setCmpLibcallCC(LC.Op, LC.Cond);
}
}
+
+ // Set the correct calling convention for ARMv7k WatchOS. It's just
+ // AAPCS_VFP for functions as simple as libcalls.
+ if (Subtarget->isTargetWatchOS()) {
+ for (int i = 0; i < RTLIB::UNKNOWN_LIBCALL; ++i)
+ setLibcallCallingConv((RTLIB::Libcall)i, CallingConv::ARM_AAPCS_VFP);
+ }
}
// These libcalls are not available in 32-bit.
setLibcallName(RTLIB::SRL_I128, nullptr);
setLibcallName(RTLIB::SRA_I128, nullptr);
- if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetMachO() &&
- !Subtarget->isTargetWindows()) {
+ // RTLIB
+ if (Subtarget->isAAPCS_ABI() &&
+ (Subtarget->isTargetAEABI() || Subtarget->isTargetGNUAEABI() ||
+ Subtarget->isTargetAndroid())) {
static const struct {
const RTLIB::Libcall Op;
const char * const Name;
{ RTLIB::UDIV_I16, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UDIV_I32, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UDIV_I64, "__aeabi_uldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
-
- // Memory operations
- // RTABI chapter 4.3.4
- { RTLIB::MEMCPY, "__aeabi_memcpy", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::MEMMOVE, "__aeabi_memmove", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::MEMSET, "__aeabi_memset", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
};
for (const auto &LC : LibraryCalls) {
if (LC.Cond != ISD::SETCC_INVALID)
setCmpLibcallCC(LC.Op, LC.Cond);
}
+
+ // EABI dependent RTLIB
+ if (TM.Options.EABIVersion == EABI::EABI4 ||
+ TM.Options.EABIVersion == EABI::EABI5) {
+ static const struct {
+ const RTLIB::Libcall Op;
+ const char *const Name;
+ const CallingConv::ID CC;
+ const ISD::CondCode Cond;
+ } MemOpsLibraryCalls[] = {
+ // Memory operations
+ // RTABI chapter 4.3.4
+ { RTLIB::MEMCPY, "__aeabi_memcpy", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::MEMMOVE, "__aeabi_memmove", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::MEMSET, "__aeabi_memset", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ };
+
+ for (const auto &LC : MemOpsLibraryCalls) {
+ setLibcallName(LC.Op, LC.Name);
+ setLibcallCallingConv(LC.Op, LC.CC);
+ if (LC.Cond != ISD::SETCC_INVALID)
+ setCmpLibcallCC(LC.Op, LC.Cond);
+ }
+ }
}
if (Subtarget->isTargetWindows()) {
{ RTLIB::SINTTOFP_I64_F64, "__i64tod", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::UINTTOFP_I64_F32, "__u64tos", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::UINTTOFP_I64_F64, "__u64tod", CallingConv::ARM_AAPCS_VFP },
+ { RTLIB::SDIV_I32, "__rt_sdiv", CallingConv::ARM_AAPCS_VFP },
+ { RTLIB::SDIV_I64, "__rt_sdiv64", CallingConv::ARM_AAPCS_VFP },
};
for (const auto &LC : LibraryCalls) {
}
// Use divmod compiler-rt calls for iOS 5.0 and later.
- if (Subtarget->getTargetTriple().isiOS() &&
- !Subtarget->getTargetTriple().isOSVersionLT(5, 0)) {
+ if (Subtarget->isTargetWatchOS() ||
+ (Subtarget->isTargetIOS() &&
+ !Subtarget->getTargetTriple().isOSVersionLT(5, 0))) {
setLibcallName(RTLIB::SDIVREM_I32, "__divmodsi4");
setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4");
}
setOperationAction(ISD::SUBE, MVT::i32, Custom);
}
+ if (!Subtarget->isThumb1Only())
+ setOperationAction(ISD::BITREVERSE, MVT::i32, Legal);
+
// ARM does not have ROTL.
