, add>, ADD_FM<0, 0x21>; -def SUBu : ArithLogicR<"subu", CPURegsOpnd, 0, IIAlu, sub>, ADD_FM<0, 0x23>; -def MUL : ArithLogicR<"mul", CPURegsOpnd, 1, IIImul, mul>, ADD_FM<0x1c, 2>; -def ADD : ArithLogicR<"add", CPURegsOpnd>, ADD_FM<0, 0x20>; -def SUB : ArithLogicR<"sub", CPURegsOpnd>, ADD_FM<0, 0x22>; -def SLT : SetCC_R<"slt", setlt, CPURegs>, ADD_FM<0, 0x2a>; -def SLTu : SetCC_R<"sltu", setult, CPURegs>, ADD_FM<0, 0x2b>; -def AND : ArithLogicR<"and", CPURegsOpnd, 1, IIAlu, and>, ADD_FM<0, 0x24>; -def OR : ArithLogicR<"or", CPURegsOpnd, 1, IIAlu, or>, ADD_FM<0, 0x25>; -def XOR : ArithLogicR<"xor", CPURegsOpnd, 1, IIAlu, xor>, ADD_FM<0, 0x26>; -def NOR : LogicNOR<"nor", CPURegsOpnd>, ADD_FM<0, 0x27>; +def ADDu : MMRel, ArithLogicR<"addu", CPURegsOpnd, 1, IIAlu, add>, + ADD_FM<0, 0x21>; +def SUBu : MMRel, ArithLogicR<"subu", CPURegsOpnd, 0, IIAlu, sub>, + ADD_FM<0, 0x23>; +def MUL : MMRel, ArithLogicR<"mul", CPURegsOpnd, 1, IIImul, mul>, + ADD_FM<0x1c, 2>; +def ADD : MMRel, ArithLogicR<"add", CPURegsOpnd>, ADD_FM<0, 0x20>; +def SUB : MMRel, ArithLogicR<"sub", CPURegsOpnd>, ADD_FM<0, 0x22>; +def SLT : MMRel, SetCC_R<"slt", setlt, CPURegs>, ADD_FM<0, 0x2a>; +def SLTu : MMRel, SetCC_R<"sltu", setult, CPURegs>, ADD_FM<0, 0x2b>; +def AND : MMRel, ArithLogicR<"and", CPURegsOpnd, 1, IIAlu, and>, + ADD_FM<0, 0x24>; +def OR : MMRel, ArithLogicR<"or", CPURegsOpnd, 1, IIAlu, or>, + ADD_FM<0, 0x25>; +def XOR : MMRel, ArithLogicR<"xor", CPURegsOpnd, 1, IIAlu, xor>, + ADD_FM<0, 0x26>; +def NOR : MMRel, LogicNOR<"nor", CPURegsOpnd>, ADD_FM<0, 0x27>; /// Shift Instructions -def SLL : shift_rotate_imm<"sll", shamt, CPURegsOpnd, shl, immZExt5>, +def SLL : MMRel, shift_rotate_imm<"sll", shamt, CPURegsOpnd, shl, immZExt5>, SRA_FM<0, 0>; -def SRL : shift_rotate_imm<"srl", shamt, CPURegsOpnd, srl, immZExt5>, +def SRL : MMRel, shift_rotate_imm<"srl", shamt, CPURegsOpnd, srl, immZExt5>, SRA_FM<2, 0>; -def SRA : shift_rotate_imm<"sra", shamt, CPURegsOpnd, sra, immZExt5>, +def SRA : MMRel, shift_rotate_imm<"sra", shamt, CPURegsOpnd, sra, immZExt5>, SRA_FM<3, 0>; -def SLLV : shift_rotate_reg<"sllv", CPURegsOpnd, shl>, SRLV_FM<4, 0>; -def SRLV : shift_rotate_reg<"srlv", CPURegsOpnd, srl>, SRLV_FM<6, 0>; -def SRAV : shift_rotate_reg<"srav", CPURegsOpnd, sra>, SRLV_FM<7, 0>; +def SLLV : MMRel, shift_rotate_reg<"sllv", CPURegsOpnd, shl>, SRLV_FM<4, 0>; +def SRLV : MMRel, shift_rotate_reg<"srlv", CPURegsOpnd, srl>, SRLV_FM<6, 0>; +def SRAV : MMRel, shift_rotate_reg<"srav", CPURegsOpnd, sra>, SRLV_FM<7, 0>; // Rotate Instructions let Predicates = [HasMips32r2, HasStdEnc] in { - def ROTR : shift_rotate_imm<"rotr", shamt, CPURegsOpnd, rotr, immZExt5>, + def ROTR : MMRel, shift_rotate_imm<"rotr", shamt, CPURegsOpnd, rotr, + immZExt5>, SRA_FM<2, 1>; - def ROTRV : shift_rotate_reg<"rotrv", CPURegsOpnd, rotr>, SRLV_FM<6, 1>; + def ROTRV : MMRel, shift_rotate_reg<"rotrv", CPURegsOpnd, rotr>, + SRLV_FM<6, 1>; } /// Load and Store Instructions /// aligned -defm LB : LoadM<"lb", CPURegs, sextloadi8>, LW_FM<0x20>; -defm LBu : LoadM<"lbu", CPURegs, zextloadi8>, LW_FM<0x24>; -defm LH : LoadM<"lh", CPURegs, sextloadi16>, LW_FM<0x21>; -defm LHu : LoadM<"lhu", CPURegs, zextloadi16>, LW_FM<0x25>; -defm LW : LoadM<"lw", CPURegs, load>, LW_FM<0x23>; -defm SB : StoreM<"sb", CPURegs, truncstorei8>, LW_FM<0x28>; -defm SH : StoreM<"sh", CPURegs, truncstorei16>, LW_FM<0x29>; -defm SW : StoreM<"sw", CPURegs, store>, LW_FM<0x2b>; +defm LB : LoadM<"lb", CPURegs, sextloadi8>, MMRel, LW_FM<0x20>; +defm LBu : LoadM<"lbu", CPURegs, zextloadi8, addrDefault>, MMRel, LW_FM<0x24>; +defm LH : LoadM<"lh", CPURegs, sextloadi16, addrDefault>, MMRel, LW_FM<0x21>; +defm LHu : LoadM<"lhu", CPURegs, zextloadi16>, MMRel, LW_FM<0x25>; +defm LW : LoadM<"lw", CPURegs, load, addrDefault>, MMRel, LW_FM<0x23>; +defm SB : StoreM<"sb", CPURegs, truncstorei8>, MMRel, LW_FM<0x28>; +defm SH : StoreM<"sh", CPURegs, truncstorei16>, MMRel, LW_FM<0x29>; +defm SW : StoreM<"sw", CPURegs, store>, MMRel, LW_FM<0x2b>; /// load/store left/right defm LWL : LoadLeftRightM<"lwl", MipsLWL, CPURegs>, LW_FM<0x22>; @@ -943,12 +1015,17 @@ let Uses = [V0, V1], isTerminator = 1, isReturn = 1, isBarrier = 1 in { } /// Multiply and Divide Instructions. -def MULT : Mult<"mult", IIImul, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x18>; -def MULTu : Mult<"multu", IIImul, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x19>; -def SDIV : Div<MipsDivRem, "div", IIIdiv, CPURegsOpnd, [HI, LO]>, - MULT_FM<0, 0x1a>; -def UDIV : Div<MipsDivRemU, "divu", IIIdiv, CPURegsOpnd, [HI, LO]>, - MULT_FM<0, 0x1b>; +def MULT : MMRel, Mult<"mult", IIImul, CPURegsOpnd, [HI, LO]>, + MULT_FM<0, 0x18>; +def MULTu : MMRel, Mult<"multu", IIImul, CPURegsOpnd, [HI, LO]>, + MULT_FM<0, 0x19>; +def PseudoMULT : MultDivPseudo<MULT, ACRegs, CPURegsOpnd, MipsMult, IIImul>; +def PseudoMULTu : MultDivPseudo<MULTu, ACRegs, CPURegsOpnd, MipsMultu, IIImul>; +def SDIV : Div<"div", IIIdiv, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x1a>; +def UDIV : Div<"divu", IIIdiv, CPURegsOpnd, [HI, LO]>, MULT_FM<0, 0x1b>; +def PseudoSDIV : MultDivPseudo<SDIV, ACRegs, CPURegsOpnd, MipsDivRem, IIIdiv, 0>; +def PseudoUDIV : MultDivPseudo<UDIV, ACRegs, CPURegsOpnd, MipsDivRemU, IIIdiv, + 0>; def MTHI : MoveToLOHI<"mthi", CPURegs, [HI]>, MTLO_FM<0x11>; def MTLO : MoveToLOHI<"mtlo", CPURegs, [LO]>, MTLO_FM<0x13>; @@ -976,10 +1053,14 @@ def NOP : PseudoSE<(outs), (ins), []>, PseudoInstExpansion<(SLL ZERO, ZERO, 0)>; def LEA_ADDiu : EffectiveAddress<"addiu", CPURegs, mem_ea>, LW_FM<9>; // MADD*/MSUB* -def MADD : MArithR<"madd", MipsMAdd, 1>, MULT_FM<0x1c, 0>; -def MADDU : MArithR<"maddu", MipsMAddu, 1>, MULT_FM<0x1c, 1>; -def MSUB : MArithR<"msub", MipsMSub>, MULT_FM<0x1c, 4>; -def MSUBU : MArithR<"msubu", MipsMSubu>, MULT_FM<0x1c, 5>; +def MADD : MArithR<"madd", 1>, MULT_FM<0x1c, 0>; +def MADDU : MArithR<"maddu", 1>, MULT_FM<0x1c, 1>; +def MSUB : MArithR<"msub">, MULT_FM<0x1c, 4>; +def MSUBU : MArithR<"msubu">, MULT_FM<0x1c, 5>; +def PseudoMADD : MAddSubPseudo<MADD, MipsMAdd>; +def PseudoMADDU : MAddSubPseudo<MADDU, MipsMAddu>; +def PseudoMSUB : MAddSubPseudo<MSUB, MipsMSub>; +def PseudoMSUBU : MAddSubPseudo<MSUBU, MipsMSubu>; def RDHWR : ReadHardware<CPURegs, HWRegsOpnd>, RDHWR_FM; @@ -1010,7 +1091,7 @@ def : InstAlias<"move $dst, $src", (ADDu CPURegsOpnd:$dst, CPURegsOpnd:$src,ZERO), 1>, Requires<[NotMips64]>; def : InstAlias<"move $dst, $src", - (OR CPURegsOpnd:$dst, CPURegsOpnd:$src,ZERO), 0>, + (OR CPURegsOpnd:$dst, CPURegsOpnd:$src,ZERO), 1>, Requires<[NotMips64]>; def : InstAlias<"bal $offset", (BGEZAL RA, brtarget:$offset), 1>; def : InstAlias<"addu $rs, $rt, $imm", @@ -1022,6 +1103,9 @@ def : InstAlias<"and $rs, $rt, $imm", def : InstAlias<"j $rs", (JR CPURegs:$rs), 0>, Requires<[NotMips64]>; def : InstAlias<"jalr $rs", (JALR RA, CPURegs:$rs)>, Requires<[NotMips64]>; +def : InstAlias<"jal $rs", (JALR RA, CPURegs:$rs), 0>, Requires<[NotMips64]>; +def : InstAlias<"jal $rd,$rs", (JALR CPURegs:$rd, CPURegs:$rs), 0>, + Requires<[NotMips64]>; def : InstAlias<"not $rt, $rs", (NOR CPURegsOpnd:$rt, CPURegsOpnd:$rs, ZERO), 1>; def : InstAlias<"neg $rt, $rs", @@ -1031,8 +1115,11 @@ def : InstAlias<"negu $rt, $rs", def : InstAlias<"slt $rs, $rt, $imm", (SLTi CPURegsOpnd:$rs, CPURegs:$rt, simm16:$imm), 0>; def : InstAlias<"xor $rs, $rt, $imm", - (XORi CPURegsOpnd:$rs, CPURegsOpnd:$rt, simm16:$imm), 0>, + (XORi CPURegsOpnd:$rs, CPURegsOpnd:$rt, uimm16:$imm), 1>, Requires<[NotMips64]>; +def : InstAlias<"or $rs, $rt, $imm", + (ORi CPURegsOpnd:$rs, CPURegsOpnd:$rt, uimm16:$imm), 1>, + Requires<[NotMips64]>; def : InstAlias<"nop", (SLL ZERO, ZERO, 0), 1>; def : InstAlias<"mfc0 $rt, $rd", (MFC0_3OP CPURegsOpnd:$rt, CPURegsOpnd:$rd, 0), 0>; @@ -1068,6 +1155,13 @@ def LoadAddr32Imm : LoadAddressImm<"la", shamt,CPURegsOpnd>; // Arbitrary patterns that map to one or more instructions //===----------------------------------------------------------------------===// +// Load/store pattern templates. +class LoadRegImmPat<Instruction LoadInst, ValueType ValTy, PatFrag Node> : + MipsPat<(ValTy (Node addrRegImm:$a)), (LoadInst addrRegImm:$a)>; + +class StoreRegImmPat<Instruction StoreInst, ValueType ValTy> : + MipsPat<(store ValTy:$v, addrRegImm:$a), (StoreInst ValTy:$v, addrRegImm:$a)>; + // Small immediates def : MipsPat<(i32 immSExt16:$in), (ADDiu ZERO, imm:$in)>; @@ -1083,10 +1177,12 @@ def : MipsPat<(i32 imm:$imm), // Carry MipsPatterns def : MipsPat<(subc CPURegs:$lhs, CPURegs:$rhs), (SUBu CPURegs:$lhs, CPURegs:$rhs)>; -def : MipsPat<(addc CPURegs:$lhs, CPURegs:$rhs), - (ADDu CPURegs:$lhs, CPURegs:$rhs)>; -def : MipsPat<(addc CPURegs:$src, immSExt16:$imm), - (ADDiu CPURegs:$src, imm:$imm)>; +let Predicates = [HasStdEnc, NotDSP] in { + def : MipsPat<(addc CPURegs:$lhs, CPURegs:$rhs), + (ADDu CPURegs:$lhs, CPURegs:$rhs)>; + def : MipsPat<(addc CPURegs:$src, immSExt16:$imm), + (ADDiu CPURegs:$src, imm:$imm)>; +} // Call def : MipsPat<(MipsJmpLink (i32 tglobaladdr:$dst)), @@ -1245,6 +1341,24 @@ defm : SetgeImmPats<CPURegs, SLTi, SLTiu>; // bswap pattern def : MipsPat<(bswap CPURegs:$rt), (ROTR (WSBH CPURegs:$rt), 16)>; +// mflo/hi patterns. +def : MipsPat<(i32 (ExtractLOHI ACRegs:$ac, imm:$lohi_idx)), + (EXTRACT_SUBREG ACRegs:$ac, imm:$lohi_idx)>; + +// Load halfword/word patterns. +let AddedComplexity = 40 in { + let Predicates = [NotN64, HasStdEnc] in { + def : LoadRegImmPat<LBu, i32, zextloadi8>; + def : LoadRegImmPat<LH, i32, sextloadi16>; + def : LoadRegImmPat<LW, i32, load>; + } + let Predicates = [IsN64, HasStdEnc] in { + def : LoadRegImmPat<LBu_P8, i32, zextloadi8>; + def : LoadRegImmPat<LH_P8, i32, sextloadi16>; + def : LoadRegImmPat<LW_P8, i32, load>; + } +} + //===----------------------------------------------------------------------===// // Floating Point Support //===----------------------------------------------------------------------===// @@ -1263,3 +1377,6 @@ include "Mips16InstrInfo.td" include "MipsDSPInstrFormats.td" include "MipsDSPInstrInfo.td" +// Micromips +include "MicroMipsInstrFormats.td" +include "MicroMipsInstrInfo.td" diff --git a/lib/Target/Mips/MipsLongBranch.cpp b/lib/Target/Mips/MipsLongBranch.cpp index 2efe534053..bf5ad37031 100644 --- a/lib/Target/Mips/MipsLongBranch.cpp +++ b/lib/Target/Mips/MipsLongBranch.cpp @@ -399,6 +399,8 @@ static void emitGPDisp(MachineFunction &F, const MipsInstrInfo *TII) { } bool MipsLongBranch::runOnMachineFunction(MachineFunction &F) { + if (TM.getSubtarget<MipsSubtarget>().inMips16Mode()) + return false; if ((TM.getRelocationModel() == Reloc::PIC_) && TM.getSubtarget<MipsSubtarget>().isABI_O32() && F.getInfo<MipsFunctionInfo>()->globalBaseRegSet()) diff --git a/lib/Target/Mips/MipsModuleISelDAGToDAG.cpp b/lib/Target/Mips/MipsModuleISelDAGToDAG.cpp new file mode 100644 index 0000000000..c6abf17df3 --- /dev/null +++ b/lib/Target/Mips/MipsModuleISelDAGToDAG.cpp @@ -0,0 +1,34 @@ +//===----------------------------------------------------------------------===// +// Instruction Selector Subtarget Control +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// This file defines a pass used to change the subtarget for the +// Mips Instruction selector. +// +//===----------------------------------------------------------------------===// + +#include "MipsISelDAGToDAG.h" +#include "MipsModuleISelDAGToDAG.h" +#include "llvm/Support/Casting.