ARMISelLowering.cpp [plain text]
#include "ARMISelLowering.h"
#include "ARMCallingConv.h"
#include "ARMConstantPoolValue.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMPerfectShuffle.h"
#include "ARMSubtarget.h"
#include "ARMTargetMachine.h"
#include "ARMTargetObjectFile.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/IntrinsicLowering.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#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"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Type.h"
#include "llvm/MC/MCSectionMachO.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include <utility>
using namespace llvm;
#define DEBUG_TYPE "arm-isel"
STATISTIC(NumTailCalls, "Number of tail calls");
STATISTIC(NumMovwMovt, "Number of GAs materialized with movw + movt");
STATISTIC(NumLoopByVals, "Number of loops generated for byval arguments");
cl::opt<bool>
EnableARMLongCalls("arm-long-calls", cl::Hidden,
cl::desc("Generate calls via indirect call instructions"),
cl::init(false));
static cl::opt<bool>
ARMInterworking("arm-interworking", cl::Hidden,
cl::desc("Enable / disable ARM interworking (for debugging only)"),
cl::init(true));
namespace {
class ARMCCState : public CCState {
public:
ARMCCState(CallingConv::ID CC, bool isVarArg, MachineFunction &MF,
SmallVectorImpl<CCValAssign> &locs, LLVMContext &C,
ParmContext PC)
: CCState(CC, isVarArg, MF, locs, C) {
assert(((PC == Call) || (PC == Prologue)) &&
"ARMCCState users must specify whether their context is call"
"or prologue generation.");
CallOrPrologue = PC;
}
};
}
static const MCPhysReg GPRArgRegs[] = {
ARM::R0, ARM::R1, ARM::R2, ARM::R3
};
void ARMTargetLowering::addTypeForNEON(MVT VT, MVT PromotedLdStVT,
MVT PromotedBitwiseVT) {
if (VT != PromotedLdStVT) {
setOperationAction(ISD::LOAD, VT, Promote);
AddPromotedToType (ISD::LOAD, VT, PromotedLdStVT);
setOperationAction(ISD::STORE, VT, Promote);
AddPromotedToType (ISD::STORE, VT, PromotedLdStVT);
}
MVT ElemTy = VT.getVectorElementType();
if (ElemTy != MVT::i64 && ElemTy != MVT::f64)
setOperationAction(ISD::SETCC, VT, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
if (ElemTy == MVT::i32) {
setOperationAction(ISD::SINT_TO_FP, VT, Custom);
setOperationAction(ISD::UINT_TO_FP, VT, Custom);
setOperationAction(ISD::FP_TO_SINT, VT, Custom);
setOperationAction(ISD::FP_TO_UINT, VT, Custom);
} else {
setOperationAction(ISD::SINT_TO_FP, VT, Expand);
setOperationAction(ISD::UINT_TO_FP, VT, Expand);
setOperationAction(ISD::FP_TO_SINT, VT, Expand);
setOperationAction(ISD::FP_TO_UINT, VT, Expand);
}
setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
setOperationAction(ISD::CONCAT_VECTORS, VT, Legal);
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
setOperationAction(ISD::SELECT, VT, Expand);
setOperationAction(ISD::SELECT_CC, VT, Expand);
setOperationAction(ISD::VSELECT, VT, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
if (VT.isInteger()) {
setOperationAction(ISD::SHL, VT, Custom);
setOperationAction(ISD::SRA, VT, Custom);
setOperationAction(ISD::SRL, VT, Custom);
}
if (VT.isInteger() && VT != PromotedBitwiseVT) {
setOperationAction(ISD::AND, VT, Promote);
AddPromotedToType (ISD::AND, VT, PromotedBitwiseVT);
setOperationAction(ISD::OR, VT, Promote);
AddPromotedToType (ISD::OR, VT, PromotedBitwiseVT);
setOperationAction(ISD::XOR, VT, Promote);
AddPromotedToType (ISD::XOR, VT, PromotedBitwiseVT);
}
setOperationAction(ISD::SDIV, VT, Expand);
setOperationAction(ISD::UDIV, VT, Expand);
setOperationAction(ISD::FDIV, VT, Expand);
setOperationAction(ISD::SREM, VT, Expand);
setOperationAction(ISD::UREM, VT, Expand);
setOperationAction(ISD::FREM, VT, Expand);
}
void ARMTargetLowering::addDRTypeForNEON(MVT VT) {
addRegisterClass(VT, &ARM::DPRRegClass);
addTypeForNEON(VT, MVT::f64, MVT::v2i32);
}
void ARMTargetLowering::addQRTypeForNEON(MVT VT) {
addRegisterClass(VT, &ARM::DPairRegClass);
addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
}
ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
const ARMSubtarget &STI)
: TargetLowering(TM), Subtarget(&STI) {
RegInfo = Subtarget->getRegisterInfo();
Itins = Subtarget->getInstrItineraryData();
setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
if (Subtarget->isTargetMachO()) {
if (Subtarget->isThumb() && Subtarget->hasVFP2() &&
Subtarget->hasARMOps() && !TM.Options.UseSoftFloat) {
setLibcallName(RTLIB::ADD_F32, "__addsf3vfp");
setLibcallName(RTLIB::SUB_F32, "__subsf3vfp");
setLibcallName(RTLIB::MUL_F32, "__mulsf3vfp");
setLibcallName(RTLIB::DIV_F32, "__divsf3vfp");
setLibcallName(RTLIB::ADD_F64, "__adddf3vfp");
setLibcallName(RTLIB::SUB_F64, "__subdf3vfp");
setLibcallName(RTLIB::MUL_F64, "__muldf3vfp");
setLibcallName(RTLIB::DIV_F64, "__divdf3vfp");
setLibcallName(RTLIB::OEQ_F32, "__eqsf2vfp");
setLibcallName(RTLIB::UNE_F32, "__nesf2vfp");
setLibcallName(RTLIB::OLT_F32, "__ltsf2vfp");
setLibcallName(RTLIB::OLE_F32, "__lesf2vfp");
setLibcallName(RTLIB::OGE_F32, "__gesf2vfp");
setLibcallName(RTLIB::OGT_F32, "__gtsf2vfp");
setLibcallName(RTLIB::UO_F32, "__unordsf2vfp");
setLibcallName(RTLIB::O_F32, "__unordsf2vfp");
setCmpLibcallCC(RTLIB::OEQ_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::UNE_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::OLT_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::OLE_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::OGE_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::OGT_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::UO_F32, ISD::SETNE);
setCmpLibcallCC(RTLIB::O_F32, ISD::SETEQ);
setLibcallName(RTLIB::OEQ_F64, "__eqdf2vfp");
setLibcallName(RTLIB::UNE_F64, "__nedf2vfp");
setLibcallName(RTLIB::OLT_F64, "__ltdf2vfp");
setLibcallName(RTLIB::OLE_F64, "__ledf2vfp");
setLibcallName(RTLIB::OGE_F64, "__gedf2vfp");
setLibcallName(RTLIB::OGT_F64, "__gtdf2vfp");
setLibcallName(RTLIB::UO_F64, "__unorddf2vfp");
setLibcallName(RTLIB::O_F64, "__unorddf2vfp");
setCmpLibcallCC(RTLIB::OEQ_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::UNE_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::OLT_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::OLE_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::OGE_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::OGT_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::UO_F64, ISD::SETNE);
setCmpLibcallCC(RTLIB::O_F64, ISD::SETEQ);
setLibcallName(RTLIB::FPTOSINT_F64_I32, "__fixdfsivfp");
setLibcallName(RTLIB::FPTOUINT_F64_I32, "__fixunsdfsivfp");
setLibcallName(RTLIB::FPTOSINT_F32_I32, "__fixsfsivfp");
setLibcallName(RTLIB::FPTOUINT_F32_I32, "__fixunssfsivfp");
setLibcallName(RTLIB::FPROUND_F64_F32, "__truncdfsf2vfp");
setLibcallName(RTLIB::FPEXT_F32_F64, "__extendsfdf2vfp");
setLibcallName(RTLIB::SINTTOFP_I32_F64, "__floatsidfvfp");
setLibcallName(RTLIB::UINTTOFP_I32_F64, "__floatunssidfvfp");
setLibcallName(RTLIB::SINTTOFP_I32_F32, "__floatsisfvfp");
setLibcallName(RTLIB::UINTTOFP_I32_F32, "__floatunssisfvfp");
}
if (Subtarget->isAAPCS_ABI() && Subtarget->isTargetHardFloat()) {
for (int i = 0; i < RTLIB::UNKNOWN_LIBCALL; ++i)
setLibcallCallingConv((RTLIB::Libcall)i, CallingConv::ARM_AAPCS_VFP);
}
}
setLibcallName(RTLIB::SHL_I128, nullptr);
setLibcallName(RTLIB::SRL_I128, nullptr);
setLibcallName(RTLIB::SRA_I128, nullptr);
if (Subtarget->isAAPCS_ABI() && !Subtarget->isTargetMachO() &&
!Subtarget->isTargetWindows()) {
static const struct {
const RTLIB::Libcall Op;
const char * const Name;
const CallingConv::ID CC;
const ISD::CondCode Cond;
} LibraryCalls[] = {
{ 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 },
{ RTLIB::SUB_F64, "__aeabi_dsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ 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::OLE_F64, "__aeabi_dcmple", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::OGE_F64, "__aeabi_dcmpge", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::OGT_F64, "__aeabi_dcmpgt", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::UO_F64, "__aeabi_dcmpun", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::O_F64, "__aeabi_dcmpun", CallingConv::ARM_AAPCS, ISD::SETEQ },
{ 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 },
{ RTLIB::SUB_F32, "__aeabi_fsub", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ 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 },
{ RTLIB::OLE_F32, "__aeabi_fcmple", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::OGE_F32, "__aeabi_fcmpge", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::OGT_F32, "__aeabi_fcmpgt", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::UO_F32, "__aeabi_fcmpun", CallingConv::ARM_AAPCS, ISD::SETNE },
{ RTLIB::O_F32, "__aeabi_fcmpun", CallingConv::ARM_AAPCS, ISD::SETEQ },
{ RTLIB::FPTOSINT_F64_I32, "__aeabi_d2iz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOUINT_F64_I32, "__aeabi_d2uiz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOSINT_F64_I64, "__aeabi_d2lz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOUINT_F64_I64, "__aeabi_d2ulz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOSINT_F32_I32, "__aeabi_f2iz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOUINT_F32_I32, "__aeabi_f2uiz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOSINT_F32_I64, "__aeabi_f2lz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPTOUINT_F32_I64, "__aeabi_f2ulz", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPROUND_F64_F32, "__aeabi_d2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPROUND_F64_F16, "__aeabi_d2h", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::FPEXT_F32_F64, "__aeabi_f2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::SINTTOFP_I32_F64, "__aeabi_i2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UINTTOFP_I32_F64, "__aeabi_ui2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::SINTTOFP_I64_F64, "__aeabi_l2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UINTTOFP_I64_F64, "__aeabi_ul2d", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::SINTTOFP_I32_F32, "__aeabi_i2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UINTTOFP_I32_F32, "__aeabi_ui2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::SINTTOFP_I64_F32, "__aeabi_l2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::UINTTOFP_I64_F32, "__aeabi_ul2f", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ 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 },
{ 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 },
{ RTLIB::MEMCPY, "__aeabi_memcpy", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::MEMMOVE, "__aeabi_memmove", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
{ RTLIB::MEMSET, "__aeabi_memset", CallingConv::ARM_AAPCS, ISD::SETCC_INVALID },
};
for (const auto &LC : LibraryCalls) {
setLibcallName(LC.Op, LC.Name);
setLibcallCallingConv(LC.Op, LC.CC);
if (LC.Cond != ISD::SETCC_INVALID)
setCmpLibcallCC(LC.Op, LC.Cond);
}
}
if (Subtarget->isTargetWindows()) {
static const struct {
const RTLIB::Libcall Op;
const char * const Name;
const CallingConv::ID CC;
} LibraryCalls[] = {
{ RTLIB::FPTOSINT_F32_I64, "__stoi64", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::FPTOSINT_F64_I64, "__dtoi64", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::FPTOUINT_F32_I64, "__stou64", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::FPTOUINT_F64_I64, "__dtou64", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::SINTTOFP_I64_F32, "__i64tos", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::SINTTOFP_I64_F64, "__i64tod", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::UINTTOFP_I64_F32, "__u64tos", CallingConv::ARM_AAPCS_VFP },
{ RTLIB::UINTTOFP_I64_F64, "__u64tod", CallingConv::ARM_AAPCS_VFP },
};
for (const auto &LC : LibraryCalls) {
setLibcallName(LC.Op, LC.Name);
setLibcallCallingConv(LC.Op, LC.CC);
}
}
if (Subtarget->isTargetWatchOS() ||
(Subtarget->isTargetIOS() &&
!Subtarget->getTargetTriple().isOSVersionLT(5, 0))) {
setLibcallName(RTLIB::SDIVREM_I32, "__divmodsi4");
setLibcallName(RTLIB::UDIVREM_I32, "__udivmodsi4");
}
if (Subtarget->isAAPCS_ABI()) {
setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_AAPCS);
setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_AAPCS);
setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_AAPCS);
} else {
setLibcallCallingConv(RTLIB::FPROUND_F32_F16, CallingConv::ARM_APCS);
setLibcallCallingConv(RTLIB::FPROUND_F64_F16, CallingConv::ARM_APCS);
setLibcallCallingConv(RTLIB::FPEXT_F16_F32, CallingConv::ARM_APCS);
}
if (Subtarget->isThumb1Only())
addRegisterClass(MVT::i32, &ARM::tGPRRegClass);
else
addRegisterClass(MVT::i32, &ARM::GPRRegClass);
if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
!Subtarget->isThumb1Only()) {
addRegisterClass(MVT::f32, &ARM::SPRRegClass);
addRegisterClass(MVT::f64, &ARM::DPRRegClass);
}
for (MVT VT : MVT::vector_valuetypes()) {
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);
}
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);
}
setOperationAction(ISD::ConstantFP, MVT::f32, Custom);
setOperationAction(ISD::ConstantFP, MVT::f64, Custom);
if (Subtarget->hasNEON()) {
addDRTypeForNEON(MVT::v2f32);
addDRTypeForNEON(MVT::v8i8);
addDRTypeForNEON(MVT::v4i16);
addDRTypeForNEON(MVT::v2i32);
addDRTypeForNEON(MVT::v1i64);
addQRTypeForNEON(MVT::v4f32);
addQRTypeForNEON(MVT::v2f64);
addQRTypeForNEON(MVT::v16i8);
addQRTypeForNEON(MVT::v8i16);
addQRTypeForNEON(MVT::v4i32);
addQRTypeForNEON(MVT::v2i64);
setOperationAction(ISD::FADD, MVT::v2f64, Expand);
setOperationAction(ISD::FSUB, MVT::v2f64, Expand);
setOperationAction(ISD::FMUL, MVT::v2f64, Expand);
setOperationAction(ISD::FDIV, MVT::v2f64, Expand);
setOperationAction(ISD::FREM, MVT::v2f64, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Expand);
setOperationAction(ISD::SETCC, MVT::v2f64, Expand);
setOperationAction(ISD::FNEG, MVT::v2f64, Expand);
setOperationAction(ISD::FABS, MVT::v2f64, Expand);
setOperationAction(ISD::FSQRT, MVT::v2f64, Expand);
setOperationAction(ISD::FSIN, MVT::v2f64, Expand);
setOperationAction(ISD::FCOS, MVT::v2f64, Expand);
setOperationAction(ISD::FPOWI, MVT::v2f64, Expand);
setOperationAction(ISD::FPOW, MVT::v2f64, Expand);
setOperationAction(ISD::FLOG, MVT::v2f64, Expand);
setOperationAction(ISD::FLOG2, MVT::v2f64, Expand);
setOperationAction(ISD::FLOG10, MVT::v2f64, Expand);
setOperationAction(ISD::FEXP, MVT::v2f64, Expand);
setOperationAction(ISD::FEXP2, MVT::v2f64, Expand);
setOperationAction(ISD::FCEIL, MVT::v2f64, Expand);
setOperationAction(ISD::FTRUNC, MVT::v2f64, Expand);
setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
setOperationAction(ISD::FMA, MVT::v2f64, Expand);
setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
setOperationAction(ISD::FCOS, MVT::v4f32, Expand);
