#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
class ARMCCState : public CCState {
public:
ARMCCState(CallingConv::ID CC, bool isVarArg, MachineFunction &MF,
- const TargetMachine &TM, SmallVectorImpl<CCValAssign> &locs,
- LLVMContext &C, ParmContext PC)
- : CCState(CC, isVarArg, MF, TM, locs, C) {
+ SmallVectorImpl<CCValAssign> &locs, LLVMContext &C,
+ ParmContext PC)
+ : CCState(CC, isVarArg, MF, locs, C) {
assert(((PC == Call) || (PC == Prologue)) &&
"ARMCCState users must specify whether their context is call"
"or prologue generation.");
ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
: TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
Subtarget = &TM.getSubtarget<ARMSubtarget>();
- RegInfo = TM.getRegisterInfo();
- Itins = TM.getInstrItineraryData();
+ RegInfo = TM.getSubtargetImpl()->getRegisterInfo();
+ Itins = TM.getSubtargetImpl()->getInstrItineraryData();
setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
// Conversions between floating types.
// RTABI chapter 4.1.2, Table 7
{ RTLIB::FPROUND_F64_F32, "__aeabi_d2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::FPROUND_F64_F16, "__aeabi_d2h", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPEXT_F32_F64, "__aeabi_f2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
// Integer to floating-point conversions.
setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4");
}
+ // The half <-> float conversion functions are always soft-float, but are
+ // needed for some targets which use a hard-float calling convention by
+ // default.
+ if (Subtarget->isAAPCS_ABI()) {
+ setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_AAPCS);
+ } else {
+ setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_APCS);
+ setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_APCS);
+ setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_APCS);
+ }
+
if (Subtarget->isThumb1Only())
addRegisterClass(MVT::i32, &ARM::tGPRRegClass);
else
if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
!Subtarget->isThumb1Only()) {
addRegisterClass(MVT::f32, &ARM::SPRRegClass);
- if (!Subtarget->isFPOnlySP())
- addRegisterClass(MVT::f64, &ARM::DPRRegClass);
-
- setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+ addRegisterClass(MVT::f64, &ARM::DPRRegClass);
}
for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
if (!Subtarget->isThumb1Only())
setTargetDAGCombine(ISD::ADDC);
+ if (Subtarget->isFPOnlySP()) {
+ // When targetting a floating-point unit with only single-precision
+ // operations, f64 is legal for the few double-precision instructions which
+ // are present However, no double-precision operations other than moves,
+ // loads and stores are provided by the hardware.
+ setOperationAction(ISD::FADD, MVT::f64, Expand);
+ setOperationAction(ISD::FSUB, MVT::f64, Expand);
+ setOperationAction(ISD::FMUL, MVT::f64, Expand);
+ setOperationAction(ISD::FMA, MVT::f64, Expand);
+ setOperationAction(ISD::FDIV, MVT::f64, Expand);
+ setOperationAction(ISD::FREM, MVT::f64, Expand);
+ setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
+ setOperationAction(ISD::FGETSIGN, MVT::f64, Expand);
+ setOperationAction(ISD::FNEG, MVT::f64, Expand);
+ setOperationAction(ISD::FABS, MVT::f64, Expand);
+ setOperationAction(ISD::FSQRT, MVT::f64, Expand);
+ setOperationAction(ISD::FSIN, MVT::f64, Expand);
+ setOperationAction(ISD::FCOS, MVT::f64, Expand);
+ setOperationAction(ISD::FPOWI, MVT::f64, Expand);
+ setOperationAction(ISD::FPOW, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG2, MVT::f64, Expand);
+ setOperationAction(ISD::FLOG10, MVT::f64, Expand);
+ setOperationAction(ISD::FEXP, MVT::f64, Expand);
+ setOperationAction(ISD::FEXP2, MVT::f64, Expand);
+ setOperationAction(ISD::FCEIL, MVT::f64, Expand);
+ setOperationAction(ISD::FTRUNC, MVT::f64, Expand);
+ setOperationAction(ISD::FRINT, MVT::f64, Expand);
+ setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand);
+ setOperationAction(ISD::FFLOOR, MVT::f64, Expand);
+ setOperationAction(ISD::FP_ROUND, MVT::f32, Custom);
+ setOperationAction(ISD::FP_EXTEND, MVT::f64, Custom);
+ }
computeRegisterProperties();
- // ARM does not have f32 extending load.
+ // ARM does not have floating-point extending loads.
setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+
+ // ... or truncating stores
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+ setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+ setTruncStoreAction(MVT::f64, MVT::f16, Expand);
// ARM does not have i1 sign extending load.
setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
setExceptionSelectorRegister(ARM::R1);
}
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
+ if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
+ else
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
+
// ARMv6 Thumb1 (except for CPUs that support dmb / dsb) and earlier use
- // the default expansion.
- if (Subtarget->hasAnyDataBarrier() && !Subtarget->isThumb1Only()) {
+ // the default expansion. If we are targeting a single threaded system,
+ // then set them all for expand so we can lower them later into their
+ // non-atomic form.
+ if (TM.Options.ThreadModel == ThreadModel::Single)
+ setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand);
+ else if (Subtarget->hasAnyDataBarrier() && !Subtarget->isThumb1Only()) {
// ATOMIC_FENCE needs custom lowering; the others should have been expanded
// to ldrex/strex loops already.
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
// On v8, we have particularly efficient implementations of atomic fences
// if they can be combined with nearby atomic loads and stores.
if (!Subtarget->hasV8Ops()) {
- // Automatically insert fences (dmb ist) around ATOMIC_SWAP etc.
