diff options
Diffstat (limited to 'lib/Target/ARM/ARMISelLowering.cpp')
| -rw-r--r-- | lib/Target/ARM/ARMISelLowering.cpp | 669 |
1 files changed, 503 insertions, 166 deletions
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index a103c94..c66618a 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -52,6 +52,7 @@ using namespace llvm; STATISTIC(NumTailCalls, "Number of tail calls"); STATISTIC(NumMovwMovt, "Number of GAs materialized with movw + movt"); +STATISTIC(NumLoopByVals, "Number of loops generated for byval arguments"); // This option should go away when tail calls fully work. static cl::opt<bool> @@ -89,76 +90,71 @@ static const uint16_t GPRArgRegs[] = { ARM::R0, ARM::R1, ARM::R2, ARM::R3 }; -void ARMTargetLowering::addTypeForNEON(EVT VT, EVT PromotedLdStVT, - EVT PromotedBitwiseVT) { +void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT, + MVT PromotedBitwiseVT) { if (VT != PromotedLdStVT) { - setOperationAction(ISD::LOAD, VT.getSimpleVT(), Promote); - AddPromotedToType (ISD::LOAD, VT.getSimpleVT(), - PromotedLdStVT.getSimpleVT()); + setOperationAction(ISD::LOAD, VT, Promote); + AddPromotedToType (ISD::LOAD, VT, PromotedLdStVT); - setOperationAction(ISD::STORE, VT.getSimpleVT(), Promote); - AddPromotedToType (ISD::STORE, VT.getSimpleVT(), - PromotedLdStVT.getSimpleVT()); + setOperationAction(ISD::STORE, VT, Promote); + AddPromotedToType (ISD::STORE, VT, PromotedLdStVT); } - EVT ElemTy = VT.getVectorElementType(); + MVT ElemTy = VT.getVectorElementType(); if (ElemTy != MVT::i64 && ElemTy != MVT::f64) - setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom); - setOperationAction(ISD::INSERT_VECTOR_ELT, VT.getSimpleVT(), Custom); - setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT.getSimpleVT(), Custom); + setOperationAction(ISD::SETCC, VT, Custom); + setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom); + setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom); if (ElemTy == MVT::i32) { - setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Custom); - setOperationAction(ISD::UINT_TO_FP, VT.getSimpleVT(), Custom); - setOperationAction(ISD::FP_TO_SINT, VT.getSimpleVT(), Custom); - setOperationAction(ISD::FP_TO_UINT, VT.getSimpleVT(), Custom); + setOperationAction(ISD::SINT_TO_FP, VT, Custom); + setOperationAction(ISD::UINT_TO_FP, VT, Custom); + setOperationAction(ISD::FP_TO_SINT, VT, Custom); + setOperationAction(ISD::FP_TO_UINT, VT, Custom); } else { - setOperationAction(ISD::SINT_TO_FP, VT.getSimpleVT(), Expand); - setOperationAction(ISD::UINT_TO_FP, VT.getSimpleVT(), Expand); - setOperationAction(ISD::FP_TO_SINT, VT.getSimpleVT(), Expand); - setOperationAction(ISD::FP_TO_UINT, VT.getSimpleVT(), Expand); - } - setOperationAction(ISD::BUILD_VECTOR, VT.getSimpleVT(), Custom); - setOperationAction(ISD::VECTOR_SHUFFLE, VT.getSimpleVT(), Custom); - setOperationAction(ISD::CONCAT_VECTORS, VT.getSimpleVT(), Legal); - setOperationAction(ISD::EXTRACT_SUBVECTOR, VT.getSimpleVT(), Legal); - setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand); - setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand); - setOperationAction(ISD::SIGN_EXTEND_INREG, VT.getSimpleVT(), Expand); + setOperationAction(ISD::SINT_TO_FP, VT, Expand); + setOperationAction(ISD::UINT_TO_FP, VT, Expand); + setOperationAction(ISD::FP_TO_SINT, VT, Expand); + setOperationAction(ISD::FP_TO_UINT, VT, Expand); + } + setOperationAction(ISD::BUILD_VECTOR, VT, Custom); + setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom); + setOperationAction(ISD::CONCAT_VECTORS, VT, Legal); + setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal); + setOperationAction(ISD::SELECT, VT, Expand); + setOperationAction(ISD::SELECT_CC, VT, Expand); + setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand); if (VT.isInteger()) { - setOperationAction(ISD::SHL, VT.getSimpleVT(), Custom); - setOperationAction(ISD::SRA, VT.getSimpleVT(), Custom); - setOperationAction(ISD::SRL, VT.getSimpleVT(), Custom); + setOperationAction(ISD::SHL, VT, Custom); + setOperationAction(ISD::SRA, VT, Custom); + setOperationAction(ISD::SRL, VT, Custom); } // Promote all bit-wise operations. if (VT.isInteger() && VT != PromotedBitwiseVT) { - setOperationAction(ISD::AND, VT.getSimpleVT(), Promote); - AddPromotedToType (ISD::AND, VT.getSimpleVT(), - PromotedBitwiseVT.getSimpleVT()); - setOperationAction(ISD::OR, VT.getSimpleVT(), Promote); - AddPromotedToType (ISD::OR, VT.getSimpleVT(), - PromotedBitwiseVT.getSimpleVT()); - setOperationAction(ISD::XOR, VT.getSimpleVT(), Promote); - AddPromotedToType (ISD::XOR, VT.getSimpleVT(), - PromotedBitwiseVT.getSimpleVT()); + setOperationAction(ISD::AND, VT, Promote); + AddPromotedToType (ISD::AND, VT, PromotedBitwiseVT); + setOperationAction(ISD::OR, VT, Promote); + AddPromotedToType (ISD::OR, VT, PromotedBitwiseVT); + setOperationAction(ISD::XOR, VT, Promote); + AddPromotedToType (ISD::XOR, VT, PromotedBitwiseVT); } // Neon does not support vector divide/remainder operations. - setOperationAction(ISD::SDIV, VT.getSimpleVT(), Expand); - setOperationAction(ISD::UDIV, VT.getSimpleVT(), Expand); - setOperationAction(ISD::FDIV, VT.getSimpleVT(), Expand); - setOperationAction(ISD::SREM, VT.getSimpleVT(), Expand); - setOperationAction(ISD::UREM, VT.getSimpleVT(), Expand); - setOperationAction(ISD::FREM, VT.getSimpleVT(), Expand); + setOperationAction(ISD::SDIV, VT, Expand); + setOperationAction(ISD::UDIV, VT, Expand); + setOperationAction(ISD::FDIV, VT, Expand); + setOperationAction(ISD::SREM, VT, Expand); + setOperationAction(ISD::UREM, VT, Expand); + setOperationAction(ISD::FREM, VT, Expand); } -void ARMTargetLowering::addDRTypeForNEON(EVT VT) { - addRegisterClass(VT, ARM::DPRRegisterClass); +void ARMTargetLowering::addDRTypeForNEON(MVT VT) { + addRegisterClass(VT, &ARM::DPRRegClass); addTypeForNEON(VT, MVT::f64, MVT::v2i32); } -void ARMTargetLowering::addQRTypeForNEON(EVT VT) { - addRegisterClass(VT, ARM::QPRRegisterClass); +void ARMTargetLowering::addQRTypeForNEON(MVT VT) { + addRegisterClass(VT, &ARM::QPRRegClass); addTypeForNEON(VT, MVT::v2f64, MVT::v4i32); } @@ -431,14 +427,14 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) } if (Subtarget->isThumb1Only()) - addRegisterClass(MVT::i32, ARM::tGPRRegisterClass); + addRegisterClass(MVT::i32, &ARM::tGPRRegClass); else - addRegisterClass(MVT::i32, ARM::GPRRegisterClass); + addRegisterClass(MVT::i32, &ARM::GPRRegClass); if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() && !Subtarget->isThumb1Only()) { - addRegisterClass(MVT::f32, ARM::SPRRegisterClass); + addRegisterClass(MVT::f32, &ARM::SPRRegClass); if (!Subtarget->isFPOnlySP()) - addRegisterClass(MVT::f64, ARM::DPRRegisterClass); + addRegisterClass(MVT::f64, &ARM::DPRRegClass); setTruncStoreAction(MVT::f64, MVT::f32, Expand); } @@ -824,6 +820,9 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) benefitFromCodePlacementOpt = true; + // Prefer likely predicted branches to selects on out-of-order cores. + predictableSelectIsExpensive = Subtarget->isCortexA9(); + setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2); } @@ -849,7 +848,7 @@ ARMTargetLowering::findRepresentativeClass(EVT VT) const{ // the cost is 1 for both f32 and f64. case MVT::f32: case MVT::f64: case MVT::v8i8: case MVT::v4i16: case MVT::v2i32: case MVT::v1i64: case MVT::v2f32: - RRC = ARM::DPRRegisterClass; + RRC = &ARM::DPRRegClass; // When NEON is used for SP, only half of the register file is available // because operations that define both SP and DP results will be constrained // to the VFP2 class (D0-D15). We currently model this constraint prior to @@ -859,15 +858,15 @@ ARMTargetLowering::findRepresentativeClass(EVT VT) const{ break; case MVT::v16i8: case MVT::v8i16: case MVT::v4i32: case MVT::v2i64: case MVT::v4f32: case MVT::v2f64: - RRC = ARM::DPRRegisterClass; + RRC = &ARM::DPRRegClass; Cost = 2; break; case MVT::v4i64: - RRC = ARM::DPRRegisterClass; + RRC = &ARM::DPRRegClass; Cost = 4; break; case MVT::v8i64: - RRC = ARM::DPRRegisterClass; + RRC = &ARM::DPRRegClass; Cost = 8; break; } @@ -891,6 +890,7 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const { case ARMISD::RET_FLAG: return "ARMISD::RET_FLAG"; case ARMISD::PIC_ADD: return "ARMISD::PIC_ADD"; case ARMISD::CMP: return "ARMISD::CMP"; + case ARMISD::CMN: return "ARMISD::CMN"; case ARMISD::CMPZ: return "ARMISD::CMPZ"; case ARMISD::CMPFP: return "ARMISD::CMPFP"; case ARMISD::CMPFPw0: return "ARMISD::CMPFPw0"; @@ -1027,17 +1027,18 @@ const TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const { // load / store 4 to 8 consecutive D registers. if (Subtarget->hasNEON()) { if (VT == MVT::v4i64) - return ARM::QQPRRegisterClass; - else if (VT == MVT::v8i64) - return ARM::QQQQPRRegisterClass; + return &ARM::QQPRRegClass; + if (VT == MVT::v8i64) + return &ARM::QQQQPRRegClass; } return TargetLowering::getRegClassFor(VT); } // Create a fast isel object. FastISel * -ARMTargetLowering::createFastISel(FunctionLoweringInfo &funcInfo) const { - return ARM::createFastISel(funcInfo); +ARMTargetLowering::createFastISel(FunctionLoweringInfo &funcInfo, + const TargetLibraryInfo *libInfo) const { + return ARM::createFastISel(funcInfo, libInfo); } /// getMaximalGlobalOffset - Returns the maximal possible offset which can @@ -1166,6 +1167,8 @@ CCAssignFn *ARMTargetLowering::CCAssignFnForNode(CallingConv::ID CC, return (Return ? RetCC_ARM_AAPCS : CC_ARM_AAPCS); case CallingConv::ARM_APCS: return (Return ? RetCC_ARM_APCS : CC_ARM_APCS); + case CallingConv::GHC: + return (Return ? RetCC_ARM_APCS : CC_ARM_APCS_GHC); } } @@ -1286,14 +1289,20 @@ void ARMTargetLowering::PassF64ArgInRegs(DebugLoc dl, SelectionDAG &DAG, /// ARMISD:CALL <- callseq_end chain. Also add input and output parameter /// nodes. SDValue -ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, - CallingConv::ID CallConv, bool isVarArg, - bool doesNotRet, bool &isTailCall, - const SmallVectorImpl<ISD::OutputArg> &Outs, - const SmallVectorImpl<SDValue> &OutVals, - const SmallVectorImpl<ISD::InputArg> &Ins, - DebugLoc dl, SelectionDAG &DAG, +ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI, SmallVectorImpl<SDValue> &InVals) const { + SelectionDAG &DAG = CLI.DAG; + 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; + CallingConv::ID CallConv = CLI.CallConv; + bool doesNotRet = CLI.DoesNotReturn; + bool isVarArg = CLI.IsVarArg; + MachineFunction &MF = DAG.getMachineFunction(); bool IsStructRet = (Outs.empty()) ? false : Outs[0].Flags.isSRet(); bool IsSibCall = false; @@ -1415,21 +1424,22 @@ ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee, CCInfo.clearFirstByValReg(); } - unsigned LocMemOffset = VA.getLocMemOffset(); - SDValue StkPtrOff = DAG.getIntPtrConstant(LocMemOffset); - SDValue Dst = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, - StkPtrOff); - SDValue SrcOffset = DAG.getIntPtrConstant(4*offset); - SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset); - SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset, - MVT::i32); - MemOpChains.push_back(DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, - Flags.getByValAlign(), - /*isVolatile=*/false, - /*AlwaysInline=*/false, - MachinePointerInfo(0), - MachinePointerInfo(0))); - + if (Flags.getByValSize() - 4*offset > 0) { + unsigned LocMemOffset = VA.getLocMemOffset(); + SDValue StkPtrOff = DAG.getIntPtrConstant(LocMemOffset); + SDValue Dst = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, + StkPtrOff); + SDValue SrcOffset = DAG.getIntPtrConstant(4*offset); + SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset); + SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset, + MVT::i32); + SDValue AlignNode = DAG.getConstant(Flags.getByValAlign(), MVT::i32); + + SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue); + SDValue Ops[] = { Chain, Dst, Src, SizeNode, AlignNode}; + MemOpChains.push_back(DAG.getNode(ARMISD::COPY_STRUCT_BYVAL, dl, VTs, + Ops, array_lengthof(Ops))); + } } else if (!IsSibCall) { assert(VA.isMemLoc()); @@ -2095,12 +2105,13 @@ ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA, Entry.Ty = (Type *) Type::getInt32Ty(*DAG.getContext()); Args.push_back(Entry); // FIXME: is there useful debug info available here? - std::pair<SDValue, SDValue> CallResult = - LowerCallTo(Chain, (Type *) Type::getInt32Ty(*DAG.getContext()), + TargetLowering::CallLoweringInfo CLI(Chain, + (Type *) Type::getInt32Ty(*DAG.getContext()), false, false, false, false, 0, CallingConv::C, /*isTailCall=*/false, /*doesNotRet=*/false, /*isReturnValueUsed=*/true, DAG.getExternalSymbol("__tls_get_addr", PtrVT), Args, DAG, dl); + std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI); return CallResult.first; } @@ -2108,7 +2119,8 @@ ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA, // "local exec" model. SDValue ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA, - SelectionDAG &DAG) const { + SelectionDAG &DAG, + TLSModel::Model model) const { const GlobalValue *GV = GA->getGlobal(); DebugLoc dl = GA->getDebugLoc(); SDValue Offset; @@ -2117,7 +2129,7 @@ ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA, // Get the Thread Pointer SDValue ThreadPointer = DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT); - if (GV->isDeclaration()) { + if (model == TLSModel::InitialExec) { MachineFunction &MF = DAG.getMachineFunction(); ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>(); unsigned ARMPCLabelIndex = AFI->createPICLabelUId(); @@ -2142,6 +2154,7 @@ ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA, false, false, false, 0); } else { // local exec model + assert(model == TLSModel::LocalExec); ARMConstantPoolValue *CPV = ARMConstantPoolConstant::Create(GV, ARMCP::TPOFF); Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4); @@ -2162,12 +2175,18 @@ ARMTargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const { assert(Subtarget->isTargetELF() && "TLS not implemented for non-ELF targets"); GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op); - // If the relocation model is PIC, use the "General Dynamic" TLS Model, - // otherwise use the "Local Exec" TLS Model - if (getTargetMachine().getRelocationModel() == Reloc::PIC_) - return LowerToTLSGeneralDynamicModel(GA, DAG); - else - return LowerToTLSExecModels(GA, DAG); + + TLSModel::Model model = getTargetMachine().getTLSModel(GA->getGlobal()); + + switch (model) { + case TLSModel::GeneralDynamic: + case TLSModel::LocalDynamic: + return LowerToTLSGeneralDynamicModel(GA, DAG); + case TLSModel::InitialExec: + case TLSModel::LocalExec: + return LowerToTLSExecModels(GA, DAG, model); + } + llvm_unreachable("bogus TLS