//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "arm-isel"
+#include "ARMISelLowering.h"
#include "ARM.h"
#include "ARMCallingConv.h"
#include "ARMConstantPoolValue.h"
-#include "ARMISelLowering.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMPerfectShuffle.h"
-#include "ARMRegisterInfo.h"
#include "ARMSubtarget.h"
#include "ARMTargetMachine.h"
#include "ARMTargetObjectFile.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
-#include <sstream>
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>
}
// The APCS parameter registers.
-static const unsigned GPRArgRegs[] = {
+static const uint16_t GPRArgRegs[] = {
ARM::R0, ARM::R1, ARM::R2, ARM::R3
};
}
void ARMTargetLowering::addDRTypeForNEON(EVT VT) {
- addRegisterClass(VT, ARM::DPRRegisterClass);
+ addRegisterClass(VT, &ARM::DPRRegClass);
addTypeForNEON(VT, MVT::f64, MVT::v2i32);
}
void ARMTargetLowering::addQRTypeForNEON(EVT VT) {
- addRegisterClass(VT, ARM::QPRRegisterClass);
+ addRegisterClass(VT, &ARM::QPRRegClass);
addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}
setLibcallName(RTLIB::SRL_I128, 0);
setLibcallName(RTLIB::SRA_I128, 0);
- if (Subtarget->isAAPCS_ABI()) {
+ if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetDarwin()) {
// Double-precision floating-point arithmetic helper functions
// RTABI chapter 4.1.2, Table 2
setLibcallName(RTLIB::ADD_F64, "__aeabi_dadd");
}
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);
}
setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType)VT, Expand);
}
+ setOperationAction(ISD::ConstantFP, MVT::f32, Custom);
+
if (Subtarget->hasNEON()) {
addDRTypeForNEON(MVT::v2f32);
addDRTypeForNEON(MVT::v8i8);
setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
-
+
setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
setOperationAction(ISD::FCOS, MVT::v4f32, Expand);
setOperationAction(ISD::SETCC, MVT::v1i64, Expand);
setOperationAction(ISD::SETCC, MVT::v2i64, Expand);
// Neon does not have single instruction SINT_TO_FP and UINT_TO_FP with
- // a destination type that is wider than the source.
+ // a destination type that is wider than the source, and nor does
+ // it have a FP_TO_[SU]INT instruction with a narrower destination than
+ // source.
setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i16, Custom);
+ setOperationAction(ISD::FP_TO_UINT, MVT::v4i16, Custom);
+ setOperationAction(ISD::FP_TO_SINT, MVT::v4i16, Custom);
setTargetDAGCombine(ISD::INTRINSIC_VOID);
setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
- setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Expand);
+ // It is legal to extload from v4i8 to v4i16 or v4i32.
+ MVT Tys[6] = {MVT::v8i8, MVT::v4i8, MVT::v2i8,
+ MVT::v4i16, MVT::v2i16,
+ MVT::v2i32};
+ for (unsigned i = 0; i < 6; ++i) {
+ setLoadExtAction(ISD::EXTLOAD, Tys[i], Legal);
+ setLoadExtAction(ISD::ZEXTLOAD, Tys[i], Legal);
+ setLoadExtAction(ISD::SEXTLOAD, Tys[i], Legal);
+ }
}
computeRegisterProperties();
setOperationAction(ISD::VAEND, MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
- setOperationAction(ISD::EHSELECTION, MVT::i32, Expand);
- setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
- setExceptionPointerRegister(ARM::R0);
- setExceptionSelectorRegister(ARM::R1);
+
+ if (!Subtarget->isTargetDarwin()) {
+ // Non-Darwin platforms may return values in these registers via the
+ // personality function.
+ setOperationAction(ISD::EHSELECTION, MVT::i32, Expand);
+ setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
+ setExceptionPointerRegister(ARM::R0);
+ setExceptionSelectorRegister(ARM::R1);
+ }
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
// ARMv6 Thumb1 (except for CPUs that support dmb / dsb) and earlier use
setOperationAction(ISD::FPOW, MVT::f64, Expand);
setOperationAction(ISD::FPOW, MVT::f32, Expand);
- setOperationAction(ISD::FMA, MVT::f64, Expand);
- setOperationAction(ISD::FMA, MVT::f32, Expand);
+ if (!Subtarget->hasVFP4()) {
+ setOperationAction(ISD::FMA, MVT::f64, Expand);
+ setOperationAction(ISD::FMA, MVT::f32, Expand);
+ }
// Various VFP goodness
if (!TM.Options.UseSoftFloat && !Subtarget->isThumb1Only()) {
setTargetDAGCombine(ISD::SUB);
setTargetDAGCombine(ISD::MUL);
- if (Subtarget->hasV6T2Ops() || Subtarget->hasNEON())
- setTargetDAGCombine(ISD::OR);
- if (Subtarget->hasNEON())
+ if (Subtarget->hasV6T2Ops() || Subtarget->hasNEON()) {
setTargetDAGCombine(ISD::AND);
+ setTargetDAGCombine(ISD::OR);
+ setTargetDAGCombine(ISD::XOR);
+ }
+
+ if (Subtarget->hasV6Ops())
+ setTargetDAGCombine(ISD::SRL);
setStackPointerRegisterToSaveRestore(ARM::SP);
benefitFromCodePlacementOpt = true;
+ // Prefer likely predicted branches to selects on out-of-order cores.
