diff options
Diffstat (limited to 'lib/Target/ARM')
26 files changed, 752 insertions, 169 deletions
diff --git a/lib/Target/ARM/ARM.h b/lib/Target/ARM/ARM.h index ff1980d..21445ad 100644 --- a/lib/Target/ARM/ARM.h +++ b/lib/Target/ARM/ARM.h @@ -109,7 +109,6 @@ FunctionPass *createNEONPreAllocPass(); FunctionPass *createNEONMoveFixPass(); FunctionPass *createThumb2ITBlockPass(); FunctionPass *createThumb2SizeReductionPass(); -FunctionPass *createARMMaxStackAlignmentCalculatorPass(); extern Target TheARMTarget, TheThumbTarget; diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp index c95d4c8..1aae369 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.cpp +++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp @@ -418,11 +418,13 @@ bool ARMBaseInstrInfo::isPredicable(MachineInstr *MI) const { return true; } -/// FIXME: Works around a gcc miscompilation with -fstrict-aliasing +/// FIXME: Works around a gcc miscompilation with -fstrict-aliasing. +DISABLE_INLINE static unsigned getNumJTEntries(const std::vector<MachineJumpTableEntry> &JT, - unsigned JTI) DISABLE_INLINE; + unsigned JTI); static unsigned getNumJTEntries(const std::vector<MachineJumpTableEntry> &JT, unsigned JTI) { + assert(JTI < JT.size()); return JT[JTI].MBBs.size(); } @@ -467,6 +469,8 @@ unsigned ARMBaseInstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const { return MI->getOperand(2).getImm(); case ARM::Int_eh_sjlj_setjmp: return 24; + case ARM::tInt_eh_sjlj_setjmp: + return 22; case ARM::t2Int_eh_sjlj_setjmp: return 22; case ARM::BR_JTr: @@ -755,7 +759,6 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, assert((RC == ARM::QPRRegisterClass || RC == ARM::QPR_VFP2RegisterClass || RC == ARM::QPR_8RegisterClass) && "Unknown regclass!"); - // FIXME: Neon instructions should support predicates if (Align >= 16 && (getRegisterInfo().needsStackRealignment(MF))) { AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLD1q64), DestReg) diff --git a/lib/Target/ARM/ARMBaseInstrInfo.h b/lib/Target/ARM/ARMBaseInstrInfo.h index 282e30c..78d9135 100644 --- a/lib/Target/ARM/ARMBaseInstrInfo.h +++ b/lib/Target/ARM/ARMBaseInstrInfo.h @@ -92,6 +92,8 @@ namespace ARMII { StMiscFrm = 9 << FormShift, LdStMulFrm = 10 << FormShift, + LdStExFrm = 28 << FormShift, + // Miscellaneous arithmetic instructions ArithMiscFrm = 11 << FormShift, @@ -190,9 +192,6 @@ public: // if there is not such an opcode. virtual unsigned getUnindexedOpcode(unsigned Opc) const =0; - // Return true if the block does not fall through. - virtual bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const =0; - virtual MachineInstr *convertToThreeAddress(MachineFunction::iterator &MFI, MachineBasicBlock::iterator &MBBI, LiveVariables *LV) const; diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp index 653328d..9b5f79f 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp +++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp @@ -471,21 +471,6 @@ ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg, } } -static unsigned calculateMaxStackAlignment(const MachineFrameInfo *FFI) { - unsigned MaxAlign = 0; - - for (int i = FFI->getObjectIndexBegin(), - e = FFI->getObjectIndexEnd(); i != e; ++i) { - if (FFI->isDeadObjectIndex(i)) - continue; - - unsigned Align = FFI->getObjectAlignment(i); - MaxAlign = std::max(MaxAlign, Align); - } - - return MaxAlign; -} - /// hasFP - Return true if the specified function should have a dedicated frame /// pointer register. This is true if the function has variable sized allocas /// or if frame pointer elimination is disabled. @@ -585,16 +570,21 @@ ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF, SmallVector<unsigned, 4> UnspilledCS2GPRs; ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>(); - MachineFrameInfo *MFI = MF.getFrameInfo(); // Calculate and set max stack object alignment early, so we can decide // whether we will need stack realignment (and thus FP). if (RealignStack) { - unsigned MaxAlign = std::max(MFI->getMaxAlignment(), - calculateMaxStackAlignment(MFI)); - MFI->setMaxAlignment(MaxAlign); + MachineFrameInfo *MFI = MF.getFrameInfo(); + MFI->calculateMaxStackAlignment(); } + // Spill R4 if Thumb2 function requires stack realignment - it will be used as + // scratch register. + // FIXME: It will be better just to find spare register here. + if (needsStackRealignment(MF) && + AFI->isThumb2Function()) + MF.getRegInfo().setPhysRegUsed(ARM::R4); + // Don't spill FP if the frame can be eliminated. This is determined // by scanning the callee-save registers to see if any is used. const unsigned *CSRegs = getCalleeSavedRegs(); @@ -1368,14 +1358,30 @@ emitPrologue(MachineFunction &MF) const { // If we need dynamic stack realignment, do it here. if (needsStackRealignment(MF)) { - unsigned Opc; unsigned MaxAlign = MFI->getMaxAlignment(); assert (!AFI->isThumb1OnlyFunction()); - Opc = AFI->isThumbFunction() ? ARM::t2BICri : ARM::BICri; - - AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(Opc), ARM::SP) + if (!AFI->isThumbFunction()) { + // Emit bic sp, sp, MaxAlign + AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, + TII.get(ARM::BICri), ARM::SP) .addReg(ARM::SP, RegState::Kill) .addImm(MaxAlign-1))); + } else { + // We cannot use sp as source/dest register here, thus we're emitting the + // following sequence: + // mov r4, sp + // bic r4, r4, MaxAlign + // mov sp, r4 + // FIXME: It will be better just to find spare register here. + BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::R4) + .addReg(ARM::SP, RegState::Kill); + AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, + TII.get(ARM::t2BICri), ARM::R4) + .addReg(ARM::R4, RegState::Kill) + .addImm(MaxAlign-1))); + BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP) + .addReg(ARM::R4, RegState::Kill); + } } } @@ -1479,48 +1485,4 @@ emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const { emitSPUpdate(isARM, MBB, MBBI, dl, TII, VARegSaveSize); } -namespace { - struct MaximalStackAlignmentCalculator : public MachineFunctionPass { - static char ID; - MaximalStackAlignmentCalculator() : MachineFunctionPass(&ID) {} - - virtual bool runOnMachineFunction(MachineFunction &MF) { - MachineFrameInfo *FFI = MF.getFrameInfo(); - MachineRegisterInfo &RI = MF.getRegInfo(); - - // Calculate max stack alignment of all already allocated stack objects. - unsigned MaxAlign = calculateMaxStackAlignment(FFI); - - // Be over-conservative: scan over all vreg defs and find, whether vector - // registers are used. If yes - there is probability, that vector register - // will be spilled and thus stack needs to be aligned properly. - for (unsigned RegNum = TargetRegisterInfo::FirstVirtualRegister; - RegNum < RI.getLastVirtReg(); ++RegNum) - MaxAlign = std::max(MaxAlign, RI.getRegClass(RegNum)->getAlignment()); - - if (FFI->getMaxAlignment() == MaxAlign) - return false; - - FFI->setMaxAlignment(MaxAlign); - return true; - } - - virtual const char *getPassName() const { - return "ARM Stack Required Alignment Auto-Detector"; - } - - virtual void getAnalysisUsage(AnalysisUsage &AU) const { - AU.setPreservesCFG(); - MachineFunctionPass::getAnalysisUsage(AU); - } - }; - - char MaximalStackAlignmentCalculator::ID = 0; -} - -FunctionPass* -llvm::createARMMaxStackAlignmentCalculatorPass() { - return new MaximalStackAlignmentCalculator(); -} - #include "ARMGenRegisterInfo.inc" diff --git a/lib/Target/ARM/ARMConstantIslandPass.cpp b/lib/Target/ARM/ARMConstantIslandPass.cpp index e59a315..acd30d2 100644 --- a/lib/Target/ARM/ARMConstantIslandPass.cpp +++ b/lib/Target/ARM/ARMConstantIslandPass.cpp @@ -418,10 +418,10 @@ void ARMConstantIslands::DoInitialPlacement(MachineFunction &MF, static bool BBHasFallthrough(MachineBasicBlock *MBB) { // Get the next machine basic block in the function. MachineFunction::iterator MBBI = MBB; - if (next(MBBI) == MBB->getParent()->end()) // Can't fall off end of function. + if (llvm::next(MBBI) == MBB->getParent()->end()) // Can't fall off end of function. return false; - MachineBasicBlock *NextBB = next(MBBI); + MachineBasicBlock *NextBB = llvm::next(MBBI); for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(), E = MBB->succ_end(); I != E; ++I) if (*I == NextBB) @@ -760,7 +760,7 @@ MachineBasicBlock *ARMConstantIslands::SplitBlockBeforeInstr(MachineInstr *MI) { CompareMBBNumbers); MachineBasicBlock* WaterBB = *IP; if (WaterBB == OrigBB) - WaterList.insert(next(IP), NewBB); + WaterList.insert(llvm::next(IP), NewBB); else WaterList.insert(IP, OrigBB); NewWaterList.insert(OrigBB); @@ -887,7 +887,7 @@ static bool BBIsJumpedOver(MachineBasicBlock *MBB) { void ARMConstantIslands::AdjustBBOffsetsAfter(MachineBasicBlock *BB, int delta) { - MachineFunction::iterator MBBI = BB; MBBI = next(MBBI); + MachineFunction::iterator MBBI = BB; MBBI = llvm::next(MBBI); for(unsigned i = BB->getNumber()+1, e = BB->getParent()->getNumBlockIDs(); i < e; ++i) { BBOffsets[i] += delta; @@ -929,7 +929,7 @@ void ARMConstantIslands::AdjustBBOffsetsAfter(MachineBasicBlock *BB, if (delta==0) return; } - MBBI = next(MBBI); + MBBI = llvm::next(MBBI); } } @@ -1096,7 +1096,7 @@ void ARMConstantIslands::CreateNewWater(unsigned CPUserIndex, DEBUG(errs() << "Split at end of block\n"); if (&UserMBB->back() == UserMI) assert(BBHasFallthrough(UserMBB) && "Expected a fallthrough BB!"); - NewMBB = next(MachineFunction::iterator(UserMBB)); + NewMBB = llvm::next(MachineFunction::iterator(UserMBB)); // Add an unconditional branch from UserMBB to fallthrough block. // Record it for branch lengthening; this new branch will not get out of // range, but if the preceding conditional branch is out of range, the @@ -1144,7 +1144,7 @@ void ARMConstantIslands::CreateNewWater(unsigned CPUserIndex, for (unsigned Offset = UserOffset+TII->GetInstSizeInBytes(UserMI); Offset < BaseInsertOffset; Offset += TII->GetInstSizeInBytes(MI), - MI = next(MI)) { + MI = llvm::next(MI)) { if (CPUIndex < CPUsers.size() && CPUsers[CPUIndex].MI == MI) { CPUser &U = CPUsers[CPUIndex]; if (!OffsetIsInRange(Offset, EndInsertOffset, @@ -1204,7 +1204,7 @@ bool ARMConstantIslands::HandleConstantPoolUser(MachineFunction &MF, NewWaterList.insert(NewIsland); } // The new CPE goes before the following block (NewMBB). - NewMBB = next(MachineFunction::iterator(WaterBB)); + NewMBB = llvm::next(MachineFunction::iterator(WaterBB)); } else { // No water found. @@ -1406,7 +1406,7 @@ ARMConstantIslands::FixUpConditionalBr(MachineFunction &MF, ImmBranch &Br) { NumCBrFixed++; if (BMI != MI) { - if (next(MachineBasicBlock::iterator(MI)) == prior(MBB->end()) && + if (llvm::next(MachineBasicBlock::iterator(MI)) == prior(MBB->end()) && BMI->getOpcode() == Br.UncondBr) { // Last MI in the BB is an unconditional branch. Can we simply invert the // condition and swap destinations: @@ -1433,12 +1433,12 @@ ARMConstantIslands::FixUpConditionalBr(MachineFunction &MF, ImmBranch &Br) { // branch to the destination. int delta = TII->GetInstSizeInBytes(&MBB->back()); BBSizes[MBB->getNumber()] -= delta; - MachineBasicBlock* SplitBB = next(MachineFunction::iterator(MBB)); + MachineBasicBlock* SplitBB = llvm::next(MachineFunction::iterator(MBB)); AdjustBBOffsetsAfter(SplitBB, -delta); MBB->back().eraseFromParent(); // BBOffsets[SplitBB] is wrong temporarily, fixed below } - MachineBasicBlock *NextBB = next(MachineFunction::iterator(MBB)); + MachineBasicBlock *NextBB = llvm::next(MachineFunction::iterator(MBB)); DEBUG(errs() << " Insert B to BB#" << DestBB->getNumber() << " also invert condition and change dest. to BB#" diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp index c929c54..1b8727d 100644 --- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp +++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp @@ -48,7 +48,7 @@ bool ARMExpandPseudo::ExpandMBB(MachineBasicBlock &MBB) { MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end(); while (MBBI != E) { MachineInstr &MI = *MBBI; - MachineBasicBlock::iterator NMBBI = next(MBBI); + MachineBasicBlock::iterator NMBBI = llvm::next(MBBI); unsigned Opcode = MI.getOpcode(); switch (Opcode) { diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp index c839fc6..655c762 100644 --- a/lib/Target/ARM/ARMISelLowering.cpp +++ b/lib/Target/ARM/ARMISelLowering.cpp @@ -42,6 +42,7 @@ #include "llvm/Support/CommandLine.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MathExtras.h" +#include "llvm/Support/raw_ostream.h" #include <sstream> using namespace llvm; @@ -377,7 +378,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM) setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom); else setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand); - setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand); + setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom); if (!Subtarget->hasV6Ops() && !Subtarget->isThumb2()) { setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand); @@ -500,6 +501,9 @@ const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const { case ARMISD::DYN_ALLOC: return "ARMISD::DYN_ALLOC"; + case ARMISD::MEMBARRIER: return "ARMISD::MEMBARRIER"; + case ARMISD::SYNCBARRIER: return "ARMISD::SYNCBARRIER"; + case ARMISD::VCEQ: return "ARMISD::VCEQ"; case ARMISD::VCGE: return "ARMISD::VCGE"; case ARMISD::VCGEU: return "ARMISD::VCGEU"; @@ -1470,6 +1474,28 @@ ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { } } +static SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG, + const ARMSubtarget *Subtarget) { + DebugLoc dl = Op.getDebugLoc(); + SDValue Op5 = Op.getOperand(5); + SDValue Res; + unsigned isDeviceBarrier = cast<ConstantSDNode>(Op5)->getZExtValue(); + if (isDeviceBarrier) { + if (Subtarget->hasV7Ops()) + Res = DAG.getNode(ARMISD::SYNCBARRIER, dl, MVT::Other, Op.getOperand(0)); + else + Res = DAG.getNode(ARMISD::SYNCBARRIER, dl, MVT::Other, Op.getOperand(0), + DAG.getConstant(0, MVT::i32)); + } else { + if (Subtarget->hasV7Ops()) + Res = DAG.getNode(ARMISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0)); + else + Res = DAG.getNode(ARMISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0), + DAG.getConstant(0, MVT::i32)); + } + return Res; +} + static SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG, unsigned VarArgsFrameIndex) { // vastart just stores the address of the VarArgsFrameIndex slot into the @@ -2528,6 +2554,25 @@ static bool isVTRNMask(const SmallVectorImpl<int> &M, EVT VT, return true; } +/// isVTRN_v_undef_Mask - Special case of isVTRNMask for canonical form of +/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef". +/// Mask is e.g., <0, 0, 2, 2> instead of <0, 4, 2, 6>. +static bool isVTRN_v_undef_Mask(const SmallVectorImpl<int> &M, EVT VT, + unsigned &WhichResult) { + unsigned EltSz = VT.getVectorElementType().getSizeInBits(); + if (EltSz == 64) + return false; + + unsigned NumElts = VT.getVectorNumElements(); + WhichResult = (M[0] == 0 ? 0 : 1); + for (unsigned i = 0; i < NumElts; i += 2) { + if ((unsigned) M[i] != i + WhichResult || + (unsigned) M[i+1] != i + WhichResult) + return false; + } + return true; +} + static bool isVUZPMask(const SmallVectorImpl<int> &M, EVT VT, unsigned &WhichResult) { unsigned EltSz = VT.getVectorElementType().getSizeInBits(); @@ -2548,6 +2593,33 @@ static bool isVUZPMask(const SmallVectorImpl<int> &M, EVT VT, return true; } +/// isVUZP_v_undef_Mask - Special case of isVUZPMask for canonical form of +/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef". +/// Mask is e.g., <0, 2, 0, 2> instead of <0, 2, 4, 6>, +static bool isVUZP_v_undef_Mask(const SmallVectorImpl<int> &M, EVT VT, + unsigned &WhichResult) { + unsigned EltSz = VT.getVectorElementType().getSizeInBits(); + if (EltSz == 64) + return false; + + unsigned Half = VT.getVectorNumElements() / 2; + WhichResult = (M[0] == 0 ? 0 : 1); + for (unsigned j = 0; j != 2; ++j) { + unsigned Idx = WhichResult; + for (unsigned i = 0; i != Half; ++i) { + if ((unsigned) M[i + j * Half] != Idx) + return false; + Idx += 2; + } + } + + // VUZP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32. + if (VT.is64BitVector() && EltSz == 32) + return false; + + return true; +} + static bool isVZIPMask(const SmallVectorImpl<int> &M, EVT VT, unsigned &WhichResult) { unsigned EltSz = VT.getVectorElementType().getSizeInBits(); @@ -2571,6 +2643,33 @@ static bool isVZIPMask(const SmallVectorImpl<int> &M, EVT VT, return true; } +/// isVZIP_v_undef_Mask - Special case of isVZIPMask for canonical form of +/// "vector_shuffle v, v", i.e., "vector_shuffle v, undef". +/// Mask is e.g., <0, 0, 1, 1> instead of <0, 4, 1, 5>. +static bool isVZIP_v_undef_Mask(const SmallVectorImpl<int> &M, EVT VT, + unsigned &WhichResult) { + unsigned EltSz = VT.getVectorElementType().getSizeInBits(); + if (EltSz == 64) + return false; + + unsigned NumElts = VT.getVectorNumElements(); + WhichResult = (M[0] == 0 ? 0 : 1); + unsigned Idx = WhichResult * NumElts / 2; + for (unsigned i = 0; i != NumElts; i += 2) { + if ((unsigned) M[i] != Idx || + (unsigned) M[i+1] != Idx) + return false; + Idx += 1; + } + + // VZIP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32. + if (VT.is64BitVector() && EltSz == 32) + return false; + + return true; +} + + static SDValue BuildSplat(SDValue Val, EVT VT, SelectionDAG &DAG, DebugLoc dl) { // Canonicalize all-zeros and all-ones vectors. ConstantSDNode *ConstVal = cast<ConstantSDNode>(Val.getNode()); @@ -2683,7 +2782,10 @@ ARMTargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M, isVEXTMask(M, VT, ReverseVEXT, Imm) || isVTRNMask(M, VT, WhichResult) || isVUZPMask(M, VT, WhichResult) || - isVZIPMask(M, VT, WhichResult)); + isVZIPMask(M, VT, WhichResult) || + isVTRN_v_undef_Mask(M, VT, WhichResult) || + isVUZP_v_undef_Mask(M, VT, WhichResult) || + isVZIP_v_undef_Mask(M, VT, WhichResult)); } /// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit @@ -2815,6 +2917,16 @@ static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) { return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT), V1, V2).getValue(WhichResult); + if (isVTRN_v_undef_Mask(ShuffleMask, VT, WhichResult)) + return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT), + V1, V1).getValue(WhichResult); + if (isVUZP_v_undef_Mask(ShuffleMask, VT, WhichResult)) + return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT), + V1, V1).getValue(WhichResult); + if (isVZIP_v_undef_Mask(ShuffleMask, VT, WhichResult)) + return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT), + V1, V1).getValue(WhichResult); + // If the shuffle is not directly supported and it has 4 elements, use // the PerfectShuffle-generated table to synthesize it from other shuffles. if (VT.getVectorNumElements() == 4 && @@ -2886,6 +2998,7 @@ SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) { case ISD::BR_JT: return LowerBR_JT(Op, DAG); case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG); case ISD::VASTART: return LowerVASTART(Op, DAG, VarArgsFrameIndex); + case ISD::MEMBARRIER: return LowerMEMBARRIER(Op, DAG, Subtarget); case ISD::SINT_TO_FP: case ISD::UINT_TO_FP: return LowerINT_TO_FP(Op, DAG); case ISD::FP_TO_SINT: @@ -2938,14 +3051,233 @@ void ARMTargetLowering::ReplaceNodeResults(SDNode *N, //===----------------------------------------------------------------------===// MachineBasicBlock * +ARMTargetLowering::EmitAtomicCmpSwap(MachineInstr *MI, + MachineBasicBlock *BB, + unsigned Size) const { + unsigned dest = MI->getOperand(0).getReg(); + unsigned ptr = MI->getOperand(1).getReg(); + unsigned oldval = MI->getOperand(2).getReg(); + unsigned newval = MI->getOperand(3).getReg(); + unsigned scratch = BB->getParent()->getRegInfo() + .createVirtualRegister(ARM::GPRRegisterClass); + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + DebugLoc dl = MI->getDebugLoc(); + bool isThumb2 = Subtarget->isThumb2(); + + unsigned ldrOpc, strOpc; + switch (Size) { + default: llvm_unreachable("unsupported size for AtomicCmpSwap!"); + case 1: + ldrOpc = isThumb2 ? ARM::t2LDREXB : ARM::LDREXB; + strOpc = isThumb2 ? ARM::t2LDREXB : ARM::STREXB; + break; + case 2: + ldrOpc = isThumb2 ? ARM::t2LDREXH : ARM::LDREXH; + strOpc = isThumb2 ? ARM::t2STREXH : ARM::STREXH; + break; + case 4: + ldrOpc = isThumb2 ? ARM::t2LDREX : ARM::LDREX; + strOpc = isThumb2 ? ARM::t2STREX : ARM::STREX; + break; + } + + MachineFunction *MF = BB->getParent(); + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction::iterator It = BB; + ++It; // insert the new blocks after the current block + + MachineBasicBlock *loop1MBB = MF->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *loop2MBB = MF->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB); + MF->insert(It, loop1MBB); + MF->insert(It, loop2MBB); + MF->insert(It, exitMBB); + exitMBB->transferSuccessors(BB); + + // thisMBB: + // ... + // fallthrough --> loop1MBB + BB->addSuccessor(loop1MBB); + + // loop1MBB: + // ldrex dest, [ptr] + // cmp dest, oldval + // bne exitMBB + BB = loop1MBB; + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr)); + AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr)) + .addReg(dest).addReg(oldval)); + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(exitMBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + BB->addSuccessor(loop2MBB); + BB->addSuccessor(exitMBB); + + // loop2MBB: + // strex scratch, newval, [ptr] + // cmp scratch, #0 + // bne loop1MBB + BB = loop2MBB; + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(newval) + .addReg(ptr)); + AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) + .addReg(scratch).addImm(0)); + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(loop1MBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + BB->addSuccessor(loop1MBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + return BB; +} + +MachineBasicBlock * +ARMTargetLowering::EmitAtomicBinary(MachineInstr *MI, MachineBasicBlock *BB, + unsigned Size, unsigned BinOpcode) const { + // This also handles ATOMIC_SWAP, indicated by BinOpcode==0. + const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); + + const BasicBlock *LLVM_BB = BB->getBasicBlock(); + MachineFunction *F = BB->getParent(); + MachineFunction::iterator It = BB; + ++It; + + unsigned dest = MI->getOperand(0).getReg(); + unsigned ptr = MI->getOperand(1).getReg(); + unsigned incr = MI->getOperand(2).getReg(); + DebugLoc dl = MI->getDebugLoc(); + bool isThumb2 = Subtarget->isThumb2(); + unsigned ldrOpc, strOpc; + switch (Size) { + default: llvm_unreachable("unsupported size for AtomicCmpSwap!"); + case 1: + ldrOpc = isThumb2 ? ARM::t2LDREXB : ARM::LDREXB; + strOpc = isThumb2 ? ARM::t2LDREXB : ARM::STREXB; + break; + case 2: + ldrOpc = isThumb2 ? ARM::t2LDREXH : ARM::LDREXH; + strOpc = isThumb2 ? ARM::t2STREXH : ARM::STREXH; + break; + case 4: + ldrOpc = isThumb2 ? ARM::t2LDREX : ARM::LDREX; + strOpc = isThumb2 ? ARM::t2STREX : ARM::STREX; + break; + } + + MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB); + MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB); + F->insert(It, loopMBB); + F->insert(It, exitMBB); + exitMBB->transferSuccessors(BB); + + MachineRegisterInfo &RegInfo = F->getRegInfo(); + unsigned scratch = RegInfo.createVirtualRegister(ARM::GPRRegisterClass); + unsigned scratch2 = (!BinOpcode) ? incr : + RegInfo.createVirtualRegister(ARM::GPRRegisterClass); + + // thisMBB: + // ... + // fallthrough --> loopMBB + BB->addSuccessor(loopMBB); + + // loopMBB: + // ldrex dest, ptr + // <binop> scratch2, dest, incr + // strex scratch, scratch2, ptr + // cmp scratch, #0 + // bne- loopMBB + // fallthrough --> exitMBB + BB = loopMBB; + AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr)); + if (BinOpcode) { + // operand order needs to go the other way for NAND + if (BinOpcode == ARM::BICrr || BinOpcode == ARM::t2BICrr) + AddDefaultPred(BuildMI(BB, dl, TII->get(BinOpcode), scratch2). + addReg(incr).addReg(dest)).addReg(0); + else + AddDefaultPred(BuildMI(BB, dl, TII->get(BinOpcode), scratch2). + addReg(dest).addReg(incr)).addReg(0); + } + + AddDefaultPred(BuildMI(BB, dl, TII->get(strOpc), scratch).addReg(scratch2) + .addReg(ptr)); + AddDefaultPred(BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri)) + .addReg(scratch).addImm(0)); + BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)) + .addMBB(loopMBB).addImm(ARMCC::NE).addReg(ARM::CPSR); + + BB->addSuccessor(loopMBB); + BB->addSuccessor(exitMBB); + + // exitMBB: + // ... + BB = exitMBB; + return BB; +} + +MachineBasicBlock * ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, MachineBasicBlock *BB, DenseMap<MachineBasicBlock*, MachineBasicBlock*> *EM) const { const TargetInstrInfo *TII = getTargetMachine().getInstrInfo(); DebugLoc dl = MI->getDebugLoc(); + bool isThumb2 = Subtarget->isThumb2(); switch (MI->getOpcode()) { default: + MI->dump(); llvm_unreachable("Unexpected instr type to insert"); + + case ARM::ATOMIC_LOAD_ADD_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr); + case ARM::ATOMIC_LOAD_ADD_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr); + case ARM::ATOMIC_LOAD_ADD_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2ADDrr : ARM::ADDrr); + + case ARM::ATOMIC_LOAD_AND_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2ANDrr : ARM::ANDrr); + case ARM::ATOMIC_LOAD_AND_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2ANDrr : ARM::ANDrr); + case ARM::ATOMIC_LOAD_AND_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2ANDrr : ARM::ANDrr); + + case ARM::ATOMIC_LOAD_OR_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2ORRrr : ARM::ORRrr); + case ARM::ATOMIC_LOAD_OR_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2ORRrr : ARM::ORRrr); + case ARM::ATOMIC_LOAD_OR_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2ORRrr : ARM::ORRrr); + + case ARM::ATOMIC_LOAD_XOR_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2EORrr : ARM::EORrr); + case ARM::ATOMIC_LOAD_XOR_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2EORrr : ARM::EORrr); + case ARM::ATOMIC_LOAD_XOR_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2EORrr : ARM::EORrr); + + case ARM::ATOMIC_LOAD_NAND_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2BICrr : ARM::BICrr); + case ARM::ATOMIC_LOAD_NAND_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2BICrr : ARM::BICrr); + case ARM::ATOMIC_LOAD_NAND_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2BICrr : ARM::BICrr); + + case ARM::ATOMIC_LOAD_SUB_I8: + return EmitAtomicBinary(MI, BB, 1, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr); + case ARM::ATOMIC_LOAD_SUB_I16: + return EmitAtomicBinary(MI, BB, 2, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr); + case ARM::ATOMIC_LOAD_SUB_I32: + return EmitAtomicBinary(MI, BB, 4, isThumb2 ? ARM::t2SUBrr : ARM::SUBrr); + + case ARM::ATOMIC_SWAP_I8: return EmitAtomicBinary(MI, BB, 1, 0); + case ARM::ATOMIC_SWAP_I16: return EmitAtomicBinary(MI, BB, 2, 0); + case ARM::ATOMIC_SWAP_I32: return EmitAtomicBinary(MI, BB, 4, 0); + + case ARM::ATOMIC_CMP_SWAP_I8: return EmitAtomicCmpSwap(MI, BB, 1); + case ARM::ATOMIC_CMP_SWAP_I16: return EmitAtomicCmpSwap(MI, BB, 2); + case ARM::ATOMIC_CMP_SWAP_I32: return EmitAtomicCmpSwap(MI, BB, 4); + case ARM::tMOVCCr_pseudo: { // To "insert" a SELECT_CC instruction, we actually have to insert the // diamond control-flow pattern. The incoming instruction knows the @@ -3935,6 +4267,8 @@ ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint, return std::make_pair(0U, ARM::SPRRegisterClass); if (VT == MVT::f64) return std::make_pair(0U, ARM::DPRRegisterClass); + if (VT.getSizeInBits() == 128) + return std::make_pair(0U, ARM::QPRRegisterClass); break; } } @@ -3973,6 +4307,9 @@ getRegClassForInlineAsmConstraint(const std::string &Constraint, ARM::D4, ARM::D5, ARM::D6, ARM::D7, ARM::D8, ARM::D9, ARM::D10,ARM::D11, ARM::D12,ARM::D13,ARM::D14,ARM::D15, 0); + if (VT.getSizeInBits() == 128) + return make_vector<unsigned>(ARM::Q0, ARM::Q1, ARM::Q2, ARM::Q3, + ARM::Q4, ARM::Q5, ARM::Q6, ARM::Q7, 0); break; } diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h index 4f31f8a..e1b3348 100644 --- a/lib/Target/ARM/ARMISelLowering.h +++ b/lib/Target/ARM/ARMISelLowering.h @@ -72,6 +72,9 @@ namespace llvm { DYN_ALLOC, // Dynamic allocation on the stack. + MEMBARRIER, // Memory barrier + SYNCBARRIER, // Memory sync barrier + VCEQ, // Vector compare equal. VCGE, // Vector compare greater than or equal. VCGEU, // Vector compare unsigned greater than or equal. @@ -328,6 +331,15 @@ namespace llvm { SDValue getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC, SDValue &ARMCC, SelectionDAG &DAG, DebugLoc dl); + + MachineBasicBlock *EmitAtomicCmpSwap(MachineInstr *MI, + MachineBasicBlock *BB, + unsigned Size) const; + MachineBasicBlock *EmitAtomicBinary(MachineInstr *MI, + MachineBasicBlock *BB, + unsigned Size, + unsigned BinOpcode) const; + }; } diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td index e76e93c..cf0edff 100644 --- a/lib/Target/ARM/ARMInstrFormats.td +++ b/lib/Target/ARM/ARMInstrFormats.td @@ -33,6 +33,8 @@ def LdMiscFrm : Format<8>; def StMiscFrm : Format<9>; def LdStMulFrm : Format<10>; +def LdStExFrm : Format<28>; + def ArithMiscFrm : Format<11>; def ExtFrm : Format<12>; @@ -199,6 +201,19 @@ class I<dag oops, dag iops, AddrMode am, SizeFlagVal sz, let Pattern = pattern; list<Predicate> Predicates = [IsARM]; } +// A few are not predicable +class InoP<dag oops, dag iops, AddrMode am, SizeFlagVal sz, + IndexMode im, Format f, InstrItinClass itin, + string opc, string asm, string cstr, + list<dag> pattern> + : InstARM<am, sz, im, f, GenericDomain, cstr, itin> { + let OutOperandList = oops; + let InOperandList = iops; + let AsmString = !strconcat(opc, asm); + let Pattern = pattern; + let isPredicable = 0; + list<Predicate> Predicates = [IsARM]; +} // Same as I except it can optionally modify CPSR. Note it's modeled as // an input operand since by default it's a zero register. It will @@ -239,6 +254,10 @@ class AXI<dag oops, dag iops, Format f, InstrItinClass itin, string asm, list<dag> pattern> : XI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, asm, "", pattern>; +class AInoP<dag oops, dag iops, Format f, InstrItinClass itin, + string opc, string asm, list<dag> pattern> + : InoP<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin, + opc, asm, "", pattern>; // Ctrl flow instructions class ABI<bits<4> opcod, dag oops, dag iops, InstrItinClass itin, @@ -264,6 +283,28 @@ class JTI<dag oops, dag iops, InstrItinClass itin, : XI<oops, iops, AddrModeNone, SizeSpecial, IndexModeNone, BrMiscFrm, itin, asm, "", pattern>; + +// Atomic load/store instructions + +class AIldrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin, + string opc, string asm, list<dag> pattern> + : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin, + opc, asm, "", pattern> { + let Inst{27-23} = 0b00011; + let Inst{22-21} = opcod; + let Inst{20} = 1; + let Inst{11-0} = 0b111110011111; +} +class AIstrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin, + string opc, string asm, list<dag> pattern> + : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin, + opc, asm, "", pattern> { + let Inst{27-23} = 0b00011; + let Inst{22-21} = opcod; + let Inst{20} = 0; + let Inst{11-4} = 0b11111001; +} + // addrmode1 instructions class AI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin, string opc, string asm, list<dag> pattern> @@ -967,6 +1008,17 @@ class Thumb2XI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, list<Predicate> Predicates = [IsThumb2]; } +class ThumbXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz, + InstrItinClass itin, + string asm, string cstr, list<dag> pattern> + : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> { + let OutOperandList = oops; + let InOperandList = iops; + let AsmString = asm; + let Pattern = pattern; + list<Predicate> Predicates = [IsThumb1Only]; +} + class T2I<dag oops, dag iops, InstrItinClass itin, string opc, string asm, list<dag> pattern> : Thumb2I<oops, iops, AddrModeNone, Size4Bytes, itin, opc, asm, "", pattern>; diff --git a/lib/Target/ARM/ARMInstrInfo.cpp b/lib/Target/ARM/ARMInstrInfo.cpp index 87bb12b..85f6b40 100644 --- a/lib/Target/ARM/ARMInstrInfo.cpp +++ b/lib/Target/ARM/ARMInstrInfo.cpp @@ -60,25 +60,6 @@ unsigned ARMInstrInfo::getUnindexedOpcode(unsigned Opc) const { return 0; } -bool ARMInstrInfo::BlockHasNoFallThrough(const MachineBasicBlock &MBB) const { - if (MBB.empty()) return false; - - switch (MBB.back().getOpcode()) { - case ARM::BX_RET: // Return. - case ARM::LDM_RET: - case ARM::B: - case ARM::BRIND: - case ARM::BR_JTr: // Jumptable branch. - case ARM::BR_JTm: // Jumptable branch through mem. - case ARM::BR_JTadd: // Jumptable branch add to pc. - return true; - default: - break; - } - - return false; -} - void ARMInstrInfo:: reMaterialize(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, unsigned DestReg, unsigned SubIdx, const MachineInstr *Orig, diff --git a/lib/Target/ARM/ARMInstrInfo.h b/lib/Target/ARM/ARMInstrInfo.h index 4319577..d4199d1 100644 --- a/lib/Target/ARM/ARMInstrInfo.h +++ b/lib/Target/ARM/ARMInstrInfo.h @@ -32,9 +32,6 @@ public: // if there is not such an opcode. unsigned getUnindexedOpcode(unsigned Opc) const; - // Return true if the block does not fall through. - bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const; - void reMaterialize(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, unsigned DestReg, unsigned SubIdx, const MachineInstr *Orig, diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td index 7516d3c..e14696a 100644 --- a/lib/Target/ARM/ARMInstrInfo.td +++ b/lib/Target/ARM/ARMInstrInfo.td @@ -46,6 +46,11 @@ def SDT_ARMPICAdd : SDTypeProfile<1, 2, [SDTCisSameAs<0, 1>, def SDT_ARMThreadPointer : SDTypeProfile<1, 0, [SDTCisPtrTy<0>]>; def SDT_ARMEH_SJLJ_Setjmp : SDTypeProfile<1, 1, [SDTCisInt<0>, SDTCisPtrTy<1>]>; +def SDT_ARMMEMBARRIERV7 : SDTypeProfile<0, 0, []>; +def SDT_ARMSYNCBARRIERV7 : SDTypeProfile<0, 0, []>; +def SDT_ARMMEMBARRIERV6 : SDTypeProfile<0, 1, [SDTCisInt<0>]>; +def SDT_ARMSYNCBARRIERV6 : SDTypeProfile<0, 1, [SDTCisInt<0>]>; + // Node definitions. def ARMWrapper : SDNode<"ARMISD::Wrapper", SDTIntUnaryOp>; def ARMWrapperJT : SDNode<"ARMISD::WrapperJT", SDTIntBinOp>; @@ -93,6 +98,15 @@ def ARMrrx : SDNode<"ARMISD::RRX" , SDTIntUnaryOp, [SDNPInFlag ]>; def ARMthread_pointer: SDNode<"ARMISD::THREAD_POINTER", SDT_ARMThreadPointer>; def ARMeh_sjlj_setjmp: SDNode<"ARMISD::EH_SJLJ_SETJMP", SDT_ARMEH_SJLJ_Setjmp>; +def ARMMemBarrierV7 : SDNode<"ARMISD::MEMBARRIER", SDT_ARMMEMBARRIERV7, + [SDNPHasChain]>; +def ARMSyncBarrierV7 : SDNode<"ARMISD::SYNCBARRIER", SDT_ARMMEMBARRIERV7, + [SDNPHasChain]>; +def ARMMemBarrierV6 : SDNode<"ARMISD::MEMBARRIER", SDT_ARMMEMBARRIERV6, + [SDNPHasChain]>; +def ARMSyncBarrierV6 : SDNode<"ARMISD::SYNCBARRIER", SDT_ARMMEMBARRIERV6, + [SDNPHasChain]>; + //===----------------------------------------------------------------------===// // ARM Instruction Predicate Definitions. // @@ -772,6 +786,7 @@ let isBranch = 1, isTerminator = 1 in { def BR_JTr : JTI<(outs), (ins GPR:$target, jtblock_operand:$jt, i32imm:$id), IIC_Br, "mov\tpc, $target \n$jt", [(ARMbrjt GPR:$target, tjumptable:$jt, imm:$id)]> { + let Inst{11-4} = 0b00000000; let Inst{15-12} = 0b1111; let Inst{20} = 0; // S Bit let Inst{24-21} = 0b1101; @@ -1561,6 +1576,189 @@ def MOVCCi : AI1<0b1101, (outs GPR:$dst), let Inst{25} = 1; } +//===----------------------------------------------------------------------===// +// Atomic operations intrinsics +// + +// memory barriers protect the atomic sequences +let hasSideEffects = 1 in { +def Int_MemBarrierV7 : AInoP<(outs), (ins), + Pseudo, NoItinerary, + "dmb", "", + [(ARMMemBarrierV7)]>, + Requires<[IsARM, HasV7]> { + let Inst{31-4} = 0xf57ff05; + // FIXME: add support for options other than a full system DMB + let Inst{3-0} = 0b1111; +} + +def Int_SyncBarrierV7 : AInoP<(outs), (ins), + Pseudo, NoItinerary, + "dsb", "", + [(ARMSyncBarrierV7)]>, + Requires<[IsARM, HasV7]> { + let Inst{31-4} = 0xf57ff04; + // FIXME: add support for options other than a full system DSB + let Inst{3-0} = 0b1111; +} + +def Int_MemBarrierV6 : AInoP<(outs), (ins GPR:$zero), + Pseudo, NoItinerary, + "mcr", "\tp15, 0, $zero, c7, c10, 5", + [(ARMMemBarrierV6 GPR:$zero)]>, + Requires<[IsARM, HasV6]> { + // FIXME: add support for options other than a full system DMB + // FIXME: add encoding +} + +def Int_SyncBarrierV6 : AInoP<(outs), (ins GPR:$zero), + Pseudo, NoItinerary, + "mcr", "\tp15, 0, $zero, c7, c10, 4", + [(ARMSyncBarrierV6 GPR:$zero)]>, + Requires<[IsARM, HasV6]> { + // FIXME: add support for options other than a full system DSB + // FIXME: add encoding +} +} + +let usesCustomInserter = 1 in { + let Uses = [CPSR] in { + def ATOMIC_LOAD_ADD_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_ADD_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_add_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_SUB_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_SUB_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_sub_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_AND_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_AND_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_and_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_OR_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_OR_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_or_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_XOR_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_XOR_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_xor_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_NAND_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_NAND_I8 PSEUDO!", + [(set GPR:$dst, (atomic_load_nand_8 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_ADD_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_ADD_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_add_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_SUB_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_SUB_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_sub_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_AND_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_AND_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_and_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_OR_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_OR_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_or_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_XOR_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_XOR_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_xor_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_NAND_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_NAND_I16 PSEUDO!", + [(set GPR:$dst, (atomic_load_nand_16 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_ADD_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_ADD_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_add_32 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_SUB_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_SUB_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_sub_32 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_AND_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_AND_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_and_32 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_OR_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_OR_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_or_32 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_XOR_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_XOR_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_xor_32 GPR:$ptr, GPR:$incr))]>; + def ATOMIC_LOAD_NAND_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$incr), NoItinerary, + "${:comment} ATOMIC_LOAD_NAND_I32 PSEUDO!", + [(set GPR:$dst, (atomic_load_nand_32 GPR:$ptr, GPR:$incr))]>; + + def ATOMIC_SWAP_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$new), NoItinerary, + "${:comment} ATOMIC_SWAP_I8 PSEUDO!", + [(set GPR:$dst, (atomic_swap_8 GPR:$ptr, GPR:$new))]>; + def ATOMIC_SWAP_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$new), NoItinerary, + "${:comment} ATOMIC_SWAP_I16 PSEUDO!", + [(set GPR:$dst, (atomic_swap_16 GPR:$ptr, GPR:$new))]>; + def ATOMIC_SWAP_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$new), NoItinerary, + "${:comment} ATOMIC_SWAP_I32 PSEUDO!", + [(set GPR:$dst, (atomic_swap_32 GPR:$ptr, GPR:$new))]>; + + def ATOMIC_CMP_SWAP_I8 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$old, GPR:$new), NoItinerary, + "${:comment} ATOMIC_CMP_SWAP_I8 PSEUDO!", + [(set GPR:$dst, (atomic_cmp_swap_8 GPR:$ptr, GPR:$old, GPR:$new))]>; + def ATOMIC_CMP_SWAP_I16 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$old, GPR:$new), NoItinerary, + "${:comment} ATOMIC_CMP_SWAP_I16 PSEUDO!", + [(set GPR:$dst, (atomic_cmp_swap_16 GPR:$ptr, GPR:$old, GPR:$new))]>; + def ATOMIC_CMP_SWAP_I32 : PseudoInst< + (outs GPR:$dst), (ins