setOperationAction(ISD::FPOWI, MVT::v4f32, Expand);
setOperationAction(ISD::FPOW, MVT::v4f32, Expand);
setOperationAction(ISD::FLOG, MVT::v4f32, Expand);
setOperationAction(ISD::FLOG2, MVT::v4f32, Expand);
setOperationAction(ISD::FLOG10, MVT::v4f32, Expand);
setOperationAction(ISD::FEXP, MVT::v4f32, Expand);
setOperationAction(ISD::FEXP2, MVT::v4f32, Expand);
setOperationAction(ISD::FCEIL, MVT::v4f32, Expand);
setOperationAction(ISD::FTRUNC, MVT::v4f32, Expand);
setOperationAction(ISD::FRINT, MVT::v4f32, Expand);
setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Expand);
setOperationAction(ISD::FFLOOR, MVT::v4f32, Expand);
setOperationAction(ISD::FSQRT, MVT::v2f32, Expand);
setOperationAction(ISD::FSIN, MVT::v2f32, Expand);
setOperationAction(ISD::FCOS, MVT::v2f32, Expand);
setOperationAction(ISD::FPOWI, MVT::v2f32, Expand);
setOperationAction(ISD::FPOW, MVT::v2f32, Expand);
setOperationAction(ISD::FLOG, MVT::v2f32, Expand);
setOperationAction(ISD::FLOG2, MVT::v2f32, Expand);
setOperationAction(ISD::FLOG10, MVT::v2f32, Expand);
setOperationAction(ISD::FEXP, MVT::v2f32, Expand);
setOperationAction(ISD::FEXP2, MVT::v2f32, Expand);
setOperationAction(ISD::FCEIL, MVT::v2f32, Expand);
setOperationAction(ISD::FTRUNC, MVT::v2f32, Expand);
setOperationAction(ISD::FRINT, MVT::v2f32, Expand);
setOperationAction(ISD::FNEARBYINT, MVT::v2f32, Expand);
setOperationAction(ISD::FFLOOR, MVT::v2f32, Expand);
setOperationAction(ISD::MUL, MVT::v1i64, Expand);
setOperationAction(ISD::MUL, MVT::v8i16, Custom);
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::MUL, MVT::v2i64, Custom);
setOperationAction(ISD::SDIV, MVT::v4i16, Custom);
setOperationAction(ISD::SDIV, MVT::v8i8, Custom);
setOperationAction(ISD::UDIV, MVT::v4i16, Custom);
setOperationAction(ISD::UDIV, MVT::v8i8, Custom);
setOperationAction(ISD::SETCC, MVT::v1i64, Expand);
setOperationAction(ISD::SETCC, MVT::v2i64, Expand);
setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::v4i16, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::v4i16, Custom);
setOperationAction(ISD::FP_ROUND, MVT::v2f32, Expand);
setOperationAction(ISD::FP_EXTEND, MVT::v2f64, Expand);
setOperationAction(ISD::CTPOP, MVT::v2i32, Custom);
setOperationAction(ISD::CTPOP, MVT::v4i32, Custom);
setOperationAction(ISD::CTPOP, MVT::v4i16, Custom);
setOperationAction(ISD::CTPOP, MVT::v8i16, Custom);
if (!Subtarget->hasVFP4()) {
setOperationAction(ISD::FMA, MVT::v2f32, Expand);
setOperationAction(ISD::FMA, MVT::v4f32, Expand);
}
setTargetDAGCombine(ISD::INTRINSIC_VOID);
setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRL);
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::SIGN_EXTEND);
setTargetDAGCombine(ISD::ZERO_EXTEND);
setTargetDAGCombine(ISD::ANY_EXTEND);
setTargetDAGCombine(ISD::SELECT_CC);
setTargetDAGCombine(ISD::BUILD_VECTOR);
setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
setTargetDAGCombine(ISD::INSERT_VECTOR_ELT);
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::FP_TO_SINT);
setTargetDAGCombine(ISD::FP_TO_UINT);
setTargetDAGCombine(ISD::FDIV);
setTargetDAGCombine(ISD::LOAD);
MVT Tys[6] = {MVT::v8i8, MVT::v4i8, MVT::v2i8,
MVT::v4i16, MVT::v2i16,
MVT::v2i32};
for (unsigned i = 0; i < 6; ++i) {
for (MVT VT : MVT::integer_vector_valuetypes()) {
setLoadExtAction(ISD::EXTLOAD, VT, Tys[i], Legal);
setLoadExtAction(ISD::ZEXTLOAD, VT, Tys[i], Legal);
setLoadExtAction(ISD::SEXTLOAD, VT, Tys[i], Legal);
}
}
}
if (!Subtarget->isThumb1Only())
setTargetDAGCombine(ISD::ADDC);
if (Subtarget->isFPOnlySP()) {
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(Subtarget->getRegisterInfo());
for (MVT VT : MVT::fp_valuetypes()) {
setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
}
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
for (MVT VT : MVT::integer_valuetypes())
setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
if (!Subtarget->isThumb1Only()) {
for (unsigned im = (unsigned)ISD::PRE_INC;
im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
setIndexedLoadAction(im, MVT::i1, Legal);
setIndexedLoadAction(im, MVT::i8, Legal);
setIndexedLoadAction(im, MVT::i16, Legal);
setIndexedLoadAction(im, MVT::i32, Legal);
setIndexedStoreAction(im, MVT::i1, Legal);
setIndexedStoreAction(im, MVT::i8, Legal);
setIndexedStoreAction(im, MVT::i16, Legal);
setIndexedStoreAction(im, MVT::i32, Legal);
}
}
setOperationAction(ISD::SADDO, MVT::i32, Custom);
setOperationAction(ISD::UADDO, MVT::i32, Custom);
setOperationAction(ISD::SSUBO, MVT::i32, Custom);
setOperationAction(ISD::USUBO, MVT::i32, Custom);
setOperationAction(ISD::MUL, MVT::i64, Expand);
setOperationAction(ISD::MULHU, MVT::i32, Expand);
if (Subtarget->isThumb1Only()) {
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
}
if (Subtarget->isThumb1Only() || !Subtarget->hasV6Ops()
|| (Subtarget->isThumb2() && !Subtarget->hasThumb2DSP()))
setOperationAction(ISD::MULHS, MVT::i32, Expand);
setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRL, MVT::i64, Custom);
setOperationAction(ISD::SRA, MVT::i64, Custom);
if (!Subtarget->isThumb1Only()) {
setOperationAction(ISD::ADDC, MVT::i32, Custom);
setOperationAction(ISD::ADDE, MVT::i32, Custom);
setOperationAction(ISD::SUBC, MVT::i32, Custom);
setOperationAction(ISD::SUBE, MVT::i32, Custom);
}
setOperationAction(ISD::ROTL, MVT::i32, Expand);
setOperationAction(ISD::CTTZ, MVT::i32, Custom);
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
if (!Subtarget->hasV5TOps() || Subtarget->isThumb1Only())
setOperationAction(ISD::CTLZ, MVT::i32, Expand);
setOperationAction(ISD::CTTZ_ZERO_UNDEF , MVT::i32 , Expand);
setOperationAction(ISD::CTLZ_ZERO_UNDEF , MVT::i32 , Expand);
setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Custom);
if (!Subtarget->hasV6Ops())
setOperationAction(ISD::BSWAP, MVT::i32, Expand);
if (!(Subtarget->hasDivide() && Subtarget->isThumb2()) &&
!(Subtarget->hasDivideInARMMode() && !Subtarget->isThumb())) {
setOperationAction(ISD::SDIV, MVT::i32, Expand);
setOperationAction(ISD::UDIV, MVT::i32, Expand);
}
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
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::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
setOperationAction(ISD::GLOBAL_OFFSET_TABLE, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
setOperationAction(ISD::TRAP, MVT::Other, Legal);
setOperationAction(ISD::VASTART, MVT::Other, Custom);
setOperationAction(ISD::VAARG, MVT::Other, Expand);
setOperationAction(ISD::VACOPY, MVT::Other, Expand);
setOperationAction(ISD::VAEND, MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
if (!Subtarget->useSjLjEH()) {
setExceptionPointerRegister(ARM::R0);
setExceptionSelectorRegister(ARM::R1);
}
if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
else
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
if (TM.Options.ThreadModel == ThreadModel::Single)
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Expand);
else if (Subtarget->hasAnyDataBarrier() && !Subtarget->isThumb1Only()) {
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
if (!Subtarget->hasV8Ops()) {
setInsertFencesForAtomic(true);
}
} else {
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other,
Subtarget->hasAnyDataBarrier() ? Custom : Expand);
setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_SWAP, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_ADD, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_SUB, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_AND, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_OR, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_XOR, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_NAND, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_MIN, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_MAX, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_UMIN, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD_UMAX, MVT::i32, Expand);
setOperationAction(ISD::ATOMIC_LOAD, MVT::i32, Custom);
setOperationAction(ISD::ATOMIC_STORE, MVT::i32, Custom);
}
setOperationAction(ISD::PREFETCH, MVT::Other, Custom);
if (!Subtarget->hasV6Ops()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8, Expand);
}
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
!Subtarget->isThumb1Only()) {
setOperationAction(ISD::BITCAST, MVT::i64, Custom);
setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
}
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
if (Subtarget->useSjLjEH()) {
setOperationAction(ISD::EH_SJLJ_SETJMP, MVT::i32, Custom);
setOperationAction(ISD::EH_SJLJ_LONGJMP, MVT::Other, Custom);
setLibcallName(RTLIB::UNWIND_RESUME, "_Unwind_SjLj_Resume");
}
setOperationAction(ISD::SETCC, MVT::i32, Expand);
setOperationAction(ISD::SETCC, MVT::f32, Expand);
setOperationAction(ISD::SETCC, MVT::f64, Expand);
setOperationAction(ISD::SELECT, MVT::i32, Custom);
setOperationAction(ISD::SELECT, MVT::f32, Custom);
setOperationAction(ISD::SELECT, MVT::f64, Custom);
setOperationAction(ISD::SELECT_CC, MVT::i32, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
setOperationAction(ISD::BRCOND, MVT::Other, Expand);
setOperationAction(ISD::BR_CC, MVT::i32, Custom);
setOperationAction(ISD::BR_CC, MVT::f32, Custom);
setOperationAction(ISD::BR_CC, MVT::f64, Custom);
setOperationAction(ISD::BR_JT, MVT::Other, Custom);
setOperationAction(ISD::FSIN, MVT::f64, Expand);
setOperationAction(ISD::FSIN, MVT::f32, Expand);
setOperationAction(ISD::FCOS, MVT::f32, Expand);
setOperationAction(ISD::FCOS, MVT::f64, Expand);
setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
setOperationAction(ISD::FREM, MVT::f64, Expand);
setOperationAction(ISD::FREM, MVT::f32, Expand);
if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
!Subtarget->isThumb1Only()) {
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);
}
setOperationAction(ISD::FPOW, MVT::f64, Expand);
setOperationAction(ISD::FPOW, MVT::f32, Expand);
if (!Subtarget->hasVFP4()) {
setOperationAction(ISD::FMA, MVT::f64, Expand);
setOperationAction(ISD::FMA, MVT::f32, Expand);
}
if (!TM.Options.UseSoftFloat && !Subtarget->isThumb1Only()) {
if (Subtarget->hasVFP2()) {
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
}
if (!Subtarget->hasFPARMv8() || Subtarget->isFPOnlySP()) {
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
}
if (!Subtarget->hasFP16()) {
setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
}
}
if (Subtarget->hasSinCos()) {
setLibcallName(RTLIB::SINCOS_F32, "sincosf");
setLibcallName(RTLIB::SINCOS_F64, "sincos");
if (Subtarget->isAAPCS_ABI() && Subtarget->isTargetHardFloat()) {
setLibcallCallingConv(RTLIB::SINCOS_F32, CallingConv::ARM_AAPCS_VFP);
setLibcallCallingConv(RTLIB::SINCOS_F64, CallingConv::ARM_AAPCS_VFP);
}
if (Subtarget->isTargetIOS() || Subtarget->isTargetWatchOS()) {
setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
}
}
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);
}
}
setTargetDAGCombine(ISD::ADD);
setTargetDAGCombine(ISD::SUB);
setTargetDAGCombine(ISD::MUL);
setTargetDAGCombine(ISD::AND);
setTargetDAGCombine(ISD::OR);
setTargetDAGCombine(ISD::XOR);
if (Subtarget->hasV6Ops())
setTargetDAGCombine(ISD::SRL);
setStackPointerRegisterToSaveRestore(ARM::SP);
if (TM.Options.UseSoftFloat || Subtarget->isThumb1Only() ||
!Subtarget->hasVFP2())
setSchedulingPreference(Sched::RegPressure);
else
setSchedulingPreference(Sched::Hybrid);
MaxStoresPerMemset = 8;
MaxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
MaxStoresPerMemcpy = 4; MaxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 4 : 2;
MaxStoresPerMemmove = 4; MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 4 : 2;
setMinStackArgumentAlignment(4);
PredictableSelectIsExpensive = Subtarget->isLikeA9();
setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2);
}
std::pair<const TargetRegisterClass *, uint8_t>
ARMTargetLowering::findRepresentativeClass(const TargetRegisterInfo *TRI,
MVT VT) const {
const TargetRegisterClass *RRC = nullptr;
uint8_t Cost = 1;
switch (VT.SimpleTy) {
default:
return TargetLowering::findRepresentativeClass(TRI, VT);
case MVT::f32: case MVT::f64: case MVT::v8i8: case MVT::v4i16:
case MVT::v2i32: case MVT::v1i64: case MVT::v2f32:
RRC = &ARM::DPRRegClass;
if (Subtarget->useNEONForSinglePrecisionFP())
Cost = 2;
break;
case MVT::v16i8: case MVT::v8i16: case MVT::v4i32: case MVT::v2i64:
case MVT::v4f32: case MVT::v2f64:
RRC = &ARM::DPRRegClass;
Cost = 2;
break;
case MVT::v4i64:
RRC = &ARM::DPRRegClass;
Cost = 4;
break;
case MVT::v8i64:
RRC = &ARM::DPRRegClass;
Cost = 8;
break;
}
return std::make_pair(RRC, Cost);
}
const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
default: return nullptr;
case ARMISD::Wrapper: return "ARMISD::Wrapper";
case ARMISD::WrapperPIC: return "ARMISD::WrapperPIC";
case ARMISD::WrapperJT: return "ARMISD::WrapperJT";
case ARMISD::CALL: return "ARMISD::CALL";
case ARMISD::CALL_PRED: return "ARMISD::CALL_PRED";
case ARMISD::CALL_NOLINK: return "ARMISD::CALL_NOLINK";
case ARMISD::tCALL: return "ARMISD::tCALL";
case ARMISD::BRCOND: return "ARMISD::BRCOND";
case ARMISD::BR_JT: return "ARMISD::BR_JT";
case ARMISD::BR2_JT: return "ARMISD::BR2_JT";
case ARMISD::RET_FLAG: return "ARMISD::RET_FLAG";
case ARMISD::INTRET_FLAG: return "ARMISD::INTRET_FLAG";
case ARMISD::PIC_ADD: return "ARMISD::PIC_ADD";
case ARMISD::CMP: return "ARMISD::CMP";
case ARMISD::CMN: return "ARMISD::CMN";
case ARMISD::CMPZ: return "ARMISD::CMPZ";
case ARMISD::CMPFP: return "ARMISD::CMPFP";
case ARMISD::CMPFPw0: return "ARMISD::CMPFPw0";
case ARMISD::BCC_i64: return "ARMISD::BCC_i64";
case ARMISD::FMSTAT: return "ARMISD::FMSTAT";
case ARMISD::CMOV: return "ARMISD::CMOV";
case ARMISD::RBIT: return "ARMISD::RBIT";
case ARMISD::FTOSI: return "ARMISD::FTOSI";
case ARMISD::FTOUI: return "ARMISD::FTOUI";
case ARMISD::SITOF: return "ARMISD::SITOF";
case ARMISD::UITOF: return "ARMISD::UITOF";
case ARMISD::SRL_FLAG: return "ARMISD::SRL_FLAG";
case ARMISD::SRA_FLAG: return "ARMISD::SRA_FLAG";
case ARMISD::RRX: return "ARMISD::RRX";
case ARMISD::ADDC: return "ARMISD::ADDC";
case ARMISD::ADDE: return "ARMISD::ADDE";
case ARMISD::SUBC: return "ARMISD::SUBC";
case ARMISD::SUBE: return "ARMISD::SUBE";
case ARMISD::VMOVRRD: return "ARMISD::VMOVRRD";
case ARMISD::VMOVDRR: return "ARMISD::VMOVDRR";
case ARMISD::EH_SJLJ_SETJMP: return "ARMISD::EH_SJLJ_SETJMP";
case ARMISD::EH_SJLJ_LONGJMP:return "ARMISD::EH_SJLJ_LONGJMP";
case ARMISD::TC_RETURN: return "ARMISD::TC_RETURN";
case ARMISD::THREAD_POINTER:return "ARMISD::THREAD_POINTER";
case ARMISD::DYN_ALLOC: return "ARMISD::DYN_ALLOC";
case ARMISD::MEMBARRIER_MCR: return "ARMISD::MEMBARRIER_MCR";
case ARMISD::PRELOAD: return "ARMISD::PRELOAD";