- setOperationAction(ISD::ROTL, MVT::i32, Expand);
+ setOperationAction(ISD::ROTL, MVT::i32, Expand);
+ for (MVT VT : MVT::vector_valuetypes()) {
+ setOperationAction(ISD::ROTL, VT, Expand);
+ setOperationAction(ISD::ROTR, VT, Expand);
+ }
setOperationAction(ISD::CTTZ, MVT::i32, Custom);
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
if (!Subtarget->hasV5TOps() || Subtarget->isThumb1Only())
}
if (Subtarget->isTargetWindows() && !Subtarget->hasDivide()) {
- setOperationAction(ISD::SDIV, MVT::i32, Custom);
setOperationAction(ISD::UDIV, MVT::i32, Custom);
- setOperationAction(ISD::SDIV, MVT::i64, Custom);
setOperationAction(ISD::UDIV, MVT::i64, Custom);
}
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
- setOperationAction(ISD::GLOBAL_OFFSET_TABLE, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
- if (!Subtarget->isTargetMachO()) {
- // Non-MachO platforms may return values in these registers via the
- // personality function.
- setExceptionPointerRegister(ARM::R0);
- setExceptionSelectorRegister(ARM::R1);
- }
-
if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
else
setOperationAction(ISD::EH_SJLJ_SETJMP, MVT::i32, Custom);
setOperationAction(ISD::EH_SJLJ_LONGJMP, MVT::Other, Custom);
setOperationAction(ISD::EH_SJLJ_SETUP_DISPATCH, MVT::Other, Custom);
- if (Subtarget->isTargetDarwin())
+ if (Subtarget->useSjLjEH())
setLibcallName(RTLIB::UNWIND_RESUME, "_Unwind_SjLj_Resume");
setOperationAction(ISD::SETCC, MVT::i32, Expand);
if (Subtarget->hasSinCos()) {
setLibcallName(RTLIB::SINCOS_F32, "sincosf");
setLibcallName(RTLIB::SINCOS_F64, "sincos");
- if (Subtarget->getTargetTriple().isiOS()) {
+ if (Subtarget->isTargetWatchOS()) {
+ setLibcallCallingConv(RTLIB::SINCOS_F32, CallingConv::ARM_AAPCS_VFP);
+ setLibcallCallingConv(RTLIB::SINCOS_F64, CallingConv::ARM_AAPCS_VFP);
+ }
+ if (Subtarget->isTargetIOS() || Subtarget->isTargetWatchOS()) {
// For iOS, we don't want to the normal expansion of a libcall to
// sincos. We want to issue a libcall to __sincos_stret.
setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
case ARMISD::CMOV: return "ARMISD::CMOV";
- case ARMISD::RBIT: return "ARMISD::RBIT";
-
case ARMISD::SRL_FLAG: return "ARMISD::SRL_FLAG";
case ARMISD::SRA_FLAG: return "ARMISD::SRA_FLAG";
case ARMISD::RRX: return "ARMISD::RRX";
SDLoc dl(Op);
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
- bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
- ARMConstantPoolValue *CPV =
- ARMConstantPoolConstant::Create(GV,
- UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT);
+ bool UseGOT_PREL =
+ !(GV->hasHiddenVisibility() || GV->hasLocalLinkage());
+
+ MachineFunction &MF = DAG.getMachineFunction();
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
+ EVT PtrVT = getPointerTy(DAG.getDataLayout());
+ SDLoc dl(Op);
+ unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
+ ARMConstantPoolValue *CPV = ARMConstantPoolConstant::Create(
+ GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj,
+ UseGOT_PREL ? ARMCP::GOT_PREL : ARMCP::no_modifier,
+ /*AddCurrentAddress=*/UseGOT_PREL);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), false,
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)
+ SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, dl, MVT::i32);
+ Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
+ if (UseGOT_PREL)
Result = DAG.getLoad(PtrVT, dl, Chain, Result,
MachinePointerInfo::getGOT(DAG.getMachineFunction()),
false, false, false, 0);
return Result;
}