model"); } SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op, @@ -2457,9 +2476,9 @@ ARMTargetLowering::GetF64FormalArgument(CCValAssign &VA, CCValAssign &NextVA, const TargetRegisterClass *RC; if (AFI->isThumb1OnlyFunction()) - RC = ARM::tGPRRegisterClass; + RC = &ARM::tGPRRegClass; else - RC = ARM::GPRRegisterClass; + RC = &ARM::GPRRegClass; // Transform the arguments stored in physical registers into virtual ones. unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC); @@ -2543,9 +2562,9 @@ ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG, for (; firstRegToSaveIndex < 4; ++firstRegToSaveIndex) { const TargetRegisterClass *RC; if (AFI->isThumb1OnlyFunction()) - RC = ARM::tGPRRegisterClass; + RC = &ARM::tGPRRegClass; else - RC = ARM::GPRRegisterClass; + RC = &ARM::GPRRegClass; unsigned VReg = MF.addLiveIn(GPRArgRegs[firstRegToSaveIndex], RC); SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32); @@ -2627,14 +2646,15 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain, const TargetRegisterClass *RC; if (RegVT == MVT::f32) - RC = ARM::SPRRegisterClass; + RC = &ARM::SPRRegClass; else if (RegVT == MVT::f64) - RC = ARM::DPRRegisterClass; + RC = &ARM::DPRRegClass; else if (RegVT == MVT::v2f64) - RC = ARM::QPRRegisterClass; + RC = &ARM::QPRRegClass; else if (RegVT == MVT::i32) - RC = (AFI->isThumb1OnlyFunction() ? - ARM::tGPRRegisterClass : ARM::GPRRegisterClass); + RC = AFI->isThumb1OnlyFunction() ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; else llvm_unreachable("RegVT not supported by FORMAL_ARGUMENTS Lowering"); @@ -4249,6 +4269,10 @@ SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op, // Record this extraction against the appropriate vector if possible... SDValue SourceVec = V.getOperand(0); + // If the element number isn't a constant, we can't effectively + // analyze what's going on. + if (!isa<ConstantSDNode>(V.getOperand(1))) + return SDValue(); unsigned EltNo = cast<ConstantSDNode>(V.getOperand(1))->getZExtValue(); bool FoundSource = false; for (unsigned j = 0; j < SourceVecs.size(); ++j) { @@ -4791,7 +4815,9 @@ static SDValue SkipExtension(SDNode *N, SelectionDAG &DAG) { for (unsigned i = 0; i != NumElts; ++i) { ConstantSDNode *C = cast<ConstantSDNode>(N->getOperand(i)); const APInt &CInt = C->getAPIntValue(); - Ops.push_back(DAG.getConstant(CInt.trunc(EltSize), TruncVT)); + // Element types smaller than 32 bits are not legal, so use i32 elements. + // The values are implicitly truncated so sext vs. zext doesn't matter. + Ops.push_back(DAG.getConstant(CInt.zextOrTrunc(32), MVT::i32)); } return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(), MVT::getVectorVT(TruncVT, NumElts), Ops.data(), NumElts); @@ -5252,14 +5278,14 @@ ARMTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, bool isThumb2 = Subtarget->isThumb2(); MachineRegisterInfo &MRI = BB->getParent()->getRegInfo(); - unsigned scratch = - MRI.createVirtualRegister(isThumb2 ? ARM::rGPRRegisterClass - : ARM::GPRRegisterClass); + unsigned scratch = MRI.createVirtualRegister(isThumb2 ? + (const TargetRegisterClass*)&ARM::rGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass); if (isThumb2) { - MRI.constrainRegClass(dest, ARM::rGPRRegisterClass); - MRI.constrainRegClass(oldval, ARM::rGPRRegisterClass); - MRI.constrainRegClass(newval, ARM::rGPRRegisterClass); + MRI.constrainRegClass(dest, &ARM::rGPRRegClass); + MRI.constrainRegClass(oldval, &ARM::rGPRRegClass); + MRI.constrainRegClass(newval, &ARM::rGPRRegClass); } unsigned ldrOpc, strOpc; @@ -5362,8 +5388,8 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, MachineRegisterInfo &MRI = BB->getParent()->getRegInfo(); if (isThumb2) { - MRI.constrainRegClass(dest, ARM::rGPRRegisterClass); - MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass); + MRI.constrainRegClass(dest, &ARM::rGPRRegClass); + MRI.constrainRegClass(ptr, &ARM::rGPRRegClass); } unsigned ldrOpc, strOpc; @@ -5394,8 +5420,9 @@ ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, BB->end()); exitMBB->transferSuccessorsAndUpdatePHIs(BB); - const TargetRegisterClass *TRC = - isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + const TargetRegisterClass *TRC = isThumb2 ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; unsigned scratch = MRI.createVirtualRegister(TRC); unsigned scratch2 = (!BinOpcode) ? incr : MRI.createVirtualRegister(TRC); @@ -5469,8 +5496,8 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, MachineRegisterInfo &MRI = BB->getParent()->getRegInfo(); if (isThumb2) { - MRI.constrainRegClass(dest, ARM::rGPRRegisterClass); - MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass); + MRI.constrainRegClass(dest, &ARM::rGPRRegClass); + MRI.constrainRegClass(ptr, &ARM::rGPRRegClass); } unsigned ldrOpc, strOpc, extendOpc; @@ -5504,8 +5531,9 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, BB->end()); exitMBB->transferSuccessorsAndUpdatePHIs(BB); - const