#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
+#include "llvm/MC/MCSectionMachO.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/Support/CommandLine.h"
#include <sstream>
using namespace llvm;
+static cl::opt<bool>
+EnableARMLongCalls("arm-long-calls", cl::Hidden,
+ cl::desc("Generate calls via indirect call instructions."),
+ cl::init(false));
+
static bool CC_ARM_APCS_Custom_f64(unsigned &ValNo, EVT &ValVT, EVT &LocVT,
CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags,
setOperationAction(ISD::VECTOR_SHUFFLE, VT.getSimpleVT(), Custom);
setOperationAction(ISD::CONCAT_VECTORS, VT.getSimpleVT(), Custom);
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand);
+ setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand);
if (VT.isInteger()) {
setOperationAction(ISD::SHL, VT.getSimpleVT(), Custom);
setOperationAction(ISD::SRA, VT.getSimpleVT(), Custom);
static TargetLoweringObjectFile *createTLOF(TargetMachine &TM) {
if (TM.getSubtarget<ARMSubtarget>().isTargetDarwin())
return new TargetLoweringObjectFileMachO();
+
return new ARMElfTargetObjectFile();
}
setTargetDAGCombine(ISD::SIGN_EXTEND);
setTargetDAGCombine(ISD::ZERO_EXTEND);
setTargetDAGCombine(ISD::ANY_EXTEND);
+ setTargetDAGCombine(ISD::SELECT_CC);
}
computeRegisterProperties();
// ARM does not have ROTL.
setOperationAction(ISD::ROTL, MVT::i32, Expand);
- setOperationAction(ISD::CTTZ, MVT::i32, Expand);
+ setOperationAction(ISD::CTTZ, MVT::i32, Custom);
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
if (!Subtarget->hasV5TOps() || Subtarget->isThumb1Only())
setOperationAction(ISD::CTLZ, MVT::i32, Expand);
// FIXME: Shouldn't need this, since no register is used, but the legalizer
// doesn't yet know how to not do that for SjLj.
setExceptionSelectorRegister(ARM::R0);
- if (Subtarget->isThumb())
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
- else
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom);
if (!Subtarget->hasV6Ops() && !Subtarget->isThumb2()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
if (!UseSoftFloat && Subtarget->hasVFP2() && !Subtarget->isThumb1Only())
- // Turn f64->i64 into VMOVRRD, i64 -> f64 to VMOVDRR iff target supports vfp2.
+ // Turn f64->i64 into VMOVRRD, i64 -> f64 to VMOVDRR
+ // iff target supports vfp2.
setOperationAction(ISD::BIT_CONVERT, MVT::i64, Custom);
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::FPOW, MVT::f64, Expand);
setOperationAction(ISD::FPOW, MVT::f32, Expand);
- // int <-> fp are custom expanded into bit_convert + ARMISD ops.
- if (!UseSoftFloat && Subtarget->hasVFP2() && !Subtarget->isThumb1Only()) {
- setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
- setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
- setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
+ // Various VFP goodness
+ if (!UseSoftFloat && !Subtarget->isThumb1Only()) {
+ // int <-> fp are custom expanded into bit_convert + ARMISD ops.
+ if (Subtarget->hasVFP2()) {
+ setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
+ setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
+ setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
+ setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
+ }
+ // Special handling for half-precision FP.
+ if (!Subtarget->hasFP16()) {
+ setOperationAction(ISD::FP16_TO_FP32, MVT::f32, Expand);
+ setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand);
+ }
}
// We have target-specific dag combine patterns for the following nodes:
// Generic (and overly aggressive) if-conversion limits.