+ // Automatically insert fences (dmb ish) around ATOMIC_SWAP etc.
setInsertFencesForAtomic(true);
}
} else {
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
}
- // Special handling for half-precision FP.
+
+ // v8 adds f64 <-> f16 conversion. Before that it should be expanded.
+ if (!Subtarget->hasV8Ops()) {
+ setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
+ setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
+ }
+
+ // fp16 is a special v7 extension that adds f16 <-> f32 conversions.
if (!Subtarget->hasFP16()) {
- setOperationAction(ISD::FP16_TO_FP32, MVT::f32, Expand);
- setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand);
+ setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
+ setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
}
}
}
}
+ // ARMv8 implements a lot of rounding-like FP operations.
+ if (Subtarget->hasV8Ops()) {
+ static MVT RoundingTypes[] = {MVT::f32, MVT::f64};
+ for (const auto Ty : RoundingTypes) {
+ setOperationAction(ISD::FFLOOR, Ty, Legal);
+ setOperationAction(ISD::FCEIL, Ty, Legal);
+ setOperationAction(ISD::FROUND, Ty, Legal);
+ setOperationAction(ISD::FTRUNC, Ty, Legal);
+ setOperationAction(ISD::FNEARBYINT, Ty, Legal);
+ setOperationAction(ISD::FRINT, Ty, Legal);
+ }
+ }
// We have target-specific dag combine patterns for the following nodes:
// ARMISD::VMOVRRD - No need to call setTargetDAGCombine
setTargetDAGCombine(ISD::ADD);
case ARMISD::PRELOAD: return "ARMISD::PRELOAD";
+ case ARMISD::WIN__CHKSTK: return "ARMISD:::WIN__CHKSTK";
+
case ARMISD::VCEQ: return "ARMISD::VCEQ";
case ARMISD::VCEQZ: return "ARMISD::VCEQZ";
case ARMISD::VCGE: return "ARMISD::VCGE";
// Load are scheduled for latency even if there instruction itinerary
// is not available.
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
const MCInstrDesc &MCID = TII->get(N->getMachineOpcode());
if (MCID.getNumDefs() == 0)
case CallingConv::C:
if (!Subtarget->isAAPCS_ABI())
return CallingConv::ARM_APCS;
- else if (Subtarget->hasVFP2() &&
+ else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() &&
getTargetMachine().Options.FloatABIType == FloatABI::Hard &&
!isVarArg)
return CallingConv::ARM_AAPCS_VFP;
return CallingConv::ARM_AAPCS;
case CallingConv::Fast:
if (!Subtarget->isAAPCS_ABI()) {
- if (Subtarget->hasVFP2() && !isVarArg)
+ if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::Fast;
return CallingConv::ARM_APCS;
- } else if (Subtarget->hasVFP2() && !isVarArg)
+ } else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::ARM_AAPCS_VFP;
else
return CallingConv::ARM_AAPCS;
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
- ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext(), Call);
+ ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
+ *DAG.getContext(), Call);
CCInfo.AnalyzeCallResult(Ins,
CCAssignFnForNode(CallConv, /* Return*/ true,
isVarArg));
// Analyze operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
- ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ArgLocs, *DAG.getContext(), Call);
+ ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
+ *DAG.getContext(), Call);
CCInfo.AnalyzeCallOperands(Outs,
CCAssignFnForNode(CallConv, /* Return*/ false,
isVarArg));
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
if (EnableARMLongCalls) {
- assert (getTargetMachine().getRelocationModel() == Reloc::Static
- && "long-calls with non-static relocation model!");
+ assert((Subtarget->isTargetWindows() ||
+ 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.
bool isExt = GV->isDeclaration() || GV->isWeakForLinker();
bool isStub = (isExt && Subtarget->isTargetMachO()) &&
getTargetMachine().getRelocationModel() != Reloc::Static;
- isARMFunc = !Subtarget->isThumb() || isStub;
+ isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
// ARM call to a local ARM function is predicable.
isLocalARMFunc = !Subtarget->isThumb() && (!isExt || !ARMInterworking);
// tBX takes a register source operand.
if (isStub && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
assert(Subtarget->isTargetMachO() && "WrapperPIC use on non-MachO?");
Callee = DAG.getNode(ARMISD::WrapperPIC, dl, getPointerTy(),
- DAG.getTargetGlobalAddress(GV, dl, getPointerTy()));
+ DAG.getTargetGlobalAddress(GV, dl, getPointerTy(),
+ 0, ARMII::MO_NONLAZY));
+ Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Callee,
+ MachinePointerInfo::getGOT(), false, false, true, 0);
+ } else if (Subtarget->isTargetCOFF()) {
+ assert(Subtarget->isTargetWindows() &&
+ "Windows is the only supported COFF target");
+ unsigned TargetFlags = GV->hasDLLImportStorageClass()
+ ? ARMII::MO_DLLIMPORT
+ : ARMII::MO_NO_FLAG;
+ Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), /*Offset=*/0,
+ TargetFlags);
+ if (GV->hasDLLImportStorageClass())
+ Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
+ DAG.getNode(ARMISD::Wrapper, dl, getPointerTy(),
+ Callee), MachinePointerInfo::getGOT(),
+ false, false, false, 0);
} else {
// On ELF targets for PIC code, direct calls should go through the PLT
unsigned OpFlags = 0;
isDirect = true;
bool isStub = Subtarget->isTargetMachO() &&
getTargetMachine().getRelocationModel() != Reloc::Static;
- isARMFunc = !Subtarget->isThumb() || isStub;
+ isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
// tBX takes a register source operand.
