//===----------------------------------------------------------------------===//
#include "AArch64ISelLowering.h"
+#include "AArch64CallingConvention.h"
#include "AArch64MachineFunctionInfo.h"
#include "AArch64PerfectShuffle.h"
#include "AArch64Subtarget.h"
cl::desc("Allow AArch64 SLI/SRI formation"),
cl::init(false));
-//===----------------------------------------------------------------------===//
-// AArch64 Lowering public interface.
-//===----------------------------------------------------------------------===//
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
- if (TT.isOSBinFormatMachO())
- return new AArch64_MachoTargetObjectFile();
-
- return new AArch64_ELFTargetObjectFile();
-}
-
-AArch64TargetLowering::AArch64TargetLowering(TargetMachine &TM)
- : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
- Subtarget = &TM.getSubtarget<AArch64Subtarget>();
+AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
+ const AArch64Subtarget &STI)
+ : TargetLowering(TM), Subtarget(&STI) {
// AArch64 doesn't have comparisons which set GPRs or setcc instructions, so
// we have to make something up. Arbitrarily, choose ZeroOrOne.
}
// Compute derived properties from the register classes
- computeRegisterProperties();
+ computeRegisterProperties(Subtarget->getRegisterInfo());
// Provide all sorts of operation actions
setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
}
+ // Make floating-point constants legal for the large code model, so they don't
+ // become loads from the constant pool.
+ if (Subtarget->isTargetMachO() && TM.getCodeModel() == CodeModel::Large) {
+ setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
+ setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
+ }
+
// AArch64 does not have floating-point extending loads, i1 sign-extending
// load, floating-point truncating stores, or v2i32->v2i16 truncating store.
- setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
- setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
- setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
- setLoadExtAction(ISD::EXTLOAD, MVT::f80, Expand);
- setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Expand);
+ for (MVT VT : MVT::fp_valuetypes()) {
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f64, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, MVT::f80, Expand);
+ }
+ for (MVT VT : MVT::integer_valuetypes())
+ setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Expand);
+
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
// AArch64 doesn't have MUL.2d:
setOperationAction(ISD::MUL, MVT::v2i64, Expand);
+ // Custom handling for some quad-vector types to detect MULL.
+ setOperationAction(ISD::MUL, MVT::v8i16, Custom);
+ setOperationAction(ISD::MUL, MVT::v4i32, Custom);
+ setOperationAction(ISD::MUL, MVT::v2i64, Custom);
+
setOperationAction(ISD::ANY_EXTEND, MVT::v4i32, Legal);
setTruncStoreAction(MVT::v2i32, MVT::v2i16, Expand);
// Likewise, narrowing and extending vector loads/stores aren't handled
// directly.
- for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
-
- setOperationAction(ISD::SIGN_EXTEND_INREG, (MVT::SimpleValueType)VT,
- Expand);
-
- setOperationAction(ISD::MULHS, (MVT::SimpleValueType)VT, Expand);
- setOperationAction(ISD::SMUL_LOHI, (MVT::SimpleValueType)VT, Expand);
- setOperationAction(ISD::MULHU, (MVT::SimpleValueType)VT, Expand);
- setOperationAction(ISD::UMUL_LOHI, (MVT::SimpleValueType)VT, Expand);
-
- setOperationAction(ISD::BSWAP, (MVT::SimpleValueType)VT, Expand);
-
- for (unsigned InnerVT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
- InnerVT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++InnerVT)
- setTruncStoreAction((MVT::SimpleValueType)VT,
- (MVT::SimpleValueType)InnerVT, Expand);
- setLoadExtAction(ISD::SEXTLOAD, (MVT::SimpleValueType)VT, Expand);
- setLoadExtAction(ISD::ZEXTLOAD, (MVT::SimpleValueType)VT, Expand);
- setLoadExtAction(ISD::EXTLOAD, (MVT::SimpleValueType)VT, Expand);
+ for (MVT VT : MVT::vector_valuetypes()) {
+ setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
+
+ setOperationAction(ISD::MULHS, VT, Expand);
+ setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+ setOperationAction(ISD::MULHU, VT, Expand);
+ setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+
+ setOperationAction(ISD::BSWAP, VT, Expand);
+
+ for (MVT InnerVT : MVT::vector_valuetypes()) {
+ setTruncStoreAction(VT, InnerVT, Expand);
+ setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
+ setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
+ setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
+ }
}
// AArch64 has implementations of a lot of rounding-like FP operations.
setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand);
setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand);
setOperationAction(ISD::VSELECT, VT.getSimpleVT(), Expand);
- setLoadExtAction(ISD::EXTLOAD, VT.getSimpleVT(), Expand);
+ for (MVT InnerVT : MVT::all_valuetypes())
+ setLoadExtAction(ISD::EXTLOAD, InnerVT, VT.getSimpleVT(), Expand);
// CNT supports only B element sizes.
if (VT != MVT::v8i8 && VT != MVT::v16i8)
return MVT::i64;
}
-unsigned AArch64TargetLowering::getMaximalGlobalOffset() const {
- // FIXME: On AArch64, this depends on the type.
- // Basically, the addressable offsets are up to 4095 * Ty.getSizeInBytes().
- // and the offset has to be a multiple of the related size in bytes.