setIfCvtBlockSizeLimit(10);
setIfCvtDupBlockSizeLimit(2);
+ } else if (Subtarget->hasV7Ops()) {
+ setIfCvtBlockSizeLimit(3);
+ setIfCvtDupBlockSizeLimit(1);
} else if (Subtarget->hasV6Ops()) {
setIfCvtBlockSizeLimit(2);
setIfCvtDupBlockSizeLimit(1);
case ARMISD::CMOV: return "ARMISD::CMOV";
case ARMISD::CNEG: return "ARMISD::CNEG";
+ case ARMISD::RBIT: return "ARMISD::RBIT";
+
case ARMISD::FTOSI: return "ARMISD::FTOSI";
case ARMISD::FTOUI: return "ARMISD::FTOUI";
case ARMISD::SITOF: return "ARMISD::SITOF";
case ARMISD::VZIP: return "ARMISD::VZIP";
case ARMISD::VUZP: return "ARMISD::VUZP";
case ARMISD::VTRN: return "ARMISD::VTRN";
+ case ARMISD::FMAX: return "ARMISD::FMAX";
+ case ARMISD::FMIN: return "ARMISD::FMIN";
}
}
DebugLoc dl) {
SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i32);
return DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, Flags.getByValAlign(),
- /*AlwaysInline=*/false, NULL, 0, NULL, 0);
+ /*isVolatile=*/false, /*AlwaysInline=*/false,
+ NULL, 0, NULL, 0);
}
/// LowerMemOpCallTo - Store the argument to the stack.
return CreateCopyOfByValArgument(Arg, PtrOff, Chain, Flags, DAG, dl);
}
return DAG.getStore(Chain, dl, Arg, PtrOff,
- PseudoSourceValue::getStack(), LocMemOffset);
+ PseudoSourceValue::getStack(), LocMemOffset,
+ false, false, 0);
}
void ARMTargetLowering::PassF64ArgInRegs(DebugLoc dl, SelectionDAG &DAG,
SDValue
ARMTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
CallingConv::ID CallConv, bool isVarArg,
- bool isTailCall,
+ bool &isTailCall,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) {
+ // ARM target does not yet support tail call optimization.
+ isTailCall = false;
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
// These operations are automatically eliminated by the prolog/epilog pass
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));
- SDValue StackPtr = DAG.getRegister(ARM::SP, MVT::i32);
+ SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());
RegsToPassVector RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
VA, ArgLocs[++i], StackPtr, MemOpChains, Flags);
} else {
assert(VA.isMemLoc());
- if (StackPtr.getNode() == 0)
- StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Op1,
dl, DAG, VA, Flags));
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
assert(VA.isMemLoc());
- if (StackPtr.getNode() == 0)
- StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
dl, DAG, VA, Flags));
bool isLocalARMFunc = false;
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
- if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
- GlobalValue *GV = G->getGlobal();
+
+ if (EnableARMLongCalls) {
+ assert (getTargetMachine().getRelocationModel() == Reloc::Static
+ && "long-calls with non-static relocation model!");
+ // Handle a global address or an external symbol. If it's not one of
+ // those, the target's already in a register, so we don't need to do
+ // anything extra.
