#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"
addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
- if (TT.isOSBinFormatMachO())
- return new TargetLoweringObjectFileMachO();
- if (TT.isOSWindows())
- return new TargetLoweringObjectFileCOFF();
- return new ARMElfTargetObjectFile();
-}
-
-ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
- : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
+ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM)
+ : TargetLowering(TM) {
Subtarget = &TM.getSubtarget<ARMSubtarget>();
RegInfo = TM.getSubtargetImpl()->getRegisterInfo();
Itins = TM.getSubtargetImpl()->getInstrItineraryData();
{ RTLIB::ADD_F64, "__aeabi_dadd", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::DIV_F64, "__aeabi_ddiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::MUL_F64, "__aeabi_dmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: double __aeabi_drsub(double x, double y) (rsub)
{ RTLIB::SUB_F64, "__aeabi_dsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
// Double-precision floating-point comparison helper functions
// RTABI chapter 4.1.2, Table 3
- // FIXME: void __aeabi_cdcmpeq(double, double)
- // FIXME: void __aeabi_cdcmple(double, double)
- // FIXME: void __aeabi_cdrcmple(double, double)
{ RTLIB::OEQ_F64, "__aeabi_dcmpeq", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::UNE_F64, "__aeabi_dcmpeq", CallingConv::ARM_AAPCS, ISD::SETEQ },
{ RTLIB::OLT_F64, "__aeabi_dcmplt", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::ADD_F32, "__aeabi_fadd", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::DIV_F32, "__aeabi_fdiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::MUL_F32, "__aeabi_fmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: void __aeabi_frsub(float x, float y)
{ RTLIB::SUB_F32, "__aeabi_fsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
// Single-precision floating-point comparison helper functions
// RTABI chapter 4.1.2, Table 5
- // FIXME: void __aeabi_cfcmpeq(float, float)
- // FIXME: void __aeabi_cfcmple(float, float)
- // FIXME: void __aeabi_cfrcmple(float, float)
{ RTLIB::OEQ_F32, "__aeabi_fcmpeq", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::UNE_F32, "__aeabi_fcmpeq", CallingConv::ARM_AAPCS, ISD::SETEQ },
{ RTLIB::OLT_F32, "__aeabi_fcmplt", CallingConv::ARM_AAPCS, ISD::SETNE },
// Conversions between floating types.
// RTABI chapter 4.1.2, Table 7
{ RTLIB::FPROUND_F64_F32, "__aeabi_d2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::FPEXT_F32_F64, "__aeabi_f2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: float __aeabi_f2f(short)
- // FIXME: float __aeabi_h2f_alt(short)
- // FIXME: short __aeabi_f2h(float)
- // FIXME: short __aeabi_f2h_alt(float)
{ RTLIB::FPROUND_F64_F16, "__aeabi_d2h", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: short __aeabi_d2h_alt(double)
+ { RTLIB::FPEXT_F32_F64, "__aeabi_f2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
// Integer to floating-point conversions.
// RTABI chapter 4.1.2, Table 8
// Long long helper functions
// RTABI chapter 4.2, Table 9
- { RTLIB::MUL_I64, "__aeabi_lmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: __aeabi_ldivmod is SDIVREM not SDIV; we should custom lower this
- { RTLIB::SDIV_I64, "__aeabi_ldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIVREM_I64, "__aeabi_ldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: __aeabi_uldivmod is UDIVREM not UDIV; we should custom lower this
- { RTLIB::UDIV_I64, "__aeabi_uldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIVREM_I64, "__aeabi_uldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SHL_I64, "__aeabi_llsl", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SRL_I64, "__aeabi_llsr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SRA_I64, "__aeabi_lasr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: int __aeabi_lcmp(long long, long long)
- // FIXME: int __aeabi_ulcmp(unsigned long long, unsigned long long)
+ { RTLIB::MUL_I64, "__aeabi_lmul", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SHL_I64, "__aeabi_llsl", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SRL_I64, "__aeabi_llsr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SRA_I64, "__aeabi_lasr", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
// Integer division functions
// RTABI chapter 4.3.1
- { RTLIB::SDIV_I8, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIV_I16, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIV_I32, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIV_I8, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIV_I16, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIV_I32, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIVREM_I8, "__aeabi_idivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIVREM_I16, "__aeabi_idivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::SDIVREM_I32, "__aeabi_idivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIVREM_I8, "__aeabi_uidivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIVREM_I16, "__aeabi_uidivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- { RTLIB::UDIVREM_I32, "__aeabi_uidivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SDIV_I8, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SDIV_I16, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SDIV_I32, "__aeabi_idiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::SDIV_I64, "__aeabi_ldivmod", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { RTLIB::UDIV_I8, "__aeabi_uidiv", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
+ { 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
- // FIXME: void __aeabi_memcpy8(void *, const void *, size_t)
- // FIXME: void __aeabi_memcpy4(void *, const void *, size_t)
{ RTLIB::MEMCPY, "__aeabi_memcpy", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: void __aeabi_memmove8(void *, const void *, size_t)
- // FIXME: void __aeabi_memmove4(void *, const void *, size_t)
{ RTLIB::MEMMOVE, "__aeabi_memmove", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: void __aeabi_memset8(void *, size_t, int)
- // FIXME: void __aeabi_memset4(void *, size_t, int)
{ RTLIB::MEMSET, "__aeabi_memset", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
- // FIXME: void __aeabi_memclr8(void *, size_t)
- // FIXME: void __aeabi_memclr4(void *, size_t)
- // FIXME: void __aeabi_memclr(void *, size_t)
};
for (const auto &LC : LibraryCalls) {
if (LC.Cond != ISD::SETCC_INVALID)
setCmpLibcallCC(LC.Op, LC.Cond);
}
-
- setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
- setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
}
if (Subtarget->isTargetWindows()) {
if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
!Subtarget->isThumb1Only()) {
addRegisterClass(MVT::f32, &ARM::SPRRegClass);
- if (!Subtarget->isFPOnlySP())
- addRegisterClass(MVT::f64, &ARM::DPRRegClass);
+ addRegisterClass(MVT::f64, &ARM::DPRRegClass);
}
for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
setTargetDAGCombine(ISD::FP_TO_SINT);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
+ setTargetDAGCombine(ISD::LOAD);
// It is legal to extload from v4i8 to v4i16 or v4i32.
