diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.h b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.h --- a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.h +++ b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.h @@ -23,12 +23,11 @@ class MachineInstrBuilder; class AMDGPUCallLowering: public CallLowering { - Register lowerParameterPtr(MachineIRBuilder &MIRBuilder, Type *ParamTy, + Register lowerParameterPtr(MachineIRBuilder &B, Type *ParamTy, uint64_t Offset) const; - void lowerParameter(MachineIRBuilder &MIRBuilder, Type *ParamTy, - uint64_t Offset, unsigned Align, - Register DstReg) const; + void lowerParameter(MachineIRBuilder &B, Type *ParamTy, uint64_t Offset, + unsigned Align, Register DstReg) const; /// A function of this type is used to perform value split action. using SplitArgTy = std::function, LLT, LLT, int)>; @@ -39,21 +38,19 @@ CallingConv::ID CallConv, SplitArgTy SplitArg) const; - bool lowerReturnVal(MachineIRBuilder &MIRBuilder, - const Value *Val, ArrayRef VRegs, - MachineInstrBuilder &Ret) const; + bool lowerReturnVal(MachineIRBuilder &B, const Value *Val, + ArrayRef VRegs, MachineInstrBuilder &Ret) const; public: AMDGPUCallLowering(const AMDGPUTargetLowering &TLI); - bool lowerReturn(MachineIRBuilder &MIRBuilder, const Value *Val, + bool lowerReturn(MachineIRBuilder &B, const Value *Val, ArrayRef VRegs) const override; - bool lowerFormalArgumentsKernel(MachineIRBuilder &MIRBuilder, - const Function &F, + bool lowerFormalArgumentsKernel(MachineIRBuilder &B, const Function &F, ArrayRef> VRegs) const; - bool lowerFormalArguments(MachineIRBuilder &MIRBuilder, const Function &F, + bool lowerFormalArguments(MachineIRBuilder &B, const Function &F, ArrayRef> VRegs) const override; static CCAssignFn *CCAssignFnForCall(CallingConv::ID CC, bool IsVarArg); static CCAssignFn *CCAssignFnForReturn(CallingConv::ID CC, bool IsVarArg); diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp --- a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp @@ -31,9 +31,9 @@ namespace { struct OutgoingValueHandler : public CallLowering::ValueHandler { - OutgoingValueHandler(MachineIRBuilder &MIRBuilder, MachineRegisterInfo &MRI, + OutgoingValueHandler(MachineIRBuilder &B, MachineRegisterInfo &MRI, MachineInstrBuilder MIB, CCAssignFn *AssignFn) - : ValueHandler(MIRBuilder, MRI, AssignFn), MIB(MIB) {} + : ValueHandler(B, MRI, AssignFn), MIB(MIB) {} MachineInstrBuilder MIB; @@ -73,9 +73,9 @@ struct IncomingArgHandler : public CallLowering::ValueHandler { uint64_t StackUsed = 0; - IncomingArgHandler(MachineIRBuilder &MIRBuilder, MachineRegisterInfo &MRI, + IncomingArgHandler(MachineIRBuilder &B, MachineRegisterInfo &MRI, CCAssignFn *AssignFn) - : ValueHandler(MIRBuilder, MRI, AssignFn) {} + : ValueHandler(B, MRI, AssignFn) {} Register getStackAddress(uint64_t Size, int64_t Offset, MachinePointerInfo &MPO) override { @@ -133,9 +133,9 @@ }; struct FormalArgHandler : public IncomingArgHandler { - FormalArgHandler(MachineIRBuilder &MIRBuilder, MachineRegisterInfo &MRI, + FormalArgHandler(MachineIRBuilder &B, MachineRegisterInfo &MRI, CCAssignFn *AssignFn) - : IncomingArgHandler(MIRBuilder, MRI, AssignFn) {} + : IncomingArgHandler(B, MRI, AssignFn) {} void