const char *Sym = S->getSymbol();
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
// FIXME: handle tail calls differently.
unsigned CallOpc;
- bool HasMinSizeAttr = Subtarget->isMinSize();
+ bool HasMinSizeAttr = MF.getFunction()->getAttributes().hasAttribute(
+ AttributeSet::FunctionIndex, Attribute::MinSize);
if (Subtarget->isThumb()) {
if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
CallOpc = ARMISD::CALL_NOLINK;
// Add a register mask operand representing the call-preserved registers.
if (!isTailCall) {
const uint32_t *Mask;
- const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+ const TargetRegisterInfo *TRI =
+ getTargetMachine().getSubtargetImpl()->getRegisterInfo();
const ARMBaseRegisterInfo *ARI = static_cast<const ARMBaseRegisterInfo*>(TRI);
if (isThisReturn) {
// For 'this' returns, use the R0-preserving mask if applicable
if (Subtarget->isThumb1Only())
return false;
+ // Externally-defined functions with weak linkage should not be
+ // tail-called on ARM when the OS does not support dynamic
+ // pre-emption of symbols, as the AAELF spec requires normal calls
+ // to undefined weak functions to be replaced with a NOP or jump to the
+ // next instruction. The behaviour of branch instructions in this
+ // situation (as used for tail calls) is implementation-defined, so we
+ // cannot rely on the linker replacing the tail call with a return.
+ if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
+ const GlobalValue *GV = G->getGlobal();
+ if (GV->hasExternalWeakLinkage())
+ return false;
+ }
+
// If the calling conventions do not match, then we'd better make sure the
// results are returned in the same way as what the caller expects.
if (!CCMatch) {
SmallVector<CCValAssign, 16> RVLocs1;
- ARMCCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs1, *DAG.getContext(), Call);
+ ARMCCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(), RVLocs1,
+ *DAG.getContext(), Call);
CCInfo1.AnalyzeCallResult(Ins, CCAssignFnForNode(CalleeCC, true, isVarArg));
SmallVector<CCValAssign, 16> RVLocs2;
- ARMCCState CCInfo2(CallerCC, false, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs2, *DAG.getContext(), Call);
+ ARMCCState CCInfo2(CallerCC, false, DAG.getMachineFunction(), RVLocs2,
+ *DAG.getContext(), Call);
CCInfo2.AnalyzeCallResult(Ins, CCAssignFnForNode(CallerCC, true, isVarArg));
if (RVLocs1.size() != RVLocs2.size())
// Check if stack adjustment is needed. For now, do not do this if any
// argument is passed on the stack.
SmallVector<CCValAssign, 16> ArgLocs;
- ARMCCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ArgLocs, *DAG.getContext(), Call);
+ ARMCCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(), ArgLocs,
+ *DAG.getContext(), Call);
CCInfo.AnalyzeCallOperands(Outs,
CCAssignFnForNode(CalleeCC, false, isVarArg));
if (CCInfo.getNextStackOffset()) {
// the caller's fixed stack objects.
MachineFrameInfo *MFI = MF.getFrameInfo();
const MachineRegisterInfo *MRI = &MF.getRegInfo();
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
i != e;
++i, ++realArgIdx) {
const SmallVectorImpl<ISD::OutputArg> &Outs,
LLVMContext &Context) const {
SmallVector<CCValAssign, 16> RVLocs;
- CCState CCInfo(CallConv, isVarArg, MF, getTargetMachine(), RVLocs, Context);
+ CCState CCInfo(CallConv, isVarArg, MF, RVLocs, Context);
return CCInfo.CheckReturn(Outs, CCAssignFnForNode(CallConv, /*Return=*/true,
isVarArg));
}
SmallVector<CCValAssign, 16> RVLocs;
// CCState - Info about the registers and stack slots.
- ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), RVLocs, *DAG.getContext(), Call);
+ ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
+ *DAG.getContext(), Call);
// Analyze outgoing return values.
CCInfo.AnalyzeReturn(Outs, CCAssignFnForNode(CallConv, /* Return */ true,
RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
bool isLittleEndian = Subtarget->isLittle();
+ MachineFunction &MF = DAG.getMachineFunction();
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ AFI->setReturnRegsCount(RVLocs.size());
+
// Copy the result values into the output registers.
for (unsigned i = 0, realRVLocIdx = 0;
i != RVLocs.size();
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(CallingConv::C, Type::getInt32Ty(*DAG.getContext()),
- DAG.getExternalSymbol("__tls_get_addr", PtrVT), &Args, 0);
+ DAG.getExternalSymbol("__tls_get_addr", PtrVT), std::move(Args),
+ 0);
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
return CallResult.first;
// If we have T2 ops, we can materialize the address directly via movt/movw
// pair. This is always cheaper.
- if (Subtarget->useMovt()) {
+ if (Subtarget->useMovt(DAG.getMachineFunction())) {
++NumMovwMovt;
// FIXME: Once remat is capable of dealing with instructions with register
// operands, expand this into two nodes.