MVT Tys[6] = {MVT::v8i8, MVT::v4i8, MVT::v2i8,
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();
}
// FIXME: Also set divmod for SREM on EABI
- setOperationAction(ISD::SREM, MVT::i32, Expand);
- setOperationAction(ISD::UREM, MVT::i32, Expand);
- if (!Subtarget->isTargetAEABI()) {
+ setOperationAction(ISD::SREM, MVT::i32, Expand);
+ setOperationAction(ISD::UREM, MVT::i32, Expand);
+ // Register based DivRem for AEABI (RTABI 4.2)
+ if (Subtarget->isTargetAEABI()) {
+ setLibcallName(RTLIB::SDIVREM_I8, "__aeabi_idivmod");
+ setLibcallName(RTLIB::SDIVREM_I16, "__aeabi_idivmod");
+ setLibcallName(RTLIB::SDIVREM_I32, "__aeabi_idivmod");
+ setLibcallName(RTLIB::SDIVREM_I64, "__aeabi_ldivmod");
+ setLibcallName(RTLIB::UDIVREM_I8, "__aeabi_uidivmod");
+ setLibcallName(RTLIB::UDIVREM_I16, "__aeabi_uidivmod");
+ setLibcallName(RTLIB::UDIVREM_I32, "__aeabi_uidivmod");
+ setLibcallName(RTLIB::UDIVREM_I64, "__aeabi_uldivmod");
+
+ setLibcallCallingConv(RTLIB::SDIVREM_I8, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::SDIVREM_I16, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::SDIVREM_I32, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::SDIVREM_I64, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::UDIVREM_I8, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::UDIVREM_I16, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::UDIVREM_I32, CallingConv::ARM_AAPCS);
+ setLibcallCallingConv(RTLIB::UDIVREM_I64, CallingConv::ARM_AAPCS);
+
+ setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
+ setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
+ } else {
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
}
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);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
}
- // v8 adds f64 <-> f16 conversion. Before that it should be expanded.
- if (!Subtarget->hasV8Ops()) {
+ // FP-ARMv8 adds f64 <-> f16 conversion. Before that it should be expanded.
+ if (!Subtarget->hasFPARMv8() || Subtarget->isFPOnlySP()) {
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
}
if (Subtarget->hasSinCos()) {
setLibcallName(RTLIB::SINCOS_F32, "sincosf");
setLibcallName(RTLIB::SINCOS_F64, "sincos");
- if (Subtarget->getTargetTriple().getOS() == Triple::IOS) {
+ if (Subtarget->getTargetTriple().isiOS()) {
// 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);
}
}
- // 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);
+ // FP-ARMv8 implements a lot of rounding-like FP operations.
+ if (Subtarget->hasFPARMv8()) {
+ setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
+ setOperationAction(ISD::FCEIL, MVT::f32, Legal);
+ setOperationAction(ISD::FROUND, MVT::f32, Legal);
+ setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
+ setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
+ setOperationAction(ISD::FRINT, MVT::f32, Legal);
+ if (!Subtarget->isFPOnlySP()) {
+ setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
+ setOperationAction(ISD::FCEIL, MVT::f64, Legal);
+ setOperationAction(ISD::FROUND, MVT::f64, Legal);
+ setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
+ setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
+ setOperationAction(ISD::FRINT, MVT::f64, Legal);
}
}
// We have target-specific dag combine patterns for the following nodes:
// True if this byval aggregate will be split between registers
// and memory.
unsigned ByValArgsCount = CCInfo.getInRegsParamsCount();
- unsigned CurByValIdx = CCInfo.getInRegsParamsProceed();
+ unsigned CurByValIdx = CCInfo.getInRegsParamsProcessed();
if (CurByValIdx < ByValArgsCount) {
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) {
if (Copies.count(UseChain.getNode()))
// Second CopyToReg
Copy = *UI;
- else
+ else {
+ // We are at the top of this chain.
+ // If the copy has a glue operand, we conservatively assume it
+ // isn't safe to perform a tail call.
+ if (UI->getOperand(UI->getNumOperands()-1).getValueType() == MVT::Glue)
+ return false;
// First CopyToReg
TCChain = UseChain;
+ }
}
} else if (Copy->getOpcode() == ISD::BITCAST) {
// f32 returned in a single GPR.
Copy = *Copy->use_begin();
if (Copy->getOpcode() != ISD::CopyToReg || !Copy->hasNUsesOfValue(1, 0))
return false;
+ // If the copy has a glue operand, we conservatively assume it isn't safe to
+ // perform a tail call.
+ if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
+ return false;
TCChain = Copy->getOperand(0);
} else {
return false;
switch (IntNo) {
default: return SDValue(); // Don't custom lower most intrinsics.
case Intrinsic::arm_rbit: {
- assert(Op.getOperand(0).getValueType() == MVT::i32 &&
+ assert(Op.getOperand(1).getValueType() == MVT::i32 &&
"RBIT intrinsic must have i32 type!");
- return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(0));
+ return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(1));
}
case Intrinsic::arm_thread_pointer: {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
ConstantSDNode *OrdN = cast<ConstantSDNode>(Op.getOperand(1));
AtomicOrdering Ord = static_cast<AtomicOrdering>(OrdN->getZExtValue());
- unsigned Domain = ARM_MB::ISH;
+ ARM_MB::MemBOpt Domain = ARM_MB::ISH;
if (Subtarget->isMClass()) {
// Only a full system barrier exists in the M-class architectures.