+ predictableSelectIsExpensive = Subtarget->isCortexA9();
+
setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2);
}
// 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
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;
}
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";
case ARMISD::BCC_i64: return "ARMISD::BCC_i64";
case ARMISD::FMSTAT: return "ARMISD::FMSTAT";
+
case ARMISD::CMOV: return "ARMISD::CMOV";
+ case ARMISD::CAND: return "ARMISD::CAND";
+ case ARMISD::COR: return "ARMISD::COR";
+ case ARMISD::CXOR: return "ARMISD::CXOR";
case ARMISD::RBIT: return "ARMISD::RBIT";
/// getRegClassFor - Return the register class that should be used for the
/// specified value type.
-TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const {
+const TargetRegisterClass *ARMTargetLowering::getRegClassFor(EVT VT) const {
// Map v4i64 to QQ registers but do not make the type legal. Similarly map
// v8i64 to QQQQ registers. v4i64 and v8i64 are only used for REG_SEQUENCE to
// 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
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);
}
}
/// 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 &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;
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());
if (Subtarget->isThumb()) {
if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
CallOpc = ARMISD::CALL_NOLINK;
+ else if (doesNotRet && isDirect && !isARMFunc &&
+ Subtarget->hasRAS() && !Subtarget->isThumb1Only())
+ // "mov lr, pc; b _foo" to avoid confusing the RSP
+ CallOpc = ARMISD::CALL_NOLINK;
else
CallOpc = isARMFunc ? ARMISD::CALL : ARMISD::tCALL;
} else {
- CallOpc = (isDirect || Subtarget->hasV5TOps())
- ? (isLocalARMFunc ? ARMISD::CALL_PRED : ARMISD::CALL)
- : ARMISD::CALL_NOLINK;
+ if (!isDirect && !Subtarget->hasV5TOps()) {
+ CallOpc = ARMISD::CALL_NOLINK;
+ } else if (doesNotRet && isDirect && Subtarget->hasRAS())
+ // "mov lr, pc; b _foo" to avoid confusing the RSP
+ CallOpc = ARMISD::CALL_NOLINK;
+ else
+ CallOpc = isLocalARMFunc ? ARMISD::CALL_PRED : ARMISD::CALL;
}
std::vector<SDValue> Ops;
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
+ // Add a register mask operand representing the call-preserved registers.
+ const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+ const uint32_t *Mask = TRI->getCallPreservedMask(CallConv);
+ assert(Mask && "Missing call preserved mask for calling convention");
+ Ops.push_back(DAG.getRegisterMask(Mask));
+
if (InFlag.getNode())
Ops.push_back(InFlag);
/// and then confiscate the rest of the parameter registers to insure
/// this.
void
-llvm::ARMTargetLowering::HandleByVal(CCState *State, unsigned &size) const {
+ARMTargetLowering::HandleByVal(CCState *State, unsigned &size) const {
unsigned reg = State->AllocateReg(GPRArgRegs, 4);
assert((State->getCallOrPrologue() == Prologue ||
State->getCallOrPrologue() == Call) &&
static
bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
MachineFrameInfo *MFI, const MachineRegisterInfo *MRI,
- const ARMInstrInfo *TII) {
+ const TargetInstrInfo *TII) {
unsigned Bytes = Arg.getValueType().getSizeInBits() / 8;
int FI = INT_MAX;
if (Arg.getOpcode() == ISD::CopyFromReg) {
// the caller's fixed stack objects.
MachineFrameInfo *MFI = MF.getFrameInfo();
const MachineRegisterInfo *MRI = &MF.getRegInfo();
- const ARMInstrInfo *TII =
- ((ARMTargetMachine&)getTargetMachine()).getInstrInfo();
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
i != e;
++i, ++realArgIdx) {
return result;
}
-bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N) const {
+bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
if (N->getNumValues() != 1)
return false;
if (!N->hasNUsesOfValue(1, 0))
return false;
- unsigned NumCopies = 0;
- SDNode* Copies[2] = { 0, 0 };
- SDNode *Use = *N->use_begin();
- if (Use->getOpcode() == ISD::CopyToReg) {
- Copies[NumCopies++] = Use;
- } else if (Use->getOpcode() == ARMISD::VMOVRRD) {
+ SDValue TCChain = Chain;
+ SDNode *Copy = *N->use_begin();
+ if (Copy->getOpcode() == ISD::CopyToReg) {
+ // If the copy has a glue operand, we conservatively assume it isn't safe to
+ // perform a tail call.