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" + +namespace llvm { + +bool MipsModuleDAGToDAGISel::runOnMachineFunction(MachineFunction &MF) { + DEBUG(errs() << "In MipsModuleDAGToDAGISel::runMachineFunction\n"); + const_cast<MipsSubtarget&>(Subtarget).resetSubtarget(&MF); + return false; +} + +char MipsModuleDAGToDAGISel::ID = 0; + +} + + +llvm::FunctionPass *llvm::createMipsModuleISelDag(MipsTargetMachine &TM) { + return new MipsModuleDAGToDAGISel(TM); +} + + diff --git a/lib/Target/Mips/MipsModuleISelDAGToDAG.h b/lib/Target/Mips/MipsModuleISelDAGToDAG.h new file mode 100644 index 0000000000..fda35ae288 --- /dev/null +++ b/lib/Target/Mips/MipsModuleISelDAGToDAG.h @@ -0,0 +1,66 @@ +//===---- MipsModuleISelDAGToDAG.h - Change Subtarget --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines a pass used to change the subtarget for the +// Mips Instruction selector. +// +//===----------------------------------------------------------------------===// + +#ifndef MIPSMODULEISELDAGTODAG_H +#define MIPSMODULEISELDAGTODAG_H + +#include "Mips.h" +#include "MipsSubtarget.h" +#include "MipsTargetMachine.h" +#include "llvm/CodeGen/SelectionDAGISel.h" + + +//===----------------------------------------------------------------------===// +// Instruction Selector Implementation +//===----------------------------------------------------------------------===// + +//===----------------------------------------------------------------------===// +// MipsModuleDAGToDAGISel - MIPS specific code to select MIPS machine +// instructions for SelectionDAG operations. +//===----------------------------------------------------------------------===// +namespace llvm { + +class MipsModuleDAGToDAGISel : public MachineFunctionPass { +public: + + static char ID; + + explicit MipsModuleDAGToDAGISel(MipsTargetMachine &TM_) + : MachineFunctionPass(ID), + TM(TM_), Subtarget(TM.getSubtarget<MipsSubtarget>()) {} + + // Pass Name + virtual const char *getPassName() const { + return "MIPS DAG->DAG Pattern Instruction Selection"; + } + + virtual bool runOnMachineFunction(MachineFunction &MF); + + virtual SDNode *Select(SDNode *N) { + llvm_unreachable("unexpected"); + } + +protected: + /// Keep a pointer to the MipsSubtarget around so that we can make the right + /// decision when generating code for different targets. + const TargetMachine &TM; + const MipsSubtarget &Subtarget; +}; + +/// createMipsISelDag - This pass converts a legalized DAG into a +/// MIPS-specific DAG, ready for instruction scheduling. +FunctionPass *createMipsModuleISelDag(MipsTargetMachine &TM); +} + +#endif diff --git a/lib/Target/Mips/MipsOs16.cpp b/lib/Target/Mips/MipsOs16.cpp new file mode 100644 index 0000000000..1919077eeb --- /dev/null +++ b/lib/Target/Mips/MipsOs16.cpp @@ -0,0 +1,113 @@ +//===---- MipsOs16.cpp for Mips Option -Os16 --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines an optimization phase for the MIPS target. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "mips-os16" +#include "MipsOs16.h" +#include "llvm/IR/Module.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" + +namespace { + + // Figure out if we need float point based on the function signature. + // We need to move variables in and/or out of floating point + // registers because of the ABI + // + bool needsFPFromSig(Function &F) { + Type* RetType = F.getReturnType(); + switch (RetType->getTypeID()) { + case Type::FloatTyID: + case Type::DoubleTyID: + return true; + default: + ; + } + if (F.arg_size() >=1) { + Argument &Arg = F.getArgumentList().front(); + switch (Arg.getType()->getTypeID()) { + case Type::FloatTyID: + case Type::DoubleTyID: + return true; + default: + ; + } + } + return false; + } + + // Figure out if the function will need floating point operations + // + bool needsFP(Function &F) { + if (needsFPFromSig(F)) + return true; + for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) + for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); + I != E; ++I) { + const Instruction &Inst = *I; + switch (Inst.getOpcode()) { + case Instruction::FAdd: + case Instruction::FSub: + case Instruction::FMul: + case Instruction::FDiv: + case Instruction::FRem: + case Instruction::FPToUI: + case Instruction::FPToSI: + case Instruction::UIToFP: + case Instruction::SIToFP: + case Instruction::FPTrunc: + case Instruction::FPExt: + case Instruction::FCmp: + return true; + default: + ; + } + if (const CallInst *CI = dyn_cast<CallInst>(I)) { + DEBUG(dbgs() << "Working on call" << "\n"); + Function &F_ = *CI->getCalledFunction(); + if (needsFPFromSig(F_)) + return true; + } + } + return false; + } +} +namespace llvm { + + +bool MipsOs16::runOnModule(Module &M) { + DEBUG(errs() << "Run on Module MipsOs16\n"); + bool modified = false; + for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) { + if (F->isDeclaration()) continue; + DEBUG(dbgs() << "Working on " << F->getName() << "\n"); + if (needsFP(*F)) { + DEBUG(dbgs() << " need to compile as nomips16 \n"); + F->addFnAttr("nomips16"); + } + else { + F->addFnAttr("mips16"); + DEBUG(dbgs() << " no need to compile as nomips16 \n"); + } + } + return modified; +} + +char MipsOs16::ID = 0; + +} + +ModulePass *llvm::createMipsOs16(MipsTargetMachine &TM) { + return new MipsOs16; +} + + diff --git a/lib/Target/Mips/MipsOs16.h b/lib/Target/Mips/MipsOs16.h new file mode 100644 index 0000000000..21beef8549 --- /dev/null +++ b/lib/Target/Mips/MipsOs16.h @@ -0,0 +1,49 @@ +//===---- MipsOs16.h for Mips Option -Os16 --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file defines an optimization phase for the MIPS target. +// +//===----------------------------------------------------------------------===// + +#include "MCTargetDesc/MipsMCTargetDesc.h" +#include "MipsTargetMachine.h" +#include "llvm/Pass.h" +#include "llvm/Target/TargetMachine.h" + + + +#ifndef MIPSOS16_H +#define MIPSOS16_H + +using namespace llvm; + +namespace llvm { + +class MipsOs16 : public ModulePass { + +public: + static char ID; + + MipsOs16() : ModulePass(ID) { + + } + + virtual const char *getPassName() const { + return "MIPS Os16 Optimization"; + } + + virtual bool runOnModule(Module &M); + +}; + +ModulePass *createMipsOs16(MipsTargetMachine &TM); + +} + +#endif diff --git a/lib/Target/Mips/MipsRegisterInfo.cpp b/lib/Target/Mips/MipsRegisterInfo.cpp index 710428b87b..cd65635f01 100644 --- a/lib/Target/Mips/MipsRegisterInfo.cpp +++ b/lib/Target/Mips/MipsRegisterInfo.cpp @@ -155,7 +155,11 @@ getReservedRegs(const MachineFunction &MF) const { Reserved.set(Mips::HWR29_64); // Reserve DSP control register. - Reserved.set(Mips::DSPCtrl); + Reserved.set(Mips::DSPPos); + Reserved.set(Mips::DSPSCount); + Reserved.set(Mips::DSPCarry); + Reserved.set(Mips::DSPEFI); + Reserved.set(Mips::DSPOutFlag); // Reserve RA if in mips16 mode. if (Subtarget.inMips16Mode()) { diff --git a/lib/Target/Mips/MipsRegisterInfo.h b/lib/Target/Mips/MipsRegisterInfo.h index 13b2a6ac17..5ed5124139 100644 --- a/lib/Target/Mips/MipsRegisterInfo.h +++ b/lib/Target/Mips/MipsRegisterInfo.h @@ -58,7 +58,8 @@ public: int SPAdj, unsigned FIOperandNum, RegScavenger *RS = NULL) const; - void processFunctionBeforeFrameFinalized(MachineFunction &MF) const; + void processFunctionBeforeFrameFinalized(MachineFunction &MF, + RegScavenger *RS = NULL) const; /// Debug information queries. unsigned getFrameRegister(const MachineFunction &MF) const; @@ -67,6 +68,9 @@ public: unsigned getEHExceptionRegister() const; unsigned getEHHandlerRegister() const; + /// \brief Return GPR register class. + virtual const TargetRegisterClass *intRegClass(unsigned Size) const = 0; + private: virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo, int FrameIndex, uint64_t StackSize, diff --git a/lib/Target/Mips/MipsRegisterInfo.td b/lib/Target/Mips/MipsRegisterInfo.td index f93dd86c17..229f1677c0 100644 --- a/lib/Target/Mips/MipsRegisterInfo.td +++ b/lib/Target/Mips/MipsRegisterInfo.td @@ -16,6 +16,15 @@ def sub_fpodd : SubRegIndex; def sub_32 : SubRegIndex; def sub_lo : SubRegIndex; def sub_hi : SubRegIndex; +def sub_dsp16_19 : SubRegIndex; +def sub_dsp20 : SubRegIndex; +def sub_dsp21 : SubRegIndex; +def sub_dsp22 : SubRegIndex; +def sub_dsp23 : SubRegIndex; +} + +class Unallocatable { + bit isAllocatable = 0; } // We have banks of 32 registers each. @@ -54,6 +63,13 @@ class AFPR64<bits<16> Enc, string n, list<Register> subregs> let SubRegIndices = [sub_32]; } +// Accumulator Registers +class ACC<bits<16> Enc, string n, list<Register> subregs> + : MipsRegWithSubRegs<Enc, n, subregs> { + let SubRegIndices = [sub_lo, sub_hi]; + let CoveredBySubRegs = 1; +} + // Mips Hardware Registers class HWR<bits<16> Enc, string n> : MipsReg<Enc, n>; @@ -218,8 +234,14 @@ let Namespace = "Mips" in { def D31_64 : AFPR64<31, "f31", [F31]>, DwarfRegNum<[63]>; // Hi/Lo registers - def HI : Register<"hi">, DwarfRegNum<[64]>; - def LO : Register<"lo">, DwarfRegNum<[65]>; + def HI : Register<"ac0">, DwarfRegNum<[64]>; + def HI1 : Register<"ac1">, DwarfRegNum<[176]>; + def HI2 : Register<"ac2">, DwarfRegNum<[178]>; + def HI3 : Register<"ac3">, DwarfRegNum<[180]>; + def LO : Register<"ac0">, DwarfRegNum<[65]>; + def LO1 : Register<"ac1">, DwarfRegNum<[177]>; + def LO2 : Register<"ac2">, DwarfRegNum<[179]>; + def LO3 : Register<"ac3">, DwarfRegNum<[181]>; let SubRegIndices = [sub_32] in { def HI64 : RegisterWithSubRegs<"hi", [HI]>; @@ -240,13 +262,30 @@ let Namespace = "Mips" in { def HWR29_64 : MipsReg<29, "29">; // Accum registers - let SubRegIndices = [sub_lo, sub_hi] in - def AC0 : MipsRegWithSubRegs<0, "ac0", [LO, HI]>; - def AC1 : MipsReg<1, "ac1">; - def AC2 : MipsReg<2, "ac2">; - def AC3 : MipsReg<3, "ac3">; - - def DSPCtrl : Register<"dspctrl">; + def AC0 : ACC<0, "ac0", [LO, HI]>; + def AC1 : ACC<1, "ac1", [LO1, HI1]>; + def AC2 : ACC<2, "ac2", [LO2, HI2]>; + def AC3 : ACC<3, "ac3", [LO3, HI3]>; + + def AC0_64 : ACC<0, "ac0", [LO64, HI64]>; + + // DSP-ASE control register fields. + def DSPPos : Register<"">; + def DSPSCount : Register<"">; + def DSPCarry : Register<"">; + def DSPEFI : Register<"">; + def DSPOutFlag16_19 : Register<"">; + def DSPOutFlag20 : Register<"">; + def DSPOutFlag21 : Register<"">; + def DSPOutFlag22 : Register<"">; + def DSPOutFlag23 : Register<"">; + def DSPCCond : Register<"">; + + let SubRegIndices = [sub_dsp16_19, sub_dsp20, sub_dsp21, sub_dsp22, + sub_dsp23] in + def DSPOutFlag : RegisterWithSubRegs<"", [DSPOutFlag16_19, DSPOutFlag20, + DSPOutFlag21, DSPOutFlag22, + DSPOutFlag23]>; } //===----------------------------------------------------------------------===// @@ -291,9 +330,9 @@ def CPU16Regs : RegisterClass<"Mips", [i32], 32, (add // Callee save S0, S1)>; -def CPURAReg : RegisterClass<"Mips", [i32], 32, (add RA)>; +def CPURAReg : RegisterClass<"Mips", [i32], 32, (add RA)>, Unallocatable; -def CPUSPReg : RegisterClass<"Mips", [i32], 32, (add SP)>; +def CPUSPReg : RegisterClass<"Mips", [i32], 32, (add SP)>, Unallocatable; // 64bit fp: // * FGR64 - 32 64-bit registers @@ -319,19 +358,36 @@ def AFGR64 : RegisterClass<"Mips", [f64], 64, (add def FGR64 : RegisterClass<"Mips", [f64], 64, (sequence "D%u_64", 0, 31)>; // Condition Register for floating point operations -def CCR : RegisterClass<"Mips", [i32], 32, (add FCR31,FCC0)>; +def CCR : RegisterClass<"Mips", [i32], 32, (add FCR31,FCC0)>, Unallocatable; // Hi/Lo Registers -def HILO : RegisterClass<"Mips", [i32], 32, (add HI, LO)>; -def HILO64 : RegisterClass<"Mips", [i64], 64, (add HI64, LO64)>; +def LORegs : RegisterClass<"Mips", [i32], 32, (add LO)>; +def HIRegs : RegisterClass<"Mips", [i32], 32, (add HI)>; +def LORegsDSP : RegisterClass<"Mips", [i32], 32, (add LO, LO1, LO2, LO3)>; +def HIRegsDSP : RegisterClass<"Mips", [i32], 32, (add HI, HI1, HI2, HI3)>; +def LORegs64 : RegisterClass<"Mips", [i64], 64, (add LO64)>; +def HIRegs64 : RegisterClass<"Mips", [i64], 64, (add HI64)>; // Hardware registers -def HWRegs : RegisterClass<"Mips", [i32], 32, (add HWR29)>; -def HWRegs64 : RegisterClass<"Mips", [i64], 32, (add HWR29_64)>; +def HWRegs : RegisterClass<"Mips", [i32], 32, (add HWR29)>, Unallocatable; +def HWRegs64 : RegisterClass<"Mips", [i64], 64, (add HWR29_64)>, Unallocatable; // Accumulator Registers -def ACRegs : RegisterClass<"Mips", [i64], 64, (sequence "AC%u", 0, 3)>; +def ACRegs : RegisterClass<"Mips", [untyped], 64, (add AC0)> { + let Size = 64; +} + +def ACRegs128 : RegisterClass<"Mips", [untyped], 128, (add AC0_64)> { + let Size = 128; +} + +def ACRegsDSP : RegisterClass<"Mips", [untyped], 64, (sequence "AC%u", 0, 3)> { + let Size = 64; +} + +def DSPCC : RegisterClass<"Mips", [v4i8, v2i16], 32, (add DSPCCond)>; +// Register Operands. def CPURegsAsmOperand : AsmOperandClass { let Name = "CPURegsAsm"; let ParserMethod = "parseCPURegs"; diff --git a/lib/Target/Mips/MipsSEFrameLowering.cpp b/lib/Target/Mips/MipsSEFrameLowering.cpp index 0dd671376f..b295e911bd 100644 --- a/lib/Target/Mips/MipsSEFrameLowering.cpp +++ b/lib/Target/Mips/MipsSEFrameLowering.cpp @@ -29,6 +29,206 @@ using namespace llvm; +namespace { +typedef MachineBasicBlock::iterator Iter; + +/// Helper class to expand pseudos. +class ExpandPseudo { +public: + ExpandPseudo(MachineFunction &MF); + bool expand(); + +private: + bool expandInstr(MachineBasicBlock &MBB, Iter I); + void expandLoadCCond(MachineBasicBlock &MBB, Iter I); + void expandStoreCCond(MachineBasicBlock &MBB, Iter I); + void expandLoadACC(MachineBasicBlock &MBB, Iter I, unsigned RegSize); + void expandStoreACC(MachineBasicBlock &MBB, Iter I, unsigned RegSize); + bool expandCopy(MachineBasicBlock &MBB, Iter