case ARMISD::WIN__CHKSTK: return "ARMISD:::WIN__CHKSTK";
case ARMISD::VCEQ: return "ARMISD::VCEQ";
case ARMISD::VCEQZ: return "ARMISD::VCEQZ";
case ARMISD::VCGE: return "ARMISD::VCGE";
case ARMISD::VCGEZ: return "ARMISD::VCGEZ";
case ARMISD::VCLEZ: return "ARMISD::VCLEZ";
case ARMISD::VCGEU: return "ARMISD::VCGEU";
case ARMISD::VCGT: return "ARMISD::VCGT";
case ARMISD::VCGTZ: return "ARMISD::VCGTZ";
case ARMISD::VCLTZ: return "ARMISD::VCLTZ";
case ARMISD::VCGTU: return "ARMISD::VCGTU";
case ARMISD::VTST: return "ARMISD::VTST";
case ARMISD::VSHL: return "ARMISD::VSHL";
case ARMISD::VSHRs: return "ARMISD::VSHRs";
case ARMISD::VSHRu: return "ARMISD::VSHRu";
case ARMISD::VRSHRs: return "ARMISD::VRSHRs";
case ARMISD::VRSHRu: return "ARMISD::VRSHRu";
case ARMISD::VRSHRN: return "ARMISD::VRSHRN";
case ARMISD::VQSHLs: return "ARMISD::VQSHLs";
case ARMISD::VQSHLu: return "ARMISD::VQSHLu";
case ARMISD::VQSHLsu: return "ARMISD::VQSHLsu";
case ARMISD::VQSHRNs: return "ARMISD::VQSHRNs";
case ARMISD::VQSHRNu: return "ARMISD::VQSHRNu";
case ARMISD::VQSHRNsu: return "ARMISD::VQSHRNsu";
case ARMISD::VQRSHRNs: return "ARMISD::VQRSHRNs";
case ARMISD::VQRSHRNu: return "ARMISD::VQRSHRNu";
case ARMISD::VQRSHRNsu: return "ARMISD::VQRSHRNsu";
case ARMISD::VGETLANEu: return "ARMISD::VGETLANEu";
case ARMISD::VGETLANEs: return "ARMISD::VGETLANEs";
case ARMISD::VMOVIMM: return "ARMISD::VMOVIMM";
case ARMISD::VMVNIMM: return "ARMISD::VMVNIMM";
case ARMISD::VMOVFPIMM: return "ARMISD::VMOVFPIMM";
case ARMISD::VDUP: return "ARMISD::VDUP";
case ARMISD::VDUPLANE: return "ARMISD::VDUPLANE";
case ARMISD::VEXT: return "ARMISD::VEXT";
case ARMISD::VREV64: return "ARMISD::VREV64";
case ARMISD::VREV32: return "ARMISD::VREV32";
case ARMISD::VREV16: return "ARMISD::VREV16";
case ARMISD::VZIP: return "ARMISD::VZIP";
case ARMISD::VUZP: return "ARMISD::VUZP";
case ARMISD::VTRN: return "ARMISD::VTRN";
case ARMISD::VTBL1: return "ARMISD::VTBL1";
case ARMISD::VTBL2: return "ARMISD::VTBL2";
case ARMISD::VMULLs: return "ARMISD::VMULLs";
case ARMISD::VMULLu: return "ARMISD::VMULLu";
case ARMISD::UMLAL: return "ARMISD::UMLAL";
case ARMISD::SMLAL: return "ARMISD::SMLAL";
case ARMISD::BUILD_VECTOR: return "ARMISD::BUILD_VECTOR";
case ARMISD::FMAX: return "ARMISD::FMAX";
case ARMISD::FMIN: return "ARMISD::FMIN";
case ARMISD::VMAXNM: return "ARMISD::VMAX";
case ARMISD::VMINNM: return "ARMISD::VMIN";
case ARMISD::BFI: return "ARMISD::BFI";
case ARMISD::VORRIMM: return "ARMISD::VORRIMM";
case ARMISD::VBICIMM: return "ARMISD::VBICIMM";
case ARMISD::VBSL: return "ARMISD::VBSL";
case ARMISD::VLD2DUP: return "ARMISD::VLD2DUP";
case ARMISD::VLD3DUP: return "ARMISD::VLD3DUP";
case ARMISD::VLD4DUP: return "ARMISD::VLD4DUP";
case ARMISD::VLD1_UPD: return "ARMISD::VLD1_UPD";
case ARMISD::VLD2_UPD: return "ARMISD::VLD2_UPD";
case ARMISD::VLD3_UPD: return "ARMISD::VLD3_UPD";
case ARMISD::VLD4_UPD: return "ARMISD::VLD4_UPD";
case ARMISD::VLD2LN_UPD: return "ARMISD::VLD2LN_UPD";
case ARMISD::VLD3LN_UPD: return "ARMISD::VLD3LN_UPD";
case ARMISD::VLD4LN_UPD: return "ARMISD::VLD4LN_UPD";
case ARMISD::VLD2DUP_UPD: return "ARMISD::VLD2DUP_UPD";
case ARMISD::VLD3DUP_UPD: return "ARMISD::VLD3DUP_UPD";
case ARMISD::VLD4DUP_UPD: return "ARMISD::VLD4DUP_UPD";
case ARMISD::VST1_UPD: return "ARMISD::VST1_UPD";
case ARMISD::VST2_UPD: return "ARMISD::VST2_UPD";
case ARMISD::VST3_UPD: return "ARMISD::VST3_UPD";
case ARMISD::VST4_UPD: return "ARMISD::VST4_UPD";
case ARMISD::VST2LN_UPD: return "ARMISD::VST2LN_UPD";
case ARMISD::VST3LN_UPD: return "ARMISD::VST3LN_UPD";
case ARMISD::VST4LN_UPD: return "ARMISD::VST4LN_UPD";
}
}
EVT ARMTargetLowering::getSetCCResultType(LLVMContext &, EVT VT) const {
if (!VT.isVector()) return getPointerTy();
return VT.changeVectorElementTypeToInteger();
}
const TargetRegisterClass *ARMTargetLowering::getRegClassFor(MVT VT) const {
if (Subtarget->hasNEON()) {
if (VT == MVT::v4i64)
return &ARM::QQPRRegClass;
if (VT == MVT::v8i64)
return &ARM::QQQQPRRegClass;
}
return TargetLowering::getRegClassFor(VT);
}
FastISel *
ARMTargetLowering::createFastISel(FunctionLoweringInfo &funcInfo,
const TargetLibraryInfo *libInfo) const {
return ARM::createFastISel(funcInfo, libInfo);
}
Sched::Preference ARMTargetLowering::getSchedulingPreference(SDNode *N) const {
unsigned NumVals = N->getNumValues();
if (!NumVals)
return Sched::RegPressure;
for (unsigned i = 0; i != NumVals; ++i) {
EVT VT = N->getValueType(i);
if (VT == MVT::Glue || VT == MVT::Other)
continue;
if (VT.isFloatingPoint() || VT.isVector())
return Sched::ILP;
}
if (!N->isMachineOpcode())
return Sched::RegPressure;
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
const MCInstrDesc &MCID = TII->get(N->getMachineOpcode());
if (MCID.getNumDefs() == 0)
return Sched::RegPressure;
if (!Itins->isEmpty() &&
Itins->getOperandCycle(MCID.getSchedClass(), 0) > 2)
return Sched::ILP;
return Sched::RegPressure;
}
static ARMCC::CondCodes IntCCToARMCC(ISD::CondCode CC) {
switch (CC) {
default: llvm_unreachable("Unknown condition code!");
case ISD::SETNE: return ARMCC::NE;
case ISD::SETEQ: return ARMCC::EQ;
case ISD::SETGT: return ARMCC::GT;
case ISD::SETGE: return ARMCC::GE;
case ISD::SETLT: return ARMCC::LT;
case ISD::SETLE: return ARMCC::LE;
case ISD::SETUGT: return ARMCC::HI;
case ISD::SETUGE: return ARMCC::HS;
case ISD::SETULT: return ARMCC::LO;
case ISD::SETULE: return ARMCC::LS;
}
}
static void FPCCToARMCC(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
ARMCC::CondCodes &CondCode2) {
CondCode2 = ARMCC::AL;
switch (CC) {
default: llvm_unreachable("Unknown FP condition!");
case ISD::SETEQ:
case ISD::SETOEQ: CondCode = ARMCC::EQ; break;
case ISD::SETGT:
case ISD::SETOGT: CondCode = ARMCC::GT; break;
case ISD::SETGE:
case ISD::SETOGE: CondCode = ARMCC::GE; break;
case ISD::SETOLT: CondCode = ARMCC::MI; break;
case ISD::SETOLE: CondCode = ARMCC::LS; break;
case ISD::SETONE: CondCode = ARMCC::MI; CondCode2 = ARMCC::GT; break;
case ISD::SETO: CondCode = ARMCC::VC; break;
case ISD::SETUO: CondCode = ARMCC::VS; break;
case ISD::SETUEQ: CondCode = ARMCC::EQ; CondCode2 = ARMCC::VS; break;
case ISD::SETUGT: CondCode = ARMCC::HI; break;
case ISD::SETUGE: CondCode = ARMCC::PL; break;
case ISD::SETLT:
case ISD::SETULT: CondCode = ARMCC::LT; break;
case ISD::SETLE:
case ISD::SETULE: CondCode = ARMCC::LE; break;
case ISD::SETNE:
case ISD::SETUNE: CondCode = ARMCC::NE; break;
}
}
#include "ARMGenCallingConv.inc"
CallingConv::ID
ARMTargetLowering::getEffectiveCallingConv(CallingConv::ID CC,
bool isVarArg) const {
switch (CC) {
default:
llvm_unreachable("Unsupported calling convention");
case CallingConv::ARM_AAPCS:
case CallingConv::ARM_APCS:
case CallingConv::GHC:
return CC;
case CallingConv::ARM_AAPCS_VFP:
return isVarArg ? CallingConv::ARM_AAPCS : CallingConv::ARM_AAPCS_VFP;
case CallingConv::C:
if (!Subtarget->isAAPCS_ABI()) {
return CallingConv::ARM_APCS;
} else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() &&
getTargetMachine().Options.FloatABIType == FloatABI::Hard &&
!isVarArg)
return CallingConv::ARM_AAPCS_VFP;
else
return CallingConv::ARM_AAPCS;
case CallingConv::Fast:
case CallingConv::Swift:
if (!Subtarget->isAAPCS_ABI()) {
if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::Fast;
return CallingConv::ARM_APCS;
} else if (Subtarget->hasVFP2() && !Subtarget->isThumb1Only() && !isVarArg)
return CallingConv::ARM_AAPCS_VFP;
else
return CallingConv::ARM_AAPCS;
}
}
CCAssignFn *ARMTargetLowering::CCAssignFnForNode(CallingConv::ID CC,
bool Return,
bool isVarArg) const {
switch (getEffectiveCallingConv(CC, isVarArg)) {
default:
llvm_unreachable("Unsupported calling convention");
case CallingConv::ARM_APCS:
return (Return ? RetCC_ARM_APCS : CC_ARM_APCS);
case CallingConv::ARM_AAPCS:
return (Return ? RetCC_ARM_AAPCS : CC_ARM_AAPCS);
case CallingConv::ARM_AAPCS_VFP:
return (Return ? RetCC_ARM_AAPCS_VFP : CC_ARM_AAPCS_VFP);
case CallingConv::Fast:
return (Return ? RetFastCC_ARM_APCS : FastCC_ARM_APCS);
case CallingConv::GHC:
return (Return ? RetCC_ARM_APCS : CC_ARM_APCS_GHC);
}
}
SDValue
ARMTargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals,
bool isThisReturn, SDValue ThisVal) const {
SmallVector<CCValAssign, 16> RVLocs;
ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext(), Call);
CCInfo.AnalyzeCallResult(Ins,
CCAssignFnForNode(CallConv, true,
isVarArg));
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign VA = RVLocs[i];
if (i == 0 && isThisReturn) {
assert(!VA.needsCustom() && VA.getLocVT() == MVT::i32 &&
"unexpected return calling convention register assignment");
InVals.push_back(ThisVal);
continue;
}
SDValue Val;
if (VA.needsCustom()) {
SDValue Lo = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
InFlag);
Chain = Lo.getValue(1);
InFlag = Lo.getValue(2);
VA = RVLocs[++i]; SDValue Hi = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
InFlag);
Chain = Hi.getValue(1);
InFlag = Hi.getValue(2);
if (!Subtarget->isLittle())
std::swap (Lo, Hi);
Val = DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);
if (VA.getLocVT() == MVT::v2f64) {
SDValue Vec = DAG.getNode(ISD::UNDEF, dl, MVT::v2f64);
Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Vec, Val,
DAG.getConstant(0, MVT::i32));
VA = RVLocs[++i]; Lo = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32, InFlag);
Chain = Lo.getValue(1);
InFlag = Lo.getValue(2);
VA = RVLocs[++i]; Hi = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32, InFlag);
Chain = Hi.getValue(1);
InFlag = Hi.getValue(2);
if (!Subtarget->isLittle())
std::swap (Lo, Hi);
Val = DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);
Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Vec, Val,
DAG.getConstant(1, MVT::i32));
}
} else {
Val = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getLocVT(),
InFlag);
Chain = Val.getValue(1);
InFlag = Val.getValue(2);
}
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::BCvt:
Val = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), Val);
break;
}
InVals.push_back(Val);
}
return Chain;
}
SDValue
ARMTargetLowering::LowerMemOpCallTo(SDValue Chain,
SDValue StackPtr, SDValue Arg,
SDLoc dl, SelectionDAG &DAG,
const CCValAssign &VA,
ISD::ArgFlagsTy Flags) const {
unsigned LocMemOffset = VA.getLocMemOffset();
SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset);
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
return DAG.getStore(Chain, dl, Arg, PtrOff,
MachinePointerInfo::getStack(LocMemOffset),
false, false, 0);
}
void ARMTargetLowering::PassF64ArgInRegs(SDLoc dl, SelectionDAG &DAG,
SDValue Chain, SDValue &Arg,
RegsToPassVector &RegsToPass,
CCValAssign &VA, CCValAssign &NextVA,
SDValue &StackPtr,
SmallVectorImpl<SDValue> &MemOpChains,
ISD::ArgFlagsTy Flags) const {
SDValue fmrrd = DAG.getNode(ARMISD::VMOVRRD, dl,
DAG.getVTList(MVT::i32, MVT::i32), Arg);
unsigned id = Subtarget->isLittle() ? 0 : 1;
RegsToPass.push_back(std::make_pair(VA.getLocReg(), fmrrd.getValue(id)));
if (NextVA.isRegLoc())
RegsToPass.push_back(std::make_pair(NextVA.getLocReg(), fmrrd.getValue(1-id)));
else {
assert(NextVA.isMemLoc());
if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, fmrrd.getValue(1-id),
dl, DAG, NextVA,
Flags));
}
}
SDValue
ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
SDLoc &dl = CLI.DL;
SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
SmallVectorImpl<SDValue> &OutVals = CLI.OutVals;
SmallVectorImpl<ISD::InputArg> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &isTailCall = CLI.IsTailCall;
CallingConv::ID CallConv = CLI.CallConv;
bool doesNotRet = CLI.DoesNotReturn;
bool isVarArg = CLI.IsVarArg;
MachineFunction &MF = DAG.getMachineFunction();
bool isStructRet = (Outs.empty()) ? false : Outs[0].Flags.isSRet();
bool isThisReturn = false;
bool isSibCall = false;
if (!Subtarget->supportsTailCall() || MF.getTarget().Options.DisableTailCalls)
isTailCall = false;
if (isTailCall) {
isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
isVarArg, isStructRet, MF.getFunction()->hasStructRetAttr(),
Outs, OutVals, Ins, DAG);
if (!isTailCall && CLI.CS && CLI.CS->isMustTailCall())
report_fatal_error("failed to perform tail call elimination on a call "
"site marked musttail");
if (isTailCall) {
++NumTailCalls;
isSibCall = true;
}
}
SmallVector<CCValAssign, 16> ArgLocs;
ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext(), Call);
CCInfo.AnalyzeCallOperands(Outs,
CCAssignFnForNode(CallConv, false,
isVarArg));
unsigned NumBytes = CCInfo.getNextStackOffset();
if (isSibCall)
NumBytes = 0;
if (!isSibCall)
Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true),
dl);
SDValue StackPtr = DAG.getCopyFromReg(Chain, dl, ARM::SP, getPointerTy());
RegsToPassVector RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
i != e;
++i, ++realArgIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[realArgIdx];
ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
bool isByVal = Flags.isByVal();
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::BCvt:
Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
break;
}
if (VA.needsCustom()) {
if (VA.getLocVT() == MVT::v2f64) {
SDValue Op0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
DAG.getConstant(0, MVT::i32));
SDValue Op1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
DAG.getConstant(1, MVT::i32));
PassF64ArgInRegs(dl, DAG, Chain, Op0, RegsToPass,
VA, ArgLocs[++i], StackPtr, MemOpChains, Flags);
VA = ArgLocs[++i]; if (VA.isRegLoc()) {
PassF64ArgInRegs(dl, DAG, Chain, Op1, RegsToPass,
VA, ArgLocs[++i], StackPtr, MemOpChains, Flags);
} else {
assert(VA.isMemLoc());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Op1,
dl, DAG, VA, Flags));
}
} else {
PassF64ArgInRegs(dl, DAG, Chain, Arg, RegsToPass, VA, ArgLocs[++i],
StackPtr, MemOpChains, Flags);
}
} else if (VA.isRegLoc()) {
if (realArgIdx == 0 && Flags.isReturned() && Outs[0].VT == MVT::i32) {
assert(VA.getLocVT() == MVT::i32 &&
"unexpected calling convention register assignment");
assert(!Ins.empty() && Ins[0].VT == MVT::i32 &&
"unexpected use of 'returned'");
isThisReturn = true;
}
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else if (isByVal) {
assert(VA.isMemLoc());
unsigned offset = 0;
unsigned ByValArgsCount = CCInfo.getInRegsParamsCount();
unsigned CurByValIdx = CCInfo.getInRegsParamsProcessed();
if (CurByValIdx < ByValArgsCount) {
unsigned RegBegin, RegEnd;
CCInfo.getInRegsParamInfo(CurByValIdx, RegBegin, RegEnd);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
unsigned int i, j;
for (i = 0, j = RegBegin; j < RegEnd; i++, j++) {
SDValue Const = DAG.getConstant(4*i, MVT::i32);
SDValue AddArg = DAG.getNode(ISD::ADD, dl, PtrVT, Arg, Const);
SDValue Load = DAG.getLoad(PtrVT, dl, Chain, AddArg,
MachinePointerInfo(),