+ if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
+ return false;
+ TCChain = Copy->getOperand(0);
+ } else if (Copy->getOpcode() == ARMISD::VMOVRRD) {
+ SDNode *VMov = Copy;
// f64 returned in a pair of GPRs.
- for (SDNode::use_iterator UI = Use->use_begin(), UE = Use->use_end();
+ SmallPtrSet<SDNode*, 2> Copies;
+ for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
UI != UE; ++UI) {
if (UI->getOpcode() != ISD::CopyToReg)
return false;
- Copies[UI.getUse().getResNo()] = *UI;
- ++NumCopies;
+ Copies.insert(*UI);
+ }
+ if (Copies.size() > 2)
+ return false;
+
+ for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
+ UI != UE; ++UI) {
+ SDValue UseChain = UI->getOperand(0);
+ if (Copies.count(UseChain.getNode()))
+ // Second CopyToReg
+ Copy = *UI;
+ else
+ // First CopyToReg
+ TCChain = UseChain;
}
- } else if (Use->getOpcode() == ISD::BITCAST) {
+ } else if (Copy->getOpcode() == ISD::BITCAST) {
// f32 returned in a single GPR.
- if (!Use->hasNUsesOfValue(1, 0))
+ if (!Copy->hasOneUse())
return false;
- Use = *Use->use_begin();
- if (Use->getOpcode() != ISD::CopyToReg || !Use->hasNUsesOfValue(1, 0))
+ Copy = *Copy->use_begin();
+ if (Copy->getOpcode() != ISD::CopyToReg || !Copy->hasNUsesOfValue(1, 0))
return false;
- Copies[NumCopies++] = Use;
+ Chain = Copy->getOperand(0);
} else {
return false;
}
- if (NumCopies != 1 && NumCopies != 2)
- return false;
-
bool HasRet = false;
- for (unsigned i = 0; i < NumCopies; ++i) {
- SDNode *Copy = Copies[i];
- for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
- UI != UE; ++UI) {
- if (UI->getOpcode() == ISD::CopyToReg) {
- SDNode *Use = *UI;
- if (Use == Copies[0] || ((NumCopies == 2) && (Use == Copies[1])))
- continue;
- return false;
- }
- if (UI->getOpcode() != ARMISD::RET_FLAG)
- return false;
- HasRet = true;
- }
+ for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
+ UI != UE; ++UI) {
+ if (UI->getOpcode() != ARMISD::RET_FLAG)
+ return false;
+ HasRet = true;
}
- return HasRet;
+ if (!HasRet)
+ return false;
+
+ Chain = TCChain;
+ return true;
}
bool ARMTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
- if (!EnableARMTailCalls)
+ if (!EnableARMTailCalls && !Subtarget->supportsTailCall())
return false;
if (!CI->isTailCall())
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, false, /*isReturnValueUsed=*/true,
+ 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;
}
// "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;
// 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();
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);
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,
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
- TargetRegisterClass *RC;
+ 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);
SmallVector<SDValue, 4> MemOps;
for (; firstRegToSaveIndex < 4; ++firstRegToSaveIndex) {
- TargetRegisterClass *RC;
+ 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);
ArgValue = GetF64FormalArgument(VA, ArgLocs[++i], Chain, DAG, dl);
} else {
- TargetRegisterClass *RC;
+ 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");
}
}
+ // ARM's BooleanContents value is UndefinedBooleanContent. Mask out the
+ // undefined bits before doing a full-word comparison with zero.
+ Cond = DAG.getNode(ISD::AND, dl, Cond.getValueType(), Cond,
+ DAG.getConstant(1, Cond.getValueType()));
+
return DAG.getSelectCC(dl, Cond,
DAG.getConstant(0, Cond.getValueType()),
SelectTrue, SelectFalse, ISD::SETNE);
SDValue Dest = Op.getOperand(4);
DebugLoc dl = Op.getDebugLoc();
- bool SeenZero = false;
- if (canChangeToInt(LHS, SeenZero, Subtarget) &&
- canChangeToInt(RHS, SeenZero, Subtarget) &&
- // If one of the operand is zero, it's safe to ignore the NaN case since
- // we only care about equality comparisons.