- return 4095;
-}
-
FastISel *
AArch64TargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
const TargetLibraryInfo *libInfo) const {
case AArch64ISD::ST2LANEpost: return "AArch64ISD::ST2LANEpost";
case AArch64ISD::ST3LANEpost: return "AArch64ISD::ST3LANEpost";
case AArch64ISD::ST4LANEpost: return "AArch64ISD::ST4LANEpost";
+ case AArch64ISD::SMULL: return "AArch64ISD::SMULL";
+ case AArch64ISD::UMULL: return "AArch64ISD::UMULL";
}
}
// EndBB:
// Dest = PHI [IfTrue, TrueBB], [IfFalse, OrigBB]
- const TargetInstrInfo *TII =
- getTargetMachine().getSubtargetImpl()->getInstrInfo();
MachineFunction *MF = MBB->getParent();
+ const TargetInstrInfo *TII = Subtarget->getInstrInfo();
const BasicBlock *LLVM_BB = MBB->getBasicBlock();
DebugLoc DL = MI->getDebugLoc();
MachineFunction::iterator It = MBB;
break;
case ISD::SETLE:
case ISD::SETGT:
- if ((VT == MVT::i32 && C != 0x7fffffff &&
+ if ((VT == MVT::i32 && C != INT32_MAX &&
isLegalArithImmed((uint32_t)(C + 1))) ||
- (VT == MVT::i64 && C != 0x7ffffffffffffffULL &&
+ (VT == MVT::i64 && C != INT64_MAX &&
isLegalArithImmed(C + 1ULL))) {
CC = (CC == ISD::SETLE) ? ISD::SETLT : ISD::SETGE;
C = (VT == MVT::i32) ? (uint32_t)(C + 1) : C + 1;
break;
case ISD::SETULE:
case ISD::SETUGT:
- if ((VT == MVT::i32 && C != 0xffffffff &&
+ if ((VT == MVT::i32 && C != UINT32_MAX &&
isLegalArithImmed((uint32_t)(C + 1))) ||
- (VT == MVT::i64 && C != 0xfffffffffffffffULL &&
+ (VT == MVT::i64 && C != UINT64_MAX &&
isLegalArithImmed(C + 1ULL))) {
CC = (CC == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
C = (VT == MVT::i32) ? (uint32_t)(C + 1) : C + 1;
SDValue AArch64TargetLowering::LowerF128Call(SDValue Op, SelectionDAG &DAG,
RTLIB::Libcall Call) const {
- SmallVector<SDValue, 2> Ops;
- for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i)
- Ops.push_back(Op.getOperand(i));
-
+ SmallVector<SDValue, 2> Ops(Op->op_begin(), Op->op_end());
return makeLibCall(DAG, Call, MVT::f128, &Ops[0], Ops.size(), false,
SDLoc(Op)).first;
}
else
LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(), Op.getValueType());
- SmallVector<SDValue, 2> Ops;
- for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i)
- Ops.push_back(Op.getOperand(i));
-
+ SmallVector<SDValue, 2> Ops(Op->op_begin(), Op->op_end());
return makeLibCall(DAG, LC, Op.getValueType(), &Ops[0], Ops.size(), false,
SDLoc(Op)).first;
}
(ArgVT == MVT::f64) ? "__sincos_stret" : "__sincosf_stret";
SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());
- StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+ StructType *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(DAG.getEntryNode())
.setCallee(CallingConv::Fast, RetTy, Callee, std::move(Args), 0);
0);
}
+static EVT getExtensionTo64Bits(const EVT &OrigVT) {
+ if (OrigVT.getSizeInBits() >= 64)
+ return OrigVT;
+
+ assert(OrigVT.isSimple() && "Expecting a simple value type");
+
+ MVT::SimpleValueType OrigSimpleTy = OrigVT.getSimpleVT().SimpleTy;
+ switch (OrigSimpleTy) {
+ default: llvm_unreachable("Unexpected Vector Type");
+ case MVT::v2i8:
+ case MVT::v2i16:
+ return MVT::v2i32;
+ case MVT::v4i8:
+ return MVT::v4i16;
+ }
+}
+
+static SDValue addRequiredExtensionForVectorMULL(SDValue N, SelectionDAG &DAG,
+ const EVT &OrigTy,
+ const EVT &ExtTy,
+ unsigned ExtOpcode) {
+ // The vector originally had a size of OrigTy. It was then extended to ExtTy.
+ // We expect the ExtTy to be 128-bits total. If the OrigTy is less than
+ // 64-bits we need to insert a new extension so that it will be 64-bits.
+ assert(ExtTy.is128BitVector() && "Unexpected extension size");
+ if (OrigTy.getSizeInBits() >= 64)
+ return N;
+
+ // Must extend size to at least 64 bits to be used as an operand for VMULL.