false, false, false,
DAG.InferPtrAlignment(AddArg));
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(j, Load));
}
offset = RegEnd - RegBegin;
CCInfo.nextInRegsParam();
}
if (Flags.getByValSize() > 4*offset) {
unsigned LocMemOffset = VA.getLocMemOffset();
SDValue StkPtrOff = DAG.getIntPtrConstant(LocMemOffset);
SDValue Dst = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr,
StkPtrOff);
SDValue SrcOffset = DAG.getIntPtrConstant(4*offset);
SDValue Src = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg, SrcOffset);
SDValue SizeNode = DAG.getConstant(Flags.getByValSize() - 4*offset,
MVT::i32);
SDValue AlignNode = DAG.getConstant(Flags.getByValAlign(), MVT::i32);
SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, Dst, Src, SizeNode, AlignNode};
MemOpChains.push_back(DAG.getNode(ARMISD::COPY_STRUCT_BYVAL, dl, VTs,
Ops));
}
} else if (!isSibCall) {
assert(VA.isMemLoc());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
dl, DAG, VA, Flags));
}
}
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
SDValue InFlag;
if (!isTailCall)
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
if (isTailCall) {
InFlag = SDValue();
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
InFlag = SDValue();
}
bool isDirect = false;
bool isARMFunc = false;
bool isLocalARMFunc = false;
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
if (EnableARMLongCalls) {
assert((Subtarget->isTargetWindows() ||
getTargetMachine().getRelocationModel() == Reloc::Static) &&
"long-calls with non-static relocation model!");
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
const GlobalValue *GV = G->getGlobal();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, ARMCP::CPValue, 0);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
} else if (ExternalSymbolSDNode *S=dyn_cast<ExternalSymbolSDNode>(Callee)) {
const char *Sym = S->getSymbol();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym,
ARMPCLabelIndex, 0);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
}
} else if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
const GlobalValue *GV = G->getGlobal();
isDirect = true;
bool isExt = GV->isDeclaration() || GV->isWeakForLinker();
bool isStub = (isExt && Subtarget->isTargetMachO()) &&
getTargetMachine().getRelocationModel() != Reloc::Static;
isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
isLocalARMFunc = !Subtarget->isThumb() && (!isExt || !ARMInterworking);
if (isStub && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
assert(Subtarget->isTargetMachO() && "WrapperPIC use on non-MachO?");
Callee = DAG.getNode(ARMISD::WrapperPIC, dl, getPointerTy(),
DAG.getTargetGlobalAddress(GV, dl, getPointerTy(),
0, ARMII::MO_NONLAZY));
Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(), Callee,
MachinePointerInfo::getGOT(), false, false, true, 0);
} else if (Subtarget->isTargetCOFF()) {
assert(Subtarget->isTargetWindows() &&
"Windows is the only supported COFF target");
unsigned TargetFlags = GV->hasDLLImportStorageClass()
? ARMII::MO_DLLIMPORT
: ARMII::MO_NO_FLAG;
Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), 0,
TargetFlags);
if (GV->hasDLLImportStorageClass())
Callee = DAG.getLoad(getPointerTy(), dl, DAG.getEntryNode(),
DAG.getNode(ARMISD::Wrapper, dl, getPointerTy(),
Callee), MachinePointerInfo::getGOT(),
false, false, false, 0);
} else {
unsigned OpFlags = 0;
if (Subtarget->isTargetELF() &&
getTargetMachine().getRelocationModel() == Reloc::PIC_)
OpFlags = ARMII::MO_PLT;
Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), 0, OpFlags);
}
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
isDirect = true;
bool isStub = Subtarget->isTargetMachO() &&
getTargetMachine().getRelocationModel() != Reloc::Static;
isARMFunc = !Subtarget->isThumb() || (isStub && !Subtarget->isMClass());
const char *Sym = S->getSymbol();
if (isARMFunc && Subtarget->isThumb1Only() && !Subtarget->hasV5TOps()) {
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
ARMConstantPoolSymbol::Create(*DAG.getContext(), Sym,
ARMPCLabelIndex, 4);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, getPointerTy(), 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
Callee = DAG.getLoad(getPointerTy(), dl,
DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Callee = DAG.getNode(ARMISD::PIC_ADD, dl,
getPointerTy(), Callee, PICLabel);
} else {
unsigned OpFlags = 0;
if (Subtarget->isTargetELF() &&
getTargetMachine().getRelocationModel() == Reloc::PIC_)
OpFlags = ARMII::MO_PLT;
Callee = DAG.getTargetExternalSymbol(Sym, getPointerTy(), OpFlags);
}
}
unsigned CallOpc;
bool HasMinSizeAttr = MF.getFunction()->hasFnAttribute(Attribute::MinSize);
if (Subtarget->isThumb()) {
if ((!isDirect || isARMFunc) && !Subtarget->hasV5TOps())
CallOpc = ARMISD::CALL_NOLINK;
else
CallOpc = isARMFunc ? ARMISD::CALL : ARMISD::tCALL;
} else {
if (!isDirect && !Subtarget->hasV5TOps())
CallOpc = ARMISD::CALL_NOLINK;
else if (doesNotRet && isDirect && Subtarget->hasRAS() &&
!HasMinSizeAttr)
CallOpc = ARMISD::CALL_NOLINK;
else
CallOpc = isLocalARMFunc ? ARMISD::CALL_PRED : ARMISD::CALL;
}
std::vector<SDValue> Ops;
Ops.push_back(Chain);
Ops.push_back(Callee);
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
if (!isTailCall) {
const uint32_t *Mask;
const ARMBaseRegisterInfo *ARI = Subtarget->getRegisterInfo();
if (isThisReturn) {
Mask = ARI->getThisReturnPreservedMask(MF, CallConv);
if (!Mask) {
isThisReturn = false;
Mask = ARI->getCallPreservedMask(MF, CallConv);
}
} else
Mask = ARI->getCallPreservedMask(MF, CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
}
if (InFlag.getNode())
Ops.push_back(InFlag);
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
if (isTailCall) {
MF.getFrameInfo()->setHasTailCall();
return DAG.getNode(ARMISD::TC_RETURN, dl, NodeTys, Ops);
}
Chain = DAG.getNode(CallOpc, dl, NodeTys, Ops);
InFlag = Chain.getValue(1);
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
DAG.getIntPtrConstant(0, true), InFlag, dl);
if (!Ins.empty())
InFlag = Chain.getValue(1);
return LowerCallResult(Chain, InFlag, CallConv, isVarArg, Ins, dl, DAG,
InVals, isThisReturn,
isThisReturn ? OutVals[0] : SDValue());
}
void ARMTargetLowering::HandleByVal(CCState *State, unsigned &Size,
unsigned Align) const {
assert((State->getCallOrPrologue() == Prologue ||
State->getCallOrPrologue() == Call) &&
"unhandled ParmContext");
Align = std::max(Align, 4U);
unsigned Reg = State->AllocateReg(GPRArgRegs);
if (!Reg)
return;
unsigned AlignInRegs = Align / 4;
unsigned Waste = (ARM::R4 - Reg) % AlignInRegs;
for (unsigned i = 0; i < Waste; ++i)
Reg = State->AllocateReg(GPRArgRegs);
if (!Reg)
return;
unsigned Excess = 4 * (ARM::R4 - Reg);
const unsigned NSAAOffset = State->getNextStackOffset();
if (NSAAOffset != 0 && Size > Excess) {
while (State->AllocateReg(GPRArgRegs))
;
return;
}
unsigned ByValRegBegin = Reg;
unsigned ByValRegEnd = std::min<unsigned>(Reg + Size / 4, ARM::R4);
State->addInRegsParamInfo(ByValRegBegin, ByValRegEnd);
for (unsigned i = Reg + 1; i != ByValRegEnd; ++i)
State->AllocateReg(GPRArgRegs);
Size = std::max<int>(Size - Excess, 0);
}
static
bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
MachineFrameInfo *MFI, const MachineRegisterInfo *MRI,
const TargetInstrInfo *TII) {
unsigned Bytes = Arg.getValueType().getSizeInBits() / 8;
int FI = INT_MAX;
if (Arg.getOpcode() == ISD::CopyFromReg) {
unsigned VR = cast<RegisterSDNode>(Arg.getOperand(1))->getReg();
if (!TargetRegisterInfo::isVirtualRegister(VR))
return false;
MachineInstr *Def = MRI->getVRegDef(VR);
if (!Def)
return false;
if (!Flags.isByVal()) {
if (!TII->isLoadFromStackSlot(Def, FI))
return false;
} else {
return false;
}
} else if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Arg)) {
if (Flags.isByVal())
return false;
SDValue Ptr = Ld->getBasePtr();
FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(Ptr);
if (!FINode)
return false;
FI = FINode->getIndex();
} else
return false;
assert(FI != INT_MAX);
if (!MFI->isFixedObjectIndex(FI))
return false;
return Offset == MFI->getObjectOffset(FI) && Bytes == MFI->getObjectSize(FI);
}
bool
ARMTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
CallingConv::ID CalleeCC,
bool isVarArg,
bool isCalleeStructRet,
bool isCallerStructRet,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins,
SelectionDAG& DAG) const {
const Function *CallerF = DAG.getMachineFunction().getFunction();
CallingConv::ID CallerCC = CallerF->getCallingConv();
bool CCMatch = CallerCC == CalleeCC;
if (isVarArg && !Outs.empty())
return false;
if (CallerF->hasFnAttribute("interrupt"))
return false;
if (isCalleeStructRet || isCallerStructRet)
return false;
if (Subtarget->isThumb1Only())
return false;
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
const GlobalValue *GV = G->getGlobal();
const Triple TT(getTargetMachine().getTargetTriple());
if (GV->hasExternalWeakLinkage() &&
(!TT.isOSWindows() || TT.isOSBinFormatELF() || TT.isOSBinFormatMachO()))
return false;
}
if (!CCMatch) {
SmallVector<CCValAssign, 16> RVLocs1;
ARMCCState CCInfo1(CalleeCC, false, DAG.getMachineFunction(), RVLocs1,
*DAG.getContext(), Call);
CCInfo1.AnalyzeCallResult(Ins, CCAssignFnForNode(CalleeCC, true, isVarArg));
SmallVector<CCValAssign, 16> RVLocs2;
ARMCCState CCInfo2(CallerCC, false, DAG.getMachineFunction(), RVLocs2,
*DAG.getContext(), Call);
CCInfo2.AnalyzeCallResult(Ins, CCAssignFnForNode(CallerCC, true, isVarArg));
if (RVLocs1.size() != RVLocs2.size())
return false;
for (unsigned i = 0, e = RVLocs1.size(); i != e; ++i) {
if (RVLocs1[i].isRegLoc() != RVLocs2[i].isRegLoc())
return false;
if (RVLocs1[i].getLocInfo() != RVLocs2[i].getLocInfo())
return false;
if (RVLocs1[i].isRegLoc()) {
if (RVLocs1[i].getLocReg() != RVLocs2[i].getLocReg())
return false;
} else {
if (RVLocs1[i].getLocMemOffset() != RVLocs2[i].getLocMemOffset())
return false;
}
}
}
const ARMFunctionInfo *AFI_Caller = DAG.getMachineFunction().
getInfo<ARMFunctionInfo>();
if (AFI_Caller->getArgRegsSaveSize())
return false;
if (!Outs.empty()) {
SmallVector<CCValAssign, 16> ArgLocs;
ARMCCState CCInfo(CalleeCC, isVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext(), Call);
CCInfo.AnalyzeCallOperands(Outs,
CCAssignFnForNode(CalleeCC, false, isVarArg));
if (CCInfo.getNextStackOffset()) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
const MachineRegisterInfo *MRI = &MF.getRegInfo();
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
for (unsigned i = 0, realArgIdx = 0, e = ArgLocs.size();
i != e;
++i, ++realArgIdx) {
CCValAssign &VA = ArgLocs[i];
EVT RegVT = VA.getLocVT();
SDValue Arg = OutVals[realArgIdx];
ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;
if (VA.getLocInfo() == CCValAssign::Indirect)
return false;
if (VA.needsCustom()) {
if (!VA.isRegLoc())
return false;
if (!ArgLocs[++i].isRegLoc())
return false;
if (RegVT == MVT::v2f64) {
if (!ArgLocs[++i].isRegLoc())
return false;
if (!ArgLocs[++i].isRegLoc())
return false;
}
} else if (!VA.isRegLoc()) {
if (!MatchingStackOffset(Arg, VA.getLocMemOffset(), Flags,
MFI, MRI, TII))
return false;
}
}
}
}
return true;
}
bool
ARMTargetLowering::CanLowerReturn(CallingConv::ID CallConv,
MachineFunction &MF, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
LLVMContext &Context) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, MF, RVLocs, Context);
return CCInfo.CheckReturn(Outs, CCAssignFnForNode(CallConv, true,
isVarArg));
}
static SDValue LowerInterruptReturn(SmallVectorImpl<SDValue> &RetOps,
SDLoc DL, SelectionDAG &DAG) {
const MachineFunction &MF = DAG.getMachineFunction();
const Function *F = MF.getFunction();
StringRef IntKind = F->getFnAttribute("interrupt").getValueAsString();
int64_t LROffset;
if (IntKind == "" || IntKind == "IRQ" || IntKind == "FIQ" ||
IntKind == "ABORT")
LROffset = 4;
else if (IntKind == "SWI" || IntKind == "UNDEF")
LROffset = 0;
else
report_fatal_error("Unsupported interrupt attribute. If present, value "
"must be one of: IRQ, FIQ, SWI, ABORT or UNDEF");
RetOps.insert(RetOps.begin() + 1, DAG.getConstant(LROffset, MVT::i32, false));
return DAG.getNode(ARMISD::INTRET_FLAG, DL, MVT::Other, RetOps);
}
SDValue
ARMTargetLowering::LowerReturn(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
SDLoc dl, SelectionDAG &DAG) const {
SmallVector<CCValAssign, 16> RVLocs;
ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext(), Call);
CCInfo.AnalyzeReturn(Outs, CCAssignFnForNode(CallConv, true,
isVarArg));
SDValue Flag;
SmallVector<SDValue, 4> RetOps;
RetOps.push_back(Chain); bool isLittleEndian = Subtarget->isLittle();
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
AFI->setReturnRegsCount(RVLocs.size());
for (unsigned i = 0, realRVLocIdx = 0;
i != RVLocs.size();
++i, ++realRVLocIdx) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
SDValue Arg = OutVals[realRVLocIdx];
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::BCvt:
Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
break;
}
if (VA.needsCustom()) {
if (VA.getLocVT() == MVT::v2f64) {
SDValue Half = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
DAG.getConstant(0, MVT::i32));
SDValue HalfGPRs = DAG.getNode(ARMISD::VMOVRRD, dl,
DAG.getVTList(MVT::i32, MVT::i32), Half);
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
HalfGPRs.getValue(isLittleEndian ? 0 : 1),
Flag);
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
VA = RVLocs[++i]; Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
HalfGPRs.getValue(isLittleEndian ? 1 : 0),
Flag);
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
VA = RVLocs[++i];
Arg = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Arg,
DAG.getConstant(1, MVT::i32));
}
SDValue fmrrd = DAG.getNode(ARMISD::VMOVRRD, dl,
DAG.getVTList(MVT::i32, MVT::i32), Arg);
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
fmrrd.getValue(isLittleEndian ? 0 : 1),
Flag);
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
VA = RVLocs[++i]; Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(),
fmrrd.getValue(isLittleEndian ? 1 : 0),
Flag);
} else
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
RetOps[0] = Chain;
if (Flag.getNode())
RetOps.push_back(Flag);
if (DAG.getMachineFunction().getFunction()->hasFnAttribute("interrupt") &&
!Subtarget->isMClass()) {
if (Subtarget->isThumb1Only())
report_fatal_error("interrupt attribute is not supported in Thumb1");
return LowerInterruptReturn(RetOps, dl, DAG);
}
return DAG.getNode(ARMISD::RET_FLAG, dl, MVT::Other, RetOps);
}
bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
if (N->getNumValues() != 1)
return false;
if (!N->hasNUsesOfValue(1, 0))
return false;
SDValue TCChain = Chain;
SDNode *Copy = *N->use_begin();
if (Copy->getOpcode() == ISD::CopyToReg) {