- (SeenZero || (DAG.isKnownNeverNaN(LHS) && DAG.isKnownNeverNaN(RHS)))) {
+ bool LHSSeenZero = false;
+ bool LHSOk = canChangeToInt(LHS, LHSSeenZero, Subtarget);
+ bool RHSSeenZero = false;
+ bool RHSOk = canChangeToInt(RHS, RHSSeenZero, Subtarget);
+ if (LHSOk && RHSOk && (LHSSeenZero || RHSSeenZero)) {
// If unsafe fp math optimization is enabled and there are no other uses of
// the CMP operands, and the condition code is EQ or NE, we can optimize it
// to an integer comparison.
else if (CC == ISD::SETUNE)
CC = ISD::SETNE;
+ SDValue Mask = DAG.getConstant(0x7fffffff, MVT::i32);
SDValue ARMcc;
if (LHS.getValueType() == MVT::f32) {
- LHS = bitcastf32Toi32(LHS, DAG);
- RHS = bitcastf32Toi32(RHS, DAG);
+ LHS = DAG.getNode(ISD::AND, dl, MVT::i32,
+ bitcastf32Toi32(LHS, DAG), Mask);
+ RHS = DAG.getNode(ISD::AND, dl, MVT::i32,
+ bitcastf32Toi32(RHS, DAG), Mask);
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
return DAG.getNode(ARMISD::BRCOND, dl, MVT::Other,
SDValue RHS1, RHS2;
expandf64Toi32(LHS, DAG, LHS1, LHS2);
expandf64Toi32(RHS, DAG, RHS1, RHS2);
+ LHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, LHS2, Mask);
+ RHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, RHS2, Mask);
ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
ARMcc = DAG.getConstant(CondCode, MVT::i32);
SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Glue);
}
static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
- assert(Op.getValueType().getVectorElementType() == MVT::i32
- && "Unexpected custom lowering");
+ EVT VT = Op.getValueType();
+ DebugLoc dl = Op.getDebugLoc();
- if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::f32)
- return Op;
- return DAG.UnrollVectorOp(Op.getNode());
+ if (Op.getValueType().getVectorElementType() == MVT::i32) {
+ if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::f32)
+ return Op;
+ return DAG.UnrollVectorOp(Op.getNode());
+ }
+
+ assert(Op.getOperand(0).getValueType() == MVT::v4f32 &&
+ "Invalid type for custom lowering!");
+ if (VT != MVT::v4i16)
+ return DAG.UnrollVectorOp(Op.getNode());
+
+ Op = DAG.getNode(Op.getOpcode(), dl, MVT::v4i32, Op.getOperand(0));
+ return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
}
static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
return DAG.getTargetConstant(EncodedVal, MVT::i32);
}
+SDValue ARMTargetLowering::LowerConstantFP(SDValue Op, SelectionDAG &DAG,
+ const ARMSubtarget *ST) const {
+ if (!ST->useNEONForSinglePrecisionFP() || !ST->hasVFP3() || ST->hasD16())
+ return SDValue();
+
+ ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Op);
+ assert(Op.getValueType() == MVT::f32 &&
+ "ConstantFP custom lowering should only occur for f32.");
+
+ // Try splatting with a VMOV.f32...
+ APFloat FPVal = CFP->getValueAPF();
+ int ImmVal = ARM_AM::getFP32Imm(FPVal);
+ if (ImmVal != -1) {
+ DebugLoc DL = Op.getDebugLoc();
+ SDValue NewVal = DAG.getTargetConstant(ImmVal, MVT::i32);
+ SDValue VecConstant = DAG.getNode(ARMISD::VMOVFPIMM, DL, MVT::v2f32,
+ NewVal);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecConstant,
+ DAG.getConstant(0, MVT::i32));
+ }
+
+ // If that fails, try a VMOV.i32
+ EVT VMovVT;
+ unsigned iVal = FPVal.bitcastToAPInt().getZExtValue();
+ SDValue NewVal = isNEONModifiedImm(iVal, 0, 32, DAG, VMovVT, false,
+ VMOVModImm);
+ if (NewVal != SDValue()) {
+ DebugLoc DL = Op.getDebugLoc();
+ SDValue VecConstant = DAG.getNode(ARMISD::VMOVIMM, DL, VMovVT,
+ NewVal);
+ SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
+ VecConstant);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
+ DAG.getConstant(0, MVT::i32));
+ }
+
+ // Finally, try a VMVN.i32
+ NewVal = isNEONModifiedImm(~iVal & 0xffffffff, 0, 32, DAG, VMovVT, false,
+ VMVNModImm);
+ if (NewVal != SDValue()) {
+ DebugLoc DL = Op.getDebugLoc();
+ SDValue VecConstant = DAG.getNode(ARMISD::VMVNIMM, DL, VMovVT, NewVal);
+ SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
+ VecConstant);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
+ DAG.getConstant(0, MVT::i32));
+ }
+
+ return SDValue();
+}
+
+
static bool isVEXTMask(ArrayRef<int> M, EVT VT,
bool &ReverseVEXT, unsigned &Imm) {
unsigned NumElts = VT.getVectorNumElements();
// 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) {
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);
case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG);
case ISD::CTTZ: return LowerCTTZ(Op.getNode(), DAG, Subtarget);
case ISD::SETCC: return LowerVSETCC(Op, DAG);
+ case ISD::ConstantFP: return LowerConstantFP(Op, DAG, Subtarget);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG, Subtarget);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
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;
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;
BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
- 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);
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;
BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
- 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);
// 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));
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;
BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
- 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:
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 =
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.