+ if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
+ const GlobalValue *GV = G->getGlobal();
+ // Create a constant pool entry for the callee address
+ unsigned ARMPCLabelIndex = AFI->createConstPoolEntryUId();
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV,
+ ARMPCLabelIndex,
+ ARMCP::CPValue, 0);
+ // Get the address of the callee into a register
+ SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
+ CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
+ Callee = DAG.getLoad(getPointerTy(), dl,
+ DAG.getEntryNode(), CPAddr,
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
+ } else if (ExternalSymbolSDNode *S=dyn_cast<ExternalSymbolSDNode>(Callee)) {
+ const char *Sym = S->getSymbol();
+
+ // Create a constant pool entry for the callee address
+ unsigned ARMPCLabelIndex = AFI->createConstPoolEntryUId();
+ ARMConstantPoolValue *CPV = new ARMConstantPoolValue(*DAG.getContext(),
+ Sym, ARMPCLabelIndex, 0);
+ // Get the address of the callee into a register
+ SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
+ CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
+ Callee = DAG.getLoad(getPointerTy(), dl,
+ DAG.getEntryNode(), CPAddr,
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
+ }
+ } else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
+ const GlobalValue *GV = G->getGlobal();
isDirect = true;
bool isExt = GV->isDeclaration() || GV->isWeakForLinker();
bool isStub = (isExt && Subtarget->isTargetDarwin()) &&
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
getPointerTy(), Callee, PICLabel);
- } else
+ } else
Callee = DAG.getTargetGlobalAddress(GV, getPointerTy());
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
isDirect = true;
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
getPointerTy(), Callee, PICLabel);
unsigned ARMPCLabelIndex = 0;
DebugLoc DL = Op.getDebugLoc();
EVT PtrVT = getPointerTy();
- BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
+ const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
SDValue CPAddr;
if (RelocM == Reloc::Static) {
}
CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
if (RelocM == Reloc::Static)
return Result;
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
SDValue Argument = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument);
Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue Chain = Argument.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
SDValue
ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA,
SelectionDAG &DAG) {
- GlobalValue *GV = GA->getGlobal();
+ const GlobalValue *GV = GA->getGlobal();
DebugLoc dl = GA->getDebugLoc();
SDValue Offset;
SDValue Chain = DAG.getEntryNode();
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
Chain = Offset.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Offset = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Offset, PICLabel);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
} else {
// local exec model
ARMConstantPoolValue *CPV = new ARMConstantPoolValue(GV, "tpoff");
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
}
// The address of the thread local variable is the add of the thread
SelectionDAG &DAG) {
EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
- GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
+ const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
if (RelocM == Reloc::PIC_) {
bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue Chain = Result.getValue(1);
SDValue GOT = DAG.getGLOBAL_OFFSET_TABLE(PtrVT);
Result = DAG.getNode(ISD::ADD, dl, PtrVT, Result, GOT);
if (!UseGOTOFF)
Result = DAG.getLoad(PtrVT, dl, Chain, Result,
- PseudoSourceValue::getGOT(), 0);
+ PseudoSourceValue::getGOT(), 0,
+ false, false, 0);
return Result;
} else {
// If we have T2 ops, we can materialize the address directly via movt/movw
SDValue CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
}
}
}
unsigned ARMPCLabelIndex = 0;
EVT PtrVT = getPointerTy();
DebugLoc dl = Op.getDebugLoc();
- GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
+ const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
SDValue CPAddr;
if (RelocM == Reloc::Static)
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue Chain = Result.getValue(1);
if (RelocM == Reloc::PIC_) {
if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
Result = DAG.getLoad(PtrVT, dl, Chain, Result,
- PseudoSourceValue::getGOT(), 0);
+ PseudoSourceValue::getGOT(), 0,
+ false, false, 0);
return Result;
}
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
SDValue
-ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
+ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
+ const ARMSubtarget *Subtarget) {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
DebugLoc dl = Op.getDebugLoc();
switch (IntNo) {
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result =
DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
- PseudoSourceValue::getConstantPool(), 0);
+ PseudoSourceValue::getConstantPool(), 0,
+ false, false, 0);
SDValue Chain = Result.getValue(1);
if (RelocM == Reloc::PIC_) {
return Result;
}
case Intrinsic::eh_sjlj_setjmp:
- return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, MVT::i32, Op.getOperand(1));
+ SDValue Val = Subtarget->isThumb() ?