if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
return false;
TCChain = Copy->getOperand(0);
} else if (Copy->getOpcode() == ARMISD::VMOVRRD) {
SDNode *VMov = Copy;
SmallPtrSet<SDNode*, 2> Copies;
for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
UI != UE; ++UI) {
if (UI->getOpcode() != ISD::CopyToReg)
return false;
Copies.insert(*UI);
}
if (Copies.size() > 2)
return false;
for (SDNode::use_iterator UI = VMov->use_begin(), UE = VMov->use_end();
UI != UE; ++UI) {
SDValue UseChain = UI->getOperand(0);
if (Copies.count(UseChain.getNode()))
Copy = *UI;
else {
if (UI->getOperand(UI->getNumOperands()-1).getValueType() == MVT::Glue)
return false;
TCChain = UseChain;
}
}
} else if (Copy->getOpcode() == ISD::BITCAST) {
if (!Copy->hasOneUse())
return false;
Copy = *Copy->use_begin();
if (Copy->getOpcode() != ISD::CopyToReg || !Copy->hasNUsesOfValue(1, 0))
return false;
if (Copy->getOperand(Copy->getNumOperands()-1).getValueType() == MVT::Glue)
return false;
TCChain = Copy->getOperand(0);
} else {
return false;
}
bool HasRet = false;
for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
UI != UE; ++UI) {
if (UI->getOpcode() != ARMISD::RET_FLAG &&
UI->getOpcode() != ARMISD::INTRET_FLAG)
return false;
HasRet = true;
}
if (!HasRet)
return false;
Chain = TCChain;
return true;
}
bool ARMTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
if (!Subtarget->supportsTailCall())
return false;
if (!CI->isTailCall() || getTargetMachine().Options.DisableTailCalls)
return false;
return !Subtarget->isThumb1Only();
}
static SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) {
EVT PtrVT = Op.getValueType();
SDLoc dl(Op);
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
SDValue Res;
if (CP->isMachineConstantPoolEntry())
Res = DAG.getTargetConstantPool(CP->getMachineCPVal(), PtrVT,
CP->getAlignment());
else
Res = DAG.getTargetConstantPool(CP->getConstVal(), PtrVT,
CP->getAlignment());
return DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Res);
}
unsigned ARMTargetLowering::getJumpTableEncoding() const {
return MachineJumpTableInfo::EK_Inline;
}
SDValue ARMTargetLowering::LowerBlockAddress(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = 0;
SDLoc DL(Op);
EVT PtrVT = getPointerTy();
const BlockAddress *BA = cast<BlockAddressSDNode>(Op)->getBlockAddress();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
SDValue CPAddr;
if (RelocM == Reloc::Static) {
CPAddr = DAG.getTargetConstantPool(BA, PtrVT, 4);
} else {
unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(BA, ARMPCLabelIndex,
ARMCP::CPBlockAddress, PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
}
CPAddr = DAG.getNode(ARMISD::Wrapper, DL, PtrVT, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
if (RelocM == Reloc::Static)
return Result;
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
return DAG.getNode(ARMISD::PIC_ADD, DL, PtrVT, Result, PICLabel);
}
SDValue
ARMTargetLowering::LowerGlobalTLSAddressDarwin(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget->isTargetDarwin() && "TLS only supported on Darwin");
SDLoc DL(Op);
SDValue DescAddr = LowerGlobalAddressDarwin(Op, DAG);
SDValue Chain = DAG.getEntryNode();
SDValue FuncTLVGet =
DAG.getLoad(MVT::i32, DL, Chain, DescAddr, MachinePointerInfo::getGOT(),
false, true, true, 4);
Chain = FuncTLVGet.getValue(1);
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
MFI->setAdjustsStack(true);
auto TRI = DAG.getMachineFunction().getSubtarget().getRegisterInfo();
auto ARI = static_cast<const ARMRegisterInfo *>(TRI);
const uint32_t *Mask = ARI->getTLSCallPreservedMask(DAG.getMachineFunction());
Chain = DAG.getCopyToReg(Chain, DL, ARM::R0, DescAddr, SDValue());
Chain =
DAG.getNode(ARMISD::CALL, DL, DAG.getVTList(MVT::Other, MVT::Glue),
Chain, FuncTLVGet, DAG.getRegister(ARM::R0, MVT::i32),
DAG.getRegisterMask(Mask), Chain.getValue(1));
return DAG.getCopyFromReg(Chain, DL, ARM::R0, MVT::i32, Chain.getValue(1));
}
SDValue
ARMTargetLowering::LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA,
SelectionDAG &DAG) const {
SDLoc dl(GA);
EVT PtrVT = getPointerTy();
unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
ARMCP::CPValue, PCAdj, ARMCP::TLSGD, true);
SDValue Argument = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Argument = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Argument);
Argument = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Argument,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
SDValue Chain = Argument.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Argument = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Argument, PICLabel);
ArgListTy Args;
ArgListEntry Entry;
Entry.Node = Argument;
Entry.Ty = (Type *) Type::getInt32Ty(*DAG.getContext());
Args.push_back(Entry);
CallingConv::ID CC =
Subtarget->isAAPCS_ABI() ? CallingConv::ARM_AAPCS : CallingConv::C;
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(CC, Type::getInt32Ty(*DAG.getContext()),
DAG.getExternalSymbol("__tls_get_addr", PtrVT), std::move(Args),
0);
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
return CallResult.first;
}
SDValue
ARMTargetLowering::LowerToTLSExecModels(GlobalAddressSDNode *GA,
SelectionDAG &DAG,
TLSModel::Model model) const {
const GlobalValue *GV = GA->getGlobal();
SDLoc dl(GA);
SDValue Offset;
SDValue Chain = DAG.getEntryNode();
EVT PtrVT = getPointerTy();
SDValue ThreadPointer = DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
if (model == TLSModel::InitialExec) {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
unsigned char PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(GA->getGlobal(), ARMPCLabelIndex,
ARMCP::CPValue, PCAdj, ARMCP::GOTTPOFF,
true);
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
Chain = Offset.getValue(1);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Offset = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Offset, PICLabel);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
} else {
assert(model == TLSModel::LocalExec);
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(GV, ARMCP::TPOFF);
Offset = DAG.getTargetConstantPool(CPV, PtrVT, 4);
Offset = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, Offset);
Offset = DAG.getLoad(PtrVT, dl, Chain, Offset,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
}
return DAG.getNode(ISD::ADD, dl, PtrVT, ThreadPointer, Offset);
}
SDValue
ARMTargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const {
if (Subtarget->isTargetDarwin())
return LowerGlobalTLSAddressDarwin(Op, DAG);
assert(Subtarget->isTargetELF() && "Only ELF implemented here");
GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
TLSModel::Model model = getTargetMachine().getTLSModel(GA->getGlobal());
switch (model) {
case TLSModel::GeneralDynamic:
case TLSModel::LocalDynamic:
return LowerToTLSGeneralDynamicModel(GA, DAG);
case TLSModel::InitialExec:
case TLSModel::LocalExec:
return LowerToTLSExecModels(GA, DAG, model);
}
llvm_unreachable("bogus TLS model");
}
SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = getPointerTy();
SDLoc dl(Op);
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(GV,
UseGOTOFF ? ARMCP::GOTOFF : ARMCP::GOT);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
SDValue Chain = Result.getValue(1);
SDValue GOT = DAG.getGLOBAL_OFFSET_TABLE(PtrVT);
Result = DAG.getNode(ISD::ADD, dl, PtrVT, Result, GOT);
if (!UseGOTOFF)
Result = DAG.getLoad(PtrVT, dl, Chain, Result,
MachinePointerInfo::getGOT(),
false, false, false, 0);
return Result;
}
if (Subtarget->useMovt(DAG.getMachineFunction())) {
++NumMovwMovt;
return DAG.getNode(ARMISD::Wrapper, dl, PtrVT,
DAG.getTargetGlobalAddress(GV, dl, PtrVT));
} else {
SDValue CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
}
}
SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = getPointerTy();
SDLoc dl(Op);
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
if (Subtarget->useMovt(DAG.getMachineFunction()))
++NumMovwMovt;
unsigned Wrapper =
RelocM == Reloc::PIC_ ? ARMISD::WrapperPIC : ARMISD::Wrapper;
SDValue G = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, ARMII::MO_NONLAZY);
SDValue Result = DAG.getNode(Wrapper, dl, PtrVT, G);
if (Subtarget->GVIsIndirectSymbol(GV, RelocM))
Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
MachinePointerInfo::getGOT(), false, false, false, 0);
return Result;
}
SDValue ARMTargetLowering::LowerGlobalAddressWindows(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget->isTargetWindows() && "non-Windows COFF is not supported");
assert(Subtarget->useMovt(DAG.getMachineFunction()) &&
"Windows on ARM expects to use movw/movt");
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
const ARMII::TOF TargetFlags =
(GV->hasDLLImportStorageClass() ? ARMII::MO_DLLIMPORT : ARMII::MO_NO_FLAG);
EVT PtrVT = getPointerTy();
SDValue Result;
SDLoc DL(Op);
++NumMovwMovt;
Result = DAG.getNode(ARMISD::Wrapper, DL, PtrVT,
DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0,
TargetFlags));
if (GV->hasDLLImportStorageClass())
Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), Result,
MachinePointerInfo::getGOT(), false, false, false, 0);
return Result;
}
SDValue ARMTargetLowering::LowerGLOBAL_OFFSET_TABLE(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget->isTargetELF() &&
"GLOBAL OFFSET TABLE not implemented for non-ELF targets");
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
EVT PtrVT = getPointerTy();
SDLoc dl(Op);
unsigned PCAdj = Subtarget->isThumb() ? 4 : 8;
ARMConstantPoolValue *CPV =
ARMConstantPoolSymbol::Create(*DAG.getContext(), "_GLOBAL_OFFSET_TABLE_",
ARMPCLabelIndex, PCAdj);
SDValue CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
return DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
SDValue
ARMTargetLowering::LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
SDValue Val = DAG.getConstant(0, MVT::i32);
return DAG.getNode(ARMISD::EH_SJLJ_SETJMP, dl,
DAG.getVTList(MVT::i32, MVT::Other), Op.getOperand(0),
Op.getOperand(1), Val);
}
SDValue
ARMTargetLowering::LowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
return DAG.getNode(ARMISD::EH_SJLJ_LONGJMP, dl, MVT::Other, Op.getOperand(0),
Op.getOperand(1), DAG.getConstant(0, MVT::i32));
}
SDValue
ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *Subtarget) const {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
SDLoc dl(Op);
switch (IntNo) {
default: return SDValue(); case Intrinsic::arm_rbit: {
assert(Op.getOperand(1).getValueType() == MVT::i32 &&
"RBIT intrinsic must have i32 type!");
return DAG.getNode(ARMISD::RBIT, dl, MVT::i32, Op.getOperand(1));
}
case Intrinsic::arm_thread_pointer: {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
return DAG.getNode(ARMISD::THREAD_POINTER, dl, PtrVT);
}
case Intrinsic::eh_sjlj_lsda: {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
EVT PtrVT = getPointerTy();
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
SDValue CPAddr;
unsigned PCAdj = (RelocM != Reloc::PIC_)
? 0 : (Subtarget->isThumb() ? 4 : 8);
ARMConstantPoolValue *CPV =
ARMConstantPoolConstant::Create(MF.getFunction(), ARMPCLabelIndex,
ARMCP::CPLSDA, PCAdj);
CPAddr = DAG.getTargetConstantPool(CPV, PtrVT, 4);
CPAddr = DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
SDValue Result =
DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), CPAddr,
MachinePointerInfo::getConstantPool(),
false, false, false, 0);
if (RelocM == Reloc::PIC_) {
SDValue PICLabel = DAG.getConstant(ARMPCLabelIndex, MVT::i32);
Result = DAG.getNode(ARMISD::PIC_ADD, dl, PtrVT, Result, PICLabel);
}
return Result;
}
case Intrinsic::arm_neon_vmulls:
case Intrinsic::arm_neon_vmullu: {
unsigned NewOpc = (IntNo == Intrinsic::arm_neon_vmulls)
? ARMISD::VMULLs : ARMISD::VMULLu;
return DAG.getNode(NewOpc, SDLoc(Op), Op.getValueType(),
Op.getOperand(1), Op.getOperand(2));
}
}
}
static SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *Subtarget) {
SDLoc dl(Op);
if (!Subtarget->hasDataBarrier()) {
assert(Subtarget->hasV6Ops() && !Subtarget->isThumb() &&
"Unexpected ISD::ATOMIC_FENCE encountered. Should be libcall!");
return DAG.getNode(ARMISD::MEMBARRIER_MCR, dl, MVT::Other, Op.getOperand(0),
DAG.getConstant(0, MVT::i32));
}
ConstantSDNode *OrdN = cast<ConstantSDNode>(Op.getOperand(1));
AtomicOrdering Ord = static_cast<AtomicOrdering>(OrdN->getZExtValue());
ARM_MB::MemBOpt Domain = ARM_MB::ISH;
if (Subtarget->isMClass()) {
Domain = ARM_MB::SY;
} else if (Subtarget->isSwift() && Ord == Release) {
Domain = ARM_MB::ISHST;
}
return DAG.getNode(ISD::INTRINSIC_VOID, dl, MVT::Other, Op.getOperand(0),
DAG.getConstant(Intrinsic::arm_dmb, MVT::i32),
DAG.getConstant(Domain, MVT::i32));
}
static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *Subtarget) {
if (!(Subtarget->isThumb2() ||
(!Subtarget->isThumb1Only() && Subtarget->hasV5TEOps())))
return Op.getOperand(0);
SDLoc dl(Op);
unsigned isRead = ~cast<ConstantSDNode>(Op.getOperand(2))->getZExtValue() & 1;
if (!isRead &&
(!Subtarget->hasV7Ops() || !Subtarget->hasMPExtension()))
return Op.getOperand(0);
unsigned isData = cast<ConstantSDNode>(Op.getOperand(4))->getZExtValue();
if (Subtarget->isThumb()) {
isRead = ~isRead & 1;
isData = ~isData & 1;
}
return DAG.getNode(ARMISD::PRELOAD, dl, MVT::Other, Op.getOperand(0),
Op.getOperand(1), DAG.getConstant(isRead, MVT::i32),
DAG.getConstant(isData, MVT::i32));
}
static SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *FuncInfo = MF.getInfo<ARMFunctionInfo>();
SDLoc dl(Op);
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1),
MachinePointerInfo(SV), false, false, 0);
}
SDValue
ARMTargetLowering::GetF64FormalArgument(CCValAssign &VA, CCValAssign &NextVA,
SDValue &Root, SelectionDAG &DAG,
SDLoc dl) const {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
const TargetRegisterClass *RC;
if (AFI->isThumb1OnlyFunction())
RC = &ARM::tGPRRegClass;
else
RC = &ARM::GPRRegClass;
unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
SDValue ArgValue = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);
SDValue ArgValue2;
if (NextVA.isMemLoc()) {
MachineFrameInfo *MFI = MF.getFrameInfo();
int FI = MFI->CreateFixedObject(4, NextVA.getLocMemOffset(), true);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue2 = DAG.getLoad(MVT::i32, dl, Root, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, false, 0);
} else {
Reg = MF.addLiveIn(NextVA.getLocReg(), RC);
ArgValue2 = DAG.getCopyFromReg(Root, dl, Reg, MVT::i32);
}
if (!Subtarget->isLittle())
std::swap (ArgValue, ArgValue2);
return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, ArgValue, ArgValue2);
}
int
ARMTargetLowering::StoreByValRegs(CCState &CCInfo, SelectionDAG &DAG,
SDLoc dl, SDValue &Chain,
const Value *OrigArg,