DenseMap<unsigned, SmallVector<MachineBasicBlock*, 2> > CallSiteNumToLPad;
unsigned MaxCSNum = 0;
MachineModuleInfo &MMI = MF->getMMI();
- for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E; ++BB) {
+ for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E;
+ ++BB) {
if (!BB->isLandingPad()) continue;
// FIXME: We should assert that the EH_LABEL is the first MI in the landing
BuildMI(DispatchBB, dl, TII->get(ARM::tInt_eh_sjlj_dispatchsetup));
else if (!Subtarget->hasVFP2())
BuildMI(DispatchBB, dl, TII->get(ARM::Int_eh_sjlj_dispatchsetup_nofp));
- else
+ else
BuildMI(DispatchBB, dl, TII->get(ARM::Int_eh_sjlj_dispatchsetup));
unsigned NumLPads = LPadList.size();
// N.B. the order the invoke BBs are processed in doesn't matter here.
const ARMBaseInstrInfo *AII = static_cast<const ARMBaseInstrInfo*>(TII);
const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
- const unsigned *SavedRegs = RI.getCalleeSavedRegs(MF);
+ const uint16_t *SavedRegs = RI.getCalleeSavedRegs(MF);
SmallVector<MachineBasicBlock*, 64> MBBLPads;
for (SmallPtrSet<MachineBasicBlock*, 64>::iterator
I = InvokeBBs.begin(), E = InvokeBBs.end(); I != E; ++I) {
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);
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 {
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,
// 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,
// 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,
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();
// return last added BB
return SinkBB;
}
+ case ARM::COPY_STRUCT_BYVAL_I32:
+ ++NumLoopByVals;
+ return EmitStructByval(MI, BB);
}
}
if (!C)
return SDValue();
- uint64_t MulAmt = C->getZExtValue();
+ int64_t MulAmt = C->getSExtValue();
unsigned ShiftAmt = CountTrailingZeros_64(MulAmt);
+
ShiftAmt = ShiftAmt & (32 - 1);
SDValue V = N->getOperand(0);
DebugLoc DL = N->getDebugLoc();
SDValue Res;
MulAmt >>= ShiftAmt;
- if (isPowerOf2_32(MulAmt - 1)) {
- // (mul x, 2^N + 1) => (add (shl x, N), x)
- Res = DAG.getNode(ISD::ADD, DL, VT,
- V, DAG.getNode(ISD::SHL, DL, VT,
- V, DAG.getConstant(Log2_32(MulAmt-1),
- MVT::i32)));
- } else if (isPowerOf2_32(MulAmt + 1)) {
- // (mul x, 2^N - 1) => (sub (shl x, N), x)
- Res = DAG.getNode(ISD::SUB, DL, VT,
- DAG.getNode(ISD::SHL, DL, VT,
- V, DAG.getConstant(Log2_32(MulAmt+1),
- MVT::i32)),
- V);
- } else
- return SDValue();
+
+ if (MulAmt >= 0) {
+ if (isPowerOf2_32(MulAmt - 1)) {
+ // (mul x, 2^N + 1) => (add (shl x, N), x)
+ Res = DAG.getNode(ISD::ADD, DL, VT,
+ V,
+ DAG.getNode(ISD::SHL, DL, VT,
+ V,
+ DAG.getConstant(Log2_32(MulAmt - 1),
+ MVT::i32)));
+ } else if (isPowerOf2_32(MulAmt + 1)) {
+ // (mul x, 2^N - 1) => (sub (shl x, N), x)
+ Res = DAG.getNode(ISD::SUB, DL, VT,
+ DAG.getNode(ISD::SHL, DL, VT,
+ V,
+ DAG.getConstant(Log2_32(MulAmt + 1),
+ MVT::i32)),
+ V);
+ } else
+ return SDValue();
+ } else {
+ uint64_t MulAmtAbs = -MulAmt;
+ if (isPowerOf2_32(MulAmtAbs + 1)) {
+ // (mul x, -(2^N - 1)) => (sub x, (shl x, N))
+ Res = DAG.getNode(ISD::SUB, DL, VT,
+ V,
+ DAG.getNode(ISD::SHL, DL, VT,
+ V,
+ DAG.getConstant(Log2_32(MulAmtAbs + 1),
+ MVT::i32)));
+ } else if (isPowerOf2_32(MulAmtAbs - 1)) {
+ // (mul x, -(2^N + 1)) => - (add (shl x, N), x)
+ Res = DAG.getNode(ISD::ADD, DL, VT,
+ V,
+ DAG.getNode(ISD::SHL, DL, VT,
+ V,
+ DAG.getConstant(Log2_32(MulAmtAbs-1),
+ MVT::i32)));
+ Res = DAG.getNode(ISD::SUB, DL, VT,
+ DAG.getConstant(0, MVT::i32),Res);
+
+ } else
+ return SDValue();
+ }
if (ShiftAmt != 0)
- Res = DAG.getNode(ISD::SHL, DL, VT, Res,
- DAG.getConstant(ShiftAmt, MVT::i32));
+ Res = DAG.getNode(ISD::SHL, DL, VT,
+ Res, DAG.getConstant(ShiftAmt, MVT::i32));
// Do not add new nodes to DAG combiner worklist.