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
- if (Subtarget->useMovt())
+ if (Subtarget->useMovt(DAG.getMachineFunction()))
++NumMovwMovt;
// FIXME: Once remat is capable of dealing with instructions with register
SDValue ARMTargetLowering::LowerGlobalAddressWindows(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget->isTargetWindows() && "non-Windows COFF is not supported");
- assert(Subtarget->useMovt() && "Windows on ARM expects to use movw/movt");
+ assert(Subtarget->useMovt(DAG.getMachineFunction()) &&
+ "Windows on ARM expects to use movw/movt");
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
+ const ARMII::TOF TargetFlags =
+ (GV->hasDLLImportStorageClass() ? ARMII::MO_DLLIMPORT : ARMII::MO_NO_FLAG);
EVT PtrVT = getPointerTy();
+ SDValue Result;
SDLoc DL(Op);
++NumMovwMovt;
// FIXME: Once remat is capable of dealing with instructions with register
// operands, expand this into two nodes.
- return DAG.getNode(ARMISD::Wrapper, DL, PtrVT,
- DAG.getTargetGlobalAddress(GV, DL, PtrVT));
+ Result = DAG.getNode(ARMISD::Wrapper, DL, PtrVT,
+ DAG.getTargetGlobalAddress(GV, DL, PtrVT, /*Offset=*/0,
+ TargetFlags));
+ if (GV->hasDLLImportStorageClass())
+ Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), Result,
+ MachinePointerInfo::getGOT(), false, false, false, 0);
+ return Result;
}
SDValue ARMTargetLowering::LowerGLOBAL_OFFSET_TABLE(SDValue Op,
SDLoc dl(Op);
switch (IntNo) {
default: return SDValue(); // Don't custom lower most intrinsics.
+ case Intrinsic::arm_rbit: {
+ assert(Op.getOperand(1).getValueType() == MVT::i32 &&
+ "RBIT intrinsic must have i32 type!");
+ return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(1));
+ }
case Intrinsic::arm_thread_pointer: {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
return DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
NumGPRs = (firstUnalloced <= 3) ? (4 - firstUnalloced) : 0;
}
- unsigned Align = MF.getTarget().getFrameLowering()->getStackAlignment();
+ unsigned Align = MF.getTarget()
+ .getSubtargetImpl()
+ ->getFrameLowering()
+ ->getStackAlignment();
ArgRegsSize = NumGPRs * 4;
// If parameter is split between stack and GPRs...
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
- ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
- getTargetMachine(), ArgLocs, *DAG.getContext(), Prologue);
+ ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
+ *DAG.getContext(), Prologue);
CCInfo.AnalyzeFormalArguments(Ins,
CCAssignFnForNode(CallConv, /* Return*/ false,
isVarArg));
}
CCInfo.rewindByValRegsInfo();
lastInsIndex = -1;
- if (isVarArg) {
+ if (isVarArg && MFI->hasVAStart()) {
unsigned ExtraArgRegsSize;
unsigned ExtraArgRegsSaveSize;
computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsCount(), 0,
}
// varargs
- if (isVarArg)
+ if (isVarArg && MFI->hasVAStart())
VarArgStyleRegisters(CCInfo, DAG, dl, Chain,
CCInfo.getNextStackOffset(),
TotalArgRegsSaveSize);
SDValue
ARMTargetLowering::getVFPCmp(SDValue LHS, SDValue RHS, SelectionDAG &DAG,
SDLoc dl) const {
+ assert(!Subtarget->isFPOnlySP() || RHS.getValueType() != MVT::f64);
SDValue Cmp;
if (!isFloatingPointZero(RHS))
Cmp = DAG.getNode(ARMISD::CMPFP, dl, MVT::Glue, LHS, RHS);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
EVT VT = Op.getValueType();
- return DAG.getNode(ARMISD::CMOV, SDLoc(Op), VT, SelectTrue, SelectFalse,
- ARMcc, CCR, OverflowCmp);
-
+ return getCMOV(SDLoc(Op), VT, SelectTrue, SelectFalse, ARMcc, CCR,
+ OverflowCmp, DAG);
}
// Convert:
SDValue CCR = Cond.getOperand(3);
SDValue Cmp = duplicateCmp(Cond.getOperand(4), DAG);
assert(True.getValueType() == VT);
- return DAG.getNode(ARMISD::CMOV, dl, VT, True, False, ARMcc, CCR, Cmp);
+ return getCMOV(dl, VT, True, False, ARMcc, CCR, Cmp, DAG);
}
}
}
}
}
+SDValue ARMTargetLowering::getCMOV(SDLoc dl, EVT VT, SDValue FalseVal,
+ SDValue TrueVal, SDValue ARMcc, SDValue CCR,
+ SDValue Cmp, SelectionDAG &DAG) const {
+ if (Subtarget->isFPOnlySP() && VT == MVT::f64) {
+ FalseVal = DAG.getNode(ARMISD::VMOVRRD, dl,
+ DAG.getVTList(MVT::i32, MVT::i32), FalseVal);
+ TrueVal = DAG.getNode(ARMISD::VMOVRRD, dl,
+ DAG.getVTList(MVT::i32, MVT::i32), TrueVal);
+
+ SDValue TrueLow = TrueVal.getValue(0);
+ SDValue TrueHigh = TrueVal.getValue(1);
+ SDValue FalseLow = FalseVal.getValue(0);
+ SDValue FalseHigh = FalseVal.getValue(1);
+
+ SDValue Low = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseLow, TrueLow,
+ ARMcc, CCR, Cmp);
+ SDValue High = DAG.getNode(ARMISD::CMOV, dl, MVT::i32, FalseHigh, TrueHigh,
+ ARMcc, CCR, duplicateCmp(Cmp, DAG));
+
+ return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Low, High);
+ } else {
+ return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,
+ Cmp);
+ }
+}
+
SDValue ARMTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
SDValue LHS = Op.getOperand(0);
SDValue FalseVal = Op.getOperand(3);
SDLoc dl(Op);
+ if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) {
+ DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC,
+ dl);
+
+ // If softenSetCCOperands only returned one value, we should compare it to
+ // zero.