TargetRegisterClass *TRC = - isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + const TargetRegisterClass *TRC = isThumb2 ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; unsigned scratch = MRI.createVirtualRegister(TRC); unsigned scratch2 = MRI.createVirtualRegister(TRC); @@ -5531,7 +5559,7 @@ ARMTargetLowering::EmitAtomicBinaryMinMax(MachineInstr *MI, // Sign extend the value, if necessary. if (signExtend && extendOpc) { - oldval = MRI.createVirtualRegister(ARM::GPRRegisterClass); + oldval = MRI.createVirtualRegister(&ARM::GPRRegClass); AddDefaultPred(BuildMI(BB, dl, TII->get(extendOpc), oldval) .addReg(dest) .addImm(0)); @@ -5586,9 +5614,9 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB, MachineRegisterInfo &MRI = BB->getParent()->getRegInfo(); if (isThumb2) { - MRI.constrainRegClass(destlo, ARM::rGPRRegisterClass); - MRI.constrainRegClass(desthi, ARM::rGPRRegisterClass); - MRI.constrainRegClass(ptr, ARM::rGPRRegisterClass); + MRI.constrainRegClass(destlo, &ARM::rGPRRegClass); + MRI.constrainRegClass(desthi, &ARM::rGPRRegClass); + MRI.constrainRegClass(ptr, &ARM::rGPRRegClass); } unsigned ldrOpc = isThumb2 ? ARM::t2LDREXD : ARM::LDREXD; @@ -5614,8 +5642,9 @@ ARMTargetLowering::EmitAtomicBinary64(MachineInstr *MI, MachineBasicBlock *BB, BB->end()); exitMBB->transferSuccessorsAndUpdatePHIs(BB); - const TargetRegisterClass *TRC = - isThumb2 ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + const TargetRegisterClass *TRC = isThumb2 ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; unsigned storesuccess = MRI.createVirtualRegister(TRC); // thisMBB: @@ -5722,8 +5751,9 @@ SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB, ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj); unsigned CPI = MCP->getConstantPoolIndex(CPV, 4); - const TargetRegisterClass *TRC = - isThumb ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + const TargetRegisterClass *TRC = isThumb ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; // Grab constant pool and fixed stack memory operands. MachineMemOperand *CPMMO = @@ -5827,8 +5857,9 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const { MachineFrameInfo *MFI = MF->getFrameInfo(); int FI = MFI->getFunctionContextIndex(); - const TargetRegisterClass *TRC = - Subtarget->isThumb() ? ARM::tGPRRegisterClass : ARM::GPRRegisterClass; + const TargetRegisterClass *TRC = Subtarget->isThumb() ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRnopcRegClass; // Get a mapping of the call site numbers to all of the landing pads they're // associated with. @@ -6176,14 +6207,12 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const { for (unsigned i = 0; SavedRegs[i] != 0; ++i) { unsigned Reg = SavedRegs[i]; if (Subtarget->isThumb2() && - !ARM::tGPRRegisterClass->contains(Reg) && - !ARM::hGPRRegisterClass->contains(Reg)) + !ARM::tGPRRegClass.contains(Reg) && + !ARM::hGPRRegClass.contains(Reg)) continue; - else if (Subtarget->isThumb1Only() && - !ARM::tGPRRegisterClass->contains(Reg)) + if (Subtarget->isThumb1Only() && !ARM::tGPRRegClass.contains(Reg)) continue; - else if (!Subtarget->isThumb() && - !ARM::GPRRegisterClass->contains(Reg)) + if (!Subtarget->isThumb() && !ARM::GPRRegClass.contains(Reg)) continue; if (!DefRegs[Reg]) MIB.addReg(Reg, RegState::ImplicitDefine | RegState::Dead); @@ -6214,6 +6243,304 @@ MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) { llvm_unreachable("Expecting a BB with two successors!"); } +MachineBasicBlock *ARMTargetLowering:: +EmitStructByval(MachineInstr *MI, MachineBasicBlock *BB) const { + // This pseudo instruction has 3 operands: dst, src, size + // We expand it to a loop if size > Subtarget->getMaxInlineSizeThreshold(). + // Otherwise, we will generate unrolled scalar copies. + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = BB; + ++It; + + unsigned dest = MI->getOperand(0).getReg(); + unsigned src = MI->getOperand(1).getReg(); + unsigned SizeVal = MI->getOperand(2).getImm(); + unsigned Align = MI->getOperand(3).getImm(); + DebugLoc dl = MI->getDebugLoc(); + + bool isThumb2 = Subtarget->isThumb2(); + MachineFunction *MF = BB->getParent(); + MachineRegisterInfo &MRI = MF->getRegInfo(); + unsigned ldrOpc, strOpc, UnitSize = 0; + + const TargetRegisterClass *TRC = isThumb2 ? + (const TargetRegisterClass*)&ARM::tGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass; + const TargetRegisterClass *TRC_Vec = 0; + + if (Align & 1) { + ldrOpc = isThumb2 ? ARM::t2LDRB_POST : ARM::LDRB_POST_IMM; + strOpc = isThumb2 ? ARM::t2STRB_POST : ARM::STRB_POST_IMM; + UnitSize = 1; + } else if (Align & 2) { + ldrOpc = isThumb2 ? ARM::t2LDRH_POST : ARM::LDRH_POST; + strOpc = isThumb2 ? ARM::t2STRH_POST : ARM::STRH_POST; + UnitSize = 2; + } else { + // Check whether we can use NEON instructions. + if (!MF->getFunction()->hasFnAttr(Attribute::NoImplicitFloat) && + Subtarget->hasNEON()) { + if ((Align % 16 == 0) && SizeVal >= 16) { + ldrOpc = ARM::VLD1q32wb_fixed; + strOpc = ARM::VST1q32wb_fixed; + UnitSize = 16; + TRC_Vec = (const TargetRegisterClass*)&ARM::DPairRegClass; + } + else if ((Align % 8 == 0) && SizeVal >= 8) { + ldrOpc = ARM::VLD1d32wb_fixed; + strOpc = ARM::VST1d32wb_fixed; + UnitSize = 8; + TRC_Vec = (const TargetRegisterClass*)&ARM::DPRRegClass; + } + } + // Can't use NEON instructions. + if (UnitSize == 0) { + ldrOpc = isThumb2 ? ARM::t2LDR_POST : ARM::LDR_POST_IMM; + strOpc = isThumb2 ? ARM::t2STR_POST : ARM::STR_POST_IMM; + UnitSize = 4; + } + } + + unsigned BytesLeft = SizeVal % UnitSize; + unsigned LoopSize = SizeVal - BytesLeft; + + if (SizeVal <= Subtarget->getMaxInlineSizeThreshold()) { + // Use LDR and STR to copy. + // [scratch, srcOut] = LDR_POST(srcIn, UnitSize) + // [destOut] = STR_POST(scratch, destIn, UnitSize) + unsigned srcIn = src; + unsigned destIn = dest; + for (unsigned i = 0; i < LoopSize; i+=UnitSize) { + unsigned scratch = MRI.createVirtualRegister(UnitSize >= 8 ? TRC_Vec:TRC); + unsigned srcOut = MRI.createVirtualRegister(TRC); + unsigned destOut = MRI.createVirtualRegister(TRC); + if (UnitSize >= 8) { + AddDefaultPred(BuildMI(*BB, MI, dl, + TII->get(ldrOpc), scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(0)); + + AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(strOpc), destOut) + .addReg(destIn).addImm(0).addReg(scratch)); + } else if (isThumb2) { + AddDefaultPred(BuildMI(*BB, MI, dl, + TII->get(ldrOpc), scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(UnitSize)); + + AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addImm(UnitSize)); + } else { + AddDefaultPred(BuildMI(*BB, MI, dl, + TII->get(ldrOpc), scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addReg(0) + .addImm(UnitSize)); + + AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addReg(0).addImm(UnitSize)); + } + srcIn = srcOut; + destIn = destOut; + } + + // Handle the leftover bytes with LDRB and STRB. + // [scratch, srcOut] = LDRB_POST(srcIn, 1) + // [destOut] = STRB_POST(scratch, destIn, 1) + ldrOpc = isThumb2 ? ARM::t2LDRB_POST : ARM::LDRB_POST_IMM; + strOpc = isThumb2 ? ARM::t2STRB_POST : ARM::STRB_POST_IMM; + for (unsigned i = 0; i < BytesLeft; i++) { + unsigned scratch = MRI.createVirtualRegister(TRC); + unsigned srcOut = MRI.createVirtualRegister(TRC); + unsigned destOut = MRI.createVirtualRegister(TRC); + if (isThumb2) { + AddDefaultPred(BuildMI(*BB, MI, dl, + TII->get(ldrOpc),scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(1)); + + AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addReg(0).addImm(1)); + } else { + AddDefaultPred(BuildMI(*BB, MI, dl, + TII->get(ldrOpc),scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(1)); + + AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addReg(0).addImm(1)); + } + srcIn = srcOut; + destIn = destOut; + } + MI->eraseFromParent(); // The instruction is gone now. + return BB; + } + + // Expand the pseudo op to a loop. + // thisMBB: + // ... + // movw varEnd, # --> with thumb2 + // movt varEnd, # + // ldrcp varEnd, idx --> without thumb2 + // fallthrough --> loopMBB + // loopMBB: + // PHI varPhi, varEnd, varLoop + // PHI srcPhi, src, srcLoop + // PHI destPhi, dst, destLoop + // [scratch, srcLoop] = LDR_POST(srcPhi, UnitSize) + // [destLoop] = STR_POST(scratch, destPhi, UnitSize) + // subs varLoop, varPhi, #UnitSize + // bne loopMBB + // fallthrough --> exitMBB + // exitMBB: + // epilogue to handle left-over bytes + // [scratch, srcOut] = LDRB_POST(srcLoop, 1) + // [destOut] = STRB_POST(scratch, destLoop, 1) + MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB); + MF->insert(It, loopMBB); + MF->insert(It, exitMBB); + + // Transfer the remainder of BB and its successor edges to exitMBB. + exitMBB->splice(exitMBB->begin(), BB, + llvm::next(MachineBasicBlock::iterator(MI)), + BB->end()); + exitMBB->transferSuccessorsAndUpdatePHIs(BB); + + // Load an immediate to varEnd. + unsigned varEnd = MRI.createVirtualRegister(TRC); + if (isThumb2) { + unsigned VReg1 = varEnd; + if ((LoopSize & 0xFFFF0000) != 0) + VReg1 = MRI.createVirtualRegister(TRC); + AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVi16), VReg1) + .addImm(LoopSize & 0xFFFF)); + + if ((LoopSize & 0xFFFF0000) != 0) + AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVTi16), varEnd) + .addReg(VReg1) + .addImm(LoopSize >> 16)); + } else { + MachineConstantPool *ConstantPool = MF->getConstantPool(); + Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext()); + const Constant *C = ConstantInt::get(Int32Ty, LoopSize); + + // MachineConstantPool wants an explicit alignment. + unsigned Align = getTargetData()->getPrefTypeAlignment(Int32Ty); + if (Align == 0) + Align = getTargetData()->getTypeAllocSize(C->getType()); + unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align); + + AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::LDRcp)) + .addReg(varEnd, RegState::Define) + .addConstantPoolIndex(Idx) + .addImm(0)); + } + BB->addSuccessor(loopMBB); + + // Generate the loop body: + // varPhi = PHI(varLoop, varEnd) + // srcPhi = PHI(srcLoop, src) + // destPhi = PHI(destLoop, dst) + MachineBasicBlock *entryBB = BB; + BB = loopMBB; + unsigned varLoop = MRI.createVirtualRegister(TRC); + unsigned varPhi = MRI.createVirtualRegister(TRC); + unsigned srcLoop = MRI.createVirtualRegister(TRC); + unsigned srcPhi = MRI.createVirtualRegister(TRC); + unsigned destLoop = MRI.createVirtualRegister(TRC); + unsigned destPhi = MRI.createVirtualRegister(TRC); + + BuildMI(*BB, BB->begin(), dl, TII->get(ARM::PHI), varPhi) + .addReg(varLoop).addMBB(loopMBB) + .addReg(varEnd).addMBB(entryBB); + BuildMI(BB, dl, TII->get(ARM::PHI), srcPhi) + .addReg(srcLoop).addMBB(loopMBB) + .addReg(src).addMBB(entryBB); + BuildMI(BB, dl, TII->get(ARM::PHI), destPhi) + .addReg(destLoop).addMBB(loopMBB) + .addReg(dest).addMBB(entryBB); + + // [scratch, srcLoop] = LDR_POST(srcPhi, UnitSize) + // [destLoop] = STR_POST(scratch, destPhi, UnitSiz) + unsigned scratch = MRI.createVirtualRegister(UnitSize >= 8 ? TRC_Vec:TRC); + if (UnitSize >= 8) { + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), scratch) + .addReg(srcLoop, RegState::Define).addReg(srcPhi).addImm(0)); + + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), destLoop) + .addReg(destPhi).addImm(0).addReg(scratch)); + } else if (isThumb2) { + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), scratch) + .addReg(srcLoop, RegState::Define).addReg(srcPhi).addImm(UnitSize)); + + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), destLoop) + .addReg(scratch).addReg(destPhi) + .addImm(UnitSize)); + } else { + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), scratch) + .addReg(srcLoop, RegState::Define).addReg(srcPhi).addReg(0) + .addImm(UnitSize)); + + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), destLoop) + .addReg(scratch).addReg(destPhi) + .addReg(0).addImm(UnitSize)); + } + + // Decrement loop variable by UnitSize. + MachineInstrBuilder MIB = BuildMI(BB, dl, + TII->get(isThumb2 ? ARM::t2SUBri : ARM::SUBri), varLoop); + AddDefaultCC(AddDefaultPred(MIB.addReg(varPhi).addImm(UnitSize))); + MIB->getOperand(5).setReg(ARM::CPSR); + MIB->getOperand(5).setIsDef(true); + + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(loopMBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + + // loopMBB can loop back to loopMBB or fall through to exitMBB. + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // Add epilogue to handle BytesLeft. + BB = exitMBB; + MachineInstr *StartOfExit = exitMBB->begin(); + ldrOpc = isThumb2 ? ARM::t2LDRB_POST : ARM::LDRB_POST_IMM; + strOpc = isThumb2 ? ARM::t2STRB_POST : ARM::STRB_POST_IMM; + + // [scratch, srcOut] = LDRB_POST(srcLoop, 1) + // [destOut] = STRB_POST(scratch, destLoop, 1) + unsigned srcIn = srcLoop; + unsigned destIn = destLoop; + for (unsigned i = 0; i < BytesLeft; i++) { + unsigned scratch = MRI.createVirtualRegister(TRC); + unsigned srcOut = MRI.createVirtualRegister(TRC); + unsigned destOut = MRI.createVirtualRegister(TRC); + if (isThumb2) { + AddDefaultPred(BuildMI(*BB, StartOfExit, dl, + TII->get(ldrOpc),scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addImm(1)); + + AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addImm(1)); + } else { + AddDefaultPred(BuildMI(*BB, StartOfExit, dl, + TII->get(ldrOpc),scratch) + .addReg(srcOut, RegState::Define).addReg(srcIn).addReg(0).addImm(1)); + + AddDefaultPred(BuildMI(*BB, StartOfExit, dl, TII->get(strOpc), destOut) + .addReg(scratch).addReg(destIn) + .addReg(0).addImm(1)); + } + srcIn = srcOut; + destIn = destOut; + } + + MI->eraseFromParent(); // The instruction is gone now. + return BB; +} + MachineBasicBlock * ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *BB) const { @@ -6517,10 +6844,9 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, MachineRegisterInfo &MRI = Fn->getRegInfo(); // In Thumb mode S must not be specified if source register is the SP or // PC and if destination register is the SP, so restrict register class - unsigned NewMovDstReg = MRI.createVirtualRegister( - isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass); - unsigned NewRsbDstReg = MRI.createVirtualRegister( - isThumb2 ? ARM::rGPRRegisterClass : ARM::GPRRegisterClass); + unsigned NewRsbDstReg = MRI.createVirtualRegister(isThumb2 ? + (const TargetRegisterClass*)&ARM::rGPRRegClass : + (const TargetRegisterClass*)&ARM::GPRRegClass); // Transfer the remainder of BB and its successor edges to sinkMBB. SinkBB->splice(SinkBB->begin(), BB, @@ -6534,12 +6860,10 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, // fall through to SinkMBB RSBBB->addSuccessor(SinkBB); - // insert a movs at the end of BB - BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2MOVr : ARM::MOVr), - NewMovDstReg) - .addReg(ABSSrcReg, RegState::Kill) - .addImm((unsigned)ARMCC::AL).addReg(0) - .addReg(ARM::CPSR, RegState::Define); + // insert a cmp at the end of BB + AddDefaultPred(BuildMI(BB, dl, + TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) + .addReg(ABSSrcReg).addImm(0)); // insert a bcc with opposite CC to ARMCC::MI at the end of BB BuildMI(BB, dl, @@ -6551,7 +6875,7 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, // by if-conversion pass BuildMI(*RSBBB, RSBBB->begin(), dl, TII->get(isThumb2 ? ARM::t2RSBri : ARM::RSBri), NewRsbDstReg) - .addReg(NewMovDstReg, RegState::Kill) + .addReg(ABSSrcReg, RegState::Kill) .addImm(0).