markPhysRegUsed(unsigned PhysReg) override { MIRBuilder.getMBB().addLiveIn(PhysReg); @@ -213,14 +213,14 @@ } // TODO: Move to generic code -static void unpackRegsToOrigType(MachineIRBuilder &MIRBuilder, +static void unpackRegsToOrigType(MachineIRBuilder &B, ArrayRef DstRegs, Register SrcReg, LLT SrcTy, LLT PartTy) { assert(DstRegs.size() > 1 && "Nothing to unpack"); - MachineFunction &MF = MIRBuilder.getMF(); + MachineFunction &MF = B.getMF(); MachineRegisterInfo &MRI = MF.getRegInfo(); const unsigned SrcSize = SrcTy.getSizeInBits(); @@ -229,40 +229,40 @@ if (SrcTy.isVector() && !PartTy.isVector() && PartSize > SrcTy.getElementType().getSizeInBits()) { // Vector was scalarized, and the elements extended. - auto UnmergeToEltTy = MIRBuilder.buildUnmerge(SrcTy.getElementType(), + auto UnmergeToEltTy = B.buildUnmerge(SrcTy.getElementType(), SrcReg); for (int i = 0, e = DstRegs.size(); i != e; ++i) - MIRBuilder.buildAnyExt(DstRegs[i], UnmergeToEltTy.getReg(i)); + B.buildAnyExt(DstRegs[i], UnmergeToEltTy.getReg(i)); return; } if (SrcSize % PartSize == 0) { - MIRBuilder.buildUnmerge(DstRegs, SrcReg); + B.buildUnmerge(DstRegs, SrcReg); return; } const int NumRoundedParts = (SrcSize + PartSize - 1) / PartSize; LLT BigTy = getMultipleType(PartTy, NumRoundedParts); - auto ImpDef = MIRBuilder.buildUndef(BigTy); + auto ImpDef = B.buildUndef(BigTy); Register BigReg = MRI.createGenericVirtualRegister(BigTy); - MIRBuilder.buildInsert(BigReg, ImpDef.getReg(0), SrcReg, 0).getReg(0); + B.buildInsert(BigReg, ImpDef.getReg(0), SrcReg, 0).getReg(0); int64_t Offset = 0; for (unsigned i = 0, e = DstRegs.size(); i != e; ++i, Offset += PartSize) - MIRBuilder.buildExtract(DstRegs[i], BigReg, Offset); + B.buildExtract(DstRegs[i], BigReg, Offset); } /// Lower the return value for the already existing \p Ret. This assumes that -/// \p MIRBuilder's insertion point is correct. -bool AMDGPUCallLowering::lowerReturnVal(MachineIRBuilder &MIRBuilder, +/// \p B's insertion point is correct. +bool AMDGPUCallLowering::lowerReturnVal(MachineIRBuilder &B, const Value *Val, ArrayRef VRegs, MachineInstrBuilder &Ret) const { if (!Val) return true; - auto &MF = MIRBuilder.getMF(); + auto &MF = B.getMF(); const auto &F = MF.getFunction(); const DataLayout &DL = MF.getDataLayout(); @@ -277,69 +277,69 @@ splitToValueTypes( OrigRetInfo, SplitRetInfos, DL, MRI, CC, [&](ArrayRef Regs, LLT LLTy, LLT PartLLT, int VTSplitIdx) { - unpackRegsToOrigType(MIRBuilder, Regs, VRegs[VTSplitIdx], LLTy, PartLLT); + unpackRegsToOrigType(B, Regs, VRegs[VTSplitIdx], LLTy, PartLLT); }); CCAssignFn *AssignFn = TLI.CCAssignFnForReturn(CC, F.isVarArg()); - OutgoingValueHandler RetHandler(MIRBuilder, MF.getRegInfo(), Ret, AssignFn); - return handleAssignments(MIRBuilder, SplitRetInfos, RetHandler); + OutgoingValueHandler RetHandler(B, MF.getRegInfo(), Ret, AssignFn); + return handleAssignments(B, SplitRetInfos, RetHandler); } -bool AMDGPUCallLowering::lowerReturn(MachineIRBuilder &MIRBuilder, +bool AMDGPUCallLowering::lowerReturn(MachineIRBuilder &B, const Value *Val, ArrayRef VRegs) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineFunction &MF = B.getMF(); MachineRegisterInfo &MRI = MF.getRegInfo(); SIMachineFunctionInfo *MFI = MF.getInfo(); MFI->setIfReturnsVoid(!Val); assert(!Val == VRegs.empty() && "Return value without a vreg"); - CallingConv::ID CC = MIRBuilder.getMF().getFunction().getCallingConv(); + CallingConv::ID CC = B.getMF().getFunction().getCallingConv(); const bool IsShader = AMDGPU::isShader(CC); const bool IsWaveEnd = (IsShader && MFI->returnsVoid()) || AMDGPU::isKernel(CC); if (IsWaveEnd) { - MIRBuilder.buildInstr(AMDGPU::S_ENDPGM) + B.buildInstr(AMDGPU::S_ENDPGM) .addImm(0); return true; } - auto const &ST = MIRBuilder.getMF().getSubtarget(); + auto const &ST = B.getMF().getSubtarget(); unsigned ReturnOpc = IsShader ? AMDGPU::SI_RETURN_TO_EPILOG : AMDGPU::S_SETPC_B64_return; - auto Ret = MIRBuilder.buildInstrNoInsert(ReturnOpc); + auto Ret = B.buildInstrNoInsert(ReturnOpc); Register ReturnAddrVReg; if (ReturnOpc == AMDGPU::S_SETPC_B64_return) { ReturnAddrVReg = MRI.createVirtualRegister(&AMDGPU::CCR_SGPR_64RegClass); Ret.addUse(ReturnAddrVReg); } - if (!lowerReturnVal(MIRBuilder, Val, VRegs, Ret)) + if (!lowerReturnVal(B, Val, VRegs, Ret)) return false; if (ReturnOpc == AMDGPU::S_SETPC_B64_return) { const SIRegisterInfo *TRI = ST.getRegisterInfo(); Register LiveInReturn = MF.addLiveIn(TRI->getReturnAddressReg(MF), &AMDGPU::SGPR_64RegClass); - MIRBuilder.buildCopy(ReturnAddrVReg, LiveInReturn); + B.buildCopy(ReturnAddrVReg, LiveInReturn); } // TODO: Handle CalleeSavedRegsViaCopy. - MIRBuilder.insertInstr(Ret); + B.insertInstr(Ret); return true; } -Register AMDGPUCallLowering::lowerParameterPtr(MachineIRBuilder &MIRBuilder, +Register AMDGPUCallLowering::lowerParameterPtr(MachineIRBuilder &B, Type *ParamTy, uint64_t Offset) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineFunction &MF = B.getMF(); const SIMachineFunctionInfo *MFI = MF.getInfo(); MachineRegisterInfo &MRI = MF.getRegInfo(); const Function &F = MF.getFunction(); @@ -352,24 +352,24 @@ Register KernArgSegmentVReg = MRI.getLiveInVirtReg(KernArgSegmentPtr); Register OffsetReg = MRI.createGenericVirtualRegister(LLT::scalar(64)); - MIRBuilder.buildConstant(OffsetReg, Offset); + B.buildConstant(OffsetReg, Offset); - MIRBuilder.buildGEP(DstReg, KernArgSegmentVReg, OffsetReg); + B.buildGEP(DstReg, KernArgSegmentVReg, OffsetReg); return DstReg; } -void AMDGPUCallLowering::lowerParameter(MachineIRBuilder &MIRBuilder, +void AMDGPUCallLowering::lowerParameter(MachineIRBuilder &B, Type *ParamTy, uint64_t Offset, unsigned Align, Register DstReg) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineFunction &MF = B.getMF(); const Function &F = MF.getFunction(); const DataLayout &DL = F.getParent()->getDataLayout(); PointerType *PtrTy = PointerType::get(ParamTy, AMDGPUAS::CONSTANT_ADDRESS); MachinePointerInfo PtrInfo(UndefValue::get(PtrTy)); unsigned TypeSize = DL.getTypeStoreSize(ParamTy); - Register PtrReg = lowerParameterPtr(MIRBuilder, ParamTy, Offset); + Register PtrReg = lowerParameterPtr(B, ParamTy, Offset); MachineMemOperand *MMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad | @@ -377,12 +377,12 @@ MachineMemOperand::MOInvariant, TypeSize, Align); - MIRBuilder.buildLoad(DstReg, PtrReg, *MMO); + B.buildLoad(DstReg, PtrReg, *MMO); } // Allocate special inputs passed in user SGPRs. static