+ DAG.getCopyFromReg(DAG.getEntryNode(), dl, ARM::SP, MVT::i32) :
+ DAG.getConstant(0, MVT::i32);
+ return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl, MVT::i32, Op.getOperand(1),
+ Val);
}
}
-static SDValue LowerMEMBARRIER(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) {
- Res = DAG.getNode(ARMISD::SYNCBARRIER, dl, MVT::Other,
- Op.getOperand(0));
+ 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 {
- Res = DAG.getNode(ARMISD::MEMBARRIER, dl, MVT::Other,
- Op.getOperand(0));
+ 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;
}
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
- return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0);
+ return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0,
+ false, false, 0);
}
SDValue
SDValue ArgValue2;
if (NextVA.isMemLoc()) {
- unsigned ArgSize = NextVA.getLocVT().getSizeInBits()/8;
MachineFrameInfo *MFI = MF.getFrameInfo();
- int FI = MFI->CreateFixedObject(ArgSize, NextVA.getLocMemOffset(),
- true, false);
+ int FI = MFI->CreateFixedObject(4, NextVA.getLocMemOffset(), true, false);
// Create load node to retrieve arguments from the stack.
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue2 = DAG.getLoad(MVT::i32, dl, Root, FIN,
- PseudoSourceValue::getFixedStack(FI), 0);
+ PseudoSourceValue::getFixedStack(FI), 0,
+ false, false, 0);
} else {
Reg = MF.addLiveIn(NextVA.getLocReg(), RC);
ArgValue2 = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);
if (VA.needsCustom()) {
// f64 and vector types are split up into multiple registers or
// combinations of registers and stack slots.
- RegVT = MVT::i32;
-
if (VA.getLocVT() == MVT::v2f64) {
SDValue ArgValue1 = GetF64FormalArgument(VA, ArgLocs[++i],
Chain, DAG, dl);
VA = ArgLocs[++i]; // skip ahead to next loc
- SDValue ArgValue2 = GetF64FormalArgument(VA, ArgLocs[++i],
- Chain, DAG, dl);
+ SDValue ArgValue2;
+ if (VA.isMemLoc()) {
+ int FI = MFI->CreateFixedObject(8, VA.getLocMemOffset(),
+ true, false);
+ SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
+ ArgValue2 = DAG.getLoad(MVT::f64, dl, Chain, FIN,
+ PseudoSourceValue::getFixedStack(FI), 0,
+ false, false, 0);
+ } else {
+ ArgValue2 = GetF64FormalArgument(VA, ArgLocs[++i],
+ Chain, DAG, dl);
+ }
ArgValue = DAG.getNode(ISD::UNDEF, dl, MVT::v2f64);
ArgValue = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64,
ArgValue, ArgValue1, DAG.getIntPtrConstant(0));
// Create load nodes to retrieve arguments from the stack.
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
- PseudoSourceValue::getFixedStack(FI), 0));
+ PseudoSourceValue::getFixedStack(FI), 0,
+ false, false, 0));
}
}
unsigned VReg = MF.addLiveIn(GPRArgRegs[NumGPRs], RC);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, FIN,
- PseudoSourceValue::getFixedStack(VarArgsFrameIndex), 0);
+ PseudoSourceValue::getFixedStack(VarArgsFrameIndex), 0,
+ false, false, 0);
MemOps.push_back(Store);
FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), FIN,
DAG.getConstant(4, getPointerTy()));
if (Op.getOperand(1).getOpcode() == ARMISD::Wrapper) {
SDValue WrapperOp = Op.getOperand(1).getOperand(0);
if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(WrapperOp))
- if (ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
+ if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
return CFP->getValueAPF().isPosZero();
}
}
}
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
Addr = DAG.getLoad((EVT)MVT::i32, dl, Chain, Addr,