unsigned InRegsParamRecordIdx,
int ArgOffset,
unsigned ArgSize) const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned RBegin, REnd;
if (InRegsParamRecordIdx < CCInfo.getInRegsParamsCount()) {
CCInfo.getInRegsParamInfo(InRegsParamRecordIdx, RBegin, REnd);
} else {
unsigned RBeginIdx = CCInfo.getFirstUnallocated(GPRArgRegs);
RBegin = RBeginIdx == 4 ? ARM::R4 : GPRArgRegs[RBeginIdx];
REnd = ARM::R4;
}
if (REnd != RBegin)
ArgOffset = -4 * (ARM::R4 - RBegin);
int FrameIndex = MFI->CreateFixedObject(ArgSize, ArgOffset, false);
SDValue FIN = DAG.getFrameIndex(FrameIndex, getPointerTy());
SmallVector<SDValue, 4> MemOps;
const TargetRegisterClass *RC =
AFI->isThumb1OnlyFunction() ? &ARM::tGPRRegClass : &ARM::GPRRegClass;
for (unsigned Reg = RBegin, i = 0; Reg < REnd; ++Reg, ++i) {
unsigned VReg = MF.addLiveIn(Reg, RC);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i32);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN,
MachinePointerInfo(OrigArg, 4 * i), false, false, 0);
MemOps.push_back(Store);
FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), FIN,
DAG.getConstant(4, getPointerTy()));
}
if (!MemOps.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
return FrameIndex;
}
void
ARMTargetLowering::VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
SDLoc dl, SDValue &Chain,
unsigned ArgOffset,
unsigned TotalArgRegsSaveSize,
bool ForceMutable) const {
MachineFunction &MF = DAG.getMachineFunction();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
int FrameIndex = StoreByValRegs(CCInfo, DAG, dl, Chain, nullptr,
CCInfo.getInRegsParamsCount(),
CCInfo.getNextStackOffset(), 4);
AFI->setVarArgsFrameIndex(FrameIndex);
}
SDValue
ARMTargetLowering::LowerFormalArguments(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg>
&Ins,
SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals)
const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
SmallVector<CCValAssign, 16> ArgLocs;
ARMCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext(), Prologue);
CCInfo.AnalyzeFormalArguments(Ins,
CCAssignFnForNode(CallConv, false,
isVarArg));
SmallVector<SDValue, 16> ArgValues;
SDValue ArgValue;
Function::const_arg_iterator CurOrigArg = MF.getFunction()->arg_begin();
unsigned CurArgIdx = 0;
AFI->setArgRegsSaveSize(0);
unsigned ArgRegBegin = ARM::R4;
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
if (CCInfo.getInRegsParamsProcessed() >= CCInfo.getInRegsParamsCount())
break;
CCValAssign &VA = ArgLocs[i];
unsigned Index = VA.getValNo();
ISD::ArgFlagsTy Flags = Ins[Index].Flags;
if (!Flags.isByVal())
continue;
assert(VA.isMemLoc() && "unexpected byval pointer in reg");
unsigned RBegin, REnd;
CCInfo.getInRegsParamInfo(CCInfo.getInRegsParamsProcessed(), RBegin, REnd);
ArgRegBegin = std::min(ArgRegBegin, RBegin);
CCInfo.nextInRegsParam();
}
CCInfo.rewindByValRegsInfo();
int lastInsIndex = -1;
if (isVarArg && MFI->hasVAStart()) {
unsigned RegIdx = CCInfo.getFirstUnallocated(GPRArgRegs);
if (RegIdx != array_lengthof(GPRArgRegs))
ArgRegBegin = std::min(ArgRegBegin, (unsigned)GPRArgRegs[RegIdx]);
}
unsigned TotalArgRegsSaveSize = 4 * (ARM::R4 - ArgRegBegin);
AFI->setArgRegsSaveSize(TotalArgRegsSaveSize);
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
if (Ins[VA.getValNo()].isOrigArg()) {
std::advance(CurOrigArg,
Ins[VA.getValNo()].getOrigArgIndex() - CurArgIdx);
CurArgIdx = Ins[VA.getValNo()].getOrigArgIndex();
}
if (VA.isRegLoc()) {
EVT RegVT = VA.getLocVT();
if (VA.needsCustom()) {
if (VA.getLocVT() == MVT::v2f64) {
SDValue ArgValue1 = GetF64FormalArgument(VA, ArgLocs[++i],
Chain, DAG, dl);
VA = ArgLocs[++i]; SDValue ArgValue2;
if (VA.isMemLoc()) {
int FI = MFI->CreateFixedObject(8, VA.getLocMemOffset(), true);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
ArgValue2 = DAG.getLoad(MVT::f64, dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, false, 0);
} else {
ArgValue2 = GetF64FormalArgument(VA, ArgLocs[++i],
Chain, DAG, dl);
}
ArgValue = DAG.getNode(ISD::UNDEF, dl, MVT::v2f64);
ArgValue = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64,
ArgValue, ArgValue1, DAG.getIntPtrConstant(0));
ArgValue = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64,
ArgValue, ArgValue2, DAG.getIntPtrConstant(1));
} else
ArgValue = GetF64FormalArgument(VA, ArgLocs[++i], Chain, DAG, dl);
} else {
const TargetRegisterClass *RC;
if (RegVT == MVT::f32)
RC = &ARM::SPRRegClass;
else if (RegVT == MVT::f64)
RC = &ARM::DPRRegClass;
else if (RegVT == MVT::v2f64)
RC = &ARM::QPRRegClass;
else if (RegVT == MVT::i32)
RC = AFI->isThumb1OnlyFunction() ? &ARM::tGPRRegClass
: &ARM::GPRRegClass;
else
llvm_unreachable("RegVT not supported by FORMAL_ARGUMENTS Lowering");
unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);
}
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::BCvt:
ArgValue = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), ArgValue);
break;
case CCValAssign::SExt:
ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
break;
case CCValAssign::ZExt:
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
break;
}
InVals.push_back(ArgValue);
} else {
assert(VA.isMemLoc());
assert(VA.getValVT() != MVT::i64 && "i64 should already be lowered");
int index = VA.getValNo();
if (index != lastInsIndex)
{
ISD::ArgFlagsTy Flags = Ins[index].Flags;
if (Flags.isByVal()) {
assert(Ins[index].isOrigArg() &&
"Byval arguments cannot be implicit");
unsigned CurByValIndex = CCInfo.getInRegsParamsProcessed();
int FrameIndex = StoreByValRegs(CCInfo, DAG, dl, Chain, CurOrigArg,
CurByValIndex, VA.getLocMemOffset(),
Flags.getByValSize());
InVals.push_back(DAG.getFrameIndex(FrameIndex, getPointerTy()));
CCInfo.nextInRegsParam();
} else {
unsigned FIOffset = VA.getLocMemOffset();
int FI = MFI->CreateFixedObject(VA.getLocVT().getSizeInBits()/8,
FIOffset, true);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
InVals.push_back(DAG.getLoad(VA.getValVT(), dl, Chain, FIN,
MachinePointerInfo::getFixedStack(FI),
false, false, false, 0));
}
lastInsIndex = index;
}
}
}
if (isVarArg && MFI->hasVAStart())
VarArgStyleRegisters(CCInfo, DAG, dl, Chain,
CCInfo.getNextStackOffset(),
TotalArgRegsSaveSize);
AFI->setArgumentStackSize(CCInfo.getNextStackOffset());
return Chain;
}
static bool isFloatingPointZero(SDValue Op) {
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op))
return CFP->getValueAPF().isPosZero();
else if (ISD::isEXTLoad(Op.getNode()) || ISD::isNON_EXTLoad(Op.getNode())) {
if (Op.getOperand(1).getOpcode() == ARMISD::Wrapper) {
SDValue WrapperOp = Op.getOperand(1).getOperand(0);
if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(WrapperOp))
if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
return CFP->getValueAPF().isPosZero();
}
} else if (Op->getOpcode() == ISD::BITCAST &&
Op->getValueType(0) == MVT::f64) {
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::getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC,
SDValue &ARMcc, SelectionDAG &DAG,
SDLoc dl) const {
if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS.getNode())) {
unsigned C = RHSC->getZExtValue();
if (!isLegalICmpImmediate(C)) {
switch (CC) {
default: break;
case ISD::SETLT:
case ISD::SETGE:
if (C != 0x80000000 && isLegalICmpImmediate(C-1)) {
CC = (CC == ISD::SETLT) ? ISD::SETLE : ISD::SETGT;
RHS = DAG.getConstant(C-1, MVT::i32);
}
break;
case ISD::SETULT:
case ISD::SETUGE:
if (C != 0 && isLegalICmpImmediate(C-1)) {
CC = (CC == ISD::SETULT) ? ISD::SETULE : ISD::SETUGT;
RHS = DAG.getConstant(C-1, MVT::i32);
}
break;
case ISD::SETLE:
case ISD::SETGT:
if (C != 0x7fffffff && isLegalICmpImmediate(C+1)) {
CC = (CC == ISD::SETLE) ? ISD::SETLT : ISD::SETGE;
RHS = DAG.getConstant(C+1, MVT::i32);
}
break;
case ISD::SETULE:
case ISD::SETUGT:
if (C != 0xffffffff && isLegalICmpImmediate(C+1)) {
CC = (CC == ISD::SETULE) ? ISD::SETULT : ISD::SETUGE;
RHS = DAG.getConstant(C+1, MVT::i32);
}
break;
}
}
}
ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
ARMISD::NodeType CompareType;
switch (CondCode) {
default:
CompareType = ARMISD::CMP;
break;
case ARMCC::EQ:
case ARMCC::NE:
CompareType = ARMISD::CMPZ;
break;
}
ARMcc = DAG.getConstant(CondCode, MVT::i32);
return DAG.getNode(CompareType, dl, MVT::Glue, LHS, RHS);
}
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);
else
Cmp = DAG.getNode(ARMISD::CMPFPw0, dl, MVT::Glue, LHS);
return DAG.getNode(ARMISD::FMSTAT, dl, MVT::Glue, Cmp);
}
SDValue
ARMTargetLowering::duplicateCmp(SDValue Cmp, SelectionDAG &DAG) const {
unsigned Opc = Cmp.getOpcode();
SDLoc DL(Cmp);
if (Opc == ARMISD::CMP || Opc == ARMISD::CMPZ)
return DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0),Cmp.getOperand(1));
assert(Opc == ARMISD::FMSTAT && "unexpected comparison operation");
Cmp = Cmp.getOperand(0);
Opc = Cmp.getOpcode();
if (Opc == ARMISD::CMPFP)
Cmp = DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0),Cmp.getOperand(1));
else {
assert(Opc == ARMISD::CMPFPw0 && "unexpected operand of FMSTAT");
Cmp = DAG.getNode(Opc, DL, MVT::Glue, Cmp.getOperand(0));
}
return DAG.getNode(ARMISD::FMSTAT, DL, MVT::Glue, Cmp);
}
std::pair<SDValue, SDValue>
ARMTargetLowering::getARMXALUOOp(SDValue Op, SelectionDAG &DAG,
SDValue &ARMcc) const {
assert(Op.getValueType() == MVT::i32 && "Unsupported value type");
SDValue Value, OverflowCmp;
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
switch (Op.getOpcode()) {
default:
llvm_unreachable("Unknown overflow instruction!");
case ISD::SADDO:
ARMcc = DAG.getConstant(ARMCC::VC, MVT::i32);
Value = DAG.getNode(ISD::ADD, SDLoc(Op), Op.getValueType(), LHS, RHS);
OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, Value, LHS);
break;
case ISD::UADDO:
ARMcc = DAG.getConstant(ARMCC::HS, MVT::i32);
Value = DAG.getNode(ISD::ADD, SDLoc(Op), Op.getValueType(), LHS, RHS);
OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, Value, LHS);
break;
case ISD::SSUBO:
ARMcc = DAG.getConstant(ARMCC::VC, MVT::i32);
Value = DAG.getNode(ISD::SUB, SDLoc(Op), Op.getValueType(), LHS, RHS);
OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, LHS, RHS);
break;
case ISD::USUBO:
ARMcc = DAG.getConstant(ARMCC::HS, MVT::i32);
Value = DAG.getNode(ISD::SUB, SDLoc(Op), Op.getValueType(), LHS, RHS);
OverflowCmp = DAG.getNode(ARMISD::CMP, SDLoc(Op), MVT::Glue, LHS, RHS);
break;
}
return std::make_pair(Value, OverflowCmp);
}
SDValue
ARMTargetLowering::LowerXALUO(SDValue Op, SelectionDAG &DAG) const {
if (!DAG.getTargetLoweringInfo().isTypeLegal(Op.getValueType()))
return SDValue();
SDValue Value, OverflowCmp;
SDValue ARMcc;
std::tie(Value, OverflowCmp) = getARMXALUOOp(Op, DAG, ARMcc);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue TVal = DAG.getConstant(1, MVT::i32);
SDValue FVal = DAG.getConstant(0, MVT::i32);
EVT VT = Op.getValueType();
SDValue Overflow = DAG.getNode(ARMISD::CMOV, SDLoc(Op), VT, TVal, FVal,
ARMcc, CCR, OverflowCmp);
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
return DAG.getNode(ISD::MERGE_VALUES, SDLoc(Op), VTs, Value, Overflow);
}
SDValue ARMTargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
SDValue Cond = Op.getOperand(0);
SDValue SelectTrue = Op.getOperand(1);
SDValue SelectFalse = Op.getOperand(2);
SDLoc dl(Op);
unsigned Opc = Cond.getOpcode();
if (Cond.getResNo() == 1 &&
(Opc == ISD::SADDO || Opc == ISD::UADDO || Opc == ISD::SSUBO ||
Opc == ISD::USUBO)) {
if (!DAG.getTargetLoweringInfo().isTypeLegal(Cond->getValueType(0)))
return SDValue();
SDValue Value, OverflowCmp;
SDValue ARMcc;
std::tie(Value, OverflowCmp) = getARMXALUOOp(Cond, DAG, ARMcc);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
EVT VT = Op.getValueType();
return getCMOV(SDLoc(Op), VT, SelectTrue, SelectFalse, ARMcc, CCR,
OverflowCmp, DAG);
}
if (Cond.getOpcode() == ARMISD::CMOV && Cond.hasOneUse()) {
const ConstantSDNode *CMOVTrue =
dyn_cast<ConstantSDNode>(Cond.getOperand(0));
const ConstantSDNode *CMOVFalse =
dyn_cast<ConstantSDNode>(Cond.getOperand(1));
if (CMOVTrue && CMOVFalse) {
unsigned CMOVTrueVal = CMOVTrue->getZExtValue();
unsigned CMOVFalseVal = CMOVFalse->getZExtValue();
SDValue True;
SDValue False;
if (CMOVTrueVal == 1 && CMOVFalseVal == 0) {
True = SelectTrue;
False = SelectFalse;
} else if (CMOVTrueVal == 0 && CMOVFalseVal == 1) {
True = SelectFalse;
False = SelectTrue;
}
if (True.getNode() && False.getNode()) {
EVT VT = Op.getValueType();
SDValue ARMcc = Cond.getOperand(2);
SDValue CCR = Cond.getOperand(3);
SDValue Cmp = duplicateCmp(Cond.getOperand(4), DAG);
assert(True.getValueType() == VT);
return getCMOV(dl, VT, True, False, ARMcc, CCR, Cmp, DAG);
}
}
}
Cond = DAG.getNode(ISD::AND, dl, Cond.getValueType(), Cond,
DAG.getConstant(1, Cond.getValueType()));
return DAG.getSelectCC(dl, Cond,
DAG.getConstant(0, Cond.getValueType()),
SelectTrue, SelectFalse, ISD::SETNE);
}
static ISD::CondCode getInverseCCForVSEL(ISD::CondCode CC) {
if (CC == ISD::SETNE)
return ISD::SETEQ;
return ISD::getSetCCInverse(CC, true);
}
static void checkVSELConstraints(ISD::CondCode CC, ARMCC::CondCodes &CondCode,
bool &swpCmpOps, bool &swpVselOps) {
if (CC == ISD::SETUGE || CC == ISD::SETOGE || CC == ISD::SETOLE ||
CC == ISD::SETULE)
CondCode = ARMCC::GE;
else if (CC == ISD::SETUGT || CC == ISD::SETOGT || CC == ISD::SETOLT ||
CC == ISD::SETULT)
CondCode = ARMCC::GT;
if (CC == ISD::SETOLE || CC == ISD::SETULE || CC == ISD::SETOLT ||
CC == ISD::SETULT)
swpCmpOps = true;
if (CC == ISD::SETULE || CC == ISD::SETULT || CC == ISD::SETUGE ||
CC == ISD::SETUGT) {
swpCmpOps = !swpCmpOps;
swpVselOps = !swpVselOps;
CondCode = CondCode == ARMCC::GT ? ARMCC::GE : ARMCC::GT;
}
if (CC == ISD::SETO) {
CondCode = ARMCC::VS;
swpVselOps = true;
}
if (CC == ISD::SETUNE) {
CondCode = ARMCC::EQ;
swpVselOps = true;
}
}
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 RHS = Op.getOperand(1);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
SDValue TrueVal = Op.getOperand(2);
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 (!RHS.getNode()) {
RHS = DAG.getConstant(0, LHS.getValueType());
CC = ISD::SETNE;
}
}
if (LHS.getValueType() == MVT::i32) {
if (Subtarget->hasFPARMv8() && (TrueVal.getValueType() == MVT::f32 ||
TrueVal.getValueType() == MVT::f64)) {
ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
if (CondCode == ARMCC::LT || CondCode == ARMCC::LE ||
CondCode == ARMCC::VC || CondCode == ARMCC::NE) {
CC = getInverseCCForVSEL(CC);
std::swap(TrueVal, FalseVal);
}
}
SDValue ARMcc;
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
return getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
}
ARMCC::CondCodes CondCode, CondCode2;
FPCCToARMCC(CC, CondCode, CondCode2);
if (Subtarget->hasFPARMv8() && (TrueVal.getValueType() == MVT::f32 ||
TrueVal.getValueType() == MVT::f64)) {
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;