+ if (!RHS.getNode()) {
+ RHS = DAG.getConstant(0, LHS.getValueType());
+ CC = ISD::SETNE;
+ }
+ }
+
if (LHS.getValueType() == MVT::i32) {
// Try to generate VSEL on ARMv8.
// The VSEL instruction can't use all the usual ARM condition
SDValue ARMcc;
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
- return DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc, CCR,
- Cmp);
+ return getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
}
ARMCC::CondCodes CondCode, CondCode2;
SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32);
SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
- SDValue Result = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal,
- ARMcc, CCR, Cmp);
+ SDValue Result = getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
if (CondCode2 != ARMCC::AL) {
SDValue ARMcc2 = DAG.getConstant(CondCode2, MVT::i32);
// FIXME: Needs another CMP because flag can have but one use.
SDValue Cmp2 = getVFPCmp(LHS, RHS, DAG, dl);
- Result = DAG.getNode(ARMISD::CMOV, dl, VT,
- Result, TrueVal, ARMcc2, CCR, Cmp2);
+ Result = getCMOV(dl, VT, Result, TrueVal, ARMcc2, CCR, Cmp2, DAG);
}
return Result;
}
SDValue Dest = Op.getOperand(4);
SDLoc dl(Op);
+ if (Subtarget->isFPOnlySP() && LHS.getValueType() == MVT::f64) {
+ DAG.getTargetLoweringInfo().softenSetCCOperands(DAG, MVT::f64, LHS, RHS, CC,
+ dl);
+
+ // If softenSetCCOperands only returned one value, we should compare it to
+ // zero.
+ if (!RHS.getNode()) {
+ RHS = DAG.getConstant(0, LHS.getValueType());
+ CC = ISD::SETNE;
+ }
+ }
+
if (LHS.getValueType() == MVT::i32) {
SDValue ARMcc;
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
}
-static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
+SDValue ARMTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
if (VT.isVector())
return LowerVectorFP_TO_INT(Op, DAG);
+ if (Subtarget->isFPOnlySP() && Op.getOperand(0).getValueType() == MVT::f64) {
+ RTLIB::Libcall LC;
+ if (Op.getOpcode() == ISD::FP_TO_SINT)
+ LC = RTLIB::getFPTOSINT(Op.getOperand(0).getValueType(),
+ Op.getValueType());
+ else
+ LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(),
+ Op.getValueType());
+ return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
+ /*isSigned*/ false, SDLoc(Op)).first;
+ }
+
SDLoc dl(Op);
unsigned Opc;
return DAG.getNode(Opc, dl, VT, Op);
}
-static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
+SDValue ARMTargetLowering::LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
if (VT.isVector())
return LowerVectorINT_TO_FP(Op, DAG);
+ if (Subtarget->isFPOnlySP() && Op.getValueType() == MVT::f64) {
+ RTLIB::Libcall LC;
+ if (Op.getOpcode() == ISD::SINT_TO_FP)
+ LC = RTLIB::getSINTTOFP(Op.getOperand(0).getValueType(),
+ Op.getValueType());
+ else
+ LC = RTLIB::getUINTTOFP(Op.getOperand(0).getValueType(),
+ Op.getValueType());
+ return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
+ /*isSigned*/ false, SDLoc(Op)).first;
+ }
+
SDLoc dl(Op);
unsigned Opc;
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
SDLoc dl(Op);
- if (Op.getOperand(1).getValueType().isFloatingPoint()) {
+ if (Op1.getValueType().isFloatingPoint()) {
switch (SetCCOpcode) {
default: llvm_unreachable("Illegal FP comparison");
case ISD::SETUNE:
BitMask <<= 8;
ImmMask <<= 1;
}
+
+ if (DAG.getTargetLoweringInfo().isBigEndian())
+ // swap higher and lower 32 bit word
+ Imm = ((Imm & 0xf) << 4) | ((Imm & 0xf0) >> 4);
+
// Op=1, Cmode=1110.
OpCmode = 0x1e;
VT = is128Bits ? MVT::v2i64 : MVT::v1i64;
bool IsDouble = Op.getValueType() == MVT::f64;
ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Op);
+ // Use the default (constant pool) lowering for double constants when we have
+ // an SP-only FPU
+ if (IsDouble && Subtarget->isFPOnlySP())
+ return SDValue();
+
// Try splatting with a VMOV.f32...
APFloat FPVal = CFP->getValueAPF();
int ImmVal = IsDouble ? ARM_AM::getFP64Imm(FPVal) : ARM_AM::getFP32Imm(FPVal);
// operation legalization where we can't create illegal types.
return DAG.getExtLoad(LD->getExtensionType(), SDLoc(LD), ExtendedTy,
LD->getChain(), LD->getBasePtr(), LD->getPointerInfo(),
- LD->getMemoryVT(), LD->isVolatile(),
+ LD->getMemoryVT(), LD->isVolatile(), LD->isInvariant(),
LD->isNonTemporal(), LD->getAlignment());
}
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
.setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()), Callee,
- &Args, 0)
+ std::move(Args), 0)
.setDiscardResult();
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
case ISD::FSINCOS: return LowerFSINCOS(Op, DAG);
case ISD::SDIVREM:
case ISD::UDIVREM: return LowerDivRem(Op, DAG);
+ case ISD::DYNAMIC_STACKALLOC:
+ if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
+ return LowerDYNAMIC_STACKALLOC(Op, DAG);
+ llvm_unreachable("Don't know how to custom lower this!");
+ case ISD::FP_ROUND: return LowerFP_ROUND(Op, DAG);
+ case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
}
}
void ARMTargetLowering::
SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB,
MachineBasicBlock *DispatchBB, int FI) const {
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
MachineBasicBlock *ARMTargetLowering::
EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
// N.B. the order the invoke BBs are processed in doesn't matter here.