- PseudoSourceValue::getJumpTable(), 0);
+ PseudoSourceValue::getJumpTable(), 0,
+ false, false, 0);
Chain = Addr.getValue(1);
Addr = DAG.getNode(ISD::ADD, dl, PTy, Addr, Table);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
} else {
Addr = DAG.getLoad(PTy, dl, Chain, Addr,
- PseudoSourceValue::getJumpTable(), 0);
+ PseudoSourceValue::getJumpTable(), 0, false, false, 0);
Chain = Addr.getValue(1);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI, UId);
}
static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
- unsigned Opc =
- Op.getOpcode() == ISD::FP_TO_SINT ? ARMISD::FTOSI : ARMISD::FTOUI;
+ unsigned Opc;
+
+ switch (Op.getOpcode()) {
+ default:
+ assert(0 && "Invalid opcode!");
+ case ISD::FP_TO_SINT:
+ Opc = ARMISD::FTOSI;
+ break;
+ case ISD::FP_TO_UINT:
+ Opc = ARMISD::FTOUI;
+ break;
+ }
Op = DAG.getNode(Opc, dl, MVT::f32, Op.getOperand(0));
return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i32, Op);
}
static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getValueType();
DebugLoc dl = Op.getDebugLoc();
- unsigned Opc =
- Op.getOpcode() == ISD::SINT_TO_FP ? ARMISD::SITOF : ARMISD::UITOF;
+ unsigned Opc;
+
+ switch (Op.getOpcode()) {
+ default:
+ assert(0 && "Invalid opcode!");
+ case ISD::SINT_TO_FP:
+ Opc = ARMISD::SITOF;
+ break;
+ case ISD::UINT_TO_FP:
+ Opc = ARMISD::UITOF;
+ break;
+ }
Op = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, Op.getOperand(0));
return DAG.getNode(Opc, dl, VT, Op);
? ARM::R7 : ARM::R11;
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, VT);
while (Depth--)
- FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr, NULL, 0);
+ FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr, NULL, 0,
+ false, false, 0);
return FrameAddr;
}
SDValue Chain,
SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
- bool AlwaysInline,
+ bool isVolatile, bool AlwaysInline,
const Value *DstSV, uint64_t DstSVOff,
const Value *SrcSV, uint64_t SrcSVOff){
// Do repeated 4-byte loads and stores. To be improved.
Loads[i] = DAG.getLoad(VT, dl, Chain,
DAG.getNode(ISD::ADD, dl, MVT::i32, Src,
DAG.getConstant(SrcOff, MVT::i32)),
- SrcSV, SrcSVOff + SrcOff);
+ SrcSV, SrcSVOff + SrcOff, isVolatile, false, 0);
TFOps[i] = Loads[i].getValue(1);
SrcOff += VTSize;
}
for (i = 0;
i < MAX_LOADS_IN_LDM && EmittedNumMemOps + i < NumMemOps; ++i) {
TFOps[i] = DAG.getStore(Chain, dl, Loads[i],
- DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
- DAG.getConstant(DstOff, MVT::i32)),
- DstSV, DstSVOff + DstOff);
+ DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
+ DAG.getConstant(DstOff, MVT::i32)),
+ DstSV, DstSVOff + DstOff, isVolatile, false, 0);
DstOff += VTSize;
}
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
Loads[i] = DAG.getLoad(VT, dl, Chain,
DAG.getNode(ISD::ADD, dl, MVT::i32, Src,
DAG.getConstant(SrcOff, MVT::i32)),
- SrcSV, SrcSVOff + SrcOff);
+ SrcSV, SrcSVOff + SrcOff, false, false, 0);
TFOps[i] = Loads[i].getValue(1);
++i;
SrcOff += VTSize;
TFOps[i] = DAG.getStore(Chain, dl, Loads[i],
DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
DAG.getConstant(DstOff, MVT::i32)),
- DstSV, DstSVOff + DstOff);
+ DstSV, DstSVOff + DstOff, false, false, 0);
++i;
DstOff += VTSize;
BytesLeft -= VTSize;
static SDValue ExpandBIT_CONVERT(SDNode *N, SelectionDAG &DAG) {
SDValue Op = N->getOperand(0);
+
+ // Do not create a VMOVDRR or VMOVRRD node if the operand type is not
+ // legal. The legalizer won't know what to do with that.