-SDValue ARMTargetLowering::LowerGLOBAL_OFFSET_TABLE(SDValue Op,
- SelectionDAG &DAG) const {
- assert(Subtarget->isTargetELF() &&
- "GLOBAL OFFSET TABLE not implemented for non-ELF targets");
- MachineFunction &MF = DAG.getMachineFunction();
- ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
- unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
- EVT PtrVT = getPointerTy(DAG.getDataLayout());
- SDLoc dl(Op);
- unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
- ARMConstantPoolValue *CPV =
- ARMConstantPoolSymbol::Create(*DAG.getContext(), "_GLOBAL_OFFSET_TABLE_",
- ARMPCLabelIndex, PCAdj);
- 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,
- MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
- false, false, false, 0);
- SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, dl, MVT::i32);
- return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
-}
-
SDValue
ARMTargetLowering::LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
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));
+ return DAG.getNode(ISD::BITREVERSE, dl, MVT::i32, Op.getOperand(1));
}
case Intrinsic::arm_thread_pointer: {
EVT PtrVT = getPointerTy(DAG.getDataLayout());
// Handle (ISD::BITCAST (ARMISD::VMOVIMM (ISD::TargetConstant 0)) MVT::f64)
// created by LowerConstantFP().
SDValue BitcastOp = Op->getOperand(0);
- if (BitcastOp->getOpcode() == ARMISD::VMOVIMM) {
- SDValue MoveOp = BitcastOp->getOperand(0);
- if (MoveOp->getOpcode() == ISD::TargetConstant &&
- cast<ConstantSDNode>(MoveOp)->getZExtValue() == 0) {
- return true;
- }
- }
+ if (BitcastOp->getOpcode() == ARMISD::VMOVIMM &&
+ isNullConstant(BitcastOp->getOperand(0)))
+ return true;
}
return false;
}
if (!ST->hasV6T2Ops())
return SDValue();
- SDValue rbit = DAG.getNode(ARMISD::RBIT, dl, VT, N->getOperand(0));
+ SDValue rbit = DAG.getNode(ISD::BITREVERSE, dl, VT, N->getOperand(0));
return DAG.getNode(ISD::CTLZ, dl, VT, rbit);
}
"Unknown shift to lower!");
// We only lower SRA, SRL of 1 here, all others use generic lowering.
- if (!isa<ConstantSDNode>(N->getOperand(1)) ||
- cast<ConstantSDNode>(N->getOperand(1))->getZExtValue() != 1)
+ if (!isOneConstant(N->getOperand(1)))
return SDValue();
// If we are in thumb mode, we don't have RRX.
// just use VDUPLANE. We can only do this if the lane being extracted
// is at a constant index, as the VDUP from lane instructions only have
// constant-index forms.
+ ConstantSDNode *constIndex;
if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
- isa<ConstantSDNode>(Value->getOperand(1))) {
+ (constIndex = dyn_cast<ConstantSDNode>(Value->getOperand(1)))) {
// We need to create a new undef vector to use for the VDUPLANE if the
// size of the vector from which we get the value is different than the
// size of the vector that we need to create. We will insert the element
// such that the register coalescer will remove unnecessary copies.
if (VT != Value->getOperand(0).getValueType()) {
- ConstantSDNode *constIndex;
- constIndex = dyn_cast<ConstantSDNode>(Value->getOperand(1));
- assert(constIndex && "The index is not a constant!");
unsigned index = constIndex->getAPIntValue().getLimitedValue() %
VT.getVectorNumElements();
N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
auto PtrVT = getPointerTy(DAG.getDataLayout());
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
// Pair of floats / doubles used to pass the result.
- StructType *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
-
- // Create stack object for sret.