const MCPhysReg *SavedRegs = RI.getCalleeSavedRegs(MF);
SmallVector<MachineBasicBlock*, 64> MBBLPads;
- for (SmallPtrSet<MachineBasicBlock*, 64>::iterator
- I = InvokeBBs.begin(), E = InvokeBBs.end(); I != E; ++I) {
- MachineBasicBlock *BB = *I;
+ for (MachineBasicBlock *BB : InvokeBBs) {
// Remove the landing pad successor from the invoke block and replace it
// with the new dispatch block.
// 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 TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = BB;
++It;
return BB;
}
+MachineBasicBlock *
+ARMTargetLowering::EmitLowered__chkstk(MachineInstr *MI,
+ MachineBasicBlock *MBB) const {
+ const TargetMachine &TM = getTargetMachine();
+ const TargetInstrInfo &TII = *TM.getSubtargetImpl()->getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+
+ assert(Subtarget->isTargetWindows() &&
+ "__chkstk is only supported on Windows");
+ assert(Subtarget->isThumb2() && "Windows on ARM requires Thumb-2 mode");
+
+ // __chkstk takes the number of words to allocate on the stack in R4, and
+ // returns the stack adjustment in number of bytes in R4. This will not
+ // clober any other registers (other than the obvious lr).
+ //
+ // Although, technically, IP should be considered a register which may be
+ // clobbered, the call itself will not touch it. Windows on ARM is a pure
+ // thumb-2 environment, so there is no interworking required. As a result, we
+ // do not expect a veneer to be emitted by the linker, clobbering IP.
+ //
+ // Each module receives its own copy of __chkstk, so no import thunk is
+ // required, again, ensuring that IP is not clobbered.
+ //
+ // Finally, although some linkers may theoretically provide a trampoline for
+ // out of range calls (which is quite common due to a 32M range limitation of
+ // branches for Thumb), we can generate the long-call version via
+ // -mcmodel=large, alleviating the need for the trampoline which may clobber
+ // IP.
+
+ switch (TM.getCodeModel()) {
+ case CodeModel::Small:
+ case CodeModel::Medium:
+ case CodeModel::Default:
+ case CodeModel::Kernel:
+ BuildMI(*MBB, MI, DL, TII.get(ARM::tBL))
+ .addImm((unsigned)ARMCC::AL).addReg(0)
+ .addExternalSymbol("__chkstk")
+ .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Define)
+ .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
+ break;
+ case CodeModel::Large:
+ case CodeModel::JITDefault: {
+ MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+ unsigned Reg = MRI.createVirtualRegister(&ARM::rGPRRegClass);
+
+ BuildMI(*MBB, MI, DL, TII.get(ARM::t2MOVi32imm), Reg)
+ .addExternalSymbol("__chkstk");
+ BuildMI(*MBB, MI, DL, TII.get(ARM::tBLXr))
+ .addImm((unsigned)ARMCC::AL).addReg(0)
+ .addReg(Reg, RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Kill)
+ .addReg(ARM::R4, RegState::Implicit | RegState::Define)
+ .addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
+ break;
+ }
+ }
+
+ AddDefaultCC(AddDefaultPred(BuildMI(*MBB, MI, DL, TII.get(ARM::t2SUBrr),
+ ARM::SP)
+ .addReg(ARM::SP).addReg(ARM::R4)));
+
+ MI->eraseFromParent();
+ return MBB;
+}
+
MachineBasicBlock *
ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) const {
- const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ const TargetInstrInfo *TII =
+ getTargetMachine().getSubtargetImpl()->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
bool isThumb2 = Subtarget->isThumb2();
switch (MI->getOpcode()) {
case ARM::COPY_STRUCT_BYVAL_I32:
++NumLoopByVals;
return EmitStructByval(MI, BB);
+ case ARM::WIN__CHKSTK:
+ return EmitLowered__chkstk(MI, BB);
}
}
// Rename pseudo opcodes.
unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode());
if (NewOpc) {
- const ARMBaseInstrInfo *TII =
- static_cast<const ARMBaseInstrInfo*>(getTargetMachine().getInstrInfo());
+ const ARMBaseInstrInfo *TII = static_cast<const ARMBaseInstrInfo *>(
+ getTargetMachine().getSubtargetImpl()->getInstrInfo());
MCID = &TII->get(NewOpc);
assert(MCID->getNumOperands() == MI->getDesc().getNumOperands() + 1 &&
/// PerformVMOVRRDCombine - Target-specific dag combine xforms for
/// ARMISD::VMOVRRD.