+ EVT NewVT = getExtensionTo64Bits(OrigTy);
+
+ return DAG.getNode(ExtOpcode, SDLoc(N), NewVT, N);
+}
+
+static bool isExtendedBUILD_VECTOR(SDNode *N, SelectionDAG &DAG,
+ bool isSigned) {
+ EVT VT = N->getValueType(0);
+
+ if (N->getOpcode() != ISD::BUILD_VECTOR)
+ return false;
+
+ for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
+ SDNode *Elt = N->getOperand(i).getNode();
+ if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Elt)) {
+ unsigned EltSize = VT.getVectorElementType().getSizeInBits();
+ unsigned HalfSize = EltSize / 2;
+ if (isSigned) {
+ if (!isIntN(HalfSize, C->getSExtValue()))
+ return false;
+ } else {
+ if (!isUIntN(HalfSize, C->getZExtValue()))
+ return false;
+ }
+ continue;
+ }
+ return false;
+ }
+
+ return true;
+}
+
+static SDValue skipExtensionForVectorMULL(SDNode *N, SelectionDAG &DAG) {
+ if (N->getOpcode() == ISD::SIGN_EXTEND || N->getOpcode() == ISD::ZERO_EXTEND)
+ return addRequiredExtensionForVectorMULL(N->getOperand(0), DAG,
+ N->getOperand(0)->getValueType(0),
+ N->getValueType(0),
+ N->getOpcode());
+
+ assert(N->getOpcode() == ISD::BUILD_VECTOR && "expected BUILD_VECTOR");
+ EVT VT = N->getValueType(0);
+ unsigned EltSize = VT.getVectorElementType().getSizeInBits() / 2;
+ unsigned NumElts = VT.getVectorNumElements();
+ MVT TruncVT = MVT::getIntegerVT(EltSize);
+ SmallVector<SDValue, 8> Ops;
+ for (unsigned i = 0; i != NumElts; ++i) {
+ ConstantSDNode *C = cast<ConstantSDNode>(N->getOperand(i));
+ const APInt &CInt = C->getAPIntValue();
+ // Element types smaller than 32 bits are not legal, so use i32 elements.
+ // The values are implicitly truncated so sext vs. zext doesn't matter.
+ Ops.push_back(DAG.getConstant(CInt.zextOrTrunc(32), MVT::i32));
+ }
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N),
+ MVT::getVectorVT(TruncVT, NumElts), Ops);
+}
+
+static bool isSignExtended(SDNode *N, SelectionDAG &DAG) {
+ if (N->getOpcode() == ISD::SIGN_EXTEND)
+ return true;
+ if (isExtendedBUILD_VECTOR(N, DAG, true))
+ return true;
+ return false;
+}
+
+static bool isZeroExtended(SDNode *N, SelectionDAG &DAG) {
+ if (N->getOpcode() == ISD::ZERO_EXTEND)
+ return true;
+ if (isExtendedBUILD_VECTOR(N, DAG, false))
+ return true;
+ return false;
+}
+
+static bool isAddSubSExt(SDNode *N, SelectionDAG &DAG) {
+ unsigned Opcode = N->getOpcode();
+ if (Opcode == ISD::ADD || Opcode == ISD::SUB) {
+ SDNode *N0 = N->getOperand(0).getNode();
+ SDNode *N1 = N->getOperand(1).getNode();
+ return N0->hasOneUse() && N1->hasOneUse() &&
+ isSignExtended(N0, DAG) && isSignExtended(N1, DAG);
+ }
+ return false;
+}
+
+static bool isAddSubZExt(SDNode *N, SelectionDAG &DAG) {
+ unsigned Opcode = N->getOpcode();
+ if (Opcode == ISD::ADD || Opcode == ISD::SUB) {
+ SDNode *N0 = N->getOperand(0).getNode();
+ SDNode *N1 = N->getOperand(1).getNode();
+ return N0->hasOneUse() && N1->hasOneUse() &&
+ isZeroExtended(N0, DAG) && isZeroExtended(N1, DAG);
+ }
+ return false;
+}
+
+static SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) {
+ // Multiplications are only custom-lowered for 128-bit vectors so that
+ // VMULL can be detected. Otherwise v2i64 multiplications are not legal.
+ EVT VT = Op.getValueType();
+ assert(VT.is128BitVector() && VT.isInteger() &&
+ "unexpected type for custom-lowering ISD::MUL");
+ SDNode *N0 = Op.getOperand(0).getNode();
+ SDNode *N1 = Op.getOperand(1).getNode();
+ unsigned NewOpc = 0;
+ bool isMLA = false;
+ bool isN0SExt = isSignExtended(N0, DAG);
+ bool isN1SExt = isSignExtended(N1, DAG);
+ if (isN0SExt && isN1SExt)
+ NewOpc = AArch64ISD::SMULL;
+ else {
+ bool isN0ZExt = isZeroExtended(N0, DAG);
+ bool isN1ZExt = isZeroExtended(N1, DAG);
+ if (isN0ZExt && isN1ZExt)
+ NewOpc = AArch64ISD::UMULL;
+ else if (isN1SExt || isN1ZExt) {
+ // Look for (s/zext A + s/zext B) * (s/zext C). We want to turn these
+ // into (s/zext A * s/zext C) + (s/zext B * s/zext C)
+ if (isN1SExt && isAddSubSExt(N0, DAG)) {
+ NewOpc = AArch64ISD::SMULL;
+ isMLA = true;
+ } else if (isN1ZExt && isAddSubZExt(N0, DAG)) {
+ NewOpc = AArch64ISD::UMULL;
+ isMLA = true;
+ } else if (isN0ZExt && isAddSubZExt(N1, DAG)) {
+ std::swap(N0, N1);
+ NewOpc = AArch64ISD::UMULL;
+ isMLA = true;
+ }
+ }
+
+ if (!NewOpc) {
+ if (VT == MVT::v2i64)
+ // Fall through to expand this. It is not legal.