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ if (!TLI.isTypeLegal(Op.getValueType()))
+ return SDValue();
+
DebugLoc dl = N->getDebugLoc();
if (N->getValueType(0) == MVT::f64) {
// Turn i64->f64 into VMOVDRR.
return DAG.getMergeValues(Ops, 2, dl);
}
+static SDValue LowerCTTZ(SDNode *N, SelectionDAG &DAG,
+ const ARMSubtarget *ST) {
+ EVT VT = N->getValueType(0);
+ DebugLoc dl = N->getDebugLoc();
+
+ if (!ST->hasV6T2Ops())
+ return SDValue();
+
+ SDValue rbit = DAG.getNode(ARMISD::RBIT, dl, VT, N->getOperand(0));
+ return DAG.getNode(ISD::CTLZ, dl, VT, rbit);
+}
+
static SDValue LowerShift(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
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);
+ 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:
case ISD::RETURNADDR: break;
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::GLOBAL_OFFSET_TABLE: return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
- case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+ case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG,
+ Subtarget);
case ISD::BIT_CONVERT: return ExpandBIT_CONVERT(Op.getNode(), DAG);
case ISD::SHL:
case ISD::SRL:
case ISD::SHL_PARTS: return LowerShiftLeftParts(Op, DAG);
case ISD::SRL_PARTS:
case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG);
+ case ISD::CTTZ: return LowerCTTZ(Op.getNode(), DAG, Subtarget);
case ISD::VSETCC: return LowerVSETCC(Op, DAG);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue>&Results,
SelectionDAG &DAG) {
+ SDValue Res;
switch (N->getOpcode()) {
default:
llvm_unreachable("Don't know how to custom expand this!");
- return;
+ break;
case ISD::BIT_CONVERT:
- Results.push_back(ExpandBIT_CONVERT(N, DAG));
- return;
+ Res = ExpandBIT_CONVERT(N, DAG);
+ break;
case ISD::SRL:
- case ISD::SRA: {
- SDValue Res = LowerShift(N, DAG, Subtarget);
- if (Res.getNode())
- Results.push_back(Res);
- return;
- }
+ case ISD::SRA:
+ Res = LowerShift(N, DAG, Subtarget);
+ break;
}
+ if (Res.getNode())
+ Results.push_back(Res);
}
//===----------------------------------------------------------------------===//
// exitMBB:
// ...
BB = exitMBB;
+
+ MF->DeleteMachineInstr(MI); // The instruction is gone now.
+
return BB;
}
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
const BasicBlock *LLVM_BB = BB->getBasicBlock();
- MachineFunction *F = BB->getParent();
+ MachineFunction *MF = BB->getParent();
MachineFunction::iterator It = BB;
++It;
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;
+ strOpc = isThumb2 ? ARM::t2STREXB : ARM::STREXB;
break;
case 2:
ldrOpc = isThumb2 ? ARM::t2LDREXH : ARM::LDREXH;
break;
}
- MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
- MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
- F->insert(It, loopMBB);
- F->insert(It, exitMBB);
+ MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ MF->insert(It, loopMBB);
+ MF->insert(It, exitMBB);
exitMBB->transferSuccessors(BB);
- MachineRegisterInfo &RegInfo = F->getRegInfo();
+ MachineRegisterInfo &RegInfo = MF->getRegInfo();
unsigned scratch = RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
unsigned scratch2 = (!BinOpcode) ? incr :
RegInfo.createVirtualRegister(ARM::GPRRegisterClass);
// fallthrough --> exitMBB
BB = loopMBB;
AddDefaultPred(BuildMI(BB, dl, TII->get(ldrOpc), dest).addReg(ptr));
- if (BinOpcode)
- AddDefaultPred(BuildMI(BB, dl, TII->get(BinOpcode), scratch2).
- addReg(dest).addReg(incr)).addReg(0);
+ 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));
// exitMBB:
// ...
BB = exitMBB;
+
+ MF->DeleteMachineInstr(MI); // The instruction is gone now.
+
return BB;
}
return SDValue();
}
-/// PerformVMOVRRDCombine - Target-specific dag combine xforms for ARMISD::VMOVRRD.