static SDValue PerformVMOVRRDCombine(SDNode *N,
- TargetLowering::DAGCombinerInfo &DCI) {
+ TargetLowering::DAGCombinerInfo &DCI,
+ const ARMSubtarget *Subtarget) {
// vmovrrd(vmovdrr x, y) -> x,y
SDValue InDouble = N->getOperand(0);
- if (InDouble.getOpcode() == ARMISD::VMOVDRR)
+ if (InDouble.getOpcode() == ARMISD::VMOVDRR && !Subtarget->isFPOnlySP())
return DCI.CombineTo(N, InDouble.getOperand(0), InDouble.getOperand(1));
// vmovrrd(load f64) -> (load i32), (load i32)
std::min(4U, LD->getAlignment() / 2));
DAG.ReplaceAllUsesOfValueWith(SDValue(LD, 1), NewLD2.getValue(1));
+ if (DCI.DAG.getTargetLoweringInfo().isBigEndian())
+ std::swap (NewLD1, NewLD2);
SDValue Result = DCI.CombineTo(N, NewLD1, NewLD2);
- DCI.RemoveFromWorklist(LD);
- DAG.DeleteNode(LD);
return Result;
}
SDLoc DL(St);
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;
+ for (unsigned i = 0; i < NumElems; ++i)
+ ShuffleVec[i] = TLI.isBigEndian() ? (i+1) * SizeRatio - 1 : i * SizeRatio;
// Can't shuffle using an illegal type.
if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
return DAG.getStore(St->getChain(), dl, V, St->getBasePtr(),
St->getPointerInfo(), St->isVolatile(),
St->isNonTemporal(), St->getAlignment(),
- St->getTBAAInfo());
+ St->getAAInfo());
}
/// hasNormalLoadOperand - Check if any of the operands of a BUILD_VECTOR node
/// PerformBUILD_VECTORCombine - Target-specific dag combine xforms for
/// ISD::BUILD_VECTOR.
static SDValue PerformBUILD_VECTORCombine(SDNode *N,
- TargetLowering::DAGCombinerInfo &DCI){
+ TargetLowering::DAGCombinerInfo &DCI,
+ const ARMSubtarget *Subtarget) {
// build_vector(N=ARMISD::VMOVRRD(X), N:1) -> bit_convert(X):
// VMOVRRD is introduced when legalizing i64 types. It forces the i64 value
// into a pair of GPRs, which is fine when the value is used as a scalar,
Tys[n] = VecTy;
Tys[n++] = MVT::i32;
Tys[n] = MVT::Other;
- SDVTList SDTys = DAG.getVTList(ArrayRef<EVT>(Tys, NumResultVecs+2));
+ SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumResultVecs+2));
SmallVector<SDValue, 8> Ops;
Ops.push_back(N->getOperand(0)); // incoming chain
Ops.push_back(N->getOperand(AddrOpIdx));
for (n = 0; n < NumVecs; ++n)
Tys[n] = VT;
Tys[n] = MVT::Other;
- SDVTList SDTys = DAG.getVTList(ArrayRef<EVT>(Tys, NumVecs+1));
+ SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumVecs+1));
SDValue Ops[] = { VLD->getOperand(0), VLD->getOperand(2) };
MemIntrinsicSDNode *VLDMemInt = cast<MemIntrinsicSDNode>(VLD);
SDValue VLDDup = DAG.getMemIntrinsicNode(NewOpc, SDLoc(VLD), SDTys,
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::VMOVRRD: return PerformVMOVRRDCombine(N, DCI, Subtarget);
case ARMISD::VMOVDRR: return PerformVMOVDRRCombine(N, DCI.DAG);
case ISD::STORE: return PerformSTORECombine(N, DCI);
- case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI);
+ case ISD::BUILD_VECTOR: return PerformBUILD_VECTORCombine(N, DCI, Subtarget);
case ISD::INSERT_VECTOR_ELT: return PerformInsertEltCombine(N, DCI);
case ISD::VECTOR_SHUFFLE: return PerformVECTOR_SHUFFLECombine(N, DCI.DAG);
case ARMISD::VDUPLANE: return PerformVDUPLANECombine(N, DCI);
return (VT == MVT::f32) && (Opc == ISD::LOAD || Opc == ISD::STORE);
}
-bool ARMTargetLowering::allowsUnalignedMemoryAccesses(EVT VT, unsigned,
- bool *Fast) const {
+bool ARMTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
+ unsigned,
+ unsigned,
+ bool *Fast) const {
// The AllowsUnaliged flag models the SCTLR.A setting in ARM cpus
bool AllowsUnaligned = Subtarget->allowsUnalignedMem();
bool Fast;
if (Size >= 16 &&
(memOpAlign(SrcAlign, DstAlign, 16) ||
- (allowsUnalignedMemoryAccesses(MVT::v2f64, 0, &Fast) && Fast))) {
+ (allowsMisalignedMemoryAccesses(MVT::v2f64, 0, 1, &Fast) && Fast))) {
return MVT::v2f64;
} else if (Size >= 8 &&
(memOpAlign(SrcAlign, DstAlign, 8) ||
- (allowsUnalignedMemoryAccesses(MVT::f64, 0, &Fast) && Fast))) {
+ (allowsMisalignedMemoryAccesses(MVT::f64, 0, 1, &Fast) &&
+ Fast))) {
return MVT::f64;
}
}
assert(Subtarget->isTargetAEABI() && "Register-based DivRem lowering only");
unsigned Opcode = Op->getOpcode();
assert((Opcode == ISD::SDIVREM || Opcode == ISD::UDIVREM) &&
- "Invalid opcode for Div/Rem lowering");
+ "Invalid opcode for Div/Rem lowering");
bool isSigned = (Opcode == ISD::SDIVREM);
EVT VT = Op->getValueType(0);
Type *Ty = VT.getTypeForEVT(*DAG.getContext());
RTLIB::Libcall LC;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
- case MVT::i8: LC= isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
- case MVT::i16: LC= isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
- case MVT::i32: LC= isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
- case MVT::i64: LC= isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
+ case MVT::i8: LC = isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
+ case MVT::i16: LC = isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
+ case MVT::i32: LC = isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
+ case MVT::i64: LC = isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
}
SDValue InChain = DAG.getEntryNode();
SDLoc dl(Op);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(InChain)
- .setCallee(getLibcallCallingConv(LC), RetTy, Callee, &Args, 0)
+ .setCallee(getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
.setInRegister().setSExtResult(isSigned).setZExtResult(!isSigned);
std::pair<SDValue, SDValue> CallInfo = LowerCallTo(CLI);
return CallInfo.first;
}
+SDValue
+ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const {
+ assert(Subtarget->isTargetWindows() && "unsupported target platform");
+ SDLoc DL(Op);
+
+ // Get the inputs.