+ Type *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
auto &DL = DAG.getDataLayout();
- const uint64_t ByteSize = DL.getTypeAllocSize(RetTy);
- const unsigned StackAlign = DL.getPrefTypeAlignment(RetTy);
- int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
- SDValue SRet = DAG.getFrameIndex(FrameIdx, getPointerTy(DL));
ArgListTy Args;
- ArgListEntry Entry;
-
- Entry.Node = SRet;
- Entry.Ty = RetTy->getPointerTo();
- Entry.isSExt = false;
- Entry.isZExt = false;
- Entry.isSRet = true;
- Args.push_back(Entry);
+ bool ShouldUseSRet = Subtarget->isAPCS_ABI();
+ SDValue SRet;
+ if (ShouldUseSRet) {
+ // Create stack object for sret.
+ const uint64_t ByteSize = DL.getTypeAllocSize(RetTy);
+ const unsigned StackAlign = DL.getPrefTypeAlignment(RetTy);
+ int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
+ SRet = DAG.getFrameIndex(FrameIdx, TLI.getPointerTy(DL));
+
+ ArgListEntry Entry;
+ Entry.Node = SRet;
+ Entry.Ty = RetTy->getPointerTo();
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Entry.isSRet = true;
+ Args.push_back(Entry);
+ RetTy = Type::getVoidTy(*DAG.getContext());
+ }
+ ArgListEntry Entry;
Entry.Node = Arg;
Entry.Ty = ArgTy;
Entry.isSExt = false;
const char *LibcallName =
(ArgVT == MVT::f64) ? "__sincos_stret" : "__sincosf_stret";
+ RTLIB::Libcall LC =
+ (ArgVT == MVT::f64) ? RTLIB::SINCOS_F64 : RTLIB::SINCOS_F32;
+ CallingConv::ID CC = getLibcallCallingConv(LC);
SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy(DL));
TargetLowering::CallLoweringInfo CLI(DAG);
- CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
- .setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()), Callee,
- std::move(Args), 0)
- .setDiscardResult();
-
+ CLI.setDebugLoc(dl)
+ .setChain(DAG.getEntryNode())
+ .setCallee(CC, RetTy, Callee, std::move(Args), 0)
+ .setDiscardResult(ShouldUseSRet);
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
+ if (!ShouldUseSRet)
+ return CallResult.first;
+
SDValue LoadSin = DAG.getLoad(ArgVT, dl, CallResult.second, SRet,
MachinePointerInfo(), false, false, false, 0);
}
SDValue ARMTargetLowering::LowerWindowsDIVLibCall(SDValue Op, SelectionDAG &DAG,
- bool Signed,
SDValue &Chain) const {
EVT VT = Op.getValueType();
assert((VT == MVT::i32 || VT == MVT::i64) &&
const auto &TLI = DAG.getTargetLoweringInfo();
const char *Name = nullptr;
- if (Signed)
- Name = (VT == MVT::i32) ? "__rt_sdiv" : "__rt_sdiv64";
- else
- Name = (VT == MVT::i32) ? "__rt_udiv" : "__rt_udiv64";
+ Name = (VT == MVT::i32) ? "__rt_udiv" : "__rt_udiv64";
SDValue ES = DAG.getExternalSymbol(Name, TLI.getPointerTy(DL));
return LowerCallTo(CLI).first;
}
-SDValue ARMTargetLowering::LowerDIV_Windows(SDValue Op, SelectionDAG &DAG,
- bool Signed) const {
+SDValue ARMTargetLowering::LowerDIV_Windows(SDValue Op,
+ SelectionDAG &DAG) const {
assert(Op.getValueType() == MVT::i32 &&
"unexpected type for custom lowering DIV");
SDLoc dl(Op);
SDValue DBZCHK = DAG.getNode(ARMISD::WIN__DBZCHK, dl, MVT::Other,
DAG.getEntryNode(), Op.getOperand(1));