void allocateHSAUserSGPRs(CCState &CCInfo, - MachineIRBuilder &MIRBuilder, + MachineIRBuilder &B, MachineFunction &MF, const SIRegisterInfo &TRI, SIMachineFunctionInfo &Info) { @@ -411,8 +411,8 @@ const LLT P4 = LLT::pointer(AMDGPUAS::CONSTANT_ADDRESS, 64); Register VReg = MRI.createGenericVirtualRegister(P4); MRI.addLiveIn(InputPtrReg, VReg); - MIRBuilder.getMBB().addLiveIn(InputPtrReg); - MIRBuilder.buildCopy(VReg, InputPtrReg); + B.getMBB().addLiveIn(InputPtrReg); + B.buildCopy(VReg, InputPtrReg); CCInfo.AllocateReg(InputPtrReg); } @@ -433,9 +433,9 @@ } bool AMDGPUCallLowering::lowerFormalArgumentsKernel( - MachineIRBuilder &MIRBuilder, const Function &F, + MachineIRBuilder &B, const Function &F, ArrayRef> VRegs) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineFunction &MF = B.getMF(); const GCNSubtarget *Subtarget = &MF.getSubtarget(); MachineRegisterInfo &MRI = MF.getRegInfo(); SIMachineFunctionInfo *Info = MF.getInfo(); @@ -447,7 +447,7 @@ SmallVector ArgLocs; CCState CCInfo(F.getCallingConv(), F.isVarArg(), MF, ArgLocs, F.getContext()); - allocateHSAUserSGPRs(CCInfo, MIRBuilder, MF, *TRI, *Info); + allocateHSAUserSGPRs(CCInfo, B, MF, *TRI, *Info); unsigned i = 0; const unsigned KernArgBaseAlign = 16; @@ -473,9 +473,9 @@ : MRI.createGenericVirtualRegister(getLLTForType(*ArgTy, DL)); unsigned Align = MinAlign(KernArgBaseAlign, ArgOffset); ArgOffset = alignTo(ArgOffset, DL.getABITypeAlignment(ArgTy)); - lowerParameter(MIRBuilder, ArgTy, ArgOffset, Align, ArgReg); + lowerParameter(B, ArgTy, ArgOffset, Align, ArgReg); if (OrigArgRegs.size() > 1) - unpackRegs(OrigArgRegs, ArgReg, ArgTy, MIRBuilder); + unpackRegs(OrigArgRegs, ArgReg, ArgTy, B); ++i; } @@ -485,13 +485,13 @@ } // TODO: Move this to generic code -static void packSplitRegsToOrigType(MachineIRBuilder &MIRBuilder, +static void packSplitRegsToOrigType(MachineIRBuilder &B, ArrayRef OrigRegs, ArrayRef Regs, LLT LLTy, LLT PartLLT) { if (!LLTy.isVector() && !PartLLT.isVector()) { - MIRBuilder.buildMerge(OrigRegs[0], Regs); + B.buildMerge(OrigRegs[0], Regs); return; } @@ -501,15 +501,15 @@ int DstElts = LLTy.getNumElements(); int PartElts = PartLLT.getNumElements(); if (DstElts % PartElts == 0) - MIRBuilder.buildConcatVectors(OrigRegs[0], Regs); + B.buildConcatVectors(OrigRegs[0], Regs); else { // Deal with v3s16 split into v2s16 assert(PartElts == 2 && DstElts % 2 != 0); int RoundedElts = PartElts * ((DstElts + PartElts - 1) / PartElts); LLT RoundedDestTy = LLT::vector(RoundedElts, PartLLT.getElementType()); - auto RoundedConcat = MIRBuilder.buildConcatVectors(RoundedDestTy, Regs); - MIRBuilder.buildExtract(OrigRegs[0], RoundedConcat, 0); + auto RoundedConcat = B.buildConcatVectors(RoundedDestTy, Regs); + B.buildExtract(OrigRegs[0], RoundedConcat, 0); } return; @@ -520,7 +520,7 @@ LLT DstEltTy = LLTy.getElementType(); if (DstEltTy == PartLLT) { // Vector was trivially scalarized. - MIRBuilder.buildBuildVector(OrigRegs[0], Regs); + B.buildBuildVector(OrigRegs[0], Regs); } else if (DstEltTy.getSizeInBits() > PartLLT.getSizeInBits()) { // Deal with vector with 64-bit elements decomposed to 32-bit // registers. Need to create intermediate 64-bit elements. @@ -530,23 +530,23 @@ assert(DstEltTy.getSizeInBits() % PartLLT.getSizeInBits() == 0); for (int I = 0, NumElts = LLTy.getNumElements(); I != NumElts; ++I) { - auto Merge = MIRBuilder.buildMerge(DstEltTy, + auto Merge = B.buildMerge(DstEltTy, Regs.take_front(PartsPerElt)); EltMerges.push_back(Merge.getReg(0)); Regs = Regs.drop_front(PartsPerElt); } - MIRBuilder.buildBuildVector(OrigRegs[0], EltMerges); + B.buildBuildVector(OrigRegs[0], EltMerges); } else { // Vector was split, and elements promoted to a wider type. LLT BVType = LLT::vector(LLTy.getNumElements(), PartLLT); - auto BV = MIRBuilder.buildBuildVector(BVType, Regs); - MIRBuilder.buildTrunc(OrigRegs[0], BV); + auto BV = B.buildBuildVector(BVType, Regs); + B.buildTrunc(OrigRegs[0], BV); } } bool AMDGPUCallLowering::lowerFormalArguments( - MachineIRBuilder &MIRBuilder, const Function &F, + MachineIRBuilder &B, const Function &F, ArrayRef> VRegs) const { CallingConv::ID CC = F.getCallingConv(); @@ -554,13 +554,13 @@ // useless for kernels. We want to avoid any kind of legalization or argument // splitting. if (CC == CallingConv::AMDGPU_KERNEL) - return lowerFormalArgumentsKernel(MIRBuilder, F, VRegs); + return lowerFormalArgumentsKernel(B, F, VRegs); const bool IsShader = AMDGPU::isShader(CC); const bool IsEntryFunc = AMDGPU::isEntryFunctionCC(CC); - MachineFunction &MF = MIRBuilder.getMF(); - MachineBasicBlock &MBB = MIRBuilder.getMBB(); + MachineFunction &MF = B.getMF(); + MachineBasicBlock &MBB = B.getMBB(); MachineRegisterInfo &MRI = MF.getRegInfo(); SIMachineFunctionInfo *Info = MF.getInfo(); const GCNSubtarget &Subtarget = MF.getSubtarget(); @@ -576,7 +576,7 @@ Register LiveInReturn = MF.addLiveIn(ReturnAddrReg, &AMDGPU::SGPR_64RegClass); MBB.addLiveIn(ReturnAddrReg); - MIRBuilder.buildCopy(LiveInReturn, ReturnAddrReg); + B.buildCopy(LiveInReturn, ReturnAddrReg); } if (Info->hasImplicitBufferPtr()) { @@ -619,7 +619,7 @@ if (SkipArg) { for (int I = 0, E = VRegs[Idx].size(); I != E; ++I) - MIRBuilder.buildUndef(VRegs[Idx][I]); + B.buildUndef(VRegs[Idx][I]); ++Idx; continue; @@ -634,7 +634,7 @@ // FIXME: We should probably be passing multiple registers to // handleAssignments to do this [&](ArrayRef Regs, LLT LLTy, LLT PartLLT, int VTSplitIdx) { - packSplitRegsToOrigType(MIRBuilder, VRegs[Idx][VTSplitIdx], Regs, + packSplitRegsToOrigType(B, VRegs[Idx][VTSplitIdx], Regs, LLTy, PartLLT); }); @@ -686,10 +686,10 @@ CCAssignFn *AssignFn = TLI.CCAssignFnForCall(CC, F.isVarArg()); if (!MBB.empty()) - MIRBuilder.setInstr(*MBB.begin()); + B.setInstr(*MBB.begin()); - FormalArgHandler Handler(MIRBuilder, MRI, AssignFn); - if (!handleAssignments(CCInfo, ArgLocs, MIRBuilder, SplitArgs, Handler)) + FormalArgHandler Handler(B, MRI, AssignFn); + if (!handleAssignments(CCInfo, ArgLocs, B, SplitArgs, Handler)) return false; if (!IsEntryFunc) { @@ -708,7 +708,7 @@ } // Move back to the end of the basic block. - MIRBuilder.setMBB(MBB); + B.setMBB(MBB); return true; } diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h --- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h +++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.h @@ -32,34 +32,34 @@ const GCNTargetMachine &TM); bool legalizeCustom(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder, + MachineIRBuilder &B, GISelChangeObserver &Observer) const override; Register getSegmentAperture(unsigned AddrSpace, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeAddrSpaceCast(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeFrint(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeFceil(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeIntrinsicTrunc(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeITOFP(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder, bool Signed) const; + MachineIRBuilder &B, bool Signed) const; bool legalizeMinNumMaxNum(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeExtractVectorElt(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeInsertVectorElt(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeSinCos(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; bool legalizeGlobalValue(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const; + MachineIRBuilder &B) const; Register getLiveInRegister(MachineRegisterInfo &MRI, Register Reg, LLT Ty) const; @@ -78,7 +78,7 @@ bool legalizeIsAddrSpace(MachineInstr &MI, MachineRegisterInfo &MRI, MachineIRBuilder &B, unsigned AddrSpace) const; bool legalizeIntrinsic(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const override; + MachineIRBuilder &B) const override; }; } // End llvm namespace. diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp --- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp @@ -823,35 +823,35 @@ bool AMDGPULegalizerInfo::legalizeCustom(MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder, + MachineIRBuilder &B, GISelChangeObserver &Observer) const { switch (MI.getOpcode()) { case TargetOpcode::G_ADDRSPACE_CAST: - return legalizeAddrSpaceCast(MI, MRI, MIRBuilder); + return legalizeAddrSpaceCast(MI, MRI, B); case TargetOpcode::G_FRINT: - return legalizeFrint(MI, MRI, MIRBuilder); + return legalizeFrint(MI, MRI, B); case TargetOpcode::G_FCEIL: - return legalizeFceil(MI, MRI, MIRBuilder); + return legalizeFceil(MI, MRI, B); case TargetOpcode::G_INTRINSIC_TRUNC: - return legalizeIntrinsicTrunc(MI, MRI, MIRBuilder); + return legalizeIntrinsicTrunc(MI, MRI, B); case TargetOpcode::G_SITOFP: - return legalizeITOFP(MI, MRI, MIRBuilder, true); + return legalizeITOFP(MI, MRI, B, true); case TargetOpcode::G_UITOFP: - return legalizeITOFP(MI, MRI, MIRBuilder, false); + return legalizeITOFP(MI, MRI, B, false); case TargetOpcode::G_FMINNUM: case TargetOpcode::G_FMAXNUM: case TargetOpcode::G_FMINNUM_IEEE: case TargetOpcode::G_FMAXNUM_IEEE: - return legalizeMinNumMaxNum(MI, MRI, MIRBuilder); + return legalizeMinNumMaxNum(MI, MRI, B); case TargetOpcode::G_EXTRACT_VECTOR_ELT: - return legalizeExtractVectorElt(MI, MRI, MIRBuilder); + return legalizeExtractVectorElt(MI, MRI, B); case TargetOpcode::G_INSERT_VECTOR_ELT: - return legalizeInsertVectorElt(MI, MRI, MIRBuilder); + return legalizeInsertVectorElt(MI, MRI, B); case TargetOpcode::G_FSIN: case TargetOpcode::G_FCOS: - return legalizeSinCos(MI, MRI, MIRBuilder); + return legalizeSinCos(MI, MRI, B); case TargetOpcode::G_GLOBAL_VALUE: - return legalizeGlobalValue(MI, MRI, MIRBuilder); + return legalizeGlobalValue(MI, MRI, B); default: return false; } @@ -862,8 +862,8 @@ Register AMDGPULegalizerInfo::getSegmentAperture( unsigned AS, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineIRBuilder &B) const { + MachineFunction &MF = B.getMF(); const GCNSubtarget &ST = MF.getSubtarget(); const LLT S32 = LLT::scalar(32); @@ -884,13 +884,13 @@ Register ApertureReg = MRI.createGenericVirtualRegister(S32); Register GetReg = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass); - MIRBuilder.buildInstr(AMDGPU::S_GETREG_B32) + B.buildInstr(AMDGPU::S_GETREG_B32) .addDef(GetReg) .addImm(Encoding); MRI.setType(GetReg, S32); - auto ShiftAmt = MIRBuilder.buildConstant(S32, WidthM1 + 1); - MIRBuilder.buildInstr(TargetOpcode::G_SHL) + auto ShiftAmt = B.buildConstant(S32, WidthM1 + 1); + B.buildInstr(TargetOpcode::G_SHL) .addDef(ApertureReg) .addUse(GetReg) .addUse(ShiftAmt.getReg(0)); @@ -902,7 +902,7 @@ LLT::pointer(AMDGPUAS::CONSTANT_ADDRESS, 64)); const SIMachineFunctionInfo *MFI = MF.getInfo(); - if (!loadInputValue(QueuePtr, MIRBuilder, &MFI->getArgInfo().QueuePtr)) + if (!loadInputValue(QueuePtr, B, &MFI->getArgInfo().QueuePtr)) return Register(); // Offset into amd_queue_t for group_segment_aperture_base_hi / @@ -926,17 +926,17 @@ Register LoadResult = MRI.createGenericVirtualRegister(S32); Register LoadAddr; - MIRBuilder.materializeGEP(LoadAddr, QueuePtr, LLT::scalar(64), StructOffset); - MIRBuilder.buildLoad(LoadResult, LoadAddr, *MMO); + B.materializeGEP(LoadAddr, QueuePtr, LLT::scalar(64), StructOffset); + B.buildLoad(LoadResult, LoadAddr, *MMO); return LoadResult; } bool AMDGPULegalizerInfo::legalizeAddrSpaceCast( MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const { - MachineFunction &MF = MIRBuilder.getMF(); + MachineIRBuilder &B) const { + MachineFunction &MF = B.getMF(); - MIRBuilder.setInstr(MI); + B.setInstr(MI); const LLT S32 = LLT::scalar(32); Register Dst = MI.getOperand(0).getReg(); @@ -956,13 +956,13 @@ const GCNSubtarget &ST = MF.getSubtarget(); if (ST.getTargetLowering()->isNoopAddrSpaceCast(SrcAS, DestAS)) { - MI.setDesc(MIRBuilder.getTII().get(TargetOpcode::G_BITCAST)); + MI.setDesc(B.getTII().get(TargetOpcode::G_BITCAST)); return true; } if (DestAS == AMDGPUAS::CONSTANT_ADDRESS_32BIT) { // Truncate. - MIRBuilder.buildExtract(Dst, Src, 0); + B.buildExtract(Dst, Src, 0); MI.eraseFromParent(); return true; } @@ -974,9 +974,9 @@ // FIXME: This is a bit ugly due to creating a merge of 2 pointers to // another. Merge operands are required to be the same type, but creating an // extra ptrtoint would be kind of pointless. - auto HighAddr = MIRBuilder.buildConstant( + auto HighAddr = B.buildConstant( LLT::pointer(AMDGPUAS::CONSTANT_ADDRESS_32BIT, 32), AddrHiVal); - MIRBuilder.buildMerge(Dst, {Src, HighAddr.getReg(0)}); + B.buildMerge(Dst, {Src, HighAddr.getReg(0)}); MI.eraseFromParent(); return true; } @@ -986,17 +986,17 @@ DestAS == AMDGPUAS::PRIVATE_ADDRESS); unsigned NullVal = TM.getNullPointerValue(DestAS); - auto SegmentNull = MIRBuilder.buildConstant(DstTy, NullVal); - auto FlatNull = MIRBuilder.buildConstant(SrcTy, 0); + auto SegmentNull = B.buildConstant(DstTy, NullVal); + auto FlatNull = B.buildConstant(SrcTy, 0); Register PtrLo32 = MRI.createGenericVirtualRegister(DstTy); // Extract low 32-bits of the