I); + bool expandCopyACC(MachineBasicBlock &MBB, Iter I, unsigned Dst, + unsigned Src, unsigned RegSize); + + MachineFunction &MF; + const MipsSEInstrInfo &TII; + const MipsRegisterInfo &RegInfo; + MachineRegisterInfo &MRI; +}; +} + +ExpandPseudo::ExpandPseudo(MachineFunction &MF_) + : MF(MF_), + TII(*static_cast<const MipsSEInstrInfo*>(MF.getTarget().getInstrInfo())), + RegInfo(TII.getRegisterInfo()), MRI(MF.getRegInfo()) {} + +bool ExpandPseudo::expand() { + bool Expanded = false; + + for (MachineFunction::iterator BB = MF.begin(), BBEnd = MF.end(); + BB != BBEnd; ++BB) + for (Iter I = BB->begin(), End = BB->end(); I != End;) + Expanded |= expandInstr(*BB, I++); + + return Expanded; +} + +bool ExpandPseudo::expandInstr(MachineBasicBlock &MBB, Iter I) { + switch(I->getOpcode()) { + case Mips::LOAD_CCOND_DSP: + case Mips::LOAD_CCOND_DSP_P8: + expandLoadCCond(MBB, I); + break; + case Mips::STORE_CCOND_DSP: + case Mips::STORE_CCOND_DSP_P8: + expandStoreCCond(MBB, I); + break; + case Mips::LOAD_AC64: + case Mips::LOAD_AC64_P8: + case Mips::LOAD_AC_DSP: + case Mips::LOAD_AC_DSP_P8: + expandLoadACC(MBB, I, 4); + break; + case Mips::LOAD_AC128: + case Mips::LOAD_AC128_P8: + expandLoadACC(MBB, I, 8); + break; + case Mips::STORE_AC64: + case Mips::STORE_AC64_P8: + case Mips::STORE_AC_DSP: + case Mips::STORE_AC_DSP_P8: + expandStoreACC(MBB, I, 4); + break; + case Mips::STORE_AC128: + case Mips::STORE_AC128_P8: + expandStoreACC(MBB, I, 8); + break; + case TargetOpcode::COPY: + if (!expandCopy(MBB, I)) + return false; + break; + default: + return false; + } + + MBB.erase(I); + return true; +} + +void ExpandPseudo::expandLoadCCond(MachineBasicBlock &MBB, Iter I) { + // load $vr, FI + // copy ccond, $vr + + assert(I->getOperand(0).isReg() && I->getOperand(1).isFI()); + + const TargetRegisterClass *RC = RegInfo.intRegClass(4); + unsigned VR = MRI.createVirtualRegister(RC); + unsigned Dst = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex(); + + TII.loadRegFromStack(MBB, I, VR, FI, RC, &RegInfo, 0); + BuildMI(MBB, I, I->getDebugLoc(), TII.get(TargetOpcode::COPY), Dst) + .addReg(VR, RegState::Kill); +} + +void ExpandPseudo::expandStoreCCond(MachineBasicBlock &MBB, Iter I) { + // copy $vr, ccond + // store $vr, FI + + assert(I->getOperand(0).isReg() && I->getOperand(1).isFI()); + + const TargetRegisterClass *RC = RegInfo.intRegClass(4); + unsigned VR = MRI.createVirtualRegister(RC); + unsigned Src = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex(); + + BuildMI(MBB, I, I->getDebugLoc(), TII.get(TargetOpcode::COPY), VR) + .addReg(Src, getKillRegState(I->getOperand(0).isKill())); + TII.storeRegToStack(MBB, I, VR, true, FI, RC, &RegInfo, 0); +} + +void ExpandPseudo::expandLoadACC(MachineBasicBlock &MBB, Iter I, + unsigned RegSize) { + // load $vr0, FI + // copy lo, $vr0 + // load $vr1, FI + 4 + // copy hi, $vr1 + + assert(I->getOperand(0).isReg() && I->getOperand(1).isFI()); + + const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize); + unsigned VR0 = MRI.createVirtualRegister(RC); + unsigned VR1 = MRI.createVirtualRegister(RC); + unsigned Dst = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex(); + unsigned Lo = RegInfo.getSubReg(Dst, Mips::sub_lo); + unsigned Hi = RegInfo.getSubReg(Dst, Mips::sub_hi); + DebugLoc DL = I->getDebugLoc(); + const MCInstrDesc &Desc = TII.get(TargetOpcode::COPY); + + TII.loadRegFromStack(MBB, I, VR0, FI, RC, &RegInfo, 0); + BuildMI(MBB, I, DL, Desc, Lo).addReg(VR0, RegState::Kill); + TII.loadRegFromStack(MBB, I, VR1, FI, RC, &RegInfo, RegSize); + BuildMI(MBB, I, DL, Desc, Hi).addReg(VR1, RegState::Kill); +} + +void ExpandPseudo::expandStoreACC(MachineBasicBlock &MBB, Iter I, + unsigned RegSize) { + // copy $vr0, lo + // store $vr0, FI + // copy $vr1, hi + // store $vr1, FI + 4 + + assert(I->getOperand(0).isReg() && I->getOperand(1).isFI()); + + const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize); + unsigned VR0 = MRI.createVirtualRegister(RC); + unsigned VR1 = MRI.createVirtualRegister(RC); + unsigned Src = I->getOperand(0).getReg(), FI = I->getOperand(1).getIndex(); + unsigned SrcKill = getKillRegState(I->getOperand(0).isKill()); + unsigned Lo = RegInfo.getSubReg(Src, Mips::sub_lo); + unsigned Hi = RegInfo.getSubReg(Src, Mips::sub_hi); + DebugLoc DL = I->getDebugLoc(); + + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR0).addReg(Lo, SrcKill); + TII.storeRegToStack(MBB, I, VR0, true, FI, RC, &RegInfo, 0); + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR1).addReg(Hi, SrcKill); + TII.storeRegToStack(MBB, I, VR1, true, FI, RC, &RegInfo, RegSize); +} + +bool ExpandPseudo::expandCopy(MachineBasicBlock &MBB, Iter I) { + unsigned Dst = I->getOperand(0).getReg(), Src = I->getOperand(1).getReg(); + + if (Mips::ACRegsDSPRegClass.contains(Dst, Src)) + return expandCopyACC(MBB, I, Dst, Src, 4); + + if (Mips::ACRegs128RegClass.contains(Dst, Src)) + return expandCopyACC(MBB, I, Dst, Src, 8); + + return false; +} + +bool ExpandPseudo::expandCopyACC(MachineBasicBlock &MBB, Iter I, unsigned Dst, + unsigned Src, unsigned RegSize) { + // copy $vr0, src_lo + // copy dst_lo, $vr0 + // copy $vr1, src_hi + // copy dst_hi, $vr1 + + const TargetRegisterClass *RC = RegInfo.intRegClass(RegSize); + unsigned VR0 = MRI.createVirtualRegister(RC); + unsigned VR1 = MRI.createVirtualRegister(RC); + unsigned SrcKill = getKillRegState(I->getOperand(1).isKill()); + unsigned DstLo = RegInfo.getSubReg(Dst, Mips::sub_lo); + unsigned DstHi = RegInfo.getSubReg(Dst, Mips::sub_hi); + unsigned SrcLo = RegInfo.getSubReg(Src, Mips::sub_lo); + unsigned SrcHi = RegInfo.getSubReg(Src, Mips::sub_hi); + DebugLoc DL = I->getDebugLoc(); + + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR0).addReg(SrcLo, SrcKill); + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), DstLo) + .addReg(VR0, RegState::Kill); + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), VR1).addReg(SrcHi, SrcKill); + BuildMI(MBB, I, DL, TII.get(TargetOpcode::COPY), DstHi) + .addReg(VR1, RegState::Kill); + return true; +} + unsigned MipsSEFrameLowering::ehDataReg(unsigned I) const { static const unsigned EhDataReg[] = { Mips::A0, Mips::A1, Mips::A2, Mips::A3 @@ -246,7 +446,10 @@ MipsSEFrameLowering::hasReservedCallFrame(const MachineFunction &MF) const { // Reserve call frame if the size of the maximum call frame fits into 16-bit // immediate field and there are no variable sized objects on the stack. - return isInt<16>(MFI->getMaxCallFrameSize()) && !MFI->hasVarSizedObjects(); + // Make sure the second register scavenger spill slot can be accessed with one + // instruction. + return isInt<16>(MFI->getMaxCallFrameSize() + getStackAlignment()) && + !MFI->hasVarSizedObjects(); } // Eliminate ADJCALLSTACKDOWN, ADJCALLSTACKUP pseudo instructions @@ -284,6 +487,18 @@ processFunctionBeforeCalleeSavedScan(MachineFunction &MF, if (MipsFI->callsEhReturn()) MipsFI->createEhDataRegsFI(); + // Expand pseudo instructions which load, store or copy accumulators. + // Add an emergency spill slot if a pseudo was expanded. + if (ExpandPseudo(MF).expand()) { + // The spill slot should be half the size of the accumulator. If target is + // mips64, it should be 64-bit, otherwise it should be 32-bt. + const TargetRegisterClass *RC = STI.hasMips64() ? + &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass; + int FI = MF.getFrameInfo()->CreateStackObject(RC->getSize(), + RC->getAlignment(), false); + RS->addScavengingFrameIndex(FI); + } + // Set scavenging frame index if necessary. uint64_t MaxSPOffset = MF.getInfo<MipsFunctionInfo>()->getIncomingArgSize() + estimateStackSize(MF); @@ -295,7 +510,7 @@ processFunctionBeforeCalleeSavedScan(MachineFunction &MF, &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass; int FI = MF.getFrameInfo()->CreateStackObject(RC->getSize(), RC->getAlignment(), false); - RS->setScavengingFrameIndex(FI); + RS->addScavengingFrameIndex(FI); } const MipsFrameLowering * diff --git a/lib/Target/Mips/MipsSEFrameLowering.h b/lib/Target/Mips/MipsSEFrameLowering.h index 7becd25455..193a66cc65 100644 --- a/lib/Target/Mips/MipsSEFrameLowering.h +++ b/lib/Target/Mips/MipsSEFrameLowering.h @@ -21,7 +21,7 @@ namespace llvm { class MipsSEFrameLowering : public MipsFrameLowering { public: explicit MipsSEFrameLowering(const MipsSubtarget &STI) - : MipsFrameLowering(STI) {} + : MipsFrameLowering(STI, STI.hasMips64() ? 16 : 8) {} /// emitProlog/emitEpilog - These methods insert prolog and epilog code into /// the function. diff --git a/lib/Target/Mips/MipsSEISelDAGToDAG.cpp b/lib/Target/Mips/MipsSEISelDAGToDAG.cpp new file mode 100644 index 0000000000..8a6523a5d4 --- /dev/null +++ b/lib/Target/Mips/MipsSEISelDAGToDAG.cpp @@ -0,0 +1,448 @@ +//===-- MipsSEISelDAGToDAG.cpp - A Dag to Dag Inst Selector for MipsSE ----===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsDAGToDAGISel specialized for mips32/64. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "mips-isel" +#include "MipsSEISelDAGToDAG.h" +#include "Mips.h" +#include "MCTargetDesc/MipsBaseInfo.h" +#include "MipsAnalyzeImmediate.h" +#include "MipsMachineFunction.h" +#include "MipsRegisterInfo.h" +#include "llvm/CodeGen/MachineConstantPool.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineFunction.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/CodeGen/SelectionDAGNodes.h" +#include "llvm/IR/GlobalValue.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/IR/Type.h" +#include "llvm/Support/CFG.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetMachine.h" +using namespace llvm; + +bool MipsSEDAGToDAGISel::runOnMachineFunction(MachineFunction &MF) { + if (Subtarget.inMips16Mode()) + return false; + return MipsDAGToDAGISel::runOnMachineFunction(MF); +} + +void MipsSEDAGToDAGISel::addDSPCtrlRegOperands(bool IsDef, MachineInstr &MI, + MachineFunction &MF) { + MachineInstrBuilder MIB(MF, &MI); + unsigned Mask = MI.getOperand(1).getImm(); + unsigned Flag = IsDef ? RegState::ImplicitDefine : RegState::Implicit; + + if (Mask & 1) + MIB.addReg(Mips::DSPPos, Flag); + + if (Mask & 2) + MIB.addReg(Mips::DSPSCount, Flag); + + if (Mask & 4) + MIB.addReg(Mips::DSPCarry, Flag); + + if (Mask & 8) + MIB.addReg(Mips::DSPOutFlag, Flag); + + if (Mask & 16) + MIB.addReg(Mips::DSPCCond, Flag); + + if (Mask & 32) + MIB.addReg(Mips::DSPEFI, Flag); +} + +bool MipsSEDAGToDAGISel::replaceUsesWithZeroReg(MachineRegisterInfo *MRI, + const MachineInstr& MI) { + unsigned DstReg = 0, ZeroReg = 0; + + // Check if MI is "addiu $dst, $zero, 0" or "daddiu $dst, $zero, 0". + if ((MI.getOpcode() == Mips::ADDiu) && + (MI.getOperand(1).getReg() == Mips::ZERO) && + (MI.getOperand(2).getImm() == 0)) { + DstReg = MI.getOperand(0).getReg(); + ZeroReg = Mips::ZERO; + } else if ((MI.getOpcode() == Mips::DADDiu) && + (MI.getOperand(1).getReg() == Mips::ZERO_64) && + (MI.getOperand(2).getImm() == 0)) { + DstReg = MI.getOperand(0).getReg(); + ZeroReg = Mips::ZERO_64; + } + + if (!DstReg) + return false; + + // Replace uses with ZeroReg. + for (MachineRegisterInfo::use_iterator U = MRI->use_begin(DstReg), + E = MRI->use_end(); U != E;) { + MachineOperand &MO = U.getOperand(); + unsigned OpNo = U.getOperandNo(); + MachineInstr *MI = MO.getParent(); + ++U; + + // Do not replace if it is a phi's operand or is tied to def operand. + if (MI->isPHI() || MI->isRegTiedToDefOperand(OpNo) || MI->isPseudo()) + continue; + + MO.setReg(ZeroReg); + } + + return true; +} + +void MipsSEDAGToDAGISel::initGlobalBaseReg(MachineFunction &MF) { + MipsFunctionInfo *MipsFI = MF.getInfo<MipsFunctionInfo>(); + + if (!MipsFI->globalBaseRegSet()) + return; + + MachineBasicBlock &MBB = MF.front(); + MachineBasicBlock::iterator I = MBB.begin(); + MachineRegisterInfo &RegInfo = MF.getRegInfo(); + const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo(); + DebugLoc DL = I != MBB.end() ? I->getDebugLoc() : DebugLoc(); + unsigned V0, V1, GlobalBaseReg = MipsFI->getGlobalBaseReg(); + const TargetRegisterClass *RC; + + if (Subtarget.isABI_N64()) + RC = (const TargetRegisterClass*)&Mips::CPU64RegsRegClass; + else + RC = (const TargetRegisterClass*)&Mips::CPURegsRegClass; + + V0 = RegInfo.createVirtualRegister(RC); + V1 = RegInfo.createVirtualRegister(RC); + + if (Subtarget.isABI_N64()) { + MF.getRegInfo().addLiveIn(Mips::T9_64); + MBB.addLiveIn(Mips::T9_64); + + // lui $v0, %hi(%neg(%gp_rel(fname))) + // daddu $v1, $v0, $t9 + // daddiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname))) + const GlobalValue *FName = MF.getFunction(); + BuildMI(MBB, I, DL, TII.get(Mips::LUi64), V0) + .