- return LowerWindowsDIVLibCall(Op, DAG, Signed, DBZCHK);
+ return LowerWindowsDIVLibCall(Op, DAG, DBZCHK);
}
void ARMTargetLowering::ExpandDIV_Windows(
- SDValue Op, SelectionDAG &DAG, bool Signed,
+ SDValue Op, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &Results) const {
const auto &DL = DAG.getDataLayout();
const auto &TLI = DAG.getTargetLoweringInfo();
SDValue DBZCHK =
DAG.getNode(ARMISD::WIN__DBZCHK, dl, MVT::Other, DAG.getEntryNode(), Or);
- SDValue Result = LowerWindowsDIVLibCall(Op, DAG, Signed, DBZCHK);
+ SDValue Result = LowerWindowsDIVLibCall(Op, DAG, DBZCHK);
SDValue Lower = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Result);
SDValue Upper = DAG.getNode(ISD::SRL, dl, MVT::i64, Result,
case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG);
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
- case ISD::GLOBAL_OFFSET_TABLE: return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
case ISD::EH_SJLJ_SETJMP: return LowerEH_SJLJ_SETJMP(Op, DAG);
case ISD::EH_SJLJ_LONGJMP: return LowerEH_SJLJ_LONGJMP(Op, DAG);
case ISD::EH_SJLJ_SETUP_DISPATCH: return LowerEH_SJLJ_SETUP_DISPATCH(Op, DAG);
case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
case ISD::MUL: return LowerMUL(Op, DAG);
- case ISD::SDIV:
- if (Subtarget->isTargetWindows())
- return LowerDIV_Windows(Op, DAG, /* Signed */ true);
- return LowerSDIV(Op, DAG);
+ case ISD::SDIV: return LowerSDIV(Op, DAG);
case ISD::UDIV:
if (Subtarget->isTargetWindows())
- return LowerDIV_Windows(Op, DAG, /* Signed */ false);
+ return LowerDIV_Windows(Op, DAG);
return LowerUDIV(Op, DAG);
case ISD::ADDC:
case ISD::ADDE:
ReplaceREADCYCLECOUNTER(N, Results, DAG, Subtarget);
return;
case ISD::UDIV:
- case ISD::SDIV:
assert(Subtarget->isTargetWindows() && "can only expand DIV on Windows");
- return ExpandDIV_Windows(SDValue(N, 0), DAG, N->getOpcode() == ISD::SDIV,
- Results);
+ return ExpandDIV_Windows(SDValue(N, 0), DAG, Results);
}
if (Res.getNode())
Results.push_back(Res);
// Helper function that checks if N is a null or all ones constant.
static inline bool isZeroOrAllOnes(SDValue N, bool AllOnes) {
- ConstantSDNode *C = dyn_cast<ConstantSDNode>(N);
- if (!C)
- return false;
- return AllOnes ? C->isAllOnesValue() : C->isNullValue();
+ return AllOnes ? isAllOnesConstant(N) : isNullConstant(N);
}
// Return true if N is conditionally 0 or all ones.
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.
+// ParseBFI - given a BFI instruction in N, extract the "from" value (Rn) and return it,
+// and fill in FromMask and ToMask with (consecutive) bits in "from" to be extracted and
+// their position in "to" (Rd).
+static SDValue ParseBFI(SDNode *N, APInt &ToMask, APInt &FromMask) {
+ assert(N->getOpcode() == ARMISD::BFI);
+
+ SDValue From = N->getOperand(1);
+ ToMask = ~cast<ConstantSDNode>(N->getOperand(2))->getAPIntValue();
+ FromMask = APInt::getLowBitsSet(ToMask.getBitWidth(), ToMask.countPopulation());
+
+ // If the Base came from a SHR #C, we can deduce that it is really testing bit
+ // #C in the base of the SHR.