DCI.CombineTo(N, Res, false);
return SDValue();
}
+static bool isCMOVWithZeroOrAllOnesLHS(SDValue N, bool AllOnes) {
+ if (N.getOpcode() != ARMISD::CMOV || !N.getNode()->hasOneUse())
+ return false;
+
+ SDValue FalseVal = N.getOperand(0);
+ ConstantSDNode *C = dyn_cast<ConstantSDNode>(FalseVal);
+ if (!C)
+ return false;
+ if (AllOnes)
+ return C->isAllOnesValue();
+ return C->isNullValue();
+}
+
+/// formConditionalOp - Combine an operation with a conditional move operand
+/// to form a conditional op. e.g. (or x, (cmov 0, y, cond)) => (or.cond x, y)
+/// (and x, (cmov -1, y, cond)) => (and.cond, x, y)
+static SDValue formConditionalOp(SDNode *N, SelectionDAG &DAG,
+ bool Commutable) {
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+
+ bool isAND = N->getOpcode() == ISD::AND;
+ bool isCand = isCMOVWithZeroOrAllOnesLHS(N1, isAND);
+ if (!isCand && Commutable) {
+ isCand = isCMOVWithZeroOrAllOnesLHS(N0, isAND);
+ if (isCand)
+ std::swap(N0, N1);
+ }
+ if (!isCand)
+ return SDValue();
+
+ unsigned Opc = 0;
+ switch (N->getOpcode()) {
+ default: llvm_unreachable("Unexpected node");
+ case ISD::AND: Opc = ARMISD::CAND; break;
+ case ISD::OR: Opc = ARMISD::COR; break;
+ case ISD::XOR: Opc = ARMISD::CXOR; break;
+ }
+ return DAG.getNode(Opc, N->getDebugLoc(), N->getValueType(0), N0,
+ N1.getOperand(1), N1.getOperand(2), N1.getOperand(3),
+ N1.getOperand(4));
+}
+
static SDValue PerformANDCombine(SDNode *N,
- TargetLowering::DAGCombinerInfo &DCI) {
+ TargetLowering::DAGCombinerInfo &DCI,
+ const ARMSubtarget *Subtarget) {
// Attempt to use immediate-form VBIC
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
}
}
+ if (!Subtarget->isThumb1Only()) {
+ // (and x, (cmov -1, y, cond)) => (and.cond x, y)
+ SDValue CAND = formConditionalOp(N, DAG, true);
+ if (CAND.getNode())
+ return CAND;
+ }
+
return SDValue();
}
}
}
+ if (!Subtarget->isThumb1Only()) {
+ // (or x, (cmov 0, y, cond)) => (or.cond x, y)
+ SDValue COR = formConditionalOp(N, DAG, true);
+ if (COR.getNode())
+ return COR;
+ }
+
SDValue N0 = N->getOperand(0);
if (N0.getOpcode() != ISD::AND)
return SDValue();
return SDValue();
}
+static SDValue PerformXORCombine(SDNode *N,
+ TargetLowering::DAGCombinerInfo &DCI,
+ const ARMSubtarget *Subtarget) {
+ EVT VT = N->getValueType(0);
+ SelectionDAG &DAG = DCI.DAG;
+
+ if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
+ return SDValue();
+
+ if (!Subtarget->isThumb1Only()) {
+ // (xor x, (cmov 0, y, cond)) => (xor.cond x, y)
+ SDValue CXOR = formConditionalOp(N, DAG, true);
+ if (CXOR.getNode())
+ return CXOR;
+ }
+
+ return SDValue();
+}
+
/// PerformBFICombine - (bfi A, (and B, Mask1), Mask2) -> (bfi A, B, Mask2) iff
/// the bits being cleared by the AND are not demanded by the BFI.
static SDValue PerformBFICombine(SDNode *N,
/// ISD::STORE.
static SDValue PerformSTORECombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
- // Bitcast an i64 store extracted from a vector to f64.