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI); + BuildMI(MBB, I, DL, TII.get(Mips::DADDu), V1).addReg(V0) + .addReg(Mips::T9_64); + BuildMI(MBB, I, DL, TII.get(Mips::DADDiu), GlobalBaseReg).addReg(V1) + .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO); + return; + } + + if (MF.getTarget().getRelocationModel() == Reloc::Static) { + // Set global register to __gnu_local_gp. + // + // lui $v0, %hi(__gnu_local_gp) + // addiu $globalbasereg, $v0, %lo(__gnu_local_gp) + BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0) + .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_HI); + BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V0) + .addExternalSymbol("__gnu_local_gp", MipsII::MO_ABS_LO); + return; + } + + MF.getRegInfo().addLiveIn(Mips::T9); + MBB.addLiveIn(Mips::T9); + + if (Subtarget.isABI_N32()) { + // lui $v0, %hi(%neg(%gp_rel(fname))) + // addu $v1, $v0, $t9 + // addiu $globalbasereg, $v1, %lo(%neg(%gp_rel(fname))) + const GlobalValue *FName = MF.getFunction(); + BuildMI(MBB, I, DL, TII.get(Mips::LUi), V0) + .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_HI); + BuildMI(MBB, I, DL, TII.get(Mips::ADDu), V1).addReg(V0).addReg(Mips::T9); + BuildMI(MBB, I, DL, TII.get(Mips::ADDiu), GlobalBaseReg).addReg(V1) + .addGlobalAddress(FName, 0, MipsII::MO_GPOFF_LO); + return; + } + + assert(Subtarget.isABI_O32()); + + // For O32 ABI, the following instruction sequence is emitted to initialize + // the global base register: + // + // 0. lui $2, %hi(_gp_disp) + // 1. addiu $2, $2, %lo(_gp_disp) + // 2. addu $globalbasereg, $2, $t9 + // + // We emit only the last instruction here. + // + // GNU linker requires that the first two instructions appear at the beginning + // of a function and no instructions be inserted before or between them. + // The two instructions are emitted during lowering to MC layer in order to + // avoid any reordering. + // + // Register $2 (Mips::V0) is added to the list of live-in registers to ensure + // the value instruction 1 (addiu) defines is valid when instruction 2 (addu) + // reads it. + MF.getRegInfo().addLiveIn(Mips::V0); + MBB.addLiveIn(Mips::V0); + BuildMI(MBB, I, DL, TII.get(Mips::ADDu), GlobalBaseReg) + .addReg(Mips::V0).addReg(Mips::T9); +} + +void MipsSEDAGToDAGISel::processFunctionAfterISel(MachineFunction &MF) { + initGlobalBaseReg(MF); + + MachineRegisterInfo *MRI = &MF.getRegInfo(); + + for (MachineFunction::iterator MFI = MF.begin(), MFE = MF.end(); MFI != MFE; + ++MFI) + for (MachineBasicBlock::iterator I = MFI->begin(); I != MFI->end(); ++I) { + if (I->getOpcode() == Mips::RDDSP) + addDSPCtrlRegOperands(false, *I, MF); + else if (I->getOpcode() == Mips::WRDSP) + addDSPCtrlRegOperands(true, *I, MF); + else + replaceUsesWithZeroReg(MRI, *I); + } +} + +SDNode *MipsSEDAGToDAGISel::selectAddESubE(unsigned MOp, SDValue InFlag, + SDValue CmpLHS, DebugLoc DL, + SDNode *Node) const { + unsigned Opc = InFlag.getOpcode(); (void)Opc; + + assert(((Opc == ISD::ADDC || Opc == ISD::ADDE) || + (Opc == ISD::SUBC || Opc == ISD::SUBE)) && + "(ADD|SUB)E flag operand must come from (ADD|SUB)C/E insn"); + + SDValue Ops[] = { CmpLHS, InFlag.getOperand(1) }; + SDValue LHS = Node->getOperand(0), RHS = Node->getOperand(1); + EVT VT = LHS.getValueType(); + + SDNode *Carry = CurDAG->getMachineNode(Mips::SLTu, DL, VT, Ops); + SDNode *AddCarry = CurDAG->getMachineNode(Mips::ADDu, DL, VT, + SDValue(Carry, 0), RHS); + return CurDAG->SelectNodeTo(Node, MOp, VT, MVT::Glue, LHS, + SDValue(AddCarry, 0)); +} + +/// ComplexPattern used on MipsInstrInfo +/// Used on Mips Load/Store instructions +bool MipsSEDAGToDAGISel::selectAddrRegImm(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + EVT ValTy = Addr.getValueType(); + + // if Address is FI, get the TargetFrameIndex. + if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode>(Addr)) { + Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); + Offset = CurDAG->getTargetConstant(0, ValTy); + return true; + } + + // on PIC code Load GA + if (Addr.getOpcode() == MipsISD::Wrapper) { + Base = Addr.getOperand(0); + Offset = Addr.getOperand(1); + return true; + } + + if (TM.getRelocationModel() != Reloc::PIC_) { + if ((Addr.getOpcode() == ISD::TargetExternalSymbol || + Addr.getOpcode() == ISD::TargetGlobalAddress)) + return false; + } + + // Addresses of the form FI+const or FI|const + if (CurDAG->isBaseWithConstantOffset(Addr)) { + ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Addr.getOperand(1)); + if (isInt<16>(CN->getSExtValue())) { + + // If the first operand is a FI, get the TargetFI Node + if (FrameIndexSDNode *FIN = dyn_cast<FrameIndexSDNode> + (Addr.getOperand(0))) + Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); + else + Base = Addr.getOperand(0); + + Offset = CurDAG->getTargetConstant(CN->getZExtValue(), ValTy); + return true; + } + } + + // Operand is a result from an ADD. + if (Addr.getOpcode() == ISD::ADD) { + // When loading from constant pools, load the lower address part in + // the instruction itself. Example, instead of: + // lui $2, %hi($CPI1_0) + // addiu $2, $2, %lo($CPI1_0) + // lwc1 $f0, 0($2) + // Generate: + // lui $2, %hi($CPI1_0) + // lwc1 $f0, %lo($CPI1_0)($2) + if (Addr.getOperand(1).getOpcode() == MipsISD::Lo || + Addr.getOperand(1).getOpcode() == MipsISD::GPRel) { + SDValue Opnd0 = Addr.getOperand(1).getOperand(0); + if (isa<ConstantPoolSDNode>(Opnd0) || isa<GlobalAddressSDNode>(Opnd0) || + isa<JumpTableSDNode>(Opnd0)) { + Base = Addr.getOperand(0); + Offset = Opnd0; + return true; + } + } + } + + return false; +} + +bool MipsSEDAGToDAGISel::selectAddrDefault(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + Base = Addr; + Offset = CurDAG->getTargetConstant(0, Addr.getValueType()); + return true; +} + +bool MipsSEDAGToDAGISel::selectIntAddr(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + return selectAddrRegImm(Addr, Base, Offset) || + selectAddrDefault(Addr, Base, Offset); +} + +std::pair<bool, SDNode*> MipsSEDAGToDAGISel::selectNode(SDNode *Node) { + unsigned Opcode = Node->getOpcode(); + DebugLoc DL = Node->getDebugLoc(); + + /// + // Instruction Selection not handled by the auto-generated + // tablegen selection should be handled here. + /// + SDNode *Result; + + switch(Opcode) { + default: break; + + case ISD::SUBE: { + SDValue InFlag = Node->getOperand(2); + Result = selectAddESubE(Mips::SUBu, InFlag, InFlag.getOperand(0), DL, Node); + return std::make_pair(true, Result); + } + + case ISD::ADDE: { + if (Subtarget.hasDSP()) // Select DSP instructions, ADDSC and ADDWC. + break; + SDValue InFlag = Node->getOperand(2); + Result = selectAddESubE(Mips::ADDu, InFlag, InFlag.getValue(0), DL, Node); + return std::make_pair(true, Result); + } + + case ISD::ConstantFP: { + ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(Node); + if (Node->getValueType(0) == MVT::f64 && CN->isExactlyValue(+0.0)) { + if (Subtarget.hasMips64()) { + SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL, + Mips::ZERO_64, MVT::i64); + Result = CurDAG->getMachineNode(Mips::DMTC1, DL, MVT::f64, Zero); + } else { + SDValue Zero = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL, + Mips::ZERO, MVT::i32); + Result = CurDAG->getMachineNode(Mips::BuildPairF64, DL, MVT::f64, Zero, + Zero); + } + + return std::make_pair(true, Result); + } + break; + } + + case ISD::Constant: { + const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Node); + unsigned Size = CN->getValueSizeInBits(0); + + if (Size == 32) + break; + + MipsAnalyzeImmediate AnalyzeImm; + int64_t Imm = CN->getSExtValue(); + + const MipsAnalyzeImmediate::InstSeq &Seq = + AnalyzeImm.Analyze(Imm, Size, false); + + MipsAnalyzeImmediate::InstSeq::const_iterator Inst = Seq.begin(); + DebugLoc DL = CN->getDebugLoc(); + SDNode *RegOpnd; + SDValue ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd), + MVT::i64); + + // The first instruction can be a LUi which is different from other + // instructions (ADDiu, ORI and SLL) in that it does not have a register + // operand. + if (Inst->Opc == Mips::LUi64) + RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, ImmOpnd); + else + RegOpnd = + CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, + CurDAG->getRegister(Mips::ZERO_64, MVT::i64), + ImmOpnd); + + // The remaining instructions in the sequence are handled here. + for (++Inst; Inst != Seq.end(); ++Inst) { + ImmOpnd = CurDAG->getTargetConstant(SignExtend64<16>(Inst->ImmOpnd), + MVT::i64); + RegOpnd = CurDAG->getMachineNode(Inst->Opc, DL, MVT::i64, + SDValue(RegOpnd, 0), ImmOpnd); + } + + return std::make_pair(true, RegOpnd); + } + + case MipsISD::ThreadPointer: { + EVT PtrVT = TLI.getPointerTy(); + unsigned RdhwrOpc, SrcReg, DestReg; + + if (PtrVT == MVT::i32) { + RdhwrOpc = Mips::RDHWR; + SrcReg = Mips::HWR29; + DestReg = Mips::V1; + } else { + RdhwrOpc = Mips::RDHWR64; + SrcReg = Mips::HWR29_64; + DestReg = Mips::V1_64; + } + + SDNode *Rdhwr = + CurDAG->getMachineNode(RdhwrOpc, Node->getDebugLoc(), + Node->getValueType(0), + CurDAG->getRegister(SrcReg, PtrVT)); + SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), DL, DestReg, + SDValue(Rdhwr, 0)); + SDValue ResNode = CurDAG->getCopyFromReg(Chain, DL, DestReg, PtrVT); + ReplaceUses(SDValue(Node, 0), ResNode); + return std::make_pair(true, ResNode.getNode()); + } + + case MipsISD::InsertLOHI: { + unsigned RCID = Subtarget.hasDSP() ? Mips::ACRegsDSPRegClassID : + Mips::ACRegsRegClassID; + SDValue RegClass = CurDAG->getTargetConstant(RCID, MVT::i32); + SDValue LoIdx = CurDAG->getTargetConstant(Mips::sub_lo, MVT::i32); + SDValue HiIdx = CurDAG->getTargetConstant(Mips::sub_hi, MVT::i32); + const SDValue Ops[] = { RegClass, Node->getOperand(0), LoIdx, + Node->getOperand(1), HiIdx }; + SDNode *Res = CurDAG->getMachineNode(TargetOpcode::REG_SEQUENCE, DL, + MVT::Untyped, Ops); + return std::make_pair(true, Res); + } + } + + return std::make_pair(false, (SDNode*)NULL); +} + +FunctionPass *llvm::createMipsSEISelDag(MipsTargetMachine &TM) { + return new MipsSEDAGToDAGISel(TM); +} diff --git a/lib/Target/Mips/MipsSEISelDAGToDAG.h b/lib/Target/Mips/MipsSEISelDAGToDAG.h new file mode 100644 index 0000000000..a235e96b96 --- /dev/null +++ b/lib/Target/Mips/MipsSEISelDAGToDAG.h @@ -0,0 +1,63 @@ +//===-- MipsSEISelDAGToDAG.h - A Dag to Dag Inst Selector for MipsSE -----===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsDAGToDAGISel specialized for mips32/64. +// +//===----------------------------------------------------------------------===// + +#ifndef MIPSSEISELDAGTODAG_H +#define MIPSSEISELDAGTODAG_H + +#include "MipsISelDAGToDAG.h" + +namespace llvm { + +class MipsSEDAGToDAGISel : public MipsDAGToDAGISel { + +public: + explicit MipsSEDAGToDAGISel(MipsTargetMachine &TM) : MipsDAGToDAGISel(TM) {} + +private: + + virtual bool runOnMachineFunction(MachineFunction &MF); + + void addDSPCtrlRegOperands(bool IsDef, MachineInstr &MI, + MachineFunction &MF); + + bool replaceUsesWithZeroReg(MachineRegisterInfo *MRI, const MachineInstr&); + + std::pair<SDNode*, SDNode*> selectMULT(SDNode *N, unsigned Opc, DebugLoc dl, + EVT Ty, bool HasLo, bool HasHi); + + SDNode *selectAddESubE(unsigned MOp, SDValue InFlag, SDValue CmpLHS, + DebugLoc DL, SDNode *Node) const; + + virtual bool selectAddrRegImm(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectAddrDefault(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectIntAddr(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual std::pair<bool, SDNode*> selectNode(SDNode *Node); + + virtual void processFunctionAfterISel(MachineFunction &MF); + + // Insert instructions to initialize the global base register in the + // first MBB of the function. + void initGlobalBaseReg(MachineFunction &MF); +}; + +FunctionPass *createMipsSEISelDag(MipsTargetMachine &TM); + +} + +#endif diff --git a/lib/Target/Mips/MipsSEISelLowering.cpp b/lib/Target/Mips/MipsSEISelLowering.cpp new file mode 100644 index 0000000000..8544bb8910 --- /dev/null +++ b/lib/Target/Mips/MipsSEISelLowering.cpp @@ -0,0 +1,748 @@ +//===-- MipsSEISelLowering.cpp - MipsSE DAG Lowering Interface --*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsTargetLowering specialized for mips32/64. +// +//===----------------------------------------------------------------------===// +#include "MipsSEISelLowering.h" +#include "MipsRegisterInfo.h" +#include "MipsTargetMachine.h" +#include "llvm/CodeGen/MachineInstrBuilder.h" +#include "llvm/CodeGen/MachineRegisterInfo.h" +#include "llvm/IR/Intrinsics.