pointer. - MIRBuilder.buildExtract(PtrLo32, Src, 0); + B.buildExtract(PtrLo32, Src, 0); Register CmpRes = MRI.createGenericVirtualRegister(LLT::scalar(1)); - MIRBuilder.buildICmp(CmpInst::ICMP_NE, CmpRes, Src, FlatNull.getReg(0)); - MIRBuilder.buildSelect(Dst, CmpRes, PtrLo32, SegmentNull.getReg(0)); + B.buildICmp(CmpInst::ICMP_NE, CmpRes, Src, FlatNull.getReg(0)); + B.buildSelect(Dst, CmpRes, PtrLo32, SegmentNull.getReg(0)); MI.eraseFromParent(); return true; @@ -1009,29 +1009,29 @@ return false; auto SegmentNull = - MIRBuilder.buildConstant(SrcTy, TM.getNullPointerValue(SrcAS)); + B.buildConstant(SrcTy, TM.getNullPointerValue(SrcAS)); auto FlatNull = - MIRBuilder.buildConstant(DstTy, TM.getNullPointerValue(DestAS)); + B.buildConstant(DstTy, TM.getNullPointerValue(DestAS)); - Register ApertureReg = getSegmentAperture(DestAS, MRI, MIRBuilder); + Register ApertureReg = getSegmentAperture(DestAS, MRI, B); if (!ApertureReg.isValid()) return false; Register CmpRes = MRI.createGenericVirtualRegister(LLT::scalar(1)); - MIRBuilder.buildICmp(CmpInst::ICMP_NE, CmpRes, Src, SegmentNull.getReg(0)); + B.buildICmp(CmpInst::ICMP_NE, CmpRes, Src, SegmentNull.getReg(0)); Register BuildPtr = MRI.createGenericVirtualRegister(DstTy); // Coerce the type of the low half of the result so we can use merge_values. Register SrcAsInt = MRI.createGenericVirtualRegister(S32); - MIRBuilder.buildInstr(TargetOpcode::G_PTRTOINT) + B.buildInstr(TargetOpcode::G_PTRTOINT) .addDef(SrcAsInt) .addUse(Src); // TODO: Should we allow mismatched types but matching sizes in merges to // avoid the ptrtoint? - MIRBuilder.buildMerge(BuildPtr, {SrcAsInt, ApertureReg}); - MIRBuilder.buildSelect(Dst, CmpRes, BuildPtr, FlatNull.getReg(0)); + B.buildMerge(BuildPtr, {SrcAsInt, ApertureReg}); + B.buildSelect(Dst, CmpRes, BuildPtr, FlatNull.getReg(0)); MI.eraseFromParent(); return true; @@ -1039,8 +1039,8 @@ bool AMDGPULegalizerInfo::legalizeFrint( MachineInstr &MI, MachineRegisterInfo &MRI, - MachineIRBuilder &MIRBuilder) const { - MIRBuilder.setInstr(MI); + MachineIRBuilder &B) const { + B.setInstr(MI); Register Src = MI.getOperand(1).getReg(); LLT Ty = MRI.getType(Src); @@ -1049,18 +1049,18 @@ APFloat C1Val(APFloat::IEEEdouble(), "0x1.0p+52"); APFloat C2Val(APFloat::IEEEdouble(), "0x1.fffffffffffffp+51"); - auto C1 = MIRBuilder.buildFConstant(Ty, C1Val); - auto CopySign = MIRBuilder.buildFCopysign(Ty, C1, Src); + auto C1 = B.buildFConstant(Ty, C1Val); + auto CopySign = B.buildFCopysign(Ty, C1, Src); // TODO: Should this propagate fast-math-flags? - auto Tmp1 = MIRBuilder.buildFAdd(Ty, Src, CopySign); - auto Tmp2 = MIRBuilder.buildFSub(Ty, Tmp1, CopySign); + auto Tmp1 = B.buildFAdd(Ty, Src, CopySign); + auto Tmp2 = B.buildFSub(Ty, Tmp1, CopySign); - auto C2 = MIRBuilder.buildFConstant(Ty, C2Val); - auto Fabs = MIRBuilder.buildFAbs(Ty, Src); + auto C2 = B.buildFConstant(Ty, C2Val); + auto Fabs = B.buildFAbs(Ty, Src); - auto Cond = MIRBuilder.buildFCmp(CmpInst::FCMP_OGT, LLT::scalar(1), Fabs, C2); - MIRBuilder.buildSelect(MI.getOperand(0).getReg(), Cond, Src, Tmp2); + auto Cond = B.buildFCmp(CmpInst::FCMP_OGT, LLT::scalar(1), Fabs, C2); + B.buildSelect(MI.getOperand(0).getReg(), Cond, Src, Tmp2); return true; }