- // Otherwise, the i64 value will be legalized to a pair of i32 values.
StoreSDNode *St = cast<StoreSDNode>(N);
+ if (St->isVolatile())
+ return SDValue();
+
+ // 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();
- if (!ISD::isNormalStore(St) || St->isVolatile())
+ EVT VT = StVal.getValueType();
+ if (St->isTruncatingStore() && VT.isVector()) {
+ SelectionDAG &DAG = DCI.DAG;
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ EVT StVT = St->getMemoryVT();
+ unsigned NumElems = VT.getVectorNumElements();
+ assert(StVT != VT && "Cannot truncate to the same type");
+ unsigned FromEltSz = VT.getVectorElementType().getSizeInBits();
+ unsigned ToEltSz = StVT.getVectorElementType().getSizeInBits();
+
+ // From, To sizes and ElemCount must be pow of two
+ if (!isPowerOf2_32(NumElems * FromEltSz * ToEltSz)) return SDValue();
+
+ // We are going to use the original vector elt for storing.
+ // Accumulated smaller vector elements must be a multiple of the store size.
+ if (0 != (NumElems * FromEltSz) % ToEltSz) return SDValue();
+
+ unsigned SizeRatio = FromEltSz / ToEltSz;
+ assert(SizeRatio * NumElems * ToEltSz == VT.getSizeInBits());
+
+ // Create a type on which we perform the shuffle.
+ EVT WideVecVT = EVT::getVectorVT(*DAG.getContext(), StVT.getScalarType(),
+ NumElems*SizeRatio);
+ assert(WideVecVT.getSizeInBits() == VT.getSizeInBits());
+
+ DebugLoc DL = St->getDebugLoc();
+ SDValue WideVec = DAG.getNode(ISD::BITCAST, DL, WideVecVT, StVal);
+ SmallVector<int, 8> ShuffleVec(NumElems * SizeRatio, -1);
+ for (unsigned i = 0; i < NumElems; ++i) ShuffleVec[i] = i * SizeRatio;
+
+ // Can't shuffle using an illegal type.
+ if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
+
+ SDValue Shuff = DAG.getVectorShuffle(WideVecVT, DL, WideVec,
+ DAG.getUNDEF(WideVec.getValueType()),
+ ShuffleVec.data());
+ // At this point all of the data is stored at the bottom of the
+ // register. We now need to save it to mem.
+
+ // Find the largest store unit
+ MVT StoreType = MVT::i8;
+ for (unsigned tp = MVT::FIRST_INTEGER_VALUETYPE;
+ tp < MVT::LAST_INTEGER_VALUETYPE; ++tp) {
+ MVT Tp = (MVT::SimpleValueType)tp;
+ if (TLI.isTypeLegal(Tp) && Tp.getSizeInBits() <= NumElems * ToEltSz)
+ StoreType = Tp;
+ }
+ // Didn't find a legal store type.
+ if (!TLI.isTypeLegal(StoreType))
+ return SDValue();
+
+ // Bitcast the original vector into a vector of store-size units
+ EVT StoreVecVT = EVT::getVectorVT(*DAG.getContext(),
+ StoreType, VT.getSizeInBits()/EVT(StoreType).getSizeInBits());
+ assert(StoreVecVT.getSizeInBits() == VT.getSizeInBits());
+ SDValue ShuffWide = DAG.getNode(ISD::BITCAST, DL, StoreVecVT, Shuff);
+ SmallVector<SDValue, 8> Chains;
+ SDValue Increment = DAG.getConstant(StoreType.getSizeInBits()/8,
+ TLI.getPointerTy());
+ SDValue BasePtr = St->getBasePtr();
+
+ // Perform one or more big stores into memory.