+ return SDValue();
+ else
+ // Other vector multiplications are legal.
+ return Op;
+ }
+ }
+
+ // Legalize to a S/UMULL instruction
+ SDLoc DL(Op);
+ SDValue Op0;
+ SDValue Op1 = skipExtensionForVectorMULL(N1, DAG);
+ if (!isMLA) {
+ Op0 = skipExtensionForVectorMULL(N0, DAG);
+ assert(Op0.getValueType().is64BitVector() &&
+ Op1.getValueType().is64BitVector() &&
+ "unexpected types for extended operands to VMULL");
+ return DAG.getNode(NewOpc, DL, VT, Op0, Op1);
+ }
+ // Optimizing (zext A + zext B) * C, to (S/UMULL A, C) + (S/UMULL B, C) during
+ // isel lowering to take advantage of no-stall back to back s/umul + s/umla.
+ // This is true for CPUs with accumulate forwarding such as Cortex-A53/A57
+ SDValue N00 = skipExtensionForVectorMULL(N0->getOperand(0).getNode(), DAG);
+ SDValue N01 = skipExtensionForVectorMULL(N0->getOperand(1).getNode(), DAG);
+ EVT Op1VT = Op1.getValueType();
+ return DAG.getNode(N0->getOpcode(), DL, VT,
+ DAG.getNode(NewOpc, DL, VT,
+ DAG.getNode(ISD::BITCAST, DL, Op1VT, N00), Op1),
+ DAG.getNode(NewOpc, DL, VT,
+ DAG.getNode(ISD::BITCAST, DL, Op1VT, N01), Op1));
+}
SDValue AArch64TargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
return LowerFP_TO_INT(Op, DAG);
case ISD::FSINCOS:
return LowerFSINCOS(Op, DAG);
+ case ISD::MUL:
+ return LowerMUL(Op, DAG);
}
}
llvm_unreachable("Unsupported calling convention.");
case CallingConv::WebKit_JS:
return CC_AArch64_WebKit_JS;
+ case CallingConv::GHC:
+ return CC_AArch64_GHC;
case CallingConv::C:
case CallingConv::Fast:
if (!Subtarget->isTargetDarwin())
unsigned CurArgIdx = 0;
for (unsigned i = 0; i != NumArgs; ++i) {
MVT ValVT = Ins[i].VT;
- std::advance(CurOrigArg, Ins[i].OrigArgIndex - CurArgIdx);
- CurArgIdx = Ins[i].OrigArgIndex;
-
- // Get type of the original argument.
- EVT ActualVT = getValueType(CurOrigArg->getType(), /*AllowUnknown*/ true);
- MVT ActualMVT = ActualVT.isSimple() ? ActualVT.getSimpleVT() : MVT::Other;
- // If ActualMVT is i1/i8/i16, we should set LocVT to i8/i8/i16.
- if (ActualMVT == MVT::i1 || ActualMVT == MVT::i8)
- ValVT = MVT::i8;
- else if (ActualMVT == MVT::i16)
- ValVT = MVT::i16;
+ if (Ins[i].isOrigArg()) {
+ std::advance(CurOrigArg, Ins[i].getOrigArgIndex() - CurArgIdx);
+ CurArgIdx = Ins[i].getOrigArgIndex();
+ // Get type of the original argument.
+ EVT ActualVT = getValueType(CurOrigArg->getType(), /*AllowUnknown*/ true);
+ MVT ActualMVT = ActualVT.isSimple() ? ActualVT.getSimpleVT() : MVT::Other;
+ // If ActualMVT is i1/i8/i16, we should set LocVT to i8/i8/i16.