Domain = ARM_MB::SY;
if (Flags.isByVal()) {
unsigned ExtraArgRegsSize;
unsigned ExtraArgRegsSaveSize;
- computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsProceed(),
+ computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsProcessed(),
Flags.getByValSize(),
ExtraArgRegsSize, ExtraArgRegsSaveSize);
}
CCInfo.rewindByValRegsInfo();
lastInsIndex = -1;
- if (isVarArg) {
+ if (isVarArg && MFI->hasVAStart()) {
unsigned ExtraArgRegsSize;
unsigned ExtraArgRegsSaveSize;
computeRegArea(CCInfo, MF, CCInfo.getInRegsParamsCount(), 0,
else if (RegVT == MVT::v2f64)
RC = &ARM::QPRRegClass;
else if (RegVT == MVT::i32)
- RC = AFI->isThumb1OnlyFunction() ?
- (const TargetRegisterClass*)&ARM::tGPRRegClass :
- (const TargetRegisterClass*)&ARM::GPRRegClass;
+ RC = AFI->isThumb1OnlyFunction() ? &ARM::tGPRRegClass
+ : &ARM::GPRRegClass;
else
llvm_unreachable("RegVT not supported by FORMAL_ARGUMENTS Lowering");
// Since they could be overwritten by lowering of arguments in case of
// a tail call.
if (Flags.isByVal()) {
- unsigned CurByValIndex = CCInfo.getInRegsParamsProceed();
+ unsigned CurByValIndex = CCInfo.getInRegsParamsProcessed();
ByValStoreOffset = RoundUpToAlignment(ByValStoreOffset, Flags.getByValAlign());
int FrameIndex = StoreByValRegs(
}
// varargs
- if (isVarArg)
+ if (isVarArg && MFI->hasVAStart())
VarArgStyleRegisters(CCInfo, DAG, dl, Chain,
CCInfo.getNextStackOffset(),
TotalArgRegsSaveSize);
if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
return CFP->getValueAPF().isPosZero();
}
+ } else if (Op->getOpcode() == ISD::BITCAST &&
+ Op->getValueType(0) == MVT::f64) {
+ // 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;
+ }
+ }
}
return false;
}
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;
// select c, a, b
// We only do this in unsafe-fp-math, because signed zeros and NaNs are
// handled differently than the original code sequence.
- if (getTargetMachine().Options.UnsafeFPMath && LHS == TrueVal &&
- RHS == FalseVal) {
- if (CC == ISD::SETOGT || CC == ISD::SETUGT)
- return DAG.getNode(ARMISD::VMAXNM, dl, VT, TrueVal, FalseVal);
- if (CC == ISD::SETOLT || CC == ISD::SETULT)
- return DAG.getNode(ARMISD::VMINNM, dl, VT, TrueVal, FalseVal);
+ if (getTargetMachine().Options.UnsafeFPMath) {
+ if (LHS == TrueVal && RHS == FalseVal) {
+ if (CC == ISD::SETOGT || CC == ISD::SETUGT)
+ return DAG.getNode(ARMISD::VMAXNM, dl, VT, TrueVal, FalseVal);
+ if (CC == ISD::SETOLT || CC == ISD::SETULT)
+ return DAG.getNode(ARMISD::VMINNM, dl, VT, TrueVal, FalseVal);
+ } else if (LHS == FalseVal && RHS == TrueVal) {
+ if (CC == ISD::SETOLT || CC == ISD::SETULT)
+ return DAG.getNode(ARMISD::VMAXNM, dl, VT, TrueVal, FalseVal);
+ if (CC == ISD::SETOGT || CC == ISD::SETUGT)
+ return DAG.getNode(ARMISD::VMINNM, dl, VT, TrueVal, FalseVal);
+ }
}
bool swpCmpOps = false;
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:
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);
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
// Pair of floats / doubles used to pass the result.
- StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+ StructType *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
// Create stack object for sret.
const uint64_t ByteSize = TLI.getDataLayout()->getTypeAllocSize(RetTy);
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);
}
}
ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj);
unsigned CPI = MCP->getConstantPoolIndex(CPV, 4);
- const TargetRegisterClass *TRC = isThumb ?
- (const TargetRegisterClass*)&ARM::tGPRRegClass :
- (const TargetRegisterClass*)&ARM::GPRRegClass;
+ const TargetRegisterClass *TRC = isThumb ? &ARM::tGPRRegClass
+ : &ARM::GPRRegClass;
// Grab constant pool and fixed stack memory operands.
MachineMemOperand *CPMMO =
.addReg(NewVReg2, RegState::Kill)
.addReg(NewVReg3, RegState::Kill));
unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
- AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tADDrSPi), NewVReg5)
- .addFrameIndex(FI)
- .addImm(36)); // &jbuf[1] :: pc
+ BuildMI(*MBB, MI, dl, TII->get(ARM::tADDframe), NewVReg5)
+ .addFrameIndex(FI)
+ .addImm(36); // &jbuf[1] :: pc
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tSTRi))
.addReg(NewVReg4, RegState::Kill)
.addReg(NewVReg5, RegState::Kill)
MachineFrameInfo *MFI = MF->getFrameInfo();
int FI = MFI->getFunctionContextIndex();
- const TargetRegisterClass *TRC = Subtarget->isThumb() ?
- (const TargetRegisterClass*)&ARM::tGPRRegClass :
- (const TargetRegisterClass*)&ARM::GPRnopcRegClass;
+ const TargetRegisterClass *TRC = Subtarget->isThumb() ? &ARM::tGPRRegClass
+ : &ARM::GPRnopcRegClass;
// Get a mapping of the call site numbers to all of the landing pads they're
// associated with.
for (std::vector<MachineBasicBlock*>::iterator
I = LPadList.begin(), E = LPadList.end(); I != E; ++I) {
MachineBasicBlock *CurMBB = *I;
- if (SeenMBBs.insert(CurMBB))
+ if (SeenMBBs.insert(CurMBB).second)
DispContBB->addSuccessor(CurMBB);
}
// 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.