GPR:$ptr, GPR:$old, GPR:$new), NoItinerary, + "${:comment} ATOMIC_CMP_SWAP_I32 PSEUDO!", + [(set GPR:$dst, (atomic_cmp_swap_32 GPR:$ptr, GPR:$old, GPR:$new))]>; +} +} + +let mayLoad = 1 in { +def LDREXB : AIldrex<0b10, (outs GPR:$dest), (ins GPR:$ptr), NoItinerary, + "ldrexb", "\t$dest, [$ptr]", + []>; +def LDREXH : AIldrex<0b11, (outs GPR:$dest), (ins GPR:$ptr), NoItinerary, + "ldrexh", "\t$dest, [$ptr]", + []>; +def LDREX : AIldrex<0b00, (outs GPR:$dest), (ins GPR:$ptr), NoItinerary, + "ldrex", "\t$dest, [$ptr]", + []>; +def LDREXD : AIldrex<0b01, (outs GPR:$dest, GPR:$dest2), (ins GPR:$ptr), + NoItinerary, + "ldrexd", "\t$dest, $dest2, [$ptr]", + []>; +} + +let mayStore = 1 in { +def STREXB : AIstrex<0b10, (outs GPR:$success), (ins GPR:$src, GPR:$ptr), + NoItinerary, + "strexb", "\t$success, $src, [$ptr]", + []>; +def STREXH : AIstrex<0b11, (outs GPR:$success), (ins GPR:$src, GPR:$ptr), + NoItinerary, + "strexh", "\t$success, $src, [$ptr]", + []>; +def STREX : AIstrex<0b00, (outs GPR:$success), (ins GPR:$src, GPR:$ptr), + NoItinerary, + "strex", "\t$success, $src, [$ptr]", + []>; +def STREXD : AIstrex<0b01, (outs GPR:$success), + (ins GPR:$src, GPR:$src2, GPR:$ptr), + NoItinerary, + "strexd", "\t$success, $src, $src2, [$ptr]", + []>; +} //===----------------------------------------------------------------------===// // TLS Instructions diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td index 3166931..61b7705 100644 --- a/lib/Target/ARM/ARMInstrNEON.td +++ b/lib/Target/ARM/ARMInstrNEON.td @@ -152,7 +152,7 @@ def VLDRQ : NI4<(outs QPR:$dst), (ins addrmode4:$addr), let Inst{24} = 0; // P bit let Inst{23} = 1; // U bit let Inst{20} = 1; - let Inst{11-9} = 0b101; + let Inst{11-8} = 0b1011; } // Use vstmia to store a Q register as a D register pair. @@ -164,7 +164,7 @@ def VSTRQ : NI4<(outs), (ins QPR:$src, addrmode4:$addr), let Inst{24} = 0; // P bit let Inst{23} = 1; // U bit let Inst{20} = 0; - let Inst{11-9} = 0b101; + let Inst{11-8} = 0b1011; } // VLD1 : Vector Load (multiple single elements) diff --git a/lib/Target/ARM/ARMInstrThumb.td b/lib/Target/ARM/ARMInstrThumb.td index b5956a3..9306bdb 100644 --- a/lib/Target/ARM/ARMInstrThumb.td +++ b/lib/Target/ARM/ARMInstrThumb.td @@ -669,6 +669,35 @@ let isCall = 1, [(set R0, ARMthread_pointer)]>; } +// SJLJ Exception handling intrinsics +// eh_sjlj_setjmp() is an instruction sequence to store the return +// address and save #0 in R0 for the non-longjmp case. +// Since by its nature we may be coming from some other function to get +// here, and we're using the stack frame for the containing function to +// save/restore registers, we can't keep anything live in regs across +// the eh_sjlj_setjmp(), else it will almost certainly have been tromped upon +// when we get here from a longjmp(). We force everthing out of registers +// except for our own input by listing the relevant registers in Defs. By +// doing so, we also cause the prologue/epilogue code to actively preserve +// all of the callee-saved resgisters, which is exactly what we want. +let Defs = + [ R0, R1, R2, R3, R4, R5, R6, R7, R12 ] in { + def tInt_eh_sjlj_setjmp : ThumbXI<(outs), (ins GPR:$src), + AddrModeNone, SizeSpecial, NoItinerary, + "mov\tr12, r1\t@ begin eh.setjmp\n" + "\tmov\tr1, sp\n" + "\tstr\tr1, [$src, #8]\n" + "\tadr\tr1, 0f\n" + "\tadds\tr1, #1\n" + "\tstr\tr1, [$src, #4]\n" + "\tmov\tr1, r12\n" + "\tmovs\tr0, #0\n" + "\tb\t1f\n" + ".align 2\n" + "0:\tmovs\tr0, #1\t@ end eh.setjmp\n" + "1:", "", + [(set R0, (ARMeh_sjlj_setjmp GPR:$src))]>; +} //===----------------------------------------------------------------------===// // Non-Instruction Patterns // diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td index 9489815..949ce73 100644 --- a/lib/Target/ARM/ARMInstrThumb2.td +++ b/lib/Target/ARM/ARMInstrThumb2.td @@ -1065,6 +1065,68 @@ def t2MOVCCror : T2I<(outs GPR:$dst), (ins GPR:$false, GPR:$true, i32imm:$rhs), RegConstraint<"$false = $dst">; //===----------------------------------------------------------------------===// +// Atomic operations intrinsics +// + +// memory barriers protect the atomic sequences +let hasSideEffects = 1 in { +def t2Int_MemBarrierV7 : AInoP<(outs), (ins), + Pseudo, NoItinerary, + "dmb", "", + [(ARMMemBarrierV7)]>, + Requires<[IsThumb2]> { + // FIXME: add support for options other than a full system DMB +} + +def t2Int_SyncBarrierV7 : AInoP<(outs), (ins), + Pseudo, NoItinerary, + "dsb", "", + [(ARMSyncBarrierV7)]>, + Requires<[IsThumb2]> { + // FIXME: add support for options other than a full system DSB +} +} + +let mayLoad = 1 in { +def t2LDREXB : Thumb2I<(outs GPR:$dest), (ins GPR:$ptr), AddrModeNone, + Size4Bytes, NoItinerary, + "ldrexb", "\t$dest, [$ptr]", "", + []>; +def t2LDREXH : Thumb2I<(outs GPR:$dest), (ins GPR:$ptr), AddrModeNone, + Size4Bytes, NoItinerary, + "ldrexh", "\t$dest, [$ptr]", "", + []>; +def t2LDREX : Thumb2I<(outs GPR:$dest), (ins GPR:$ptr), AddrModeNone, + Size4Bytes, NoItinerary, + "ldrex", "\t$dest, [$ptr]", "", + []>; +def t2LDREXD : Thumb2I<(outs GPR:$dest, GPR:$dest2), (ins GPR:$ptr), + AddrModeNone, Size4Bytes, NoItinerary, + "ldrexd", "\t$dest, $dest2, [$ptr]", "", + []>; +} + +let mayStore = 1 in { +def t2STREXB : Thumb2I<(outs GPR:$success), (ins GPR:$src, GPR:$ptr), + AddrModeNone, Size4Bytes, NoItinerary, + "strexb", "\t$success, $src, [$ptr]", "", + []>; +def t2STREXH : Thumb2I<(outs GPR:$success), (ins GPR:$src, GPR:$ptr), + AddrModeNone, Size4Bytes, NoItinerary, + "strexh", "\t$success, $src, [$ptr]", "", + []>; +def t2STREX : Thumb2I<(outs GPR:$success), (ins GPR:$src, GPR:$ptr), + AddrModeNone, Size4Bytes, NoItinerary, + "strex", "\t$success, $src, [$ptr]", "", + []>; +def t2STREXD : Thumb2I<(outs GPR:$success), + (ins GPR:$src, GPR:$src2, GPR:$ptr), + AddrModeNone, Size4Bytes, NoItinerary, + "strexd", "\t$success, $src, $src2, [$ptr]", "", + []>; +} + +//===----------------------------------------------------------------------===// // TLS Instructions // diff --git a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp index 304d0ef..22bd80e 100644 --- a/lib/Target/ARM/ARMLoadStoreOptimizer.cpp +++ b/lib/Target/ARM/ARMLoadStoreOptimizer.cpp @@ -449,7 +449,7 @@ bool ARMLoadStoreOpt::MergeBaseUpdateLSMultiple(MachineBasicBlock &MBB, } if (MBBI != MBB.end()) { - MachineBasicBlock::iterator NextMBBI = next(MBBI); + MachineBasicBlock::iterator NextMBBI = llvm::next(MBBI); if ((Mode == ARM_AM::ia || Mode == ARM_AM::ib) && isMatchingIncrement(NextMBBI, Base, Bytes, 0, Pred, PredReg)) { MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(Mode, true)); @@ -494,7 +494,7 @@ bool ARMLoadStoreOpt::MergeBaseUpdateLSMultiple(MachineBasicBlock &MBB, } if (MBBI != MBB.end()) { - MachineBasicBlock::iterator NextMBBI = next(MBBI); + MachineBasicBlock::iterator NextMBBI = llvm::next(MBBI); if (Mode == ARM_AM::ia && isMatchingIncrement(NextMBBI, Base, Bytes, 0, Pred, PredReg)) { MI->getOperand(1).setImm(ARM_AM::getAM5Opc(ARM_AM::ia, true, Offset)); @@ -604,7 +604,7 @@ bool ARMLoadStoreOpt::MergeBaseUpdateLoadStore(MachineBasicBlock &MBB, } if (!DoMerge && MBBI != MBB.end()) { - MachineBasicBlock::iterator NextMBBI = next(MBBI); + MachineBasicBlock::iterator NextMBBI = llvm::next(MBBI); if (!isAM5 && isMatchingDecrement(NextMBBI, Base, Bytes, Limit, Pred, PredReg)) { DoMerge = true; diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp index 2564ed9..1c6fca7 100644 --- a/lib/Target/ARM/ARMTargetMachine.cpp +++ b/lib/Target/ARM/ARMTargetMachine.cpp @@ -95,7 +95,7 @@ bool ARMBaseTargetMachine::addPreRegAlloc(PassManagerBase &PM, // Calculate and set max stack object alignment early, so we can decide // whether we will need stack realignment (and thus FP). - PM.add(createARMMaxStackAlignmentCalculatorPass()); + PM.add(createMaxStackAlignmentCalculatorPass()); // FIXME: temporarily disabling load / store optimization pass for Thumb1. if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only()) diff --git a/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp b/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp index 692bb19..362bbf1 100644 --- a/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp +++ b/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp @@ -1045,6 +1045,7 @@ bool ARMAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNum, printNoHashImmediate(MI, OpNum); return false; case 'P': // Print a VFP double precision register. + case 'q': // Print a NEON quad precision register. printOperand(MI, OpNum); return false; case 'Q': diff --git a/lib/Target/ARM/NEONMoveFix.cpp b/lib/Target/ARM/NEONMoveFix.cpp index 50abcf4..3c0414d 100644 --- a/lib/Target/ARM/NEONMoveFix.cpp +++ b/lib/Target/ARM/NEONMoveFix.cpp @@ -51,7 +51,7 @@ bool NEONMoveFixPass::InsertMoves(MachineBasicBlock &MBB) { MachineBasicBlock::iterator MII = MBB.begin(), E = MBB.end(); MachineBasicBlock::iterator NextMII; for (; MII != E; MII = NextMII) { - NextMII = next(MII); + NextMII = llvm::next(MII); MachineInstr *MI = &*MII; if (MI->getOpcode() == ARM::VMOVD && diff --git a/lib/Target/ARM/NEONPreAllocPass.cpp b/lib/Target/ARM/NEONPreAllocPass.cpp index 206677b..d9942c8 100644 --- a/lib/Target/ARM/NEONPreAllocPass.cpp +++ b/lib/Target/ARM/NEONPreAllocPass.cpp @@ -338,7 +338,7 @@ bool NEONPreAllocPass::PreAllocNEONRegisters(MachineBasicBlock &MBB) { if (!isNEONMultiRegOp(MI->getOpcode(), FirstOpnd, NumRegs, Offset, Stride)) continue; - MachineBasicBlock::iterator NextI = next(MBBI); + MachineBasicBlock::iterator NextI = llvm::next(MBBI); for (unsigned R = 0; R < NumRegs; ++R) { MachineOperand &MO = MI->getOperand(FirstOpnd + R); assert(MO.isReg() && MO.getSubReg() == 0 && "unexpected operand"); diff --git a/lib/Target/ARM/Thumb1InstrInfo.cpp b/lib/Target/ARM/Thumb1InstrInfo.cpp index 7602b6d..66d3b83 100644 --- a/lib/Target/ARM/Thumb1InstrInfo.cpp +++ b/lib/Target/ARM/Thumb1InstrInfo.cpp @@ -32,25 +32,6 @@ unsigned Thumb1InstrInfo::getUnindexedOpcode(unsigned Opc) const { return 0; } -bool -Thumb1InstrInfo::BlockHasNoFallThrough(const MachineBasicBlock &MBB) const { - if (MBB.empty()) return false; - - switch (MBB.back().getOpcode()) { - case ARM::tBX_RET: - case ARM::tBX_RET_vararg: - case ARM::tPOP_RET: - case ARM::tB: - case ARM::tBRIND: - case ARM::tBR_JTr: - return true; - default: - break; - } - - return false; -} - bool Thumb1InstrInfo::copyRegToReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, unsigned DestReg, unsigned SrcReg, diff --git a/lib/Target/ARM/Thumb1InstrInfo.h b/lib/Target/ARM/Thumb1InstrInfo.h index b28229d..516ddf1 100644 --- a/lib/Target/ARM/Thumb1InstrInfo.h +++ b/lib/Target/ARM/Thumb1InstrInfo.h @@ -31,9 +31,6 @@ public: // if there is not such an opcode. unsigned getUnindexedOpcode(unsigned Opc) const; - // Return true if the block does not fall through. - bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const; - /// getRegisterInfo - TargetInstrInfo is a superset of MRegister info. As /// such, whenever a client has an instance of instruction info, it should /// always be able to get register info as well (through this method). diff --git a/lib/Target/ARM/Thumb1RegisterInfo.cpp b/lib/Target/ARM/Thumb1RegisterInfo.cpp index 37adf37..9f3816a 100644 --- a/lib/Target/ARM/Thumb1RegisterInfo.cpp +++ b/lib/Target/ARM/Thumb1RegisterInfo.cpp @@ -528,7 +528,7 @@ Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, MI.getOperand(i+1).ChangeToImmediate(Mask); } Offset = (Offset - Mask * Scale); - MachineBasicBlock::iterator NII = next(II); + MachineBasicBlock::iterator NII = llvm::next(II); emitThumbRegPlusImmediate(MBB, NII, DestReg, DestReg, Offset, TII, *this, dl); } else { diff --git a/lib/Target/ARM/Thumb2InstrInfo.cpp b/lib/Target/ARM/Thumb2InstrInfo.cpp index 16c1e6f..f4a8c27 100644 --- a/lib/Target/ARM/Thumb2InstrInfo.cpp +++ b/lib/Target/ARM/Thumb2InstrInfo.cpp @@ -36,30 +36,6 @@ unsigned Thumb2InstrInfo::getUnindexedOpcode(unsigned Opc) const { } bool -Thumb2InstrInfo::BlockHasNoFallThrough(const MachineBasicBlock &MBB) const { - if (MBB.empty()) return false; - - switch (MBB.back().getOpcode()) { - case ARM::t2LDM_RET: - case ARM::t2B: // Uncond branch. - case ARM::t2BR_JT: // Jumptable branch. - case ARM::t2TBB: // Table branch byte. - case ARM::t2TBH: // Table branch halfword. - case ARM::tBR_JTr: // Jumptable branch (16-bit version). - case ARM::tBX_RET: - case ARM::tBX_RET_vararg: - case ARM::tPOP_RET: - case ARM::tB: - case ARM::tBRIND: - return true; - default: - break; - } - - return false; -} - -bool Thumb2InstrInfo::copyRegToReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, unsigned DestReg, unsigned SrcReg, diff --git a/lib/Target/ARM/Thumb2InstrInfo.h b/lib/Target/ARM/Thumb2InstrInfo.h index 663a60b..a0f89a6 100644 --- a/lib/Target/ARM/Thumb2InstrInfo.h +++ b/lib/Target/ARM/Thumb2InstrInfo.h @@ -31,9 +31,6 @@ public: // if there is not such an opcode. unsigned getUnindexedOpcode(unsigned Opc) const; - // Return true if the block does not fall through. - bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const; - bool copyRegToReg(MachineBasicBlock &MBB, MachineBasicBlock::iterator I, unsigned DestReg, unsigned SrcReg, diff --git a/lib/Target/ARM/Thumb2SizeReduction.cpp b/lib/Target/ARM/Thumb2SizeReduction.cpp index b2fd7b3..35359aa 100644 --- a/lib/Target/ARM/Thumb2SizeReduction.cpp +++ b/lib/Target/ARM/Thumb2SizeReduction.cpp @@ -649,7 +649,7 @@ bool Thumb2SizeReduce::ReduceMBB(MachineBasicBlock &MBB) { MachineBasicBlock::iterator MII = MBB.begin(), E = MBB.end(); MachineBasicBlock::iterator NextMII; for (; MII != E; MII = NextMII) { - NextMII = next(MII); + NextMII = llvm::next(MII); MachineInstr *MI = &*MII; LiveCPSR = UpdateCPSRUse(*MI, LiveCPSR); |