+/// PerformVMOVRRDCombine - Target-specific dag combine xforms for
+/// ARMISD::VMOVRRD.
static SDValue PerformVMOVRRDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
// fmrrd(fmdrr x, y) -> x,y
return SDValue();
}
+/// PerformSELECT_CCCombine - Target-specific DAG combining for ISD::SELECT_CC
+/// to match f32 max/min patterns to use NEON vmax/vmin instructions.
+static SDValue PerformSELECT_CCCombine(SDNode *N, SelectionDAG &DAG,
+ const ARMSubtarget *ST) {
+ // If the target supports NEON, try to use vmax/vmin instructions for f32
+ // selects like "x < y ? x : y". Unless the FiniteOnlyFPMath option is set,
+ // be careful about NaNs: NEON's vmax/vmin return NaN if either operand is
+ // a NaN; only do the transformation when it matches that behavior.
+
+ // For now only do this when using NEON for FP operations; if using VFP, it
+ // is not obvious that the benefit outweighs the cost of switching to the
+ // NEON pipeline.
+ if (!ST->hasNEON() || !ST->useNEONForSinglePrecisionFP() ||
+ N->getValueType(0) != MVT::f32)
+ return SDValue();
+
+ SDValue CondLHS = N->getOperand(0);
+ SDValue CondRHS = N->getOperand(1);
+ SDValue LHS = N->getOperand(2);
+ SDValue RHS = N->getOperand(3);
+ ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(4))->get();
+
+ unsigned Opcode = 0;
+ bool IsReversed;
+ if (DAG.isEqualTo(LHS, CondLHS) && DAG.isEqualTo(RHS, CondRHS)) {
+ IsReversed = false; // x CC y ? x : y
+ } else if (DAG.isEqualTo(LHS, CondRHS) && DAG.isEqualTo(RHS, CondLHS)) {
+ IsReversed = true ; // x CC y ? y : x
+ } else {
+ return SDValue();
+ }
+
+ bool IsUnordered;
+ switch (CC) {
+ default: break;
+ case ISD::SETOLT:
+ case ISD::SETOLE:
+ case ISD::SETLT:
+ case ISD::SETLE:
+ case ISD::SETULT:
+ case ISD::SETULE:
+ // If LHS is NaN, an ordered comparison will be false and the result will
+ // be the RHS, but vmin(NaN, RHS) = NaN. Avoid this by checking that LHS
+ // != NaN. Likewise, for unordered comparisons, check for RHS != NaN.
+ IsUnordered = (CC == ISD::SETULT || CC == ISD::SETULE);
+ if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
+ break;
+ // For less-than-or-equal comparisons, "+0 <= -0" will be true but vmin
+ // will return -0, so vmin can only be used for unsafe math or if one of
+ // the operands is known to be nonzero.
+ if ((CC == ISD::SETLE || CC == ISD::SETOLE || CC == ISD::SETULE) &&
+ !UnsafeFPMath &&
+ !(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
+ break;
+ Opcode = IsReversed ? ARMISD::FMAX : ARMISD::FMIN;
+ break;
+
+ case ISD::SETOGT:
+ case ISD::SETOGE:
+ case ISD::SETGT:
+ case ISD::SETGE:
+ case ISD::SETUGT:
+ case ISD::SETUGE:
+ // If LHS is NaN, an ordered comparison will be false and the result will
+ // be the RHS, but vmax(NaN, RHS) = NaN. Avoid this by checking that LHS
+ // != NaN. Likewise, for unordered comparisons, check for RHS != NaN.
+ IsUnordered = (CC == ISD::SETUGT || CC == ISD::SETUGE);
+ if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
+ break;
+ // For greater-than-or-equal comparisons, "-0 >= +0" will be true but vmax
+ // will return +0, so vmax can only be used for unsafe math or if one of
+ // the operands is known to be nonzero.