// Select the correct opcode and register class for unit size load/store
bool IsNeon = UnitSize >= 8;
- TRC = (IsThumb1 || IsThumb2) ? (const TargetRegisterClass *)&ARM::tGPRRegClass
- : (const TargetRegisterClass *)&ARM::GPRRegClass;
+ TRC = (IsThumb1 || IsThumb2) ? &ARM::tGPRRegClass : &ARM::GPRRegClass;
if (IsNeon)
- VecTRC = UnitSize == 16
- ? (const TargetRegisterClass *)&ARM::DPairRegClass
- : UnitSize == 8
- ? (const TargetRegisterClass *)&ARM::DPRRegClass
- : nullptr;
+ VecTRC = UnitSize == 16 ? &ARM::DPairRegClass
+ : UnitSize == 8 ? &ARM::DPRRegClass
+ : nullptr;
unsigned BytesLeft = SizeVal % UnitSize;
unsigned LoopSize = SizeVal - BytesLeft;
MachineRegisterInfo &MRI = Fn->getRegInfo();
// In Thumb mode S must not be specified if source register is the SP or
// PC and if destination register is the SP, so restrict register class
- unsigned NewRsbDstReg = MRI.createVirtualRegister(isThumb2 ?
- (const TargetRegisterClass*)&ARM::rGPRRegClass :
- (const TargetRegisterClass*)&ARM::GPRRegClass);
+ unsigned NewRsbDstReg =
+ MRI.createVirtualRegister(isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRRegClass);
// Transfer the remainder of BB and its successor edges to sinkMBB.
SinkBB->splice(SinkBB->begin(), BB,
void ARMTargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI,
SDNode *Node) const {
- if (!MI->hasPostISelHook()) {
- assert(!convertAddSubFlagsOpcode(MI->getOpcode()) &&
- "Pseudo flag-setting opcodes must be marked with 'hasPostISelHook'");
- return;
- }
-
const MCInstrDesc *MCID = &MI->getDesc();
// Adjust potentially 's' setting instructions after isel, i.e. ADC, SBC, RSB,
// RSC. Coming out of isel, they have an implicit CPSR def, but the optional
unsigned InvMask = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue();
unsigned LSB = countTrailingZeros(~InvMask);
unsigned Width = (32 - countLeadingZeros(~InvMask)) - LSB;
- unsigned Mask = (1 << Width)-1;
+ assert(Width <
+ static_cast<unsigned>(std::numeric_limits<unsigned>::digits) &&
+ "undefined behavior");
+ unsigned Mask = (1u << Width) - 1;
unsigned Mask2 = N11C->getZExtValue();
if ((Mask & (~Mask2)) == 0)
return DCI.DAG.getNode(ARMISD::BFI, SDLoc(N), N->getValueType(0),
/// 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)
return SDValue();
}
-/// PerformSTORECombine - Target-specific dag combine xforms for
-/// ISD::STORE.
-static SDValue PerformSTORECombine(SDNode *N,
- TargetLowering::DAGCombinerInfo &DCI) {
- StoreSDNode *St = cast<StoreSDNode>(N);
- if (St->isVolatile())
- return SDValue();
-
- // Optimize trunc store (of multiple scalars) to shuffle and store. First,
- // pack all of the elements in one place. Next, store to memory in fewer
- // chunks.
- SDValue StVal = St->getValue();
- EVT VT = StVal.getValueType();
- if (St->isTruncatingStore() && VT.isVector()) {
- SelectionDAG &DAG = DCI.DAG;
- const TargetLowering &TLI = DAG.getTargetLoweringInfo();
- EVT StVT = St->getMemoryVT();
- unsigned NumElems = VT.getVectorNumElements();
- assert(StVT != VT && "Cannot truncate to the same type");
- unsigned FromEltSz = VT.getVectorElementType().getSizeInBits();
- unsigned ToEltSz = StVT.getVectorElementType().getSizeInBits();
-
- // From, To sizes and ElemCount must be pow of two
- if (!isPowerOf2_32(NumElems * FromEltSz * ToEltSz)) return SDValue();
-
- // We are going to use the original vector elt for storing.
- // Accumulated smaller vector elements must be a multiple of the store size.
- if (0 != (NumElems * FromEltSz) % ToEltSz) return SDValue();
-
- unsigned SizeRatio = FromEltSz / ToEltSz;
- assert(SizeRatio * NumElems * ToEltSz == VT.getSizeInBits());
-
- // Create a type on which we perform the shuffle.
- EVT WideVecVT = EVT::getVectorVT(*DAG.getContext(), StVT.getScalarType(),
- NumElems*SizeRatio);
- assert(WideVecVT.getSizeInBits() == VT.getSizeInBits());
-
- 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] = TLI.isBigEndian() ? (i+1) * SizeRatio - 1 : i * SizeRatio;
-
- // Can't shuffle using an illegal type.
- if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
-
- SDValue Shuff = DAG.getVectorShuffle(WideVecVT, DL, WideVec,
- DAG.getUNDEF(WideVec.getValueType()),
- ShuffleVec.data());
- // At this point all of the data is stored at the bottom of the
- // register. We now need to save it to mem.
-
- // Find the largest store unit
- MVT StoreType = MVT::i8;
- for (unsigned tp = MVT::FIRST_INTEGER_VALUETYPE;
- tp < MVT::LAST_INTEGER_VALUETYPE; ++tp) {
- MVT Tp = (MVT::SimpleValueType)tp;
- if (TLI.isTypeLegal(Tp) && Tp.getSizeInBits() <= NumElems * ToEltSz)
- StoreType = Tp;
- }
- // Didn't find a legal store type.
- if (!TLI.isTypeLegal(StoreType))
- return SDValue();
-
- // Bitcast the original vector into a vector of store-size units
- EVT StoreVecVT = EVT::getVectorVT(*DAG.getContext(),
- StoreType, VT.getSizeInBits()/EVT(StoreType).getSizeInBits());
- assert(StoreVecVT.getSizeInBits() == VT.getSizeInBits());
- SDValue ShuffWide = DAG.getNode(ISD::BITCAST, DL, StoreVecVT, Shuff);
- SmallVector<SDValue, 8> Chains;
- SDValue Increment = DAG.getConstant(StoreType.getSizeInBits()/8,
- TLI.getPointerTy());
- SDValue BasePtr = St->getBasePtr();
-
- // Perform one or more big stores into memory.