+ unsigned E = (ToEltSz*NumElems)/StoreType.getSizeInBits();
+ for (unsigned I = 0; I < E; I++) {
+ SDValue SubVec = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
+ StoreType, ShuffWide,
+ DAG.getIntPtrConstant(I));
+ SDValue Ch = DAG.getStore(St->getChain(), DL, SubVec, BasePtr,
+ St->getPointerInfo(), St->isVolatile(),
+ St->isNonTemporal(), St->getAlignment());
+ BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr,
+ Increment);
+ Chains.push_back(Ch);
+ }
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, &Chains[0],
+ Chains.size());
+ }
+
+ if (!ISD::isNormalStore(St))
return SDValue();
+ // Split a store of a VMOVDRR into two integer stores to avoid mixing NEON and
+ // ARM stores of arguments in the same cache line.
if (StVal.getNode()->getOpcode() == ARMISD::VMOVDRR &&
- StVal.getNode()->hasOneUse() && !St->isVolatile()) {
+ StVal.getNode()->hasOneUse()) {
SelectionDAG &DAG = DCI.DAG;
DebugLoc DL = St->getDebugLoc();
SDValue BasePtr = St->getBasePtr();
StVal.getNode()->getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return SDValue();
+ // Bitcast an i64 store extracted from a vector to f64.
+ // Otherwise, the i64 value will be legalized to a pair of i32 values.
SelectionDAG &DAG = DCI.DAG;
DebugLoc dl = StVal.getDebugLoc();
SDValue IntVec = StVal.getOperand(0);
static SDValue PerformShiftCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
+ if (N->getOpcode() == ISD::SRL && VT == MVT::i32 && ST->hasV6Ops()) {
+ // Canonicalize (srl (bswap x), 16) to (rotr (bswap x), 16) if the high
+ // 16-bits of x is zero. This optimizes rev + lsr 16 to rev16.
+ SDValue N1 = N->getOperand(1);
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N1)) {
+ SDValue N0 = N->getOperand(0);
+ if (C->getZExtValue() == 16 && N0.getOpcode() == ISD::BSWAP &&
+ DAG.MaskedValueIsZero(N0.getOperand(0),
+ APInt::getHighBitsSet(32, 16)))
+ return DAG.getNode(ISD::ROTR, N->getDebugLoc(), VT, N0, N1);
+ }
+ }
// Nothing to be done for scalar shifts.
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (Res.getNode()) {
APInt KnownZero, KnownOne;
- APInt Mask = APInt::getAllOnesValue(VT.getScalarType().getSizeInBits());
- DAG.ComputeMaskedBits(SDValue(N,0), Mask, KnownZero, KnownOne);
+ DAG.ComputeMaskedBits(SDValue(N,0), KnownZero, KnownOne);
// Capture demanded bits information that would be otherwise lost.
if (KnownZero == 0xfffffffe)
Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
case ISD::SUB: return PerformSUBCombine(N, DCI);
case ISD::MUL: return PerformMULCombine(N, DCI, Subtarget);
case ISD::OR: return PerformORCombine(N, DCI, Subtarget);
- case ISD::AND: return PerformANDCombine(N, DCI);
+ case ISD::XOR: return PerformXORCombine(N, DCI, Subtarget);
+ case ISD::AND: return PerformANDCombine(N, DCI, Subtarget);
case ARMISD::BFI: return PerformBFICombine(N, DCI);
case ARMISD::VMOVRRD: return PerformVMOVRRDCombine(N, DCI);
case ARMISD::VMOVDRR: return PerformVMOVDRRCombine(N, DCI.DAG);
/// a register against the immediate without having to materialize the
/// immediate into a register.
bool ARMTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
+ // Thumb2 and ARM modes can use cmn for negative immediates.
if (!Subtarget->isThumb())
- return ARM_AM::getSOImmVal(Imm) != -1;
+ return ARM_AM::getSOImmVal(llvm::abs64(Imm)) != -1;
if (Subtarget->isThumb2())
- return ARM_AM::getT2SOImmVal(Imm) != -1;
+ return ARM_AM::getT2SOImmVal(llvm::abs64(Imm)) != -1;
+ // Thumb1 doesn't have cmn, and only 8-bit immediates.
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,
}
void ARMTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
- const APInt &Mask,
APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
- KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
+ KnownZero = KnownOne = APInt(KnownOne.getBitWidth(), 0);
switch (Op.getOpcode()) {
default: break;
case ARMISD::CMOV: {
// Bits are known zero/one if known on the LHS and RHS.
- DAG.ComputeMaskedBits(Op.getOperand(0), Mask, KnownZero, KnownOne, Depth+1);
+ DAG.ComputeMaskedBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
if (KnownZero == 0 && KnownOne == 0) return;
APInt KnownZeroRHS, KnownOneRHS;
- DAG.ComputeMaskedBits(Op.getOperand(1), Mask,
- KnownZeroRHS, KnownOneRHS, Depth+1);
+ DAG.ComputeMaskedBits(Op.getOperand(1), KnownZeroRHS, KnownOneRHS, Depth+1);
KnownZero &= KnownZeroRHS;
KnownOne &= KnownOneRHS;
return;
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);
}