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Target/TargetInstrInfo.h" + +using namespace llvm; + +static cl::opt<bool> +EnableMipsTailCalls("enable-mips-tail-calls", cl::Hidden, + cl::desc("MIPS: Enable tail calls."), cl::init(false)); + +MipsSETargetLowering::MipsSETargetLowering(MipsTargetMachine &TM) + : MipsTargetLowering(TM) { + // Set up the register classes + + clearRegisterClasses(); + + addRegisterClass(MVT::i32, &Mips::CPURegsRegClass); + + if (HasMips64) + addRegisterClass(MVT::i64, &Mips::CPU64RegsRegClass); + + if (Subtarget->hasDSP()) { + MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8}; + + for (unsigned i = 0; i < array_lengthof(VecTys); ++i) { + addRegisterClass(VecTys[i], &Mips::DSPRegsRegClass); + + // Expand all builtin opcodes. + for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc) + setOperationAction(Opc, VecTys[i], Expand); + + setOperationAction(ISD::ADD, VecTys[i], Legal); + setOperationAction(ISD::SUB, VecTys[i], Legal); + setOperationAction(ISD::LOAD, VecTys[i], Legal); + setOperationAction(ISD::STORE, VecTys[i], Legal); + setOperationAction(ISD::BITCAST, VecTys[i], Legal); + } + + setTargetDAGCombine(ISD::SHL); + setTargetDAGCombine(ISD::SRA); + setTargetDAGCombine(ISD::SRL); + setTargetDAGCombine(ISD::SETCC); + setTargetDAGCombine(ISD::VSELECT); + } + + if (Subtarget->hasDSPR2()) + setOperationAction(ISD::MUL, MVT::v2i16, Legal); + + if (!TM.Options.UseSoftFloat) { + addRegisterClass(MVT::f32, &Mips::FGR32RegClass); + + // When dealing with single precision only, use libcalls + if (!Subtarget->isSingleFloat()) { + if (HasMips64) + addRegisterClass(MVT::f64, &Mips::FGR64RegClass); + else + addRegisterClass(MVT::f64, &Mips::AFGR64RegClass); + } + } + + setOperationAction(ISD::SMUL_LOHI, MVT::i32, Custom); + setOperationAction(ISD::UMUL_LOHI, MVT::i32, Custom); + setOperationAction(ISD::MULHS, MVT::i32, Custom); + setOperationAction(ISD::MULHU, MVT::i32, Custom); + + if (HasMips64) { + setOperationAction(ISD::MULHS, MVT::i64, Custom); + setOperationAction(ISD::MULHU, MVT::i64, Custom); + setOperationAction(ISD::MUL, MVT::i64, Custom); + } + + setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i64, Custom); + setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i64, Custom); + + setOperationAction(ISD::SDIVREM, MVT::i32, Custom); + setOperationAction(ISD::UDIVREM, MVT::i32, Custom); + setOperationAction(ISD::SDIVREM, MVT::i64, Custom); + setOperationAction(ISD::UDIVREM, MVT::i64, Custom); + setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom); + setOperationAction(ISD::LOAD, MVT::i32, Custom); + setOperationAction(ISD::STORE, MVT::i32, Custom); + + setTargetDAGCombine(ISD::ADDE); + setTargetDAGCombine(ISD::SUBE); + + computeRegisterProperties(); +} + +const MipsTargetLowering * +llvm::createMipsSETargetLowering(MipsTargetMachine &TM) { + return new MipsSETargetLowering(TM); +} + + +bool +MipsSETargetLowering::allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const { + MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy; + + switch (SVT) { + case MVT::i64: + case MVT::i32: + if (Fast) + *Fast = true; + return true; + default: + return false; + } +} + +SDValue MipsSETargetLowering::LowerOperation(SDValue Op, + SelectionDAG &DAG) const { + switch(Op.getOpcode()) { + case ISD::SMUL_LOHI: return lowerMulDiv(Op, MipsISD::Mult, true, true, DAG); + case ISD::UMUL_LOHI: return lowerMulDiv(Op, MipsISD::Multu, true, true, DAG); + case ISD::MULHS: return lowerMulDiv(Op, MipsISD::Mult, false, true, DAG); + case ISD::MULHU: return lowerMulDiv(Op, MipsISD::Multu, false, true, DAG); + case ISD::MUL: return lowerMulDiv(Op, MipsISD::Mult, true, false, DAG); + case ISD::SDIVREM: return lowerMulDiv(Op, MipsISD::DivRem, true, true, DAG); + case ISD::UDIVREM: return lowerMulDiv(Op, MipsISD::DivRemU, true, true, + DAG); + case ISD::INTRINSIC_WO_CHAIN: return lowerINTRINSIC_WO_CHAIN(Op, DAG); + case ISD::INTRINSIC_W_CHAIN: return lowerINTRINSIC_W_CHAIN(Op, DAG); + } + + return MipsTargetLowering::LowerOperation(Op, DAG); +} + +// selectMADD - +// Transforms a subgraph in CurDAG if the following pattern is found: +// (addc multLo, Lo0), (adde multHi, Hi0), +// where, +// multHi/Lo: product of multiplication +// Lo0: initial value of Lo register +// Hi0: initial value of Hi register +// Return true if pattern matching was successful. +static bool selectMADD(SDNode *ADDENode, SelectionDAG *CurDAG) { + // ADDENode's second operand must be a flag output of an ADDC node in order + // for the matching to be successful. + SDNode *ADDCNode = ADDENode->getOperand(2).getNode(); + + if (ADDCNode->getOpcode() != ISD::ADDC) + return false; + + SDValue MultHi = ADDENode->getOperand(0); + SDValue MultLo = ADDCNode->getOperand(0); + SDNode *MultNode = MultHi.getNode(); + unsigned MultOpc = MultHi.getOpcode(); + + // MultHi and MultLo must be generated by the same node, + if (MultLo.getNode() != MultNode) + return false; + + // and it must be a multiplication. + if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI) + return false; + + // MultLo amd MultHi must be the first and second output of MultNode + // respectively. + if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0) + return false; + + // Transform this to a MADD only if ADDENode and ADDCNode are the only users + // of the values of MultNode, in which case MultNode will be removed in later + // phases. + // If there exist users other than ADDENode or ADDCNode, this function returns + // here, which will result in MultNode being mapped to a single MULT + // instruction node rather than a pair of MULT and MADD instructions being + // produced. + if (!MultHi.hasOneUse() || !MultLo.hasOneUse()) + return false; + + DebugLoc DL = ADDENode->getDebugLoc(); + + // Initialize accumulator. + SDValue ACCIn = CurDAG->getNode(MipsISD::InsertLOHI, DL, MVT::Untyped, + ADDCNode->getOperand(1), + ADDENode->getOperand(1)); + + // create MipsMAdd(u) node + MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MAddu : MipsISD::MAdd; + + SDValue MAdd = CurDAG->getNode(MultOpc, DL, MVT::Untyped, + MultNode->getOperand(0),// Factor 0 + MultNode->getOperand(1),// Factor 1 + ACCIn); + + // replace uses of adde and addc here + if (!SDValue(ADDCNode, 0).use_empty()) { + SDValue LoIdx = CurDAG->getConstant(Mips::sub_lo, MVT::i32); + SDValue LoOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MAdd, + LoIdx); + CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDCNode, 0), LoOut); + } + if (!SDValue(ADDENode, 0).use_empty()) { + SDValue HiIdx = CurDAG->getConstant(Mips::sub_hi, MVT::i32); + SDValue HiOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MAdd, + HiIdx); + CurDAG->ReplaceAllUsesOfValueWith(SDValue(ADDENode, 0), HiOut); + } + + return true; +} + +// selectMSUB - +// Transforms a subgraph in CurDAG if the following pattern is found: +// (addc Lo0, multLo), (sube Hi0, multHi), +// where, +// multHi/Lo: product of multiplication +// Lo0: initial value of Lo register +// Hi0: initial value of Hi register +// Return true if pattern matching was successful. +static bool selectMSUB(SDNode *SUBENode, SelectionDAG *CurDAG) { + // SUBENode's second operand must be a flag output of an SUBC node in order + // for the matching to be successful. + SDNode *SUBCNode = SUBENode->getOperand(2).getNode(); + + if (SUBCNode->getOpcode() != ISD::SUBC) + return false; + + SDValue MultHi = SUBENode->getOperand(1); + SDValue MultLo = SUBCNode->getOperand(1); + SDNode *MultNode = MultHi.getNode(); + unsigned MultOpc = MultHi.getOpcode(); + + // MultHi and MultLo must be generated by the same node, + if (MultLo.getNode() != MultNode) + return false; + + // and it must be a multiplication. + if (MultOpc != ISD::SMUL_LOHI && MultOpc != ISD::UMUL_LOHI) + return false; + + // MultLo amd MultHi must be the first and second output of MultNode + // respectively. + if (MultHi.getResNo() != 1 || MultLo.getResNo() != 0) + return false; + + // Transform this to a MSUB only if SUBENode and SUBCNode are the only users + // of the values of MultNode, in which case MultNode will be removed in later + // phases. + // If there exist users other than SUBENode or SUBCNode, this function returns + // here, which will result in MultNode being mapped to a single MULT + // instruction node rather than a pair of MULT and MSUB instructions being + // produced. + if (!MultHi.hasOneUse() || !MultLo.hasOneUse()) + return false; + + DebugLoc DL = SUBENode->getDebugLoc(); + + // Initialize accumulator. + SDValue ACCIn = CurDAG->getNode(MipsISD::InsertLOHI, DL, MVT::Untyped, + SUBCNode->getOperand(0), + SUBENode->getOperand(0)); + + // create MipsSub(u) node + MultOpc = MultOpc == ISD::UMUL_LOHI ? MipsISD::MSubu : MipsISD::MSub; + + SDValue MSub = CurDAG->getNode(MultOpc, DL, MVT::Glue, + MultNode->getOperand(0),// Factor 0 + MultNode->getOperand(1),// Factor 1 + ACCIn); + + // replace uses of sube and subc here + if (!SDValue(SUBCNode, 0).use_empty()) { + SDValue LoIdx = CurDAG->getConstant(Mips::sub_lo, MVT::i32); + SDValue LoOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MSub, + LoIdx); + CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBCNode, 0), LoOut); + } + if (!SDValue(SUBENode, 0).use_empty()) { + SDValue HiIdx = CurDAG->getConstant(Mips::sub_hi, MVT::i32); + SDValue HiOut = CurDAG->getNode(MipsISD::ExtractLOHI, DL, MVT::i32, MSub, + HiIdx); + CurDAG->ReplaceAllUsesOfValueWith(SDValue(SUBENode, 0), HiOut); + } + + return true; +} + +static SDValue performADDECombine(SDNode *N, SelectionDAG &DAG, + TargetLowering::DAGCombinerInfo &DCI, + const MipsSubtarget *Subtarget) { + if (DCI.isBeforeLegalize()) + return SDValue(); + + if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 && + selectMADD(N, &DAG)) + return SDValue(N, 0); + + return SDValue(); +} + +static SDValue performSUBECombine(SDNode *N, SelectionDAG &DAG, + TargetLowering::DAGCombinerInfo &DCI, + const MipsSubtarget *Subtarget) { + if (DCI.isBeforeLegalize()) + return SDValue(); + + if (Subtarget->hasMips32() && N->getValueType(0) == MVT::i32 && + selectMSUB(N, &DAG)) + return SDValue(N, 0); + + return SDValue(); +} + +static SDValue performDSPShiftCombine(unsigned Opc, SDNode *N, EVT Ty, + SelectionDAG &DAG, + const MipsSubtarget *Subtarget) { + // See if this is a vector splat immediate node. + APInt SplatValue, SplatUndef; + unsigned SplatBitSize; + bool HasAnyUndefs; + unsigned EltSize = Ty.getVectorElementType().getSizeInBits(); + BuildVectorSDNode *BV = dyn_cast<BuildVectorSDNode>(N->getOperand(1)); + + if (!BV || + !BV->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs, + EltSize, !Subtarget->isLittle()) || + (SplatBitSize != EltSize) || + (SplatValue.getZExtValue() >= EltSize)) + return SDValue(); + + return DAG.getNode(Opc, N->getDebugLoc(), Ty, N->getOperand(0), + DAG.getConstant(SplatValue.getZExtValue(), MVT::i32)); +} + +static SDValue performSHLCombine(SDNode *N, SelectionDAG &DAG, + TargetLowering::DAGCombinerInfo &DCI, + const MipsSubtarget *Subtarget) { + EVT Ty = N->getValueType(0); + + if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8)) + return SDValue(); + + return performDSPShiftCombine(MipsISD::SHLL_DSP, N, Ty, DAG, Subtarget); +} + +static SDValue performSRACombine(SDNode *N, SelectionDAG &DAG, + TargetLowering::DAGCombinerInfo &DCI, + const MipsSubtarget *Subtarget) { + EVT Ty = N->getValueType(0); + + if ((Ty != MVT::v2i16) && ((Ty != MVT::v4i8) || !Subtarget->hasDSPR2())) + return SDValue(); + + return performDSPShiftCombine(MipsISD::SHRA_DSP, N, Ty, DAG, Subtarget); +} + + +static SDValue performSRLCombine(SDNode *N, SelectionDAG &DAG, + TargetLowering::DAGCombinerInfo &DCI, + const MipsSubtarget *Subtarget) { + EVT Ty = N->getValueType(0); + + if (((Ty != MVT::v2i16) || !Subtarget->hasDSPR2()) && (Ty != MVT::v4i8)) + return SDValue(); + + return performDSPShiftCombine(MipsISD::SHRL_DSP, N, Ty, DAG, Subtarget); +} + +static bool isLegalDSPCondCode(EVT Ty, ISD::CondCode CC) { + bool IsV216 = (Ty == MVT::v2i16); + + switch (CC) { + case ISD::SETEQ: + case ISD::SETNE: return true; + case ISD::SETLT: + case ISD::SETLE: + case ISD::SETGT: + case ISD::SETGE: return IsV216; + case ISD::SETULT: + case ISD::SETULE: + case ISD::SETUGT: + case ISD::SETUGE: return !IsV216; + default: return false; + } +} + +static SDValue performSETCCCombine(SDNode *N, SelectionDAG &DAG) { + EVT Ty = N->getValueType(0); + + if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8)) + return SDValue(); + + if (!isLegalDSPCondCode(Ty, cast<CondCodeSDNode>(N->getOperand(2))->get())) + return SDValue(); + + return DAG.getNode(MipsISD::SETCC_DSP, N->getDebugLoc(), Ty, N->getOperand(0), + N->getOperand(1), N->getOperand(2)); +} + +static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG) { + EVT Ty = N->getValueType(0); + + if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8)) + return SDValue(); + + SDValue SetCC = N->getOperand(0); + + if (SetCC.getOpcode() != MipsISD::SETCC_DSP) + return SDValue(); + + return DAG.getNode(MipsISD::SELECT_CC_DSP, N->getDebugLoc(), Ty, + SetCC.getOperand(0), SetCC.getOperand(1), N->getOperand(1), + N->getOperand(2), SetCC.getOperand(2)); +} + +SDValue +MipsSETargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const { + SelectionDAG &DAG = DCI.DAG; + SDValue Val; + + switch (N->getOpcode()) { + case ISD::ADDE: + return performADDECombine(N, DAG, DCI, Subtarget); + case ISD::SUBE: + return performSUBECombine(N, DAG, DCI, Subtarget); + case ISD::SHL: + return performSHLCombine(N, DAG, DCI, Subtarget); + case ISD::SRA: + return performSRACombine(N, DAG, DCI, Subtarget); + case ISD::SRL: + return performSRLCombine(N, DAG, DCI, Subtarget); + case ISD::VSELECT: + return performVSELECTCombine(N, DAG); + case ISD::SETCC: { + Val = performSETCCCombine(N, DAG); + break; + } + } + + if (Val.getNode()) + return Val; + + return MipsTargetLowering::PerformDAGCombine(N, DCI); +} + +MachineBasicBlock * +MipsSETargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, + MachineBasicBlock *BB) const { + switch (MI->getOpcode()) { + default: + return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB); + case Mips::BPOSGE32_PSEUDO: + return emitBPOSGE32(MI, BB); + } +} + +bool MipsSETargetLowering:: +isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, + unsigned NextStackOffset, + const MipsFunctionInfo& FI) const { + if (!EnableMipsTailCalls) + return false; + + // Return false if either the callee or caller has a byval argument. + if (MipsCCInfo.hasByValArg() || FI.hasByvalArg()) + return false; + + // Return true if the callee's argument area is no larger than the + // caller's. + return NextStackOffset <= FI.getIncomingArgSize(); +} + +void MipsSETargetLowering:: +getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const { + // T9 should contain the address of the callee function if + // -reloction-model=pic or it is an indirect call. + if (IsPICCall || !GlobalOrExternal) { + unsigned T9Reg = IsN64 ? Mips::T9_64 : Mips::T9; + RegsToPass.push_front(std::make_pair(T9Reg, Callee)); + } else + Ops.push_back(Callee); + + MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal, + InternalLinkage, CLI, Callee, Chain); +} + +SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc, + bool HasLo, bool HasHi, + SelectionDAG &DAG) const { + EVT Ty = Op.getOperand(0).getValueType(); + DebugLoc DL = Op.getDebugLoc(); + SDValue Mult = DAG.getNode(NewOpc, DL, MVT::Untyped, + Op.getOperand(0), Op.getOperand(1)); + SDValue Lo, Hi; + + if (HasLo) + Lo = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult, + DAG.getConstant(Mips::sub_lo, MVT::i32)); + if (HasHi) + Hi = DAG.getNode(MipsISD::ExtractLOHI, DL, Ty, Mult, + DAG.getConstant(Mips::sub_hi, MVT::i32)); + + if (!HasLo || !HasHi) + return HasLo ? Lo : Hi; + + SDValue Vals[] = { Lo, Hi }; + return DAG.getMergeValues(Vals, 2, DL); +} + + +static SDValue initAccumulator(SDValue In, DebugLoc DL, SelectionDAG &DAG) { + SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In, + DAG.getConstant(0, MVT::i32)); + SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In, + DAG.getConstant(1, MVT::i32)); + return DAG.getNode(MipsISD::InsertLOHI, DL, MVT::Untyped, InLo, InHi); +} + +static SDValue extractLOHI(SDValue Op, DebugLoc DL, SelectionDAG &DAG) { + SDValue Lo = DAG.getNode(MipsISD::ExtractLOHI, DL, MVT::i32, Op, + DAG.getConstant(Mips::sub_lo, MVT::i32)); + SDValue Hi = DAG.getNode(MipsISD::ExtractLOHI, DL, MVT::i32, Op, + DAG.getConstant(Mips::sub_hi, MVT::i32)); + return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi); +} + +// This function expands mips intrinsic nodes which have 64-bit input operands +// or output values. +// +// out64 = intrinsic-node in64 +// => +// lo = copy (extract-element (in64, 0)) +// hi = copy (extract-element (in64, 1)) +// mips-specific-node +// v0 = copy lo +// v1 = copy hi +// out64 = merge-values (v0, v1) +// +static SDValue lowerDSPIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) { + DebugLoc DL = Op.getDebugLoc(); + bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other; + SmallVector<SDValue, 3> Ops; + unsigned OpNo = 0; + + // See if Op has a chain input. + if (HasChainIn) + Ops.push_back(Op->getOperand(OpNo++)); + + // The next operand is the intrinsic opcode. + assert(Op->getOperand(OpNo).getOpcode() == ISD::TargetConstant); + + // See if the next operand has type i64. + SDValue Opnd = Op->getOperand(++OpNo), In64; + + if (Opnd.getValueType() == MVT::i64) + In64 = initAccumulator(Opnd, DL, DAG); + else + Ops.push_back(Opnd); + + // Push the remaining operands. + for (++OpNo ; OpNo < Op->getNumOperands(); ++OpNo) + Ops.push_back(Op->getOperand(OpNo)); + + // Add In64 to the end of the list. + if (In64.getNode()) + Ops.push_back(In64); + + // Scan output. + SmallVector<EVT, 2> ResTys; + + for (SDNode::value_iterator I = Op->value_begin(), E = Op->value_end(); + I != E; ++I) + ResTys.push_back((*I == MVT::i64) ? MVT::Untyped : *I); + + // Create node. + SDValue Val = DAG.getNode(Opc, DL, ResTys, &Ops[0], Ops.size()); + SDValue Out = (ResTys[0] == MVT::Untyped) ? extractLOHI(Val, DL, DAG) : Val; + + if (!HasChainIn) + return Out; + + assert(Val->getValueType(1) == MVT::Other); + SDValue Vals[] = { Out, SDValue(Val.getNode(), 1) }; + return DAG.getMergeValues(Vals, 2, DL); +} + +SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op, + SelectionDAG &DAG) const { + switch (cast<ConstantSDNode>(Op->getOperand(0))->getZExtValue()) { + default: + return SDValue(); + case Intrinsic::mips_shilo: + return lowerDSPIntr(Op, DAG, MipsISD::SHILO); + case Intrinsic::mips_dpau_h_qbl: + return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL); + case Intrinsic::mips_dpau_h_qbr: + return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR); + case Intrinsic::mips_dpsu_h_qbl: + return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL); + case Intrinsic::mips_dpsu_h_qbr: + return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR); + case Intrinsic::mips_dpa_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH); + case Intrinsic::mips_dps_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH); + case Intrinsic::mips_dpax_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH); + case Intrinsic::mips_dpsx_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH); + case Intrinsic::mips_mulsa_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH); + case Intrinsic::mips_mult: + return lowerDSPIntr(Op, DAG, MipsISD::Mult); + case Intrinsic::mips_multu: + return lowerDSPIntr(Op, DAG, MipsISD::Multu); + case Intrinsic::mips_madd: + return lowerDSPIntr(Op, DAG, MipsISD::MAdd); + case Intrinsic::mips_maddu: + return lowerDSPIntr(Op, DAG, MipsISD::MAddu); + case Intrinsic::mips_msub: + return lowerDSPIntr(Op, DAG, MipsISD::MSub); + case Intrinsic::mips_msubu: + return lowerDSPIntr(Op, DAG, MipsISD::MSubu); + } +} + +SDValue MipsSETargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op, + SelectionDAG &DAG) const { + switch (cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue()) { + default: + return SDValue(); + case Intrinsic::mips_extp: + return lowerDSPIntr(Op, DAG, MipsISD::EXTP); + case Intrinsic::mips_extpdp: + return lowerDSPIntr(Op, DAG, MipsISD::EXTPDP); + case Intrinsic::mips_extr_w: + return lowerDSPIntr(Op, DAG, MipsISD::EXTR_W); + case Intrinsic::mips_extr_r_w: + return lowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W); + case Intrinsic::mips_extr_rs_w: + return lowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W); + case Intrinsic::mips_extr_s_h: + return lowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H); + case Intrinsic::mips_mthlip: + return lowerDSPIntr(Op, DAG, MipsISD::MTHLIP); + case Intrinsic::mips_mulsaq_s_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH); + case Intrinsic::mips_maq_s_w_phl: + return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL); + case Intrinsic::mips_maq_s_w_phr: + return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR); + case Intrinsic::mips_maq_sa_w_phl: + return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL); + case Intrinsic::mips_maq_sa_w_phr: + return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR); + case Intrinsic::mips_dpaq_s_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH); + case Intrinsic::mips_dpsq_s_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH); + case Intrinsic::mips_dpaq_sa_l_w: + return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W); + case Intrinsic::mips_dpsq_sa_l_w: + return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W); + case Intrinsic::mips_dpaqx_s_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH); + case Intrinsic::mips_dpaqx_sa_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH); + case Intrinsic::mips_dpsqx_s_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH); + case Intrinsic::mips_dpsqx_sa_w_ph: + return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH); + } +} + +MachineBasicBlock * MipsSETargetLowering:: +emitBPOSGE32(MachineInstr *MI, MachineBasicBlock *BB) const{ + // $bb: + // bposge32_pseudo $vr0 + // => + // $bb: + // bposge32 $tbb + // $fbb: + // li $vr2, 0 + // b $sink + // $tbb: + // li $vr1, 1 + // $sink: + // $vr0 = phi($vr2, $fbb, $vr1, $tbb) + + MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo(); + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + const TargetRegisterClass *RC = &Mips::CPURegsRegClass; + DebugLoc DL = MI->getDebugLoc(); + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = llvm::next(MachineFunction::iterator(BB)); + MachineFunction *F = BB->getParent(); + MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *Sink = F->CreateMachineBasicBlock(LLVM_BB); + F->insert(It, FBB); + F->insert(It, TBB); + F->insert(It, Sink); + + // Transfer the remainder of BB and its successor edges to Sink. + Sink->splice(Sink->begin(), BB, llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + Sink->transferSuccessorsAndUpdatePHIs(BB); + + // Add successors. + BB->addSuccessor(FBB); + BB->addSuccessor(TBB); + FBB->addSuccessor(Sink); + TBB->addSuccessor(Sink); + + // Insert the real bposge32 instruction to $BB. + BuildMI(BB, DL, TII->get(Mips::BPOSGE32)).addMBB(TBB); + + // Fill $FBB. + unsigned VR2 = RegInfo.createVirtualRegister(RC); + BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), VR2) + .addReg(Mips::ZERO).addImm(0); + BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink); + + // Fill $TBB. + unsigned VR1 = RegInfo.createVirtualRegister(RC); + BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), VR1) + .addReg(Mips::ZERO).addImm(1); + + // Insert phi function to $Sink. + BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI), + MI->getOperand(0).getReg()) + .addReg(VR2).addMBB(FBB).addReg(VR1).addMBB(TBB); + + MI->eraseFromParent(); // The pseudo instruction is gone now. + return Sink; +} diff --git a/lib/Target/Mips/MipsSEISelLowering.h b/lib/Target/Mips/MipsSEISelLowering.h new file mode 100644 index 0000000000..ec8a5c73f1 --- /dev/null +++ b/lib/Target/Mips/MipsSEISelLowering.h @@ -0,0 +1,70 @@ +//===-- MipsSEISelLowering.h - MipsSE DAG Lowering Interface ----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Subclass of MipsTargetLowering specialized for mips32/64. +// +//===----------------------------------------------------------------------===// + +#ifndef MipsSEISELLOWERING_H +#define MipsSEISELLOWERING_H + +#include "MipsISelLowering.h" +#include "MipsRegisterInfo.h" + +namespace llvm { + class MipsSETargetLowering : public MipsTargetLowering { + public: + explicit MipsSETargetLowering(MipsTargetMachine &TM); + + virtual bool allowsUnalignedMemoryAccesses(EVT VT, bool *Fast) const; + + virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const; + + virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const; + + virtual MachineBasicBlock * + EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *MBB) const; + + virtual bool isShuffleMaskLegal(const SmallVectorImpl<int> &Mask, + EVT VT) const { + return false; + } + + virtual const TargetRegisterClass *getRepRegClassFor(MVT VT) const { + if (VT == MVT::Untyped) + return Subtarget->hasDSP() ? &Mips::ACRegsDSPRegClass : + &Mips::ACRegsRegClass; + + return TargetLowering::getRepRegClassFor(VT); + } + + private: + virtual bool + isEligibleForTailCallOptimization(const MipsCC &MipsCCInfo, + unsigned NextStackOffset, + const MipsFunctionInfo& FI) const; + + virtual void + getOpndList(SmallVectorImpl<SDValue> &Ops, + std::deque< std::pair<unsigned, SDValue> > &RegsToPass, + bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage, + CallLoweringInfo &CLI, SDValue Callee, SDValue Chain) const; + + SDValue lowerMulDiv(SDValue Op, unsigned NewOpc, bool HasLo, bool HasHi, + SelectionDAG &DAG) const; + + SDValue lowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const; + SDValue lowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const; + + MachineBasicBlock *emitBPOSGE32(MachineInstr *MI, + MachineBasicBlock *BB) const; + }; +} + +#endif // MipsSEISELLOWERING_H diff --git a/lib/Target/Mips/MipsSEInstrInfo.cpp b/lib/Target/Mips/MipsSEInstrInfo.cpp index a9809ef712..a0768e51c0 100644 --- a/lib/Target/Mips/MipsSEInstrInfo.cpp +++ b/lib/Target/Mips/MipsSEInstrInfo.cpp @@ -95,20 +95,39 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB, Opc = Mips::CFC1; else if (Mips::FGR32RegClass.contains(SrcReg)) Opc = Mips::MFC1; - else if (SrcReg == Mips::HI) + else if (Mips::HIRegsRegClass.contains(SrcReg)) Opc = Mips::MFHI, SrcReg = 0; - else if (SrcReg == Mips::LO) + else if (Mips::LORegsRegClass.contains(SrcReg)) Opc = Mips::MFLO, SrcReg = 0; + else if (Mips::HIRegsDSPRegClass.contains(SrcReg)) + Opc = Mips::MFHI_DSP; + else if (Mips::LORegsDSPRegClass.contains(SrcReg)) + Opc = Mips::MFLO_DSP; + else if (Mips::DSPCCRegClass.contains(SrcReg)) { + BuildMI(MBB, I, DL, get(Mips::RDDSP), DestReg).addImm(1 << 4) + .addReg(SrcReg, RegState::Implicit | getKillRegState(KillSrc)); + return; + } } else if (Mips::CPURegsRegClass.contains(SrcReg)) { // Copy from CPU Reg. if (Mips::CCRRegClass.contains(DestReg)) Opc = Mips::CTC1; else if (Mips::FGR32RegClass.contains(DestReg)) Opc = Mips::MTC1; - else if (DestReg == Mips::HI) + else if (Mips::HIRegsRegClass.contains(DestReg)) Opc = Mips::MTHI, DestReg = 0; - else if (DestReg == Mips::LO) + else if (Mips::LORegsRegClass.contains(DestReg)) Opc = Mips::MTLO, DestReg = 0; + else if (Mips::HIRegsDSPRegClass.contains(DestReg)) + Opc = Mips::MTHI_DSP; + else if (Mips::LORegsDSPRegClass.contains(DestReg)) + Opc = Mips::MTLO_DSP; + else if (Mips::DSPCCRegClass.contains(DestReg)) { + BuildMI(MBB, I, DL, get(Mips::WRDSP)) + .addReg(SrcReg, getKillRegState(KillSrc)).addImm(1 << 4) + .addReg(DestReg, RegState::ImplicitDefine); + return; + } } else if (Mips::FGR32RegClass.contains(DestReg, SrcReg)) Opc = Mips::FMOV_S; @@ -121,17 +140,17 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB, else if (Mips::CPU64RegsRegClass.contains(DestReg)) { // Copy to CPU64 Reg. if (Mips::CPU64RegsRegClass.contains(SrcReg)) Opc = Mips::OR64, ZeroReg = Mips::ZERO_64; - else if (SrcReg == Mips::HI64) + else if (Mips::HIRegs64RegClass.contains(SrcReg)) Opc = Mips::MFHI64, SrcReg = 0; - else if (SrcReg == Mips::LO64) + else if (Mips::LORegs64RegClass.contains(SrcReg)) Opc = Mips::MFLO64, SrcReg = 0; else if (Mips::FGR64RegClass.contains(SrcReg)) Opc = Mips::DMFC1; } else if (Mips::CPU64RegsRegClass.contains(SrcReg)) { // Copy from CPU64 Reg. - if (DestReg == Mips::HI64) + if (Mips::HIRegs64RegClass.contains(DestReg)) Opc = Mips::MTHI64, DestReg = 0; - else if (DestReg == Mips::LO64) + else if (Mips::LORegs64RegClass.contains(DestReg)) Opc = Mips::MTLO64, DestReg = 0; else if (Mips::FGR64RegClass.contains(DestReg)) Opc = Mips::DMTC1; @@ -152,10 +171,10 @@ void MipsSEInstrInfo::copyPhysReg(MachineBasicBlock &MBB, } void MipsSEInstrInfo:: -storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, - unsigned SrcReg, bool isKill, int FI, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const { +storeRegToStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, + unsigned SrcReg, bool isKill, int FI, + const TargetRegisterClass *RC, const TargetRegisterInfo *TRI, + int64_t Offset) const { DebugLoc DL; if (I != MBB.end()) DL = I->getDebugLoc(); MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOStore); @@ -166,6 +185,14 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Opc = IsN64 ? Mips::SW_P8 : Mips::SW; else if (Mips::CPU64RegsRegClass.hasSubClassEq(RC)) Opc = IsN64 ? Mips::SD_P8 : Mips::SD; + else if (Mips::ACRegsRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::STORE_AC64_P8 : Mips::STORE_AC64; + else if (Mips::ACRegsDSPRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::STORE_AC_DSP_P8 : Mips::STORE_AC_DSP; + else if (Mips::ACRegs128RegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::STORE_AC128_P8 : Mips::STORE_AC128; + else if (Mips::DSPCCRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::STORE_CCOND_DSP_P8 : Mips::STORE_CCOND_DSP; else if (Mips::FGR32RegClass.hasSubClassEq(RC)) Opc = IsN64 ? Mips::SWC1_P8 : Mips::SWC1; else if (Mips::AFGR64RegClass.hasSubClassEq(RC)) @@ -175,15 +202,13 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, assert(Opc && "Register class not handled!"); BuildMI(MBB, I, DL, get(Opc)).addReg(SrcReg, getKillRegState(isKill)) - .addFrameIndex(FI).addImm(0).addMemOperand(MMO); + .addFrameIndex(FI).addImm(Offset).addMemOperand(MMO); } void MipsSEInstrInfo:: -loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, - unsigned DestReg, int FI, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const -{ +loadRegFromStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, + unsigned DestReg, int FI, const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, int64_t Offset) const { DebugLoc DL; if (I != MBB.end()) DL = I->getDebugLoc(); MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOLoad); @@ -193,6 +218,14 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Opc = IsN64 ? Mips::LW_P8 : Mips::LW; else if (Mips::CPU64RegsRegClass.hasSubClassEq(RC)) Opc = IsN64 ? Mips::LD_P8 : Mips::LD; + else if (Mips::ACRegsRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::LOAD_AC64_P8 : Mips::LOAD_AC64; + else if (Mips::ACRegsDSPRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::LOAD_AC_DSP_P8 : Mips::LOAD_AC_DSP; + else if (Mips::ACRegs128RegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::LOAD_AC128_P8 : Mips::LOAD_AC128; + else if (Mips::DSPCCRegClass.hasSubClassEq(RC)) + Opc = IsN64 ? Mips::LOAD_CCOND_DSP_P8 : Mips::LOAD_CCOND_DSP; else if (Mips::FGR32RegClass.hasSubClassEq(RC)) Opc = IsN64 ? Mips::LWC1_P8 : Mips::LWC1; else if (Mips::AFGR64RegClass.hasSubClassEq(RC)) @@ -201,7 +234,7 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, Opc = IsN64 ? Mips::LDC164_P8 : Mips::LDC164; assert(Opc && "Register class not handled!"); - BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(0) + BuildMI(MBB, I, DL, get(Opc), DestReg).addFrameIndex(FI).addImm(Offset) .addMemOperand(MMO); } @@ -371,6 +404,7 @@ void MipsSEInstrInfo::ExpandEhReturn(MachineBasicBlock &MBB, unsigned JR = STI.isABI_N64() ? Mips::JR64 : Mips::JR; unsigned SP = STI.isABI_N64() ? Mips::SP_64 : Mips::SP; unsigned RA = STI.isABI_N64() ? Mips::RA_64 : Mips::RA; + unsigned T9 = STI.isABI_N64() ? Mips::T9_64 : Mips::T9; unsigned ZERO = STI.isABI_N64() ? Mips::ZERO_64 : Mips::ZERO; unsigned OffsetReg = I->getOperand(0).getReg(); unsigned TargetReg = I->getOperand(1).getReg(); @@ -378,6 +412,9 @@ void MipsSEInstrInfo::ExpandEhReturn(MachineBasicBlock &MBB, // or $ra, $v0, $zero // addu $sp, $sp, $v1 // jr $ra + if (TM.getRelocationModel() == Reloc::PIC_) + BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(OR), T9) + .addReg(TargetReg).addReg(ZERO); BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(OR), RA) .addReg(TargetReg).addReg(ZERO); BuildMI(MBB, I, I->getDebugLoc(), TM.getInstrInfo()->get(ADDU), SP) diff --git a/lib/Target/Mips/MipsSEInstrInfo.h b/lib/Target/Mips/MipsSEInstrInfo.h index 3e22b33ed7..0bf7876f0f 100644 --- a/lib/Target/Mips/MipsSEInstrInfo.h +++ b/lib/Target/Mips/MipsSEInstrInfo.h @@ -49,17 +49,19 @@ public: unsigned DestReg, unsigned SrcReg, bool KillSrc) const; - virtual void storeRegToStackSlot(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MBBI, - unsigned SrcReg, bool isKill, int FrameIndex, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const; - - virtual void loadRegFromStackSlot(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MBBI, - unsigned DestReg, int FrameIndex, - const TargetRegisterClass *RC, - const TargetRegisterInfo *TRI) const; + virtual void storeRegToStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + unsigned SrcReg, bool isKill, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const; + + virtual void loadRegFromStack(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, + unsigned DestReg, int FrameIndex, + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI, + int64_t Offset) const; virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const; diff --git a/lib/Target/Mips/MipsSERegisterInfo.cpp b/lib/Target/Mips/MipsSERegisterInfo.cpp index a39b393e4e..96967380b2 100644 --- a/lib/Target/Mips/MipsSERegisterInfo.cpp +++ b/lib/Target/Mips/MipsSERegisterInfo.cpp @@ -54,6 +54,15 @@ requiresFrameIndexScavenging(const MachineFunction &MF) const { return true; } +const TargetRegisterClass * +MipsSERegisterInfo::intRegClass(unsigned Size) const { + if (Size == 4) + return &Mips::CPURegsRegClass; + + assert(Size == 8); + return &Mips::CPU64RegsRegClass; +} + void MipsSERegisterInfo::eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo, int FrameIndex, uint64_t StackSize, diff --git a/lib/Target/Mips/MipsSERegisterInfo.h b/lib/Target/Mips/MipsSERegisterInfo.h index f6827e9663..2f7c37bb46 100644 --- a/lib/Target/Mips/MipsSERegisterInfo.h +++ b/lib/Target/Mips/MipsSERegisterInfo.h @@ -31,6 +31,8 @@ public: bool requiresFrameIndexScavenging(const MachineFunction &MF) const; + virtual const TargetRegisterClass *intRegClass(unsigned Size) const; + private: virtual void eliminateFI(MachineBasicBlock::iterator II, unsigned OpNo, int FrameIndex, uint64_t StackSize, diff --git a/lib/Target/Mips/MipsSubtarget.cpp b/lib/Target/Mips/MipsSubtarget.cpp index f368e17e3b..1d34b61bad 100644 --- a/lib/Target/Mips/MipsSubtarget.cpp +++ b/lib/Target/Mips/MipsSubtarget.cpp @@ -11,29 +11,56 @@ // //===----------------------------------------------------------------------===// +#define DEBUG_TYPE "mips-subtarget" + +#include "MipsMachineFunction.h" #include "MipsSubtarget.h" +#include "MipsTargetMachine.h" #include "Mips.h" #include "MipsRegisterInfo.h" +#include "llvm/IR/Attributes.h" +#include "llvm/IR/Function.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" #include "llvm/Support/TargetRegistry.h" +#include "llvm/Support/raw_ostream.h" #define GET_SUBTARGETINFO_TARGET_DESC #define GET_SUBTARGETINFO_CTOR #include "MipsGenSubtargetInfo.inc" + using namespace llvm; +// FIXME: Maybe this should be on by default when Mips16 is specified +// +static cl::opt<bool> Mixed16_32( + "mips-mixed-16-32", + cl::init(false), + cl::desc("Allow for a mixture of Mips16 " + "and Mips32 code in a single source file"), + cl::Hidden); + +static cl::opt<bool> Mips_Os16( + "mips-os16", + cl::init(false), + cl::desc("Compile all functions that don' use " + "floating point as Mips 16"), + cl::Hidden); + void MipsSubtarget::anchor() { } MipsSubtarget::MipsSubtarget(const std::string &TT, const std::string &CPU, const std::string &FS, bool little, - Reloc::Model _RM) : + Reloc::Model _RM, MipsTargetMachine *_TM) : MipsGenSubtargetInfo(TT, CPU, FS), MipsArchVersion(Mips32), MipsABI(UnknownABI), IsLittle(little), IsSingleFloat(false), IsFP64bit(false), IsGP64bit(false), HasVFPU(false), IsLinux(true), HasSEInReg(false), HasCondMov(false), HasSwap(false), HasBitCount(false), HasFPIdx(false), InMips16Mode(false), InMicroMipsMode(false), HasDSP(false), HasDSPR2(false), - IsAndroid(false), RM(_RM) + AllowMixed16_32(Mixed16_32 | Mips_Os16), Os16(Mips_Os16), + RM(_RM), OverrideMode(NoOverride), TM(_TM) // @LOCALMOD-START , TargetTriple(TT) // @LOCALMOD-END @@ -45,6 +72,8 @@ MipsSubtarget::MipsSubtarget(const std::string &TT, const std::string &CPU, // Parse features string. ParseSubtargetFeatures(CPUName, FS); + PreviousInMips16Mode = InMips16Mode; + // Initialize scheduling itinerary for the specified CPU. InstrItins = getInstrItineraryForCPU(CPUName); @@ -75,3 +104,48 @@ MipsSubtarget::enablePostRAScheduler(CodeGenOpt::Level OptLevel, &Mips::CPU64RegsRegClass : &Mips::CPURegsRegClass); return OptLevel >= CodeGenOpt::Aggressive; } + +//FIXME: This logic for reseting the subtarget along with +// the helper classes can probably be simplified but there are a lot of +// cases so we will defer rewriting this to later. +// +void MipsSubtarget::resetSubtarget(MachineFunction *MF) { + bool ChangeToMips16 = false, ChangeToNoMips16 = false; + DEBUG(dbgs() << "resetSubtargetFeatures" << "\n"); + AttributeSet FnAttrs = MF->getFunction()->getAttributes(); + ChangeToMips16 = FnAttrs.hasAttribute(AttributeSet::FunctionIndex, + "mips16"); + ChangeToNoMips16 = FnAttrs.hasAttribute(AttributeSet::FunctionIndex, + "nomips16"); + assert (!