bool swpVselOps = false;
checkVSELConstraints(CC, CondCode, swpCmpOps, swpVselOps);
if (CondCode == ARMCC::GT || CondCode == ARMCC::GE ||
CondCode == ARMCC::VS || CondCode == ARMCC::EQ) {
if (swpCmpOps)
std::swap(LHS, RHS);
if (swpVselOps)
std::swap(TrueVal, FalseVal);
}
}
SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32);
SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Result = getCMOV(dl, VT, FalseVal, TrueVal, ARMcc, CCR, Cmp, DAG);
if (CondCode2 != ARMCC::AL) {
SDValue ARMcc2 = DAG.getConstant(CondCode2, MVT::i32);
SDValue Cmp2 = getVFPCmp(LHS, RHS, DAG, dl);
Result = getCMOV(dl, VT, Result, TrueVal, ARMcc2, CCR, Cmp2, DAG);
}
return Result;
}
static bool canChangeToInt(SDValue Op, bool &SeenZero,
const ARMSubtarget *Subtarget) {
SDNode *N = Op.getNode();
if (!N->hasOneUse())
return false;
if (!N->getNumValues())
return false;
EVT VT = Op.getValueType();
if (VT != MVT::f32 && !Subtarget->isFPBrccSlow())
return false;
if (isFloatingPointZero(Op)) {
SeenZero = true;
return true;
}
return ISD::isNormalLoad(N);
}
static SDValue bitcastf32Toi32(SDValue Op, SelectionDAG &DAG) {
if (isFloatingPointZero(Op))
return DAG.getConstant(0, MVT::i32);
if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Op))
return DAG.getLoad(MVT::i32, SDLoc(Op),
Ld->getChain(), Ld->getBasePtr(), Ld->getPointerInfo(),
Ld->isVolatile(), Ld->isNonTemporal(),
Ld->isInvariant(), Ld->getAlignment());
llvm_unreachable("Unknown VFP cmp argument!");
}
static void expandf64Toi32(SDValue Op, SelectionDAG &DAG,
SDValue &RetVal1, SDValue &RetVal2) {
if (isFloatingPointZero(Op)) {
RetVal1 = DAG.getConstant(0, MVT::i32);
RetVal2 = DAG.getConstant(0, MVT::i32);
return;
}
if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Op)) {
SDValue Ptr = Ld->getBasePtr();
RetVal1 = DAG.getLoad(MVT::i32, SDLoc(Op),
Ld->getChain(), Ptr,
Ld->getPointerInfo(),
Ld->isVolatile(), Ld->isNonTemporal(),
Ld->isInvariant(), Ld->getAlignment());
EVT PtrType = Ptr.getValueType();
unsigned NewAlign = MinAlign(Ld->getAlignment(), 4);
SDValue NewPtr = DAG.getNode(ISD::ADD, SDLoc(Op),
PtrType, Ptr, DAG.getConstant(4, PtrType));
RetVal2 = DAG.getLoad(MVT::i32, SDLoc(Op),
Ld->getChain(), NewPtr,
Ld->getPointerInfo().getWithOffset(4),
Ld->isVolatile(), Ld->isNonTemporal(),
Ld->isInvariant(), NewAlign);
return;
}
llvm_unreachable("Unknown VFP cmp argument!");
}
SDValue
ARMTargetLowering::OptimizeVFPBrcond(SDValue Op, SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
SDValue LHS = Op.getOperand(2);
SDValue RHS = Op.getOperand(3);
SDValue Dest = Op.getOperand(4);
SDLoc dl(Op);
bool LHSSeenZero = false;
bool LHSOk = canChangeToInt(LHS, LHSSeenZero, Subtarget);
bool RHSSeenZero = false;
bool RHSOk = canChangeToInt(RHS, RHSSeenZero, Subtarget);
if (LHSOk && RHSOk && (LHSSeenZero || RHSSeenZero)) {
if (CC == ISD::SETOEQ)
CC = ISD::SETEQ;
else if (CC == ISD::SETUNE)
CC = ISD::SETNE;
SDValue Mask = DAG.getConstant(0x7fffffff, MVT::i32);
SDValue ARMcc;
if (LHS.getValueType() == MVT::f32) {
LHS = DAG.getNode(ISD::AND, dl, MVT::i32,
bitcastf32Toi32(LHS, DAG), Mask);
RHS = DAG.getNode(ISD::AND, dl, MVT::i32,
bitcastf32Toi32(RHS, DAG), Mask);
SDValue Cmp = getARMCmp(LHS, RHS, CC, ARMcc, DAG, dl);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
return DAG.getNode(ARMISD::BRCOND, dl, MVT::Other,
Chain, Dest, ARMcc, CCR, Cmp);
}
SDValue LHS1, LHS2;
SDValue RHS1, RHS2;
expandf64Toi32(LHS, DAG, LHS1, LHS2);
expandf64Toi32(RHS, DAG, RHS1, RHS2);
LHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, LHS2, Mask);
RHS2 = DAG.getNode(ISD::AND, dl, MVT::i32, RHS2, Mask);
ARMCC::CondCodes CondCode = IntCCToARMCC(CC);
ARMcc = DAG.getConstant(CondCode, MVT::i32);
SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, ARMcc, LHS1, LHS2, RHS1, RHS2, Dest };
return DAG.getNode(ARMISD::BCC_i64, dl, VTList, Ops);
}
return SDValue();
}
SDValue ARMTargetLowering::LowerBR_CC(SDValue Op, SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(1))->get();
SDValue LHS = Op.getOperand(2);
SDValue RHS = Op.getOperand(3);
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 (!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);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
return DAG.getNode(ARMISD::BRCOND, dl, MVT::Other,
Chain, Dest, ARMcc, CCR, Cmp);
}
assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);
if (getTargetMachine().Options.UnsafeFPMath &&
(CC == ISD::SETEQ || CC == ISD::SETOEQ ||
CC == ISD::SETNE || CC == ISD::SETUNE)) {
SDValue Result = OptimizeVFPBrcond(Op, DAG);
if (Result.getNode())
return Result;
}
ARMCC::CondCodes CondCode, CondCode2;
FPCCToARMCC(CC, CondCode, CondCode2);
SDValue ARMcc = DAG.getConstant(CondCode, MVT::i32);
SDValue Cmp = getVFPCmp(LHS, RHS, DAG, dl);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, Dest, ARMcc, CCR, Cmp };
SDValue Res = DAG.getNode(ARMISD::BRCOND, dl, VTList, Ops);
if (CondCode2 != ARMCC::AL) {
ARMcc = DAG.getConstant(CondCode2, MVT::i32);
SDValue Ops[] = { Res, Dest, ARMcc, CCR, Res.getValue(1) };
Res = DAG.getNode(ARMISD::BRCOND, dl, VTList, Ops);
}
return Res;
}
SDValue ARMTargetLowering::LowerBR_JT(SDValue Op, SelectionDAG &DAG) const {
SDValue Chain = Op.getOperand(0);
SDValue Table = Op.getOperand(1);
SDValue Index = Op.getOperand(2);
SDLoc dl(Op);
EVT PTy = getPointerTy();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Table);
SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PTy);
Table = DAG.getNode(ARMISD::WrapperJT, dl, MVT::i32, JTI);
Index = DAG.getNode(ISD::MUL, dl, PTy, Index, DAG.getConstant(4, PTy));
SDValue Addr = DAG.getNode(ISD::ADD, dl, PTy, Index, Table);
if (Subtarget->isThumb2()) {
return DAG.getNode(ARMISD::BR2_JT, dl, MVT::Other, Chain,
Addr, Op.getOperand(2), JTI);
}
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
Addr = DAG.getLoad((EVT)MVT::i32, dl, Chain, Addr,
MachinePointerInfo::getJumpTable(),
false, false, false, 0);
Chain = Addr.getValue(1);
Addr = DAG.getNode(ISD::ADD, dl, PTy, Addr, Table);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI);
} else {
Addr = DAG.getLoad(PTy, dl, Chain, Addr,
MachinePointerInfo::getJumpTable(),
false, false, false, 0);
Chain = Addr.getValue(1);
return DAG.getNode(ARMISD::BR_JT, dl, MVT::Other, Chain, Addr, JTI);
}
}
static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getValueType();
SDLoc dl(Op);
if (Op.getValueType().getVectorElementType() == MVT::i32) {
if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::f32)
return Op;
return DAG.UnrollVectorOp(Op.getNode());
}
assert(Op.getOperand(0).getValueType() == MVT::v4f32 &&
"Invalid type for custom lowering!");
if (VT != MVT::v4i16)
return DAG.UnrollVectorOp(Op.getNode());
Op = DAG.getNode(Op.getOpcode(), dl, MVT::v4i32, Op.getOperand(0));
return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
}
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,
false, SDLoc(Op)).first;
}
SDLoc dl(Op);
unsigned Opc;
switch (Op.getOpcode()) {
default: llvm_unreachable("Invalid opcode!");
case ISD::FP_TO_SINT:
Opc = ARMISD::FTOSI;
break;
case ISD::FP_TO_UINT:
Opc = ARMISD::FTOUI;
break;
}
Op = DAG.getNode(Opc, dl, MVT::f32, Op.getOperand(0));
return DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
}
static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getValueType();
SDLoc dl(Op);
if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::i32) {
if (VT.getVectorElementType() == MVT::f32)
return Op;
return DAG.UnrollVectorOp(Op.getNode());
}
assert(Op.getOperand(0).getValueType() == MVT::v4i16 &&
"Invalid type for custom lowering!");
if (VT != MVT::v4f32)
return DAG.UnrollVectorOp(Op.getNode());
unsigned CastOpc;
unsigned Opc;
switch (Op.getOpcode()) {
default: llvm_unreachable("Invalid opcode!");
case ISD::SINT_TO_FP:
CastOpc = ISD::SIGN_EXTEND;
Opc = ISD::SINT_TO_FP;
break;
case ISD::UINT_TO_FP:
CastOpc = ISD::ZERO_EXTEND;
Opc = ISD::UINT_TO_FP;
break;
}
Op = DAG.getNode(CastOpc, dl, MVT::v4i32, Op.getOperand(0));
return DAG.getNode(Opc, dl, VT, Op);
}
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,
false, SDLoc(Op)).first;
}
SDLoc dl(Op);
unsigned Opc;
switch (Op.getOpcode()) {
default: llvm_unreachable("Invalid opcode!");
case ISD::SINT_TO_FP:
Opc = ARMISD::SITOF;
break;
case ISD::UINT_TO_FP:
Opc = ARMISD::UITOF;
break;
}
Op = DAG.getNode(ISD::BITCAST, dl, MVT::f32, Op.getOperand(0));
return DAG.getNode(Opc, dl, VT, Op);
}
SDValue ARMTargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
SDValue Tmp0 = Op.getOperand(0);
SDValue Tmp1 = Op.getOperand(1);
SDLoc dl(Op);
EVT VT = Op.getValueType();
EVT SrcVT = Tmp1.getValueType();
bool InGPR = Tmp0.getOpcode() == ISD::BITCAST ||
Tmp0.getOpcode() == ARMISD::VMOVDRR;
bool UseNEON = !InGPR && Subtarget->hasNEON();
if (UseNEON) {
unsigned EncodedVal = ARM_AM::createNEONModImm(0x6, 0x80);
SDValue Mask = DAG.getNode(ARMISD::VMOVIMM, dl, MVT::v2i32,
DAG.getTargetConstant(EncodedVal, MVT::i32));
EVT OpVT = (VT == MVT::f32) ? MVT::v2i32 : MVT::v1i64;
if (VT == MVT::f64)
Mask = DAG.getNode(ARMISD::VSHL, dl, OpVT,
DAG.getNode(ISD::BITCAST, dl, OpVT, Mask),
DAG.getConstant(32, MVT::i32));
else
Tmp0 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2f32, Tmp0);
if (SrcVT == MVT::f32) {
Tmp1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2f32, Tmp1);
if (VT == MVT::f64)
Tmp1 = DAG.getNode(ARMISD::VSHL, dl, OpVT,
DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp1),
DAG.getConstant(32, MVT::i32));
} else if (VT == MVT::f32)
Tmp1 = DAG.getNode(ARMISD::VSHRu, dl, MVT::v1i64,
DAG.getNode(ISD::BITCAST, dl, MVT::v1i64, Tmp1),
DAG.getConstant(32, MVT::i32));
Tmp0 = DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp0);
Tmp1 = DAG.getNode(ISD::BITCAST, dl, OpVT, Tmp1);
SDValue AllOnes = DAG.getTargetConstant(ARM_AM::createNEONModImm(0xe, 0xff),
MVT::i32);
AllOnes = DAG.getNode(ARMISD::VMOVIMM, dl, MVT::v8i8, AllOnes);
SDValue MaskNot = DAG.getNode(ISD::XOR, dl, OpVT, Mask,
DAG.getNode(ISD::BITCAST, dl, OpVT, AllOnes));
SDValue Res = DAG.getNode(ISD::OR, dl, OpVT,
DAG.getNode(ISD::AND, dl, OpVT, Tmp1, Mask),
DAG.getNode(ISD::AND, dl, OpVT, Tmp0, MaskNot));
if (VT == MVT::f32) {
Res = DAG.getNode(ISD::BITCAST, dl, MVT::v2f32, Res);
Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f32, Res,
DAG.getConstant(0, MVT::i32));
} else {
Res = DAG.getNode(ISD::BITCAST, dl, MVT::f64, Res);
}
return Res;
}
if (SrcVT == MVT::f64)
Tmp1 = DAG.getNode(ARMISD::VMOVRRD, dl, DAG.getVTList(MVT::i32, MVT::i32),
Tmp1).getValue(1);
Tmp1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Tmp1);
SDValue Mask1 = DAG.getConstant(0x80000000, MVT::i32);
SDValue Mask2 = DAG.getConstant(0x7fffffff, MVT::i32);
Tmp1 = DAG.getNode(ISD::AND, dl, MVT::i32, Tmp1, Mask1);
if (VT == MVT::f32) {
Tmp0 = DAG.getNode(ISD::AND, dl, MVT::i32,
DAG.getNode(ISD::BITCAST, dl, MVT::i32, Tmp0), Mask2);
return DAG.getNode(ISD::BITCAST, dl, MVT::f32,
DAG.getNode(ISD::OR, dl, MVT::i32, Tmp0, Tmp1));
}
Tmp0 = DAG.getNode(ARMISD::VMOVRRD, dl, DAG.getVTList(MVT::i32, MVT::i32),
Tmp0);
SDValue Lo = Tmp0.getValue(0);
SDValue Hi = DAG.getNode(ISD::AND, dl, MVT::i32, Tmp0.getValue(1), Mask2);
Hi = DAG.getNode(ISD::OR, dl, MVT::i32, Hi, Tmp1);
return DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi);
}
SDValue ARMTargetLowering::LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const{
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setReturnAddressIsTaken(true);
if (verifyReturnAddressArgumentIsConstant(Op, DAG))
return SDValue();
EVT VT = Op.getValueType();
SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
if (Depth) {
SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
SDValue Offset = DAG.getConstant(4, MVT::i32);
return DAG.getLoad(VT, dl, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, dl, VT, FrameAddr, Offset),
MachinePointerInfo(), false, false, false, 0);
}
unsigned Reg = MF.addLiveIn(ARM::LR, getRegClassFor(MVT::i32));
return DAG.getCopyFromReg(DAG.getEntryNode(), dl, Reg, VT);
}
SDValue ARMTargetLowering::LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
const ARMBaseRegisterInfo &ARI =
*static_cast<const ARMBaseRegisterInfo*>(RegInfo);
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MFI->setFrameAddressIsTaken(true);
EVT VT = Op.getValueType();
SDLoc dl(Op); unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
unsigned FrameReg = ARI.getFrameRegister(MF);
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg, VT);
while (Depth--)
FrameAddr = DAG.getLoad(VT, dl, DAG.getEntryNode(), FrameAddr,
MachinePointerInfo(),
false, false, false, 0);
return FrameAddr;
}
unsigned ARMTargetLowering::getRegisterByName(const char* RegName,
EVT VT) const {
unsigned Reg = StringSwitch<unsigned>(RegName)
.Case("sp", ARM::SP)
.Default(0);
if (Reg)
return Reg;
report_fatal_error("Invalid register name global variable");
}
static SDValue ExpandBITCAST(SDNode *N, SelectionDAG &DAG) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDLoc dl(N);
SDValue Op = N->getOperand(0);
EVT SrcVT = Op.getValueType();
EVT DstVT = N->getValueType(0);
assert((SrcVT == MVT::i64 || DstVT == MVT::i64) &&
"ExpandBITCAST called for non-i64 type");
if (SrcVT == MVT::i64 && TLI.isTypeLegal(DstVT)) {
SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, Op,
DAG.getConstant(0, MVT::i32));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, Op,
DAG.getConstant(1, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, DstVT,
DAG.getNode(ARMISD::VMOVDRR, dl, MVT::f64, Lo, Hi));
}
if (DstVT == MVT::i64 && TLI.isTypeLegal(SrcVT)) {
SDValue Cvt;
if (TLI.isBigEndian() && SrcVT.isVector() &&
SrcVT.getVectorNumElements() > 1)
Cvt = DAG.getNode(ARMISD::VMOVRRD, dl,
DAG.getVTList(MVT::i32, MVT::i32),
DAG.getNode(ARMISD::VREV64, dl, SrcVT, Op));
else
Cvt = DAG.getNode(ARMISD::VMOVRRD, dl,
DAG.getVTList(MVT::i32, MVT::i32), Op);
return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Cvt, Cvt.getValue(1));
}
return SDValue();
}
static SDValue getZeroVector(EVT VT, SelectionDAG &DAG, SDLoc dl) {
assert(VT.isVector() && "Expected a vector type");
SDValue EncodedVal = DAG.getTargetConstant(0, MVT::i32);
EVT VmovVT = VT.is128BitVector() ? MVT::v4i32 : MVT::v2i32;
SDValue Vmov = DAG.getNode(ARMISD::VMOVIMM, dl, VmovVT, EncodedVal);
return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
}
SDValue ARMTargetLowering::LowerShiftRightParts(SDValue Op,
SelectionDAG &DAG) const {
assert(Op.getNumOperands() == 3 && "Not a double-shift!");
EVT VT = Op.getValueType();
unsigned VTBits = VT.getSizeInBits();
SDLoc dl(Op);
SDValue ShOpLo = Op.getOperand(0);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
SDValue ARMcc;
unsigned Opc = (Op.getOpcode() == ISD::SRA_PARTS) ? ISD::SRA : ISD::SRL;