+ if (From->getOpcode() == ISD::SRL &&
+ isa<ConstantSDNode>(From->getOperand(1))) {
+ APInt Shift = cast<ConstantSDNode>(From->getOperand(1))->getAPIntValue();
+ assert(Shift.getLimitedValue() < 32 && "Shift too large!");
+ FromMask <<= Shift.getLimitedValue(31);
+ From = From->getOperand(0);
+ }
+
+ return From;
+}
+
+// If A and B contain one contiguous set of bits, does A | B == A . B?
+//
+// Neither A nor B must be zero.
+static bool BitsProperlyConcatenate(const APInt &A, const APInt &B) {
+ unsigned LastActiveBitInA = A.countTrailingZeros();
+ unsigned FirstActiveBitInB = B.getBitWidth() - B.countLeadingZeros() - 1;
+ return LastActiveBitInA - 1 == FirstActiveBitInB;
+}
+
+static SDValue FindBFIToCombineWith(SDNode *N) {
+ // We have a BFI in N. Follow a possible chain of BFIs and find a BFI it can combine with,
+ // if one exists.
+ APInt ToMask, FromMask;
+ SDValue From = ParseBFI(N, ToMask, FromMask);
+ SDValue To = N->getOperand(0);
+
+ // Now check for a compatible BFI to merge with. We can pass through BFIs that
+ // aren't compatible, but not if they set the same bit in their destination as
+ // we do (or that of any BFI we're going to combine with).
+ SDValue V = To;
+ APInt CombinedToMask = ToMask;
+ while (V.getOpcode() == ARMISD::BFI) {
+ APInt NewToMask, NewFromMask;
+ SDValue NewFrom = ParseBFI(V.getNode(), NewToMask, NewFromMask);
+ if (NewFrom != From) {
+ // This BFI has a different base. Keep going.
+ CombinedToMask |= NewToMask;
+ V = V.getOperand(0);
+ continue;
+ }
+
+ // Do the written bits conflict with any we've seen so far?
+ if ((NewToMask & CombinedToMask).getBoolValue())
+ // Conflicting bits - bail out because going further is unsafe.
+ return SDValue();
+
+ // Are the new bits contiguous when combined with the old bits?
+ if (BitsProperlyConcatenate(ToMask, NewToMask) &&
+ BitsProperlyConcatenate(FromMask, NewFromMask))
+ return V;
+ if (BitsProperlyConcatenate(NewToMask, ToMask) &&
+ BitsProperlyConcatenate(NewFromMask, FromMask))
+ return V;
+
+ // We've seen a write to some bits, so track it.
+ CombinedToMask |= NewToMask;
+ // Keep going...
+ V = V.getOperand(0);
+ }
+
+ return SDValue();
+}
+
static SDValue PerformBFICombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SDValue N1 = N->getOperand(1);
if (N1.getOpcode() == ISD::AND) {
+ // (bfi A, (and B, Mask1), Mask2) -> (bfi A, B, Mask2) iff
+ // the bits being cleared by the AND are not demanded by the BFI.
ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1));
if (!N11C)
return SDValue();
return DCI.DAG.getNode(ARMISD::BFI, SDLoc(N), N->getValueType(0),
N->getOperand(0), N1.getOperand(0),
N->getOperand(2));
+ } else if (N->getOperand(0).getOpcode() == ARMISD::BFI) {
+ // We have a BFI of a BFI. Walk up the BFI chain to see how long it goes.
+ // Keep track of any consecutive bits set that all come from the same base
+ // value. We can combine these together into a single BFI.
+ SDValue CombineBFI = FindBFIToCombineWith(N);
+ if (CombineBFI == SDValue())
+ return SDValue();
+
+ // We've found a BFI.
+ APInt ToMask1, FromMask1;
+ SDValue From1 = ParseBFI(N, ToMask1, FromMask1);
+
+ APInt ToMask2, FromMask2;
+ SDValue From2 = ParseBFI(CombineBFI.getNode(), ToMask2, FromMask2);
+ assert(From1 == From2);
+ (void)From2;
+
+ // First, unlink CombineBFI.