+ if (ActualMVT == MVT::i1 || ActualMVT == MVT::i8)
+ ValVT = MVT::i8;
+ else if (ActualMVT == MVT::i16)
+ ValVT = MVT::i16;
+ }
CCAssignFn *AssignFn = CCAssignFnForCall(CallConv, /*IsVarArg=*/false);
bool Res =
AssignFn(i, ValVT, ValVT, CCValAssign::Full, Ins[i].Flags, CCInfo);
unsigned ArgSize = VA.getValVT().getSizeInBits() / 8;
uint32_t BEAlign = 0;
- if (ArgSize < 8 && !Subtarget->isLittleEndian())
+ if (!Subtarget->isLittleEndian() && ArgSize < 8 &&
+ !Ins[i].Flags.isInConsecutiveRegs())
BEAlign = 8 - ArgSize;
int FI = MFI->CreateFixedObject(ArgSize, ArgOffset + BEAlign, true);
ArgValue = DAG.getExtLoad(ExtType, DL, VA.getLocVT(), Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
- MemVT, false, false, false, 0, nullptr);
+ MemVT, false, false, false, 0);
InVals.push_back(ArgValue);
}
AArch64::X3, AArch64::X4, AArch64::X5,
AArch64::X6, AArch64::X7 };
static const unsigned NumGPRArgRegs = array_lengthof(GPRArgRegs);
- unsigned FirstVariadicGPR =
- CCInfo.getFirstUnallocated(GPRArgRegs, NumGPRArgRegs);
+ unsigned FirstVariadicGPR = CCInfo.getFirstUnallocated(GPRArgRegs);
unsigned GPRSaveSize = 8 * (NumGPRArgRegs - FirstVariadicGPR);
int GPRIdx = 0;
AArch64::Q0, AArch64::Q1, AArch64::Q2, AArch64::Q3,
AArch64::Q4, AArch64::Q5, AArch64::Q6, AArch64::Q7};
static const unsigned NumFPRArgRegs = array_lengthof(FPRArgRegs);
- unsigned FirstVariadicFPR =
- CCInfo.getFirstUnallocated(FPRArgRegs, NumFPRArgRegs);
+ unsigned FirstVariadicFPR = CCInfo.getFirstUnallocated(FPRArgRegs);
unsigned FPRSaveSize = 16 * (NumFPRArgRegs - FirstVariadicFPR);
int FPRIdx = 0;
// 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())
+ const Triple TT(getTargetMachine().getTargetTriple());
+ if (GV->hasExternalWeakLinkage() &&
+ (!TT.isOSWindows() || TT.isOSBinFormatELF() || TT.isOSBinFormatMachO()))
return false;
}
unsigned OpSize = Flags.isByVal() ? Flags.getByValSize() * 8
: VA.getValVT().getSizeInBits();
OpSize = (OpSize + 7) / 8;
- if (!Subtarget->isLittleEndian() && !Flags.isByVal()) {
+ if (!Subtarget->isLittleEndian() && !Flags.isByVal() &&
+ !Flags.isInConsecutiveRegs()) {
if (OpSize < 8)
BEAlign = 8 - OpSize;
}
// Add a register mask operand representing the call-preserved registers.
const uint32_t *Mask;
- const TargetRegisterInfo *TRI =
- getTargetMachine().getSubtargetImpl()->getRegisterInfo();
- const AArch64RegisterInfo *ARI =
- static_cast<const AArch64RegisterInfo *>(TRI);
+ const AArch64RegisterInfo *TRI = Subtarget->getRegisterInfo();
if (IsThisReturn) {
// For 'this' returns, use the X0-preserving mask if applicable
- Mask = ARI->getThisReturnPreservedMask(CallConv);
+ Mask = TRI->getThisReturnPreservedMask(CallConv);
if (!Mask) {
IsThisReturn = false;
- Mask = ARI->getCallPreservedMask(CallConv);
+ Mask = TRI->getCallPreservedMask(CallConv);
}
} else
- Mask = ARI->getCallPreservedMask(CallConv);
+ Mask = TRI->getCallPreservedMask(CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
// TLS calls preserve all registers except those that absolutely must be
// trashed: X0 (it takes an argument), LR (it's a call) and NZCV (let's not be
// silly).
- const TargetRegisterInfo *TRI =
- getTargetMachine().getSubtargetImpl()->getRegisterInfo();
- const AArch64RegisterInfo *ARI =
- static_cast<const AArch64RegisterInfo *>(TRI);
- const uint32_t *Mask = ARI->getTLSCallPreservedMask();
+ const uint32_t *Mask =
+ Subtarget->getRegisterInfo()->getTLSCallPreservedMask();
// Finally, we can make the call. This is just a degenerate version of a
// normal AArch64 call node: x0 takes the address of the descriptor, and
// TLS calls preserve all registers except those that absolutely must be
// trashed: X0 (it takes an argument), LR (it's a call) and NZCV (let's not be
// silly).
- const TargetRegisterInfo *TRI =
- getTargetMachine().getSubtargetImpl()->getRegisterInfo();
- const AArch64RegisterInfo *ARI =
- static_cast<const AArch64RegisterInfo *>(TRI);
- const uint32_t *Mask = ARI->getTLSCallPreservedMask();
+ const uint32_t *Mask =
+ Subtarget->getRegisterInfo()->getTLSCallPreservedMask();
// The function takes only one argument: the address of the descriptor itself
// in X0.
OFCC = getInvertedCondCode(OFCC);
SDValue CCVal = DAG.getConstant(OFCC, MVT::i32);
- return DAG.getNode(AArch64ISD::BRCOND, SDLoc(LHS), MVT::Other, Chain, Dest,
- CCVal, Overflow);
+ return DAG.getNode(AArch64ISD::BRCOND, dl, MVT::Other, Chain, Dest, CCVal,
+ Overflow);
}
if (LHS.getValueType().isInteger()) {
EVT VecVT;
EVT EltVT;
- SDValue EltMask, VecVal1, VecVal2;
+ uint64_t EltMask;
+ SDValue VecVal1, VecVal2;
if (VT == MVT::f32 || VT == MVT::v2f32 || VT == MVT::v4f32) {
EltVT = MVT::i32;
VecVT = MVT::v4i32;
- EltMask = DAG.getConstant(0x80000000ULL, EltVT);
+ EltMask = 0x80000000ULL;
if (!VT.isVector()) {
VecVal1 = DAG.getTargetInsertSubreg(AArch64::ssub, DL, VecVT,
// We want to materialize a mask with the the high bit set, but the AdvSIMD
// immediate moves cannot materialize that in a single instruction for
// 64-bit elements. Instead, materialize zero and then negate it.
