diff options
Diffstat (limited to 'contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp')
-rw-r--r-- | contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp | 137 |
1 files changed, 89 insertions, 48 deletions
diff --git a/contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp b/contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp index 13bd0c7..a561d5b 100644 --- a/contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp +++ b/contrib/llvm/lib/Target/PowerPC/PPCFastISel.cpp @@ -17,6 +17,7 @@ #include "MCTargetDesc/PPCPredicates.h" #include "PPCCallingConv.h" #include "PPCISelLowering.h" +#include "PPCMachineFunctionInfo.h" #include "PPCSubtarget.h" #include "PPCTargetMachine.h" #include "llvm/ADT/Optional.h" @@ -85,18 +86,20 @@ typedef struct Address { class PPCFastISel final : public FastISel { const TargetMachine &TM; + const PPCSubtarget *PPCSubTarget; + PPCFunctionInfo *PPCFuncInfo; const TargetInstrInfo &TII; const TargetLowering &TLI; - const PPCSubtarget *PPCSubTarget; LLVMContext *Context; public: explicit PPCFastISel(FunctionLoweringInfo &FuncInfo, const TargetLibraryInfo *LibInfo) : FastISel(FuncInfo, LibInfo), TM(FuncInfo.MF->getTarget()), - TII(*TM.getSubtargetImpl()->getInstrInfo()), - TLI(*TM.getSubtargetImpl()->getTargetLowering()), - PPCSubTarget(&TM.getSubtarget<PPCSubtarget>()), + PPCSubTarget(&FuncInfo.MF->getSubtarget<PPCSubtarget>()), + PPCFuncInfo(FuncInfo.MF->getInfo<PPCFunctionInfo>()), + TII(*PPCSubTarget->getInstrInfo()), + TLI(*PPCSubTarget->getTargetLowering()), Context(&FuncInfo.Fn->getContext()) {} // Backend specific FastISel code. @@ -141,9 +144,13 @@ class PPCFastISel final : public FastISel { private: bool isTypeLegal(Type *Ty, MVT &VT); bool isLoadTypeLegal(Type *Ty, MVT &VT); + bool isValueAvailable(const Value *V) const; bool isVSFRCRegister(unsigned Register) const { return MRI.getRegClass(Register)->getID() == PPC::VSFRCRegClassID; } + bool isVSSRCRegister(unsigned Register) const { + return MRI.getRegClass(Register)->getID() == PPC::VSSRCRegClassID; + } bool PPCEmitCmp(const Value *Src1Value, const Value *Src2Value, bool isZExt, unsigned DestReg); bool PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr, @@ -280,6 +287,17 @@ bool PPCFastISel::isLoadTypeLegal(Type *Ty, MVT &VT) { return false; } +bool PPCFastISel::isValueAvailable(const Value *V) const { + if (!isa<Instruction>(V)) + return true; + + const auto *I = cast<Instruction>(V); + if (FuncInfo.MBBMap[I->getParent()] == FuncInfo.MBB) + return true; + + return false; +} + // Given a value Obj, create an Address object Addr that represents its // address. Return false if we can't handle it. bool PPCFastISel::PPCComputeAddress(const Value *Obj, Address &Addr) { @@ -488,8 +506,11 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr, // If this is a potential VSX load with an offset of 0, a VSX indexed load can // be used. + bool IsVSSRC = (ResultReg != 0) && isVSSRCRegister(ResultReg); bool IsVSFRC = (ResultReg != 0) && isVSFRCRegister(ResultReg); - if (IsVSFRC && (Opc == PPC::LFD) && + bool Is32VSXLoad = IsVSSRC && Opc == PPC::LFS; + bool Is64VSXLoad = IsVSSRC && Opc == PPC::LFD; + if ((Is32VSXLoad || Is64VSXLoad) && (Addr.BaseType != Address::FrameIndexBase) && UseOffset && (Addr.Offset == 0)) { UseOffset = false; @@ -503,7 +524,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr, // into a RegBase. if (Addr.BaseType == Address::FrameIndexBase) { // VSX only provides an indexed load. - if (IsVSFRC && Opc == PPC::LFD) return false; + if (Is32VSXLoad || Is64VSXLoad) return false; MachineMemOperand *MMO = FuncInfo.MF->getMachineMemOperand( @@ -517,7 +538,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr, // Base reg with offset in range. } else if (UseOffset) { // VSX only provides an indexed load. - if (IsVSFRC && Opc == PPC::LFD) return false; + if (Is32VSXLoad || Is64VSXLoad) return false; BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg) .addImm(Addr.Offset).addReg(Addr.Base.Reg); @@ -540,7 +561,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr, case PPC::LWA: Opc = PPC::LWAX; break; case PPC::LWA_32: Opc = PPC::LWAX_32; break; case PPC::LD: Opc = PPC::LDX; break; - case PPC::LFS: Opc = PPC::LFSX; break; + case PPC::LFS: Opc = IsVSSRC ? PPC::LXSSPX : PPC::LFSX; break; case