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0); // insert PHI in SinkBB, @@ -6559,7 +6883,7 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, BuildMI(*SinkBB, SinkBB->begin(), dl, TII->get(ARM::PHI), ABSDstReg) .addReg(NewRsbDstReg).addMBB(RSBBB) - .addReg(NewMovDstReg).addMBB(BB); + .addReg(ABSSrcReg).addMBB(BB); // remove ABS instruction MI->eraseFromParent(); @@ -6567,6 +6891,9 @@ ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, // return last added BB return SinkBB; } + case ARM::COPY_STRUCT_BYVAL_I32: + ++NumLoopByVals; + return EmitStructByval(MI, BB); } } @@ -7095,8 +7422,12 @@ static SDValue PerformORCombine(SDNode *N, return COR; } + + // The code below optimizes (or (and X, Y), Z). + // The AND operand needs to have a single user to make these optimizations + // profitable. SDValue N0 = N->getOperand(0); - if (N0.getOpcode() != ISD::AND) + if (N0.getOpcode() != ISD::AND || !N0.hasOneUse()) return SDValue(); SDValue N1 = N->getOperand(1); @@ -7353,7 +7684,7 @@ static SDValue PerformSTORECombine(SDNode *N, if (St->isVolatile()) return SDValue(); - // Optimize trunc store (of multiple scalars) to shuffle and store. First, + // Optimize trunc store (of multiple scalars) to shuffle and store. First, // pack all of the elements in one place. Next, store to memory in fewer // chunks. SDValue StVal = St->getValue(); @@ -8721,12 +9052,19 @@ bool ARMTargetLowering::isLegalICmpImmediate(int64_t Imm) const { return Imm >= 0 && Imm <= 255; } -/// isLegalAddImmediate - Return true if the specified immediate is legal -/// add immediate, that is the target has add instructions which can add -/// a register with the immediate without having to materialize the +/// isLegalAddImmediate - Return true if the specified immediate is a legal add +/// *or sub* immediate, that is the target has add or sub instructions which can +/// add a register with the immediate without having to materialize the /// immediate into a register. bool ARMTargetLowering::isLegalAddImmediate(int64_t Imm) const { - return ARM_AM::getSOImmVal(Imm) != -1; + // Same encoding for add/sub, just flip the sign. + int64_t AbsImm = llvm::abs64(Imm); + if (!Subtarget->isThumb()) + return ARM_AM::getSOImmVal(AbsImm) != -1; + if (Subtarget->isThumb2()) + return ARM_AM::getT2SOImmVal(AbsImm) != -1; + // Thumb1 only has 8-bit unsigned immediate. + return AbsImm >= 0 && AbsImm <= 255; } static bool getARMIndexedAddressParts(SDNode *Ptr, EVT VT, @@ -9030,39 +9368,38 @@ ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, switch (Constraint[0]) { case 'l': // Low regs or general regs. if (Subtarget->isThumb()) - return RCPair(0U, ARM::tGPRRegisterClass); - else - return RCPair(0U, ARM::GPRRegisterClass); + return RCPair(0U, &ARM::tGPRRegClass); + return RCPair(0U, &ARM::GPRRegClass); case 'h': // High regs or no regs. if (Subtarget->isThumb()) - return RCPair(0U, ARM::hGPRRegisterClass); + return RCPair(0U, &ARM::hGPRRegClass); break; case 'r': - return RCPair(0U, ARM::GPRRegisterClass); + return RCPair(0U, &ARM::GPRRegClass); case 'w': if (VT == MVT::f32) - return RCPair(0U, ARM::SPRRegisterClass); + return RCPair(0U, &ARM::SPRRegClass); if (VT.getSizeInBits() == 64) - return RCPair(0U, ARM::DPRRegisterClass); + return RCPair(0U, &ARM::DPRRegClass); if (VT.getSizeInBits() == 128) - return RCPair(0U, ARM::QPRRegisterClass); + return RCPair(0U, &ARM::QPRRegClass); break; case 'x': if (VT == MVT::f32) - return RCPair(0U, ARM::SPR_8RegisterClass); + return RCPair(0U, &ARM::SPR_8RegClass); if (VT.getSizeInBits() == 64) - return RCPair(0U, ARM::DPR_8RegisterClass); + return RCPair(0U, &ARM::DPR_8RegClass); if (VT.getSizeInBits() == 128) - return RCPair(0U, ARM::QPR_8RegisterClass); + return RCPair(0U, &ARM::QPR_8RegClass); break; case 't': if (VT == MVT::f32) - return RCPair(0U, ARM::SPRRegisterClass); + return RCPair(0U, &ARM::SPRRegClass); break; } } if (StringRef("{cc}").equals_lower(Constraint)) - return std::make_pair(unsigned(ARM::CPSR), ARM::CCRRegisterClass); + return std::make_pair(unsigned(ARM::CPSR), &ARM::CCRRegClass); return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT); } |