- EltMask = DAG.getConstant(0, EltVT);
+ EltMask = 0;
if (!VT.isVector()) {
VecVal1 = DAG.getTargetInsertSubreg(AArch64::dsub, DL, VecVT,
llvm_unreachable("Invalid type for copysign!");
}
- std::vector<SDValue> BuildVectorOps;
- for (unsigned i = 0; i < VecVT.getVectorNumElements(); ++i)
- BuildVectorOps.push_back(EltMask);
-
- SDValue BuildVec = DAG.getNode(ISD::BUILD_VECTOR, DL, VecVT, BuildVectorOps);
+ SDValue BuildVec = DAG.getConstant(EltMask, VecVT);
// If we couldn't materialize the mask above, then the mask vector will be
// the zero vector, and we need to negate it here.
}
SDValue AArch64TargetLowering::LowerCTPOP(SDValue Op, SelectionDAG &DAG) const {
- if (DAG.getMachineFunction().getFunction()->getAttributes().hasAttribute(
- AttributeSet::FunctionIndex, Attribute::NoImplicitFloat))
+ if (DAG.getMachineFunction().getFunction()->hasFnAttribute(
+ Attribute::NoImplicitFloat))
+ return SDValue();
+
+ if (!Subtarget->hasNEON())
return SDValue();
// While there is no integer popcount instruction, it can
SDValue Val = Op.getOperand(0);
SDLoc DL(Op);
EVT VT = Op.getValueType();
- SDValue ZeroVec = DAG.getUNDEF(MVT::v8i8);
- SDValue VecVal;
- if (VT == MVT::i32) {
- VecVal = DAG.getNode(ISD::BITCAST, DL, MVT::f32, Val);
- VecVal = DAG.getTargetInsertSubreg(AArch64::ssub, DL, MVT::v8i8, ZeroVec,
- VecVal);
- } else {
- VecVal = DAG.getNode(ISD::BITCAST, DL, MVT::v8i8, Val);
- }
+ if (VT == MVT::i32)
+ Val = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, Val);
+ Val = DAG.getNode(ISD::BITCAST, DL, MVT::v8i8, Val);
- SDValue CtPop = DAG.getNode(ISD::CTPOP, DL, MVT::v8i8, VecVal);
+ SDValue CtPop = DAG.getNode(ISD::CTPOP, DL, MVT::v8i8, Val);
SDValue UaddLV = DAG.getNode(
ISD::INTRINSIC_WO_CHAIN, DL, MVT::i32,
DAG.getConstant(Intrinsic::aarch64_neon_uaddlv, MVT::i32), CtPop);
std::pair<unsigned, const TargetRegisterClass *>
AArch64TargetLowering::getRegForInlineAsmConstraint(
- const std::string &Constraint, MVT VT) const {
+ const TargetRegisterInfo *TRI, const std::string &Constraint,
+ MVT VT) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
case 'r':
// Use the default implementation in TargetLowering to convert the register
// constraint into a member of a register class.
std::pair<unsigned, const TargetRegisterClass *> Res;
- Res = TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+ Res = TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
// Not found as a standard register?
if (!Res.second) {
SDValue SourceVec = V.getOperand(0);
auto Source = std::find(Sources.begin(), Sources.end(), SourceVec);
if (Source == Sources.end())
- Sources.push_back(ShuffleSourceInfo(SourceVec));
+ Source = Sources.insert(Sources.end(), ShuffleSourceInfo(SourceVec));
// Update the minimum and maximum lane number seen.
unsigned EltNo = cast<ConstantSDNode>(V.getOperand(1))->getZExtValue();
// This stage of the search produces a source with the same element type as
// the original, but with a total width matching the BUILD_VECTOR output.