PPC::LFD: Opc = IsVSFRC ? PPC::LXSDX : PPC::LFDX; break; } BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg) @@ -621,9 +642,12 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) { // If this is a potential VSX store with an offset of 0, a VSX indexed store // can be used. + bool IsVSSRC = isVSSRCRegister(SrcReg); bool IsVSFRC = isVSFRCRegister(SrcReg); - if (IsVSFRC && (Opc == PPC::STFD) && - (Addr.BaseType != Address::FrameIndexBase) && UseOffset && + bool Is32VSXStore = IsVSSRC && Opc == PPC::STFS; + bool Is64VSXStore = IsVSFRC && Opc == PPC::STFD; + if ((Is32VSXStore || Is64VSXStore) && + (Addr.BaseType != Address::FrameIndexBase) && UseOffset && (Addr.Offset == 0)) { UseOffset = false; } @@ -633,7 +657,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) { // into a RegBase. if (Addr.BaseType == Address::FrameIndexBase) { // VSX only provides an indexed store. - if (IsVSFRC && Opc == PPC::STFD) return false; + if (Is32VSXStore || Is64VSXStore) return false; MachineMemOperand *MMO = FuncInfo.MF->getMachineMemOperand( @@ -650,7 +674,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) { // Base reg with offset in range. } else if (UseOffset) { // VSX only provides an indexed store. - if (IsVSFRC && Opc == PPC::STFD) return false; + if (Is32VSXStore || Is64VSXStore) return false; BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc)) .addReg(SrcReg).addImm(Addr.Offset).addReg(Addr.Base.Reg); @@ -669,11 +693,21 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) { case PPC::STH8: Opc = PPC::STHX8; break; case PPC::STW8: Opc = PPC::STWX8; break; case PPC::STD: Opc = PPC::STDX; break; - case PPC::STFS: Opc = PPC::STFSX; break; + case PPC::STFS: Opc = IsVSSRC ? PPC::STXSSPX : PPC::STFSX; break; case PPC::STFD: Opc = IsVSFRC ? PPC::STXSDX : PPC::STFDX; break; } - BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc)) - .addReg(SrcReg).addReg(Addr.Base.Reg).addReg(IndexReg); + + auto MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc)) + .addReg(SrcReg); + + // If we have an index register defined we use it in the store inst, + // otherwise we use X0 as base as it makes the vector instructions to + // use zero in the computation of the effective address regardless the + // content of the register. + if (IndexReg) + MIB.addReg(Addr.Base.Reg).addReg(IndexReg); + else + MIB.addReg(PPC::ZERO8).addReg(Addr.Base.Reg); } return true; @@ -718,30 +752,31 @@ bool PPCFastISel::SelectBranch(const Instruction *I) { // For now, just try the simplest case where it's fed by a compare. if (const CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition())) { - Optional<PPC::Predicate> OptPPCPred = getComparePred(CI->getPredicate()); - if (!OptPPCPred) - return false; + if (isValueAvailable(CI)) { + Optional<PPC::Predicate> OptPPCPred = getComparePred(CI->getPredicate()); + if (!OptPPCPred) + return false; - PPC::Predicate PPCPred = OptPPCPred.getValue(); + PPC::Predicate PPCPred = OptPPCPred.getValue(); - // Take advantage of fall-through opportunities. - if (FuncInfo.MBB->isLayoutSuccessor(TBB)) { - std::swap(TBB, FBB); - PPCPred = PPC::InvertPredicate(PPCPred); - } - - unsigned CondReg = createResultReg(&PPC::CRRCRegClass); + // Take advantage of fall-through opportunities. + if (FuncInfo.MBB->isLayoutSuccessor(TBB)) { + std::swap(TBB, FBB); + PPCPred = PPC::InvertPredicate(PPCPred); + } - if (!PPCEmitCmp(CI->getOperand(0), CI->getOperand(1), CI->isUnsigned(), - CondReg)) - return false; + unsigned CondReg = createResultReg(&PPC::CRRCRegClass); - BuildMI(*BrBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCC)) - .addImm(PPCPred).addReg(CondReg).addMBB(TBB); - fastEmitBranch(FBB, DbgLoc); - FuncInfo.MBB->addSuccessor(TBB); - return true; + if (!PPCEmitCmp(CI->getOperand(0), CI->getOperand(1), CI->isUnsigned(), + CondReg)) + return false; + BuildMI(*BrBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCC)) + .addImm(PPCPred).addReg(CondReg).addMBB(TBB); + fastEmitBranch(FBB, DbgLoc); + FuncInfo.MBB->addSuccessor(TBB); + return true; + } } else if (const ConstantInt *CI = dyn_cast<ConstantInt>(BI->getCondition())) { uint64_t Imm = CI->getZExtValue(); @@ -945,6 +980,8 @@ unsigned PPCFastISel::PPCMoveToFPReg(MVT SrcVT, unsigned SrcReg, } // Attempt to fast-select an integer-to-floating-point