assert(Op.getOpcode() == ISD::SRA_PARTS || Op.getOpcode() == ISD::SRL_PARTS);
SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
DAG.getConstant(VTBits, MVT::i32), ShAmt);
SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, ShAmt);
SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
DAG.getConstant(VTBits, MVT::i32));
SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, RevShAmt);
SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
SDValue TrueVal = DAG.getNode(Opc, dl, VT, ShOpHi, ExtraShAmt);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Cmp = getARMCmp(ExtraShAmt, DAG.getConstant(0, MVT::i32), ISD::SETGE,
ARMcc, DAG, dl);
SDValue Hi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
SDValue Lo = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, TrueVal, ARMcc,
CCR, Cmp);
SDValue Ops[2] = { Lo, Hi };
return DAG.getMergeValues(Ops, dl);
}
SDValue ARMTargetLowering::LowerShiftLeftParts(SDValue Op,
SelectionDAG &DAG) const {
assert(Op.getNumOperands() == 3 && "Not a double-shift!");
EVT VT = Op.getValueType();
unsigned VTBits = VT.getSizeInBits();
SDLoc dl(Op);
SDValue ShOpLo = Op.getOperand(0);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
SDValue ARMcc;
assert(Op.getOpcode() == ISD::SHL_PARTS);
SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
DAG.getConstant(VTBits, MVT::i32), ShAmt);
SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, RevShAmt);
SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
DAG.getConstant(VTBits, MVT::i32));
SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, ShAmt);
SDValue Tmp3 = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ExtraShAmt);
SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
SDValue CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDValue Cmp = getARMCmp(ExtraShAmt, DAG.getConstant(0, MVT::i32), ISD::SETGE,
ARMcc, DAG, dl);
SDValue Lo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
SDValue Hi = DAG.getNode(ARMISD::CMOV, dl, VT, FalseVal, Tmp3, ARMcc,
CCR, Cmp);
SDValue Ops[2] = { Lo, Hi };
return DAG.getMergeValues(Ops, dl);
}
SDValue ARMTargetLowering::LowerFLT_ROUNDS_(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
SDValue FPSCR = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::i32,
DAG.getConstant(Intrinsic::arm_get_fpscr,
MVT::i32));
SDValue FltRounds = DAG.getNode(ISD::ADD, dl, MVT::i32, FPSCR,
DAG.getConstant(1U << 22, MVT::i32));
SDValue RMODE = DAG.getNode(ISD::SRL, dl, MVT::i32, FltRounds,
DAG.getConstant(22, MVT::i32));
return DAG.getNode(ISD::AND, dl, MVT::i32, RMODE,
DAG.getConstant(3, MVT::i32));
}
static SDValue LowerCTTZ(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
SDLoc dl(N);
if (!ST->hasV6T2Ops())
return SDValue();
SDValue rbit = DAG.getNode(ARMISD::RBIT, dl, VT, N->getOperand(0));
return DAG.getNode(ISD::CTLZ, dl, VT, rbit);
}
static SDValue getCTPOP16BitCounts(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
SDLoc DL(N);
EVT VT8Bit = VT.is64BitVector() ? MVT::v8i8 : MVT::v16i8;
SDValue N0 = DAG.getNode(ISD::BITCAST, DL, VT8Bit, N->getOperand(0));
SDValue N1 = DAG.getNode(ISD::CTPOP, DL, VT8Bit, N0);
SDValue N2 = DAG.getNode(ARMISD::VREV16, DL, VT8Bit, N1);
SDValue N3 = DAG.getNode(ISD::ADD, DL, VT8Bit, N1, N2);
return DAG.getNode(ARMISD::VUZP, DL, VT8Bit, N3, N3);
}
static SDValue lowerCTPOP16BitElements(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
SDLoc DL(N);
SDValue BitCounts = getCTPOP16BitCounts(N, DAG);
if (VT.is64BitVector()) {
SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, BitCounts);
return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, Extended,
DAG.getIntPtrConstant(0));
} else {
SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v8i8,
BitCounts, DAG.getIntPtrConstant(0));
return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v8i16, Extracted);
}
}
static SDValue lowerCTPOP32BitElements(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
SDLoc DL(N);
EVT VT16Bit = VT.is64BitVector() ? MVT::v4i16 : MVT::v8i16;
SDValue Bitcast = DAG.getNode(ISD::BITCAST, DL, VT16Bit, N->getOperand(0));
SDValue Counts16 = lowerCTPOP16BitElements(Bitcast.getNode(), DAG);
SDValue N0 = DAG.getNode(ARMISD::VREV32, DL, VT16Bit, Counts16);
SDValue N1 = DAG.getNode(ISD::ADD, DL, VT16Bit, Counts16, N0);
SDValue N2 = DAG.getNode(ARMISD::VUZP, DL, VT16Bit, N1, N1);
if (VT.is64BitVector()) {
SDValue Extended = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, N2);
return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v2i32, Extended,
DAG.getIntPtrConstant(0));
} else {
SDValue Extracted = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, MVT::v4i16, N2,
DAG.getIntPtrConstant(0));
return DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::v4i32, Extracted);
}
}
static SDValue LowerCTPOP(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
assert(ST->hasNEON() && "Custom ctpop lowering requires NEON.");
assert((VT == MVT::v2i32 || VT == MVT::v4i32 ||
VT == MVT::v4i16 || VT == MVT::v8i16) &&
"Unexpected type for custom ctpop lowering");
if (VT.getVectorElementType() == MVT::i32)
return lowerCTPOP32BitElements(N, DAG);
else
return lowerCTPOP16BitElements(N, DAG);
}
static SDValue LowerShift(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
SDLoc dl(N);
if (!VT.isVector())
return SDValue();
assert(ST->hasNEON() && "unexpected vector shift");
if (N->getOpcode() == ISD::SHL)
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
DAG.getConstant(Intrinsic::arm_neon_vshiftu, MVT::i32),
N->getOperand(0), N->getOperand(1));
assert((N->getOpcode() == ISD::SRA ||
N->getOpcode() == ISD::SRL) && "unexpected vector shift opcode");
EVT ShiftVT = N->getOperand(1).getValueType();
SDValue NegatedCount = DAG.getNode(ISD::SUB, dl, ShiftVT,
getZeroVector(ShiftVT, DAG, dl),
N->getOperand(1));
Intrinsic::ID vshiftInt = (N->getOpcode() == ISD::SRA ?
Intrinsic::arm_neon_vshifts :
Intrinsic::arm_neon_vshiftu);
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
DAG.getConstant(vshiftInt, MVT::i32),
N->getOperand(0), NegatedCount);
}
static SDValue Expand64BitShift(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
SDLoc dl(N);
if (VT != MVT::i64)
return SDValue();
assert((N->getOpcode() == ISD::SRL || N->getOpcode() == ISD::SRA) &&
"Unknown shift to lower!");
if (!isa<ConstantSDNode>(N->getOperand(1)) ||
cast<ConstantSDNode>(N->getOperand(1))->getZExtValue() != 1)
return SDValue();
if (ST->isThumb1Only()) return SDValue();
SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
DAG.getConstant(0, MVT::i32));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32, N->getOperand(0),
DAG.getConstant(1, MVT::i32));
unsigned Opc = N->getOpcode() == ISD::SRL ? ARMISD::SRL_FLAG:ARMISD::SRA_FLAG;
Hi = DAG.getNode(Opc, dl, DAG.getVTList(MVT::i32, MVT::Glue), Hi);
Lo = DAG.getNode(ARMISD::RRX, dl, MVT::i32, Lo, Hi.getValue(1));
return DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Lo, Hi);
}
static SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) {
SDValue TmpOp0, TmpOp1;
bool Invert = false;
bool Swap = false;
unsigned Opc = 0;
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue CC = Op.getOperand(2);
EVT CmpVT = Op0.getValueType().changeVectorElementTypeToInteger();
EVT VT = Op.getValueType();
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
SDLoc dl(Op);
if (Op1.getValueType().isFloatingPoint()) {
switch (SetCCOpcode) {
default: llvm_unreachable("Illegal FP comparison");
case ISD::SETUNE:
case ISD::SETNE: Invert = true; case ISD::SETOEQ:
case ISD::SETEQ: Opc = ARMISD::VCEQ; break;
case ISD::SETOLT:
case ISD::SETLT: Swap = true; case ISD::SETOGT:
case ISD::SETGT: Opc = ARMISD::VCGT; break;
case ISD::SETOLE:
case ISD::SETLE: Swap = true; case ISD::SETOGE:
case ISD::SETGE: Opc = ARMISD::VCGE; break;
case ISD::SETUGE: Swap = true; case ISD::SETULE: Invert = true; Opc = ARMISD::VCGT; break;
case ISD::SETUGT: Swap = true; case ISD::SETULT: Invert = true; Opc = ARMISD::VCGE; break;
case ISD::SETUEQ: Invert = true; case ISD::SETONE:
TmpOp0 = Op0;
TmpOp1 = Op1;
Opc = ISD::OR;
Op0 = DAG.getNode(ARMISD::VCGT, dl, CmpVT, TmpOp1, TmpOp0);
Op1 = DAG.getNode(ARMISD::VCGT, dl, CmpVT, TmpOp0, TmpOp1);
break;
case ISD::SETUO: Invert = true; case ISD::SETO:
TmpOp0 = Op0;
TmpOp1 = Op1;
Opc = ISD::OR;
Op0 = DAG.getNode(ARMISD::VCGT, dl, CmpVT, TmpOp1, TmpOp0);
Op1 = DAG.getNode(ARMISD::VCGE, dl, CmpVT, TmpOp0, TmpOp1);
break;
}
} else {
switch (SetCCOpcode) {
default: llvm_unreachable("Illegal integer comparison");
case ISD::SETNE: Invert = true;
case ISD::SETEQ: Opc = ARMISD::VCEQ; break;
case ISD::SETLT: Swap = true;
case ISD::SETGT: Opc = ARMISD::VCGT; break;
case ISD::SETLE: Swap = true;
case ISD::SETGE: Opc = ARMISD::VCGE; break;
case ISD::SETULT: Swap = true;
case ISD::SETUGT: Opc = ARMISD::VCGTU; break;
case ISD::SETULE: Swap = true;
case ISD::SETUGE: Opc = ARMISD::VCGEU; break;
}
if (Opc == ARMISD::VCEQ) {
SDValue AndOp;
if (ISD::isBuildVectorAllZeros(Op1.getNode()))
AndOp = Op0;
else if (ISD::isBuildVectorAllZeros(Op0.getNode()))
AndOp = Op1;
if (AndOp.getNode() && AndOp.getOpcode() == ISD::BITCAST)
AndOp = AndOp.getOperand(0);
if (AndOp.getNode() && AndOp.getOpcode() == ISD::AND) {
Opc = ARMISD::VTST;
Op0 = DAG.getNode(ISD::BITCAST, dl, CmpVT, AndOp.getOperand(0));
Op1 = DAG.getNode(ISD::BITCAST, dl, CmpVT, AndOp.getOperand(1));
Invert = !Invert;
}
}
}
if (Swap)
std::swap(Op0, Op1);
SDValue SingleOp;
if (ISD::isBuildVectorAllZeros(Op1.getNode()))
SingleOp = Op0;
else if (ISD::isBuildVectorAllZeros(Op0.getNode())) {
if (Opc == ARMISD::VCGE)
Opc = ARMISD::VCLEZ;
else if (Opc == ARMISD::VCGT)
Opc = ARMISD::VCLTZ;
SingleOp = Op1;
}
SDValue Result;
if (SingleOp.getNode()) {
switch (Opc) {
case ARMISD::VCEQ:
Result = DAG.getNode(ARMISD::VCEQZ, dl, CmpVT, SingleOp); break;
case ARMISD::VCGE:
Result = DAG.getNode(ARMISD::VCGEZ, dl, CmpVT, SingleOp); break;
case ARMISD::VCLEZ:
Result = DAG.getNode(ARMISD::VCLEZ, dl, CmpVT, SingleOp); break;
case ARMISD::VCGT:
Result = DAG.getNode(ARMISD::VCGTZ, dl, CmpVT, SingleOp); break;
case ARMISD::VCLTZ:
Result = DAG.getNode(ARMISD::VCLTZ, dl, CmpVT, SingleOp); break;
default:
Result = DAG.getNode(Opc, dl, CmpVT, Op0, Op1);
}
} else {
Result = DAG.getNode(Opc, dl, CmpVT, Op0, Op1);
}
Result = DAG.getSExtOrTrunc(Result, dl, VT);
if (Invert)
Result = DAG.getNOT(dl, Result, VT);
return Result;
}
static SDValue isNEONModifiedImm(uint64_t SplatBits, uint64_t SplatUndef,
unsigned SplatBitSize, SelectionDAG &DAG,
EVT &VT, bool is128Bits, NEONModImmType type) {
unsigned OpCmode, Imm;
if (SplatBits == 0)
SplatBitSize = 32;
switch (SplatBitSize) {
case 8:
if (type != VMOVModImm)
return SDValue();
assert((SplatBits & ~0xff) == 0 && "one byte splat value is too big");
OpCmode = 0xe;
Imm = SplatBits;
VT = is128Bits ? MVT::v16i8 : MVT::v8i8;
break;
case 16:
VT = is128Bits ? MVT::v8i16 : MVT::v4i16;
if ((SplatBits & ~0xff) == 0) {
OpCmode = 0x8;
Imm = SplatBits;
break;
}
if ((SplatBits & ~0xff00) == 0) {
OpCmode = 0xa;
Imm = SplatBits >> 8;
break;
}
return SDValue();
case 32:
VT = is128Bits ? MVT::v4i32 : MVT::v2i32;
if ((SplatBits & ~0xff) == 0) {
OpCmode = 0;
Imm = SplatBits;
break;
}
if ((SplatBits & ~0xff00) == 0) {
OpCmode = 0x2;
Imm = SplatBits >> 8;
break;
}
if ((SplatBits & ~0xff0000) == 0) {
OpCmode = 0x4;
Imm = SplatBits >> 16;
break;
}
if ((SplatBits & ~0xff000000) == 0) {
OpCmode = 0x6;
Imm = SplatBits >> 24;
break;
}
if (type == OtherModImm) return SDValue();
if ((SplatBits & ~0xffff) == 0 &&
((SplatBits | SplatUndef) & 0xff) == 0xff) {
OpCmode = 0xc;
Imm = SplatBits >> 8;
break;
}
if ((SplatBits & ~0xffffff) == 0 &&
((SplatBits | SplatUndef) & 0xffff) == 0xffff) {
OpCmode = 0xd;
Imm = SplatBits >> 16;
break;
}
return SDValue();
case 64: {
if (type != VMOVModImm)
return SDValue();
uint64_t BitMask = 0xff;
uint64_t Val = 0;
unsigned ImmMask = 1;
Imm = 0;
for (int ByteNum = 0; ByteNum < 8; ++ByteNum) {
if (((SplatBits | SplatUndef) & BitMask) == BitMask) {
Val |= BitMask;
Imm |= ImmMask;
} else if ((SplatBits & BitMask) != 0) {
return SDValue();
}
BitMask <<= 8;
ImmMask <<= 1;
}
if (DAG.getTargetLoweringInfo().isBigEndian())
Imm = ((Imm & 0xf) << 4) | ((Imm & 0xf0) >> 4);
OpCmode = 0x1e;
VT = is128Bits ? MVT::v2i64 : MVT::v1i64;
break;
}
default:
llvm_unreachable("unexpected size for isNEONModifiedImm");
}
unsigned EncodedVal = ARM_AM::createNEONModImm(OpCmode, Imm);
return DAG.getTargetConstant(EncodedVal, MVT::i32);
}
SDValue ARMTargetLowering::LowerConstantFP(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *ST) const {
if (!ST->hasVFP3())
return SDValue();
bool IsDouble = Op.getValueType() == MVT::f64;
ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Op);
if (IsDouble && Subtarget->isFPOnlySP())
return SDValue();
APFloat FPVal = CFP->getValueAPF();
int ImmVal = IsDouble ? ARM_AM::getFP64Imm(FPVal) : ARM_AM::getFP32Imm(FPVal);
if (ImmVal != -1) {
if (IsDouble || !ST->useNEONForSinglePrecisionFP()) {
return Op;
}
SDLoc DL(Op);
SDValue NewVal = DAG.getTargetConstant(ImmVal, MVT::i32);
SDValue VecConstant = DAG.getNode(ARMISD::VMOVFPIMM, DL, MVT::v2f32,
NewVal);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecConstant,
DAG.getConstant(0, MVT::i32));
}
if (!ST->hasNEON() || (!IsDouble && !ST->useNEONForSinglePrecisionFP()))
return SDValue();
EVT VMovVT;
uint64_t iVal = FPVal.bitcastToAPInt().getZExtValue();
if (IsDouble && (iVal & 0xffffffff) != (iVal >> 32))
return SDValue();
SDValue NewVal = isNEONModifiedImm(iVal & 0xffffffffU, 0, 32, DAG, VMovVT,
false, VMOVModImm);
if (NewVal != SDValue()) {
SDLoc DL(Op);
SDValue VecConstant = DAG.getNode(ARMISD::VMOVIMM, DL, VMovVT,
NewVal);
if (IsDouble)
return DAG.getNode(ISD::BITCAST, DL, MVT::f64, VecConstant);
SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
VecConstant);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
DAG.getConstant(0, MVT::i32));
}
NewVal = isNEONModifiedImm(~iVal & 0xffffffffU, 0, 32, DAG, VMovVT,
false, VMVNModImm);
if (NewVal != SDValue()) {
SDLoc DL(Op);
SDValue VecConstant = DAG.getNode(ARMISD::VMVNIMM, DL, VMovVT, NewVal);
if (IsDouble)
return DAG.getNode(ISD::BITCAST, DL, MVT::f64, VecConstant);
SDValue VecFConstant = DAG.getNode(ISD::BITCAST, DL, MVT::v2f32,
VecConstant);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32, VecFConstant,
DAG.getConstant(0, MVT::i32));
}
return SDValue();
}
static bool isSingletonVEXTMask(ArrayRef<int> M, EVT VT, unsigned &Imm) {
unsigned NumElts = VT.getVectorNumElements();
if (M[0] < 0)
return false;
Imm = M[0];
unsigned ExpectedElt = Imm;
for (unsigned i = 1; i < NumElts; ++i) {
++ExpectedElt;
if (ExpectedElt == NumElts)
ExpectedElt = 0;
if (M[i] < 0) continue; if (