EVT EltVT = SrcVT.getVectorElementType();
- EVT DestVT = EVT::getVectorVT(*DAG.getContext(), EltVT,
- VT.getSizeInBits() / EltVT.getSizeInBits());
+ unsigned NumSrcElts = VT.getSizeInBits() / EltVT.getSizeInBits();
+ EVT DestVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumSrcElts);
if (SrcVT.getSizeInBits() < VT.getSizeInBits()) {
assert(2 * SrcVT.getSizeInBits() == VT.getSizeInBits());
assert(SrcVT.getSizeInBits() == 2 * VT.getSizeInBits());
- if (Src.MaxElt - Src.MinElt >= NumElts) {
+ if (Src.MaxElt - Src.MinElt >= NumSrcElts) {
// Span too large for a VEXT to cope
return SDValue();
}
- if (Src.MinElt >= NumElts) {
+ if (Src.MinElt >= NumSrcElts) {
// The extraction can just take the second half
Src.ShuffleVec =
DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
- DAG.getIntPtrConstant(NumElts));
- Src.WindowBase = -NumElts;
- } else if (Src.MaxElt < NumElts) {
+ DAG.getConstant(NumSrcElts, MVT::i64));
+ Src.WindowBase = -NumSrcElts;
+ } else if (Src.MaxElt < NumSrcElts) {
// The extraction can just take the first half
- Src.ShuffleVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT,
- Src.ShuffleVec, DAG.getIntPtrConstant(0));
+ Src.ShuffleVec =
+ DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
+ DAG.getConstant(0, MVT::i64));
} else {
// An actual VEXT is needed
- SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT,
- Src.ShuffleVec, DAG.getIntPtrConstant(0));
+ SDValue VEXTSrc1 =
+ DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
+ DAG.getConstant(0, MVT::i64));
SDValue VEXTSrc2 =
DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, DestVT, Src.ShuffleVec,
- DAG.getIntPtrConstant(NumElts));
+ DAG.getConstant(NumSrcElts, MVT::i64));
unsigned Imm = Src.MinElt * getExtFactor(VEXTSrc1);
Src.ShuffleVec = DAG.getNode(AArch64ISD::EXT, dl, DestVT, VEXTSrc1,
AArch64CC::CondCode CC, bool NoNans, EVT VT,
SDLoc dl, SelectionDAG &DAG) {
EVT SrcVT = LHS.getValueType();
+ assert(VT.getSizeInBits() == SrcVT.getSizeInBits() &&
+ "function only supposed to emit natural comparisons");
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(RHS.getNode());
APInt CnstBits(VT.getSizeInBits(), 0);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
+ EVT CmpVT = LHS.getValueType().changeVectorElementTypeToInteger();
SDLoc dl(Op);
if (LHS.getValueType().getVectorElementType().isInteger()) {
assert(LHS.getValueType() == RHS.getValueType());
AArch64CC::CondCode AArch64CC = changeIntCCToAArch64CC(CC);
- return EmitVectorComparison(LHS, RHS, AArch64CC, false, Op.getValueType(),
- dl, DAG);
+ SDValue Cmp =
+ EmitVectorComparison(LHS, RHS, AArch64CC, false, CmpVT, dl, DAG);
+ return DAG.getSExtOrTrunc(Cmp, dl, Op.getValueType());
}
assert(LHS.getValueType().getVectorElementType() == MVT::f32 ||
bool NoNaNs = getTargetMachine().Options.NoNaNsFPMath;
SDValue Cmp =
- EmitVectorComparison(LHS, RHS, CC1, NoNaNs, Op.getValueType(), dl, DAG);
+ EmitVectorComparison(LHS, RHS, CC1, NoNaNs, CmpVT, dl, DAG);
if (!Cmp.getNode())
return SDValue();
if (CC2 != AArch64CC::AL) {
SDValue Cmp2 =
- EmitVectorComparison(LHS, RHS, CC2, NoNaNs, Op.getValueType(), dl, DAG);
+ EmitVectorComparison(LHS, RHS, CC2, NoNaNs, CmpVT, dl, DAG);
if (!Cmp2.getNode())
return SDValue();
- Cmp = DAG.getNode(ISD::OR, dl, Cmp.getValueType(), Cmp, Cmp2);
+ Cmp = DAG.getNode(ISD::OR, dl, CmpVT, Cmp, Cmp2);
}
+ Cmp = DAG.getSExtOrTrunc(Cmp, dl, Op.getValueType());
+
if (ShouldInvert)
return Cmp = DAG.getNOT(dl, Cmp, Cmp.getValueType());
return NumBits1 > NumBits2;
}
+/// Check if it is profitable to hoist instruction in then/else to if.
+/// Not profitable if I and it's user can form a FMA instruction
+/// because we prefer FMSUB/FMADD.
+bool AArch64TargetLowering::isProfitableToHoist(Instruction *I) const {
+ if (I->getOpcode() != Instruction::FMul)
+ return true;
+
+ if (I->getNumUses() != 1)
+ return true;
+
+ Instruction *User = I->user_back();
+
+ if (User &&
+ !(User->getOpcode() == Instruction::FSub ||
+ User->getOpcode() == Instruction::FAdd))
+ return true;
+
+ const TargetOptions &Options = getTargetMachine().Options;
+ EVT VT = getValueType(User->getOperand(0)->getType());
+
+ if (isFMAFasterThanFMulAndFAdd(VT) &&
+ isOperationLegalOrCustom(ISD::FMA, VT) &&
+ (Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath))
+ return false;
+
+ return true;
+}
+
// All 32-bit GPR operations implicitly zero the high-half of the corresponding
// 64-bit GPR.
bool AArch64TargetLowering::isZExtFree(Type *Ty1, Type *Ty2) const {
bool Fast;
const Function *F = MF.getFunction();
if (Subtarget->hasFPARMv8() && !IsMemset && Size >= 16 &&
- !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
- Attribute::NoImplicitFloat) &&
+ !F->hasFnAttribute(Attribute::NoImplicitFloat) &&
(memOpAlign(SrcAlign, DstAlign, 16) ||
(allowsMisalignedMemoryAccesses(MVT::f128, 0, 1, &Fast) && Fast)))
return MVT::f128;
SDValue N0 = N->getOperand(0);
unsigned Lg2 = Divisor.countTrailingZeros();
SDValue Zero = DAG.getConstant(0, VT);
- SDValue Pow2MinusOne = DAG.getConstant((1 << Lg2) - 1, VT);
+ SDValue Pow2MinusOne = DAG.getConstant((1ULL << Lg2) - 1, VT);
// Add (N0 < 0) ? Pow2 - 1 : 0;
SDValue CCVal;
return SDValue();
}
-static SDValue performIntToFpCombine(SDNode *N, SelectionDAG &DAG) {
+static SDValue performIntToFpCombine(SDNode *N, SelectionDAG &DAG,
+ const AArch64Subtarget *Subtarget) {
// First try to optimize away the conversion when it's conditionally from
// a constant. Vectors only.