conversion. +// FIXME: Once fast-isel has better support for VSX, conversions using +// direct moves should be implemented. bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) { MVT DstVT; Type *DstTy = I->getType(); @@ -1052,6 +1089,8 @@ unsigned PPCFastISel::PPCMoveToIntReg(const Instruction *I, MVT VT, } // Attempt to fast-select a floating-point-to-integer conversion. +// FIXME: Once fast-isel has better support for VSX, conversions using +// direct moves should be implemented. bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) { MVT DstVT, SrcVT; Type *DstTy = I->getType(); @@ -1234,9 +1273,7 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args, CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, ArgLocs, *Context); // Reserve space for the linkage area on the stack. - bool isELFv2ABI = PPCSubTarget->isELFv2ABI(); - unsigned LinkageSize = PPCFrameLowering::getLinkageSize(true, false, - isELFv2ABI); + unsigned LinkageSize = PPCSubTarget->getFrameLowering()->getLinkageSize(); CCInfo.AllocateStack(LinkageSize, 8); CCInfo.AnalyzeCallOperands(ArgVTs, ArgFlags, CC_PPC64_ELF_FIS); @@ -1275,7 +1312,7 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args, // Prepare to assign register arguments. Every argument uses up a // GPR protocol register even if it's passed in a floating-point - // register. + // register (unless we're using the fast calling convention). unsigned NextGPR = PPC::X3; unsigned NextFPR = PPC::F1; @@ -1325,7 +1362,8 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args, unsigned ArgReg; if (ArgVT == MVT::f32 || ArgVT == MVT::f64) { ArgReg = NextFPR++; - ++NextGPR; + if (CC != CallingConv::Fast) + ++NextGPR; } else ArgReg = NextGPR++; @@ -1432,6 +1470,9 @@ bool PPCFastISel::fastLowerCall(CallLoweringInfo &CLI) { else if (!isTypeLegal(RetTy, RetVT) && RetVT != MVT::i16 && RetVT != MVT::i8) return false; + else if (RetVT == MVT::i1 && PPCSubTarget->useCRBits()) + // We can't handle boolean returns when CR bits are in use. + return false; // FIXME: No multi-register return values yet. if (RetVT != MVT::isVoid && RetVT != MVT::i8 && RetVT != MVT::i16 && @@ -1523,13 +1564,14 @@ bool PPCFastISel::fastLowerCall(CallLoweringInfo &CLI) { for (unsigned II = 0, IE = RegArgs.size(); II != IE; ++II) MIB.addReg(RegArgs[II], RegState::Implicit); - // Direct calls in the ELFv2 ABI need the TOC register live into the call. - if (PPCSubTarget->isELFv2ABI()) - MIB.addReg(PPC::X2, RegState::Implicit); + // Direct calls, in both the ELF V1 and V2 ABIs, need the TOC register live + // into the call. + PPCFuncInfo->setUsesTOCBasePtr(); + MIB.addReg(PPC::X2, RegState::Implicit); // Add a register mask with the call-preserved registers. Proper // defs for return values will be added by setPhysRegsDeadExcept(). - MIB.addRegMask(TRI.getCallPreservedMask(CC)); + MIB.addRegMask(TRI.getCallPreservedMask(*FuncInfo.MF, CC)); CLI.Call = MIB; @@ -1863,6 +1905,7 @@ unsigned PPCFastISel::PPCMaterializeFP(const ConstantFP *CFP, MVT VT) { unsigned Opc = (VT == MVT::f32) ? PPC::LFS : PPC::LFD; unsigned TmpReg = createResultReg(&PPC::G8RC_and_G8RC_NOX0RegClass); + PPCFuncInfo->setUsesTOCBasePtr(); // For small code model, generate a LF[SD](0, LDtocCPT(Idx, X2)). if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault) { BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocCPT), @@ -1912,6 +1955,7 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) { if (GV->isThreadLocal()) return 0; + PPCFuncInfo->setUsesTOCBasePtr(); // For small code model, generate a simple TOC load. if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault) BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtoc), @@ -2297,13 +2341,10 @@ namespace llvm { // Create the fast instruction selector for PowerPC64 ELF. FastISel *PPC::createFastISel(FunctionLoweringInfo &FuncInfo, const TargetLibraryInfo *LibInfo) { - const TargetMachine &TM = FuncInfo.MF->getTarget(); - // Only available on 64-bit ELF for now. - const PPCSubtarget *Subtarget = &TM.getSubtarget<PPCSubtarget>(); - if (Subtarget->isPPC64() && Subtarget->isSVR4ABI()) + const PPCSubtarget &Subtarget = FuncInfo.MF->getSubtarget<PPCSubtarget>(); + if (Subtarget.isPPC64() && Subtarget.isSVR4ABI()) return new PPCFastISel(FuncInfo, LibInfo); - return nullptr; } } |