- unsigned E = (ToEltSz*NumElems)/StoreType.getSizeInBits();
- for (unsigned I = 0; I < E; I++) {
- SDValue SubVec = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
- StoreType, ShuffWide,
- DAG.getIntPtrConstant(I));
- SDValue Ch = DAG.getStore(St->getChain(), DL, SubVec, BasePtr,
- St->getPointerInfo(), St->isVolatile(),
- St->isNonTemporal(), St->getAlignment());
- BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr,
- Increment);
- Chains.push_back(Ch);
- }
- return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
- }
-
- if (!ISD::isNormalStore(St))
- return SDValue();
-
- // Split a store of a VMOVDRR into two integer stores to avoid mixing NEON and
- // ARM stores of arguments in the same cache line.
- if (StVal.getNode()->getOpcode() == ARMISD::VMOVDRR &&
- StVal.getNode()->hasOneUse()) {
- SelectionDAG &DAG = DCI.DAG;
- bool isBigEndian = DAG.getTargetLoweringInfo().isBigEndian();
- SDLoc DL(St);
- SDValue BasePtr = St->getBasePtr();
- SDValue NewST1 = DAG.getStore(St->getChain(), DL,
- StVal.getNode()->getOperand(isBigEndian ? 1 : 0 ),
- BasePtr, St->getPointerInfo(), St->isVolatile(),
- St->isNonTemporal(), St->getAlignment());
-
- SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
- DAG.getConstant(4, MVT::i32));
- return DAG.getStore(NewST1.getValue(0), DL,
- StVal.getNode()->getOperand(isBigEndian ? 0 : 1),
- OffsetPtr, St->getPointerInfo(), St->isVolatile(),
- St->isNonTemporal(),
- std::min(4U, St->getAlignment() / 2));
- }
-
- if (StVal.getValueType() != MVT::i64 ||
- StVal.getNode()->getOpcode() != ISD::EXTRACT_VECTOR_ELT)
- return SDValue();
-
- // Bitcast an i64 store extracted from a vector to f64.
- // Otherwise, the i64 value will be legalized to a pair of i32 values.
- SelectionDAG &DAG = DCI.DAG;
- SDLoc dl(StVal);
- SDValue IntVec = StVal.getOperand(0);
- EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64,
- IntVec.getValueType().getVectorNumElements());
- SDValue Vec = DAG.getNode(ISD::BITCAST, dl, FloatVT, IntVec);
- SDValue ExtElt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
- Vec, StVal.getOperand(1));
- dl = SDLoc(N);
- SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::i64, ExtElt);
- // Make the DAGCombiner fold the bitcasts.
- DCI.AddToWorklist(Vec.getNode());
- DCI.AddToWorklist(ExtElt.getNode());
- DCI.AddToWorklist(V.getNode());
- return DAG.getStore(St->getChain(), dl, V, St->getBasePtr(),
- St->getPointerInfo(), St->isVolatile(),
- St->isNonTemporal(), St->getAlignment(),
- St->getAAInfo());
-}
-
/// hasNormalLoadOperand - Check if any of the operands of a BUILD_VECTOR node
/// are normal, non-volatile loads. If so, it is profitable to bitcast an
/// i64 vector to have f64 elements, since the value can then be loaded
/// 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,
DAG.getUNDEF(VT), NewMask.data());
}
-/// CombineBaseUpdate - Target-specific DAG combine function for VLDDUP and
-/// NEON load/store intrinsics to merge base address updates.
+/// CombineBaseUpdate - Target-specific DAG combine function for VLDDUP,
+/// NEON load/store intrinsics, and generic vector load/stores, to merge
+/// base address updates.
+/// For generic load/stores, the memory type is assumed to be a vector.
+/// The caller is assumed to have checked legality.
static SDValue CombineBaseUpdate(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
- if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
- return SDValue();
-
SelectionDAG &DAG = DCI.DAG;
bool isIntrinsic = (N->getOpcode() == ISD::INTRINSIC_VOID ||
N->getOpcode() == ISD::INTRINSIC_W_CHAIN);
- unsigned AddrOpIdx = (isIntrinsic ? 2 : 1);
+ bool isStore = N->getOpcode() == ISD::STORE;
+ unsigned AddrOpIdx = ((isIntrinsic || isStore) ? 2 : 1);
SDValue Addr = N->getOperand(AddrOpIdx);
// Search for a use of the address operand that is an increment.
case ARMISD::VLD2DUP: NewOpc = ARMISD::VLD2DUP_UPD; NumVecs = 2; break;
case ARMISD::VLD3DUP: NewOpc = ARMISD::VLD3DUP_UPD; NumVecs = 3; break;
case ARMISD::VLD4DUP: NewOpc = ARMISD::VLD4DUP_UPD; NumVecs = 4; break;
+ case ISD::LOAD: NewOpc = ARMISD::VLD1_UPD;
+ NumVecs = 1; isLaneOp = false; break;
+ case ISD::STORE: NewOpc = ARMISD::VST1_UPD;
+ NumVecs = 1; isLoad = false; isLaneOp = false; break;
}
}
EVT VecTy;
if (isLoad)
VecTy = N->getValueType(0);
- else
+ else if (isIntrinsic)
VecTy = N->getOperand(AddrOpIdx+1).getValueType();
+ else
+ VecTy = N->getOperand(1).getValueType();
+
unsigned NumBytes = NumVecs * VecTy.getSizeInBits() / 8;
if (isLaneOp)
NumBytes /= VecTy.getVectorNumElements();
}
// Create the new updating load/store node.
+ // First, create an SDVTList for the new updating node's results.
EVT Tys[6];
unsigned NumResultVecs = (isLoad ? NumVecs : 0);
unsigned n;
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));
+
+ // Then, gather the new node's operands.