+ DCI.DAG.ReplaceAllUsesWith(CombineBFI, CombineBFI.getOperand(0));
+ // Then create a new BFI, combining the two together.
+ APInt NewFromMask = FromMask1 | FromMask2;
+ APInt NewToMask = ToMask1 | ToMask2;
+
+ EVT VT = N->getValueType(0);
+ SDLoc dl(N);
+
+ if (NewFromMask[0] == 0)
+ From1 = DCI.DAG.getNode(
+ ISD::SRL, dl, VT, From1,
+ DCI.DAG.getConstant(NewFromMask.countTrailingZeros(), dl, VT));
+ return DCI.DAG.getNode(ARMISD::BFI, dl, VT, N->getOperand(0), From1,
+ DCI.DAG.getConstant(~NewToMask, dl, VT));
}
return SDValue();
}
return SDValue();
}
+static void computeKnownBits(SelectionDAG &DAG, SDValue Op, APInt &KnownZero,
+ APInt &KnownOne) {
+ if (Op.getOpcode() == ARMISD::BFI) {
+ // Conservatively, we can recurse down the first operand
+ // and just mask out all affected bits.
+ computeKnownBits(DAG, Op.getOperand(0), KnownZero, KnownOne);
+
+ // The operand to BFI is already a mask suitable for removing the bits it
+ // sets.
+ ConstantSDNode *CI = cast<ConstantSDNode>(Op.getOperand(2));
+ APInt Mask = CI->getAPIntValue();
+ KnownZero &= Mask;
+ KnownOne &= Mask;
+ return;
+ }
+ if (Op.getOpcode() == ARMISD::CMOV) {
+ APInt KZ2(KnownZero.getBitWidth(), 0);
+ APInt KO2(KnownOne.getBitWidth(), 0);
+ computeKnownBits(DAG, Op.getOperand(1), KnownZero, KnownOne);
+ computeKnownBits(DAG, Op.getOperand(2), KZ2, KO2);
+
+ KnownZero &= KZ2;
+ KnownOne &= KO2;
+ return;
+ }
+ return DAG.computeKnownBits(Op, KnownZero, KnownOne);
+}
+
+SDValue ARMTargetLowering::PerformCMOVToBFICombine(SDNode *CMOV, SelectionDAG &DAG) const {
+ // If we have a CMOV, OR and AND combination such as:
+ // if (x & CN)
+ // y |= CM;
+ //
+ // And:
+ // * CN is a single bit;
+ // * All bits covered by CM are known zero in y
+ //
+ // Then we can convert this into a sequence of BFI instructions. This will
+ // always be a win if CM is a single bit, will always be no worse than the
+ // TST&OR sequence if CM is two bits, and for thumb will be no worse if CM is
+ // three bits (due to the extra IT instruction).
+
+ SDValue Op0 = CMOV->getOperand(0);
+ SDValue Op1 = CMOV->getOperand(1);
+ auto CCNode = cast<ConstantSDNode>(CMOV->getOperand(2));
+ auto CC = CCNode->getAPIntValue().getLimitedValue();
+ SDValue CmpZ = CMOV->getOperand(4);
+
+ // The compare must be against zero.
+ if (!isNullConstant(CmpZ->getOperand(1)))
+ return SDValue();
+
+ assert(CmpZ->getOpcode() == ARMISD::CMPZ);
+ SDValue And = CmpZ->getOperand(0);
+ if (And->getOpcode() != ISD::AND)
+ return SDValue();
+ ConstantSDNode *AndC = dyn_cast<ConstantSDNode>(And->getOperand(1));
+ if (!AndC || !AndC->getAPIntValue().isPowerOf2())
+ return SDValue();
+ SDValue X = And->getOperand(0);
+
+ if (CC == ARMCC::EQ) {
+ // We're performing an "equal to zero" compare. Swap the operands so we
+ // canonicalize on a "not equal to zero" compare.