+ if ((CC == ISD::SETGE || CC == ISD::SETOGE || CC == ISD::SETUGE) &&
+ !UnsafeFPMath &&
+ !(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
+ break;
+ Opcode = IsReversed ? ARMISD::FMIN : ARMISD::FMAX;
+ break;
+ }
+
+ if (!Opcode)
+ return SDValue();
+ return DAG.getNode(Opcode, N->getDebugLoc(), N->getValueType(0), LHS, RHS);
+}
+
SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
switch (N->getOpcode()) {
default: break;
- case ISD::ADD: return PerformADDCombine(N, DCI);
- case ISD::SUB: return PerformSUBCombine(N, DCI);
+ case ISD::ADD: return PerformADDCombine(N, DCI);
+ case ISD::SUB: return PerformSUBCombine(N, DCI);
case ARMISD::VMOVRRD: return PerformVMOVRRDCombine(N, DCI);
- case ISD::INTRINSIC_WO_CHAIN:
- return PerformIntrinsicCombine(N, DCI.DAG);
+ case ISD::INTRINSIC_WO_CHAIN: return PerformIntrinsicCombine(N, DCI.DAG);
case ISD::SHL:
case ISD::SRA:
- case ISD::SRL:
- return PerformShiftCombine(N, DCI.DAG, Subtarget);
+ case ISD::SRL: return PerformShiftCombine(N, DCI.DAG, Subtarget);
case ISD::SIGN_EXTEND:
case ISD::ZERO_EXTEND:
- case ISD::ANY_EXTEND:
- return PerformExtendCombine(N, DCI.DAG, Subtarget);
+ case ISD::ANY_EXTEND: return PerformExtendCombine(N, DCI.DAG, Subtarget);
+ case ISD::SELECT_CC: return PerformSELECT_CCCombine(N, DCI.DAG, Subtarget);
}
return SDValue();
}
if (!Subtarget->hasV6Ops())
// Pre-v6 does not support unaligned mem access.
return false;
- else if (!Subtarget->hasV6Ops()) {
- // v6 may or may not support unaligned mem access.
+ else {
+ // v6+ may or may not support unaligned mem access depending on the system
+ // configuration.
+ // FIXME: This is pretty conservative. Should we provide cmdline option to
+ // control the behaviour?
if (!Subtarget->isTargetDarwin())
return false;
}
ARMTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
EVT VT) const {
if (Constraint.size() == 1) {
- // GCC RS6000 Constraint Letters
+ // GCC ARM Constraint Letters
switch (Constraint[0]) {
case 'l':
- if (Subtarget->isThumb1Only())
+ if (Subtarget->isThumb())
return std::make_pair(0U, ARM::tGPRRegisterClass);
else
return std::make_pair(0U, ARM::GPRRegisterClass);
case 'w':
if (VT == MVT::f32)
return std::make_pair(0U, ARM::SPRRegisterClass);
- if (VT == MVT::f64)
+ if (VT.getSizeInBits() == 64)
return std::make_pair(0U, ARM::DPRRegisterClass);
if (VT.getSizeInBits() == 128)
return std::make_pair(0U, ARM::QPRRegisterClass);
break;
}
}
+ if (StringRef("{cc}").equals_lower(Constraint))
+ return std::make_pair(0U, ARM::CCRRegisterClass);
+
return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
}
ARM::S20,ARM::S21,ARM::S22,ARM::S23,
ARM::S24,ARM::S25,ARM::S26,ARM::S27,
ARM::S28,ARM::S29,ARM::S30,ARM::S31, 0);
- if (VT == MVT::f64)
+ if (VT.getSizeInBits() == 64)
return make_vector<unsigned>(ARM::D0, ARM::D1, ARM::D2, ARM::D3,
ARM::D4, ARM::D5, ARM::D6, ARM::D7,
ARM::D8, ARM::D9, ARM::D10,ARM::D11,