SmallVector<SDValue, 8> Ops;
Ops.push_back(N->getOperand(0)); // incoming chain
Ops.push_back(N->getOperand(AddrOpIdx));
Ops.push_back(Inc);
- for (unsigned i = AddrOpIdx + 1; i < N->getNumOperands(); ++i) {
- Ops.push_back(N->getOperand(i));
+ if (StoreSDNode *StN = dyn_cast<StoreSDNode>(N)) {
+ // Try to match the intrinsic's signature
+ Ops.push_back(StN->getValue());
+ Ops.push_back(DAG.getConstant(StN->getAlignment(), MVT::i32));
+ } else {
+ for (unsigned i = AddrOpIdx + 1; i < N->getNumOperands(); ++i)
+ Ops.push_back(N->getOperand(i));
}
- MemIntrinsicSDNode *MemInt = cast<MemIntrinsicSDNode>(N);
+ MemSDNode *MemInt = cast<MemSDNode>(N);
SDValue UpdN = DAG.getMemIntrinsicNode(NewOpc, SDLoc(N), SDTys,
Ops, MemInt->getMemoryVT(),
MemInt->getMemOperand());
return SDValue();
}
+static SDValue PerformVLDCombine(SDNode *N,
+ TargetLowering::DAGCombinerInfo &DCI) {
+ if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
+ return SDValue();
+
+ return CombineBaseUpdate(N, DCI);
+}
+
/// CombineVLDDUP - For a VDUPLANE node N, check if its source operand is a
/// vldN-lane (N > 1) intrinsic, and if all the other uses of that intrinsic
/// are also VDUPLANEs. If so, combine them to a vldN-dup operation and
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,
return DCI.DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Op);
}
+static SDValue PerformLOADCombine(SDNode *N,
+ TargetLowering::DAGCombinerInfo &DCI) {
+ EVT VT = N->getValueType(0);
+
+ // If this is a legal vector load, try to combine it into a VLD1_UPD.
+ if (ISD::isNormalLoad(N) && VT.isVector() &&
+ DCI.DAG.getTargetLoweringInfo().isTypeLegal(VT))
+ return CombineBaseUpdate(N, DCI);
+
+ return SDValue();
+}
+
+/// PerformSTORECombine - Target-specific dag combine xforms for
+/// ISD::STORE.
+static SDValue PerformSTORECombine(SDNode *N,
+ TargetLowering::DAGCombinerInfo &DCI) {
+ StoreSDNode *St = cast<StoreSDNode>(N);
+ if (St->isVolatile())
+ return SDValue();
+
+ // Optimize trunc store (of multiple scalars) to shuffle and store. First,
+ // pack all of the elements in one place. Next, store to memory in fewer
+ // chunks.
+ SDValue StVal = St->getValue();
+ EVT VT = StVal.getValueType();
+ if (St->isTruncatingStore() && VT.isVector()) {
+ SelectionDAG &DAG = DCI.DAG;
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ EVT StVT = St->getMemoryVT();
+ unsigned NumElems = VT.getVectorNumElements();
+ assert(StVT != VT && "Cannot truncate to the same type");
+ unsigned FromEltSz = VT.getVectorElementType().getSizeInBits();
+ unsigned ToEltSz = StVT.getVectorElementType().getSizeInBits();
+
+ // From, To sizes and ElemCount must be pow of two
+ if (!isPowerOf2_32(NumElems * FromEltSz * ToEltSz)) return SDValue();
+
+ // We are going to use the original vector elt for storing.
+ // Accumulated smaller vector elements must be a multiple of the store size.
+ if (0 != (NumElems * FromEltSz) % ToEltSz) return SDValue();
+
+ unsigned SizeRatio = FromEltSz / ToEltSz;
+ assert(SizeRatio * NumElems * ToEltSz == VT.getSizeInBits());
+
+ // Create a type on which we perform the shuffle.
+ EVT WideVecVT = EVT::getVectorVT(*DAG.getContext(), StVT.getScalarType(),
+ NumElems*SizeRatio);
+ assert(WideVecVT.getSizeInBits() == VT.getSizeInBits());
+
+ 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] = TLI.isBigEndian() ? (i+1) * SizeRatio - 1 : i * SizeRatio;
+
+ // Can't shuffle using an illegal type.
+ if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
+
+ SDValue Shuff = DAG.getVectorShuffle(WideVecVT, DL, WideVec,
+ DAG.getUNDEF(WideVec.getValueType()),
+ ShuffleVec.data());
+ // At this point all of the data is stored at the bottom of the
+ // register. We now need to save it to mem.
+
+ // Find the largest store unit
+ MVT StoreType = MVT::i8;
+ for (unsigned tp = MVT::FIRST_INTEGER_VALUETYPE;
+ tp < MVT::LAST_INTEGER_VALUETYPE; ++tp) {
+ MVT Tp = (MVT::SimpleValueType)tp;
+ if (TLI.isTypeLegal(Tp) && Tp.getSizeInBits() <= NumElems * ToEltSz)
+ StoreType = Tp;
+ }
+ // Didn't find a legal store type.
+ if (!TLI.isTypeLegal(StoreType))
+ return SDValue();
+
+ // Bitcast the original vector into a vector of store-size units
+ EVT StoreVecVT = EVT::getVectorVT(*DAG.getContext(),
+ StoreType, VT.getSizeInBits()/EVT(StoreType).getSizeInBits());
+ assert(StoreVecVT.getSizeInBits() == VT.getSizeInBits());
+ SDValue ShuffWide = DAG.getNode(ISD::BITCAST, DL, StoreVecVT, Shuff);
+ SmallVector<SDValue, 8> Chains;
+ SDValue Increment = DAG.getConstant(StoreType.getSizeInBits()/8,
+ TLI.getPointerTy());
+ SDValue BasePtr = St->getBasePtr();
+
+ // Perform one or more big stores into memory.