+ std::swap(Op0, Op1);
+ } else {
+ assert(CC == ARMCC::NE && "How can a CMPZ node not be EQ or NE?");
+ }
+
+ if (Op1->getOpcode() != ISD::OR)
+ return SDValue();
+
+ ConstantSDNode *OrC = dyn_cast<ConstantSDNode>(Op1->getOperand(1));
+ if (!OrC)
+ return SDValue();
+ SDValue Y = Op1->getOperand(0);
+
+ if (Op0 != Y)
+ return SDValue();
+
+ // Now, is it profitable to continue?
+ APInt OrCI = OrC->getAPIntValue();
+ unsigned Heuristic = Subtarget->isThumb() ? 3 : 2;
+ if (OrCI.countPopulation() > Heuristic)
+ return SDValue();
+
+ // Lastly, can we determine that the bits defined by OrCI
+ // are zero in Y?
+ APInt KnownZero, KnownOne;
+ computeKnownBits(DAG, Y, KnownZero, KnownOne);
+ if ((OrCI & KnownZero) != OrCI)
+ return SDValue();
+
+ // OK, we can do the combine.
+ SDValue V = Y;
+ SDLoc dl(X);
+ EVT VT = X.getValueType();
+ unsigned BitInX = AndC->getAPIntValue().logBase2();
+
+ if (BitInX != 0) {
+ // We must shift X first.
+ X = DAG.getNode(ISD::SRL, dl, VT, X,
+ DAG.getConstant(BitInX, dl, VT));
+ }
+
+ for (unsigned BitInY = 0, NumActiveBits = OrCI.getActiveBits();
+ BitInY < NumActiveBits; ++BitInY) {
+ if (OrCI[BitInY] == 0)
+ continue;
+ APInt Mask(VT.getSizeInBits(), 0);
+ Mask.setBit(BitInY);
+ V = DAG.getNode(ARMISD::BFI, dl, VT, V, X,
+ // Confusingly, the operand is an *inverted* mask.
+ DAG.getConstant(~Mask, dl, VT));
+ }
+
+ return V;
+}
+
/// PerformCMOVCombine - Target-specific DAG combining for ARMISD::CMOV.
SDValue
ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const {
ARMCC::CondCodes CC =
(ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue();
+ // BFI is only available on V6T2+.
+ if (!Subtarget->isThumb1Only() && Subtarget->hasV6T2Ops()) {
+ SDValue R = PerformCMOVToBFICombine(N, DAG);
+ if (R)
+ return R;
+ }
+
// Simplify
// mov r1, r0
// cmp r1, x
return false;
}
+bool ARMTargetLowering::isCheapToSpeculateCttz() const {
+ return Subtarget->hasV6T2Ops();
+}
+
+bool ARMTargetLowering::isCheapToSpeculateCtlz() const {
+ return Subtarget->hasV6T2Ops();
+}
+
Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
AtomicOrdering Ord) const {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
bool IsIntArray = Ty->isArrayTy() && Ty->getArrayElementType()->isIntegerTy();
return IsHA || IsIntArray;
}
+
+unsigned ARMTargetLowering::getExceptionPointerRegister(
+ const Constant *PersonalityFn) const {
+ // Platforms which do not use SjLj EH may return values in these registers
+ // via the personality function.
+ return Subtarget->useSjLjEH() ? ARM::NoRegister : ARM::R0;
+}
+
+unsigned ARMTargetLowering::getExceptionSelectorRegister(
+ const Constant *PersonalityFn) const {
+ // Platforms which do not use SjLj EH may return values in these registers
+ // via the personality function.
+ return Subtarget->useSjLjEH() ? ARM::NoRegister : ARM::R1;
+}
projects / oota-llvm.git / blobdiff