(ChangeToMips16 & ChangeToNoMips16) && + "mips16 and nomips16 specified on the same function"); + if (ChangeToMips16) { + if (PreviousInMips16Mode) + return; + OverrideMode = Mips16Override; + PreviousInMips16Mode = true; + TM->setHelperClassesMips16(); + return; + } else if (ChangeToNoMips16) { + if (!PreviousInMips16Mode) + return; + OverrideMode = NoMips16Override; + PreviousInMips16Mode = false; + TM->setHelperClassesMipsSE(); + return; + } else { + if (OverrideMode == NoOverride) + return; + OverrideMode = NoOverride; + DEBUG(dbgs() << "back to default" << "\n"); + if (inMips16Mode() && !PreviousInMips16Mode) { + TM->setHelperClassesMips16(); + PreviousInMips16Mode = true; + } else if (!inMips16Mode() && PreviousInMips16Mode) { + TM->setHelperClassesMipsSE(); + PreviousInMips16Mode = false; + } + return; + } +} + diff --git a/lib/Target/Mips/MipsSubtarget.h b/lib/Target/Mips/MipsSubtarget.h index 87d8345f71..864a3392af 100644 --- a/lib/Target/Mips/MipsSubtarget.h +++ b/lib/Target/Mips/MipsSubtarget.h @@ -16,7 +16,9 @@ #include "MCTargetDesc/MipsReginfo.h" #include "llvm/MC/MCInstrItineraries.h" +#include "llvm/Support/ErrorHandling.h" #include "llvm/Target/TargetSubtargetInfo.h" + #include <string> #define GET_SUBTARGETINFO_HEADER @@ -25,6 +27,8 @@ namespace llvm { class StringRef; +class MipsTargetMachine; + class MipsSubtarget : public MipsGenSubtargetInfo { virtual void anchor(); @@ -89,14 +93,22 @@ protected: // InMips16 -- can process Mips16 instructions bool InMips16Mode; + // PreviousInMips16 -- the function we just processed was in Mips 16 Mode + bool PreviousInMips16Mode; + // InMicroMips -- can process MicroMips instructions bool InMicroMipsMode; // HasDSP, HasDSPR2 -- supports DSP ASE. bool HasDSP, HasDSPR2; - // IsAndroid -- target is android - bool IsAndroid; + // Allow mixed Mips16 and Mips32 in one source file + bool AllowMixed16_32; + + // Optimize for space by compiling all functions as Mips 16 unless + // it needs floating point. Functions needing floating point are + // compiled as Mips32 + bool Os16; InstrItineraryData InstrItins; @@ -108,6 +120,11 @@ protected: Triple TargetTriple; // @LOCALMOD + // We can override the determination of whether we are in mips16 mode + // as from the command line + enum {NoOverride, Mips16Override, NoMips16Override} OverrideMode; + + MipsTargetMachine *TM; public: virtual bool enablePostRAScheduler(CodeGenOpt::Level OptLevel, AntiDepBreakMode& Mode, @@ -123,7 +140,8 @@ public: /// This constructor initializes the data members to match that /// of the specified triple. MipsSubtarget(const std::string &TT, const std::string &CPU, - const std::string &FS, bool little, Reloc::Model RM); + const std::string &FS, bool little, Reloc::Model RM, + MipsTargetMachine *TM); /// ParseSubtargetFeatures - Parses features string setting specified /// subtarget options. Definition of function is auto generated by tblgen. @@ -142,11 +160,23 @@ public: bool isSingleFloat() const { return IsSingleFloat; } bool isNotSingleFloat() const { return !IsSingleFloat; } bool hasVFPU() const { return HasVFPU; } - bool inMips16Mode() const { return InMips16Mode; } + bool inMips16Mode() const { + switch (OverrideMode) { + case NoOverride: + return InMips16Mode; + case Mips16Override: + return true; + case NoMips16Override: + return false; + } + llvm_unreachable("Unexpected mode"); + } + bool inMips16ModeDefault() { + return InMips16Mode; + } bool inMicroMipsMode() const { return InMicroMipsMode; } bool hasDSP() const { return HasDSP; } bool hasDSPR2() const { return HasDSPR2; } - bool isAndroid() const { return IsAndroid; } bool isLinux() const { return IsLinux; } bool useSmallSection() const { return UseSmallSection; } @@ -159,6 +189,10 @@ public: bool hasBitCount() const { return HasBitCount; } bool hasFPIdx() const { return HasFPIdx; } + bool allowMixed16_32() const { return AllowMixed16_32;}; + + bool os16() const { return Os16;}; + // @LOCALMOD-START bool isTargetNaCl() const { return TargetTriple.isOSNaCl(); } bool isNotTargetNaCl() const { return !TargetTriple.isOSNaCl(); } @@ -169,6 +203,11 @@ public: // Grab relocation model Reloc::Model getRelocationModel() const {return RM;} + + /// \brief Reset the subtarget for the Mips target. + void resetSubtarget(MachineFunction *MF); + + }; } // End llvm namespace diff --git a/lib/Target/Mips/MipsTargetMachine.cpp b/lib/Target/Mips/MipsTargetMachine.cpp index b3c6321e8f..a2fb63d74c 100644 --- a/lib/Target/Mips/MipsTargetMachine.cpp +++ b/lib/Target/Mips/MipsTargetMachine.cpp @@ -15,11 +15,26 @@ #include "Mips.h" #include "MipsFrameLowering.h" #include "MipsInstrInfo.h" +#include "MipsModuleISelDAGToDAG.h" +#include "MipsOs16.h" +#include "MipsSEFrameLowering.h" +#include "MipsSEInstrInfo.h" +#include "MipsSEISelLowering.h" +#include "MipsSEISelDAGToDAG.h" +#include "Mips16FrameLowering.h" +#include "Mips16InstrInfo.h" +#include "Mips16ISelDAGToDAG.h" +#include "Mips16ISelLowering.h" +#include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/CodeGen/Passes.h" #include "llvm/PassManager.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/raw_ostream.h" #include "llvm/Support/TargetRegistry.h" using namespace llvm; + + extern "C" void LLVMInitializeMipsTarget() { // Register the target. RegisterTargetMachine<MipsebTargetMachine> X(TheMipsTarget); @@ -42,7 +57,7 @@ MipsTargetMachine(const Target &T, StringRef TT, CodeGenOpt::Level OL, bool isLittle) : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL), - Subtarget(TT, CPU, FS, isLittle, RM), + Subtarget(TT, CPU, FS, isLittle, RM, this), DL(isLittle ? (Subtarget.isABI_N64() ? "e-p:64:64:64-i8:8:32-i16:16:32-i64:64:64-f128:128:128-" @@ -54,9 +69,46 @@ MipsTargetMachine(const Target &T, StringRef TT, "E-p:32:32:32-i8:8:32-i16:16:32-i64:64:64-n32-S64")), InstrInfo(MipsInstrInfo::create(*this)), FrameLowering(MipsFrameLowering::create(*this, Subtarget)), - TLInfo(*this), TSInfo(*this), JITInfo() { + TLInfo(MipsTargetLowering::create(*this)), + TSInfo(*this), JITInfo() { +} + + +void MipsTargetMachine::setHelperClassesMips16() { + InstrInfoSE.swap(InstrInfo); + FrameLoweringSE.swap(FrameLowering); + TLInfoSE.swap(TLInfo); + if (!InstrInfo16) { + InstrInfo.reset(MipsInstrInfo::create(*this)); + FrameLowering.reset(MipsFrameLowering::create(*this, Subtarget)); + TLInfo.reset(MipsTargetLowering::create(*this)); + } else { + InstrInfo16.swap(InstrInfo); + FrameLowering16.swap(FrameLowering); + TLInfo16.swap(TLInfo); + } + assert(TLInfo && "null target lowering 16"); + assert(InstrInfo && "null instr info 16"); + assert(FrameLowering && "null frame lowering 16"); } +void MipsTargetMachine::setHelperClassesMipsSE() { + InstrInfo16.swap(InstrInfo); + FrameLowering16.swap(FrameLowering); + TLInfo16.swap(TLInfo); + if (!InstrInfoSE) { + InstrInfo.reset(MipsInstrInfo::create(*this)); + FrameLowering.reset(MipsFrameLowering::create(*this, Subtarget)); + TLInfo.reset(MipsTargetLowering::create(*this)); + } else { + InstrInfoSE.swap(InstrInfo); + FrameLoweringSE.swap(FrameLowering); + TLInfoSE.swap(TLInfo); + } + assert(TLInfo && "null target lowering in SE"); + assert(InstrInfo && "null instr info SE"); + assert(FrameLowering && "null frame lowering SE"); +} void MipsebTargetMachine::anchor() { } MipsebTargetMachine:: @@ -90,6 +142,7 @@ public: return *getMipsTargetMachine().getSubtargetImpl(); } + virtual void addIRPasses(); virtual bool addInstSelector(); virtual bool addPreEmitPass(); }; @@ -99,24 +152,50 @@ TargetPassConfig *MipsTargetMachine::createPassConfig(PassManagerBase &PM) { return new MipsPassConfig(this, PM); } +void MipsPassConfig::addIRPasses() { + TargetPassConfig::addIRPasses(); + if (getMipsSubtarget().os16()) + addPass(createMipsOs16(getMipsTargetMachine())); +} // Install an instruction selector pass using // the ISelDag to gen Mips code. bool MipsPassConfig::addInstSelector() { - addPass(createMipsISelDag(getMipsTargetMachine())); + if (getMipsSubtarget().allowMixed16_32()) { + addPass(createMipsModuleISelDag(getMipsTargetMachine())); + addPass(createMips16ISelDag(getMipsTargetMachine())); + addPass(createMipsSEISelDag(getMipsTargetMachine())); + } else { + addPass(createMipsISelDag(getMipsTargetMachine())); + } return false; } +void MipsTargetMachine::addAnalysisPasses(PassManagerBase &PM) { + if (Subtarget.allowMixed16_32()) { + DEBUG(errs() << "No "); + //FIXME: The Basic Target Transform Info + // pass needs to become a function pass instead of + // being an immutable pass and then this method as it exists now + // would be unnecessary. + PM.add(createNoTargetTransformInfoPass()); + } else + LLVMTargetMachine::addAnalysisPasses(PM); + DEBUG(errs() << "Target Transform Info Pass Added\n"); +} + // Implemented by targets that want to run passes immediately before // machine code is emitted. return true if -print-machineinstrs should // print out the code after the passes. bool MipsPassConfig::addPreEmitPass() { MipsTargetMachine &TM = getMipsTargetMachine(); + const MipsSubtarget &Subtarget = TM.getSubtarget<MipsSubtarget>(); addPass(createMipsDelaySlotFillerPass(TM)); - // NOTE: long branch has not been implemented for mips16. - if (TM.getSubtarget<MipsSubtarget>().hasStandardEncoding()) + if (Subtarget.hasStandardEncoding() || + Subtarget.allowMixed16_32()) addPass(createMipsLongBranchPass(TM)); - if (TM.getSubtarget<MipsSubtarget>().inMips16Mode()) + if (Subtarget.inMips16Mode() || + Subtarget.allowMixed16_32()) addPass(createMipsConstantIslandPass(TM)); diff --git a/lib/Target/Mips/MipsTargetMachine.h b/lib/Target/Mips/MipsTargetMachine.h index c4928c21eb..ee557084fb 100644 --- a/lib/Target/Mips/MipsTargetMachine.h +++ b/lib/Target/Mips/MipsTargetMachine.h @@ -21,6 +21,8 @@ #include "MipsSelectionDAGInfo.h" #include "MipsSubtarget.h" #include "llvm/ADT/OwningPtr.h" +#include "llvm/CodeGen/Passes.h" +#include "llvm/CodeGen/SelectionDAGISel.h" #include "llvm/IR/DataLayout.h" #include "llvm/Target/TargetFrameLowering.h" #include "llvm/Target/TargetMachine.h" @@ -34,7 +36,13 @@ class MipsTargetMachine : public LLVMTargetMachine { const DataLayout DL; // Calculates type size & alignment OwningPtr<const MipsInstrInfo> InstrInfo; OwningPtr<const MipsFrameLowering> FrameLowering; - MipsTargetLowering TLInfo; + OwningPtr<const MipsTargetLowering> TLInfo; + OwningPtr<const MipsInstrInfo> InstrInfo16; + OwningPtr<const MipsFrameLowering> FrameLowering16; + OwningPtr<const MipsTargetLowering> TLInfo16; + OwningPtr<const MipsInstrInfo> InstrInfoSE; + OwningPtr<const MipsFrameLowering> FrameLoweringSE; + OwningPtr<const MipsTargetLowering> TLInfoSE; MipsSelectionDAGInfo TSInfo; MipsJITInfo JITInfo; @@ -47,6 +55,8 @@ public: virtual ~MipsTargetMachine() {} + virtual void addAnalysisPasses(PassManagerBase &PM); + virtual const MipsInstrInfo *getInstrInfo() const { return InstrInfo.get(); } virtual const TargetFrameLowering *getFrameLowering() const @@ -63,7 +73,7 @@ public: } virtual const MipsTargetLowering *getTargetLowering() const { - return &TLInfo; + return TLInfo.get(); } virtual const MipsSelectionDAGInfo* getSelectionDAGInfo() const { @@ -73,6 +83,13 @@ public: // Pass Pipeline Configuration virtual TargetPassConfig *createPassConfig(PassManagerBase &PM); virtual bool addCodeEmitter(PassManagerBase &PM, JITCodeEmitter &JCE); + + // Set helper classes + void setHelperClassesMips16(); + + void setHelperClassesMipsSE(); + + }; /// MipsebTargetMachine - Mips32/64 big endian target machine. |