+ unsigned E = (ToEltSz*NumElems)/StoreType.getSizeInBits();
+ for (unsigned I = 0; I < E; I++) {
+ SDValue SubVec = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
+ StoreType, ShuffWide,
+ DAG.getIntPtrConstant(I));
+ SDValue Ch = DAG.getStore(St->getChain(), DL, SubVec, BasePtr,
+ St->getPointerInfo(), St->isVolatile(),
+ St->isNonTemporal(), St->getAlignment());
+ BasePtr = DAG.getNode(ISD::ADD, DL, BasePtr.getValueType(), BasePtr,
+ Increment);
+ Chains.push_back(Ch);
+ }
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
+ }
+
+ if (!ISD::isNormalStore(St))
+ return SDValue();
+
+ // Split a store of a VMOVDRR into two integer stores to avoid mixing NEON and
+ // ARM stores of arguments in the same cache line.
+ if (StVal.getNode()->getOpcode() == ARMISD::VMOVDRR &&
+ StVal.getNode()->hasOneUse()) {
+ SelectionDAG &DAG = DCI.DAG;
+ bool isBigEndian = DAG.getTargetLoweringInfo().isBigEndian();
+ SDLoc DL(St);
+ SDValue BasePtr = St->getBasePtr();
+ SDValue NewST1 = DAG.getStore(St->getChain(), DL,
+ StVal.getNode()->getOperand(isBigEndian ? 1 : 0 ),
+ BasePtr, St->getPointerInfo(), St->isVolatile(),
+ St->isNonTemporal(), St->getAlignment());
+
+ SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
+ DAG.getConstant(4, MVT::i32));
+ return DAG.getStore(NewST1.getValue(0), DL,
+ StVal.getNode()->getOperand(isBigEndian ? 0 : 1),
+ OffsetPtr, St->getPointerInfo(), St->isVolatile(),
+ St->isNonTemporal(),
+ std::min(4U, St->getAlignment() / 2));
+ }
+
+ if (StVal.getValueType() == MVT::i64 &&
+ StVal.getNode()->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+
+ // Bitcast an i64 store extracted from a vector to f64.
+ // Otherwise, the i64 value will be legalized to a pair of i32 values.
+ SelectionDAG &DAG = DCI.DAG;
+ SDLoc dl(StVal);
+ SDValue IntVec = StVal.getOperand(0);
+ EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64,
+ IntVec.getValueType().getVectorNumElements());
+ SDValue Vec = DAG.getNode(ISD::BITCAST, dl, FloatVT, IntVec);
+ SDValue ExtElt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
+ Vec, StVal.getOperand(1));
+ dl = SDLoc(N);
+ SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::i64, ExtElt);
+ // Make the DAGCombiner fold the bitcasts.
+ DCI.AddToWorklist(Vec.getNode());
+ DCI.AddToWorklist(ExtElt.getNode());
+ DCI.AddToWorklist(V.getNode());
+ return DAG.getStore(St->getChain(), dl, V, St->getBasePtr(),
+ St->getPointerInfo(), St->isVolatile(),
+ St->isNonTemporal(), St->getAlignment(),
+ St->getAAInfo());
+ }
+
+ // If this is a legal vector store, try to combine it into a VST1_UPD.
+ if (ISD::isNormalStore(N) && VT.isVector() &&
+ DCI.DAG.getTargetLoweringInfo().isTypeLegal(VT))
+ return CombineBaseUpdate(N, DCI);
+
+ return SDValue();
+}
+
// isConstVecPow2 - Return true if each vector element is a power of 2, all
// elements are the same constant, C, and Log2(C) ranges from 1 to 32.
static bool isConstVecPow2(SDValue ConstVec, bool isSigned, uint64_t &C)
MVT FloatTy = Op.getSimpleValueType().getVectorElementType();
MVT IntTy = N->getSimpleValueType(0).getVectorElementType();
- if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32) {
+ unsigned NumLanes = Op.getValueType().getVectorNumElements();
+ if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32 ||
+ NumLanes > 4) {
// These instructions only exist converting from f32 to i32. We can handle
// smaller integers by generating an extra truncate, but larger ones would
- // be lossy.
+ // be lossy. We also can't handle more then 4 lanes, since these intructions
+ // only support v2i32/v4i32 types.
return SDValue();
}
unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfp2fxs :
Intrinsic::arm_neon_vcvtfp2fxu;
- unsigned NumLanes = Op.getValueType().getVectorNumElements();
SDValue FixConv = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N),
NumLanes == 2 ? MVT::v2i32 : MVT::v4i32,
DAG.getConstant(IntrinsicOpcode, MVT::i32), N0,
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);
case ISD::ANY_EXTEND: return PerformExtendCombine(N, DCI.DAG, Subtarget);
case ISD::SELECT_CC: return PerformSELECT_CCCombine(N, DCI.DAG, Subtarget);
case ARMISD::CMOV: return PerformCMOVCombine(N, DCI.DAG);
+ case ISD::LOAD: return PerformLOADCombine(N, DCI);
case ARMISD::VLD2DUP:
case ARMISD::VLD3DUP:
case ARMISD::VLD4DUP:
- return CombineBaseUpdate(N, DCI);
+ return PerformVLDCombine(N, DCI);
case ARMISD::BUILD_VECTOR:
return PerformARMBUILD_VECTORCombine(N, DCI);
case ISD::INTRINSIC_VOID:
case Intrinsic::arm_neon_vst2lane:
case Intrinsic::arm_neon_vst3lane:
case Intrinsic::arm_neon_vst4lane:
- return CombineBaseUpdate(N, DCI);
+ return PerformVLDCombine(N, DCI);
default: break;
}
break;
return RCPair(0U, &ARM::hGPRRegClass);
break;
case 'r':
+ if (Subtarget->isThumb1Only())
+ return RCPair(0U, &ARM::tGPRRegClass);
return RCPair(0U, &ARM::GPRRegClass);
case 'w':
if (VT == MVT::Other)
SDValue Callee = DAG.getExternalSymbol(getLibcallName(LC),
getPointerTy());
- Type *RetTy = (Type*)StructType::get(Ty, Ty, NULL);
+ Type *RetTy = (Type*)StructType::get(Ty, Ty, nullptr);
SDLoc dl(Op);
TargetLowering::CallLoweringInfo CLI(DAG);
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;
}
return true;
}
-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. 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;
+bool ARMTargetLowering::hasLoadLinkedStoreConditional() const { return true; }
+
+Instruction* ARMTargetLowering::makeDMB(IRBuilder<> &Builder,
+ ARM_MB::MemBOpt Domain) const {
+ Module *M = Builder.GetInsertBlock()->getParent()->getParent();
+
+ // First, if the target has no DMB, see what fallback we can use.