+ SDValue Chain = Op.getOperand(0);
+ SDValue Size = Op.getOperand(1);
+
+ SDValue Words = DAG.getNode(ISD::SRL, DL, MVT::i32, Size,
+ DAG.getConstant(2, MVT::i32));
+
+ SDValue Flag;
+ Chain = DAG.getCopyToReg(Chain, DL, ARM::R4, Words, Flag);
+ Flag = Chain.getValue(1);
+
+ SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+ Chain = DAG.getNode(ARMISD::WIN__CHKSTK, DL, NodeTys, Chain, Flag);
+
+ SDValue NewSP = DAG.getCopyFromReg(Chain, DL, ARM::SP, MVT::i32);
+ Chain = NewSP.getValue(1);
+
+ SDValue Ops[2] = { NewSP, Chain };
+ return DAG.getMergeValues(Ops, DL);
+}
+
+SDValue ARMTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
+ assert(Op.getValueType() == MVT::f64 && Subtarget->isFPOnlySP() &&
+ "Unexpected type for custom-lowering FP_EXTEND");
+
+ RTLIB::Libcall LC;
+ LC = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType());
+
+ SDValue SrcVal = Op.getOperand(0);
+ return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
+ /*isSigned*/ false, SDLoc(Op)).first;
+}
+
+SDValue ARMTargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
+ assert(Op.getOperand(0).getValueType() == MVT::f64 &&
+ Subtarget->isFPOnlySP() &&
+ "Unexpected type for custom-lowering FP_ROUND");
+
+ RTLIB::Libcall LC;
+ LC = RTLIB::getFPROUND(Op.getOperand(0).getValueType(), Op.getValueType());
+
+ SDValue SrcVal = Op.getOperand(0);
+ return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
+ /*isSigned*/ false, SDLoc(Op)).first;
+}
+
bool
ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
// The ARM target isn't yet aware of offsets.
return false;
if (VT == MVT::f32)
return ARM_AM::getFP32Imm(Imm) != -1;
- if (VT == MVT::f64)
+ if (VT == MVT::f64 && !Subtarget->isFPOnlySP())
return ARM_AM::getFP64Imm(Imm) != -1;
return false;
}
bool ARMTargetLowering::shouldExpandAtomicInIR(Instruction *Inst) const {
// Loads and stores less than 64-bits are already atomic; ones above that
// are doomed anyway, so defer to the default libcall and blame the OS when
- // things go wrong:
- if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
- return SI->getValueOperand()->getType()->getPrimitiveSizeInBits() == 64;
- else if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
- return LI->getType()->getPrimitiveSizeInBits() == 64;
+ // things go wrong. Cortex M doesn't have ldrexd/strexd though, so don't emit
+ // anything for those.
+ bool IsMClass = Subtarget->isMClass();
+ if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+ unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
+ return Size == 64 && !IsMClass;
+ } else if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+ return LI->getType()->getPrimitiveSizeInBits() == 64 && !IsMClass;
+ }
+
+ // For the real atomic operations, we have ldrex/strex up to 32 bits,
+ // and up to 64 bits on the non-M profiles
+ unsigned AtomicLimit = IsMClass ? 32 : 64;
+ return Inst->getType()->getPrimitiveSizeInBits() <= AtomicLimit;
+}
- // For the real atomic operations, we have ldrex/strex up to 64 bits.
- return Inst->getType()->getPrimitiveSizeInBits() <= 64;
+// This has so far only been implemented for MachO.
+bool ARMTargetLowering::useLoadStackGuardNode() const {
+ return Subtarget->getTargetTriple().getObjectFormat() == Triple::MachO;
}
Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
AtomicOrdering Ord) const {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
Type *ValTy = cast<PointerType>(Addr->getType())->getElementType();
- bool IsAcquire =
- Ord == Acquire || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+ bool IsAcquire = isAtLeastAcquire(Ord);
// Since i64 isn't legal and intrinsics don't get type-lowered, the ldrexd
// intrinsic must return {i32, i32} and we have to recombine them into a
Value *Addr,
AtomicOrdering Ord) const {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
- bool IsRelease =
- Ord == Release || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+ bool IsRelease = isAtLeastRelease(Ord);
// Since the intrinsics must have legal type, the i64 intrinsics take two
// parameters: "i32, i32". We must marshal Val into the appropriate form
HABaseType Base = HA_UNKNOWN;
uint64_t Members = 0;
bool result = isHomogeneousAggregate(Ty, Base, Members);
- DEBUG(dbgs() << "isHA: " << result << " "; Ty->dump(); dbgs() << "\n");
+ DEBUG(dbgs() << "isHA: " << result << " "; Ty->dump());
return result;
}