SDValue Res = performVectorCompareAndMaskUnaryOpCombine(N, DAG);
// conversion, use a fp load instead and a AdvSIMD scalar {S|U}CVTF instead.
// This eliminates an "integer-to-vector-move UOP and improve throughput.
SDValue N0 = N->getOperand(0);
- if (ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
+ if (Subtarget->hasNEON() && ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
// Do not change the width of a volatile load.
!cast<LoadSDNode>(N0)->isVolatile()) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
EVT InNVT = EVT::getVectorVT(*DAG.getContext(), SrcVT.getVectorElementType(),
LoVT.getVectorNumElements());
Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, Src,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, MVT::i64));
Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, InNVT, Src,
- DAG.getIntPtrConstant(InNVT.getVectorNumElements()));
+ DAG.getConstant(InNVT.getVectorNumElements(), MVT::i64));
Lo = DAG.getNode(N->getOpcode(), DL, LoVT, Lo);
Hi = DAG.getNode(N->getOpcode(), DL, HiVT, Hi);
return SDValue();
// Cyclone has bad performance on unaligned 16B stores when crossing line and
- // page boundries. We want to split such stores.
+ // page boundaries. We want to split such stores.
if (!Subtarget->isCyclone())
return SDValue();
// Don't split at Oz.
MachineFunction &MF = DAG.getMachineFunction();
- bool IsMinSize = MF.getFunction()->getAttributes().hasAttribute(
- AttributeSet::FunctionIndex, Attribute::MinSize);
+ bool IsMinSize = MF.getFunction()->hasFnAttribute(Attribute::MinSize);
if (IsMinSize)
return SDValue();
EVT HalfVT =
EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(), NumElts);
SDValue SubVector0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, HalfVT, StVal,
- DAG.getIntPtrConstant(0));
+ DAG.getConstant(0, MVT::i64));
SDValue SubVector1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, HalfVT, StVal,
- DAG.getIntPtrConstant(NumElts));
+ DAG.getConstant(NumElts, MVT::i64));
SDValue BasePtr = S->getBasePtr();
SDValue NewST1 =
DAG.getStore(S->getChain(), DL, SubVector0, BasePtr, S->getPointerInfo(),
LoadSDN->getMemOperand());
// Update the uses.
- std::vector<SDValue> NewResults;
+ SmallVector<SDValue, 2> NewResults;
NewResults.push_back(SDValue(LD, 0)); // The result of load
NewResults.push_back(SDValue(UpdN.getNode(), 2)); // Chain
DCI.CombineTo(LD, NewResults);
// largest real NEON comparison is 64-bits per lane, which means the result is
// at most 32-bits and an illegal vector. Just bail out for now.
EVT SrcVT = N0.getOperand(0).getValueType();
+
+ // Don't try to do this optimization when the setcc itself has i1 operands.
+ // There are no legal vectors of i1, so this would be pointless.
+ if (SrcVT == MVT::i1)
+ return SDValue();
+
int NumMaskElts = ResVT.getSizeInBits() / SrcVT.getSizeInBits();
if (!ResVT.isVector() || NumMaskElts == 0)
return SDValue();
return performMulCombine(N, DAG, DCI, Subtarget);
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
- return performIntToFpCombine(N, DAG);
+ return performIntToFpCombine(N, DAG, Subtarget);
case ISD::OR:
return performORCombine(N, DCI, Subtarget);
case ISD::INTRINSIC_WO_CHAIN:
static void ReplaceBITCASTResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) {
- if (N->getValueType(0) != MVT::i16)
- return;
-
SDLoc DL(N);
SDValue Op = N->getOperand(0);
- assert(Op.getValueType() == MVT::f16 &&
- "Inconsistent bitcast? Only 16-bit types should be i16 or f16");
+
+ if (N->getValueType(0) != MVT::i16 || Op.getValueType() != MVT::f16)
+ return;
+
Op = SDValue(
DAG.getMachineNode(TargetOpcode::INSERT_SUBREG, DL, MVT::f32,
DAG.getUNDEF(MVT::i32), Op,
return true;
}
+bool AArch64TargetLowering::combineRepeatedFPDivisors(unsigned NumUsers) const {
+ // Combine multiple FDIVs with the same divisor into multiple FMULs by the
+ // reciprocal if there are three or more FDIVs.
+ return NumUsers > 2;
+}
+
TargetLoweringBase::LegalizeTypeAction
AArch64TargetLowering::getPreferredVectorAction(EVT VT) const {
MVT SVT = VT.getSimpleVT();
Val, Stxr->getFunctionType()->getParamType(0)),
Addr);
}
+
+bool AArch64TargetLowering::functionArgumentNeedsConsecutiveRegisters(
+ Type *Ty, CallingConv::ID CallConv, bool isVarArg) const {
+ return Ty->isArrayTy();
+}
projects / oota-llvm.git / blobdiff