+ if (!Subtarget->hasDataBarrier()) {
+ // Some ARMv6 cpus can support data barriers with an mcr instruction.
+ // Thumb1 and pre-v6 ARM mode use a libcall instead and should never get
+ // here.
+ if (Subtarget->hasV6Ops() && !Subtarget->isThumb()) {
+ Function *MCR = llvm::Intrinsic::getDeclaration(M, Intrinsic::arm_mcr);
+ Value* args[6] = {Builder.getInt32(15), Builder.getInt32(0),
+ Builder.getInt32(0), Builder.getInt32(7),
+ Builder.getInt32(10), Builder.getInt32(5)};
+ return Builder.CreateCall(MCR, args);
+ } else {
+ // Instead of using barriers, atomic accesses on these subtargets use
+ // libcalls.
+ llvm_unreachable("makeDMB on a target so old that it has no barriers");
+ }
+ } else {
+ Function *DMB = llvm::Intrinsic::getDeclaration(M, Intrinsic::arm_dmb);
+ // Only a full system barrier exists in the M-class architectures.
+ Domain = Subtarget->isMClass() ? ARM_MB::SY : Domain;
+ Constant *CDomain = Builder.getInt32(Domain);
+ return Builder.CreateCall(DMB, CDomain);
+ }
+}
+
+// Based on http://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
+Instruction* ARMTargetLowering::emitLeadingFence(IRBuilder<> &Builder,
+ AtomicOrdering Ord, bool IsStore,
+ bool IsLoad) const {
+ if (!getInsertFencesForAtomic())
+ return nullptr;
+
+ switch (Ord) {
+ case NotAtomic:
+ case Unordered:
+ llvm_unreachable("Invalid fence: unordered/non-atomic");
+ case Monotonic:
+ case Acquire:
+ return nullptr; // Nothing to do
+ case SequentiallyConsistent:
+ if (!IsStore)
+ return nullptr; // Nothing to do
+ /*FALLTHROUGH*/
+ case Release:
+ case AcquireRelease:
+ if (Subtarget->isSwift())
+ return makeDMB(Builder, ARM_MB::ISHST);
+ // FIXME: add a comment with a link to documentation justifying this.
+ else
+ return makeDMB(Builder, ARM_MB::ISH);
+ }
+ llvm_unreachable("Unknown fence ordering in emitLeadingFence");
+}
+
+Instruction* ARMTargetLowering::emitTrailingFence(IRBuilder<> &Builder,
+ AtomicOrdering Ord, bool IsStore,
+ bool IsLoad) const {
+ if (!getInsertFencesForAtomic())
+ return nullptr;
+
+ switch (Ord) {
+ case NotAtomic:
+ case Unordered:
+ llvm_unreachable("Invalid fence: unordered/not-atomic");
+ case Monotonic:
+ case Release:
+ return nullptr; // Nothing to do
+ case Acquire:
+ case AcquireRelease:
+ case SequentiallyConsistent:
+ return makeDMB(Builder, ARM_MB::ISH);
}
+ llvm_unreachable("Unknown fence ordering in emitTrailingFence");
+}
+
+// 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. Cortex M doesn't have ldrexd/strexd though, so don't emit
+// anything for those.
+bool ARMTargetLowering::shouldExpandAtomicStoreInIR(StoreInst *SI) const {
+ unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
+ return (Size == 64) && !Subtarget->isMClass();
+}
+
+// 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. Cortex M doesn't have ldrexd/strexd though, so don't emit
+// anything for those.
+// FIXME: ldrd and strd are atomic if the CPU has LPAE (e.g. A15 has that
+// guarantee, see DDI0406C ARM architecture reference manual,
+// sections A8.8.72-74 LDRD)
+bool ARMTargetLowering::shouldExpandAtomicLoadInIR(LoadInst *LI) const {
+ unsigned Size = LI->getType()->getPrimitiveSizeInBits();
+ return (Size == 64) && !Subtarget->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 32 bits,
+// and up to 64 bits on the non-M profiles
+bool ARMTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
+ unsigned Size = AI->getType()->getPrimitiveSizeInBits();
+ return Size <= (Subtarget->isMClass() ? 32U : 64U);
}
// This has so far only been implemented for MachO.
bool ARMTargetLowering::useLoadStackGuardNode() const {
- return Subtarget->getTargetTriple().getObjectFormat() == Triple::MachO;
+ return Subtarget->isTargetMachO();
+}
+
+bool ARMTargetLowering::canCombineStoreAndExtract(Type *VectorTy, Value *Idx,
+ unsigned &Cost) const {
+ // If we do not have NEON, vector types are not natively supported.
+ if (!Subtarget->hasNEON())
+ return false;
+
+ // Floating point values and vector values map to the same register file.
+ // Therefore, althought we could do a store extract of a vector type, this is
+ // better to leave at float as we have more freedom in the addressing mode for
+ // those.
+ if (VectorTy->isFPOrFPVectorTy())
+ return false;
+
+ // If the index is unknown at compile time, this is very expensive to lower
+ // and it is not possible to combine the store with the extract.
+ if (!isa<ConstantInt>(Idx))
+ return false;
+
+ assert(VectorTy->isVectorTy() && "VectorTy is not a vector type");
+ unsigned BitWidth = cast<VectorType>(VectorTy)->getBitWidth();
+ // We can do a store + vector extract on any vector that fits perfectly in a D
+ // or Q register.
+ if (BitWidth == 64 || BitWidth == 128) {
+ Cost = 0;
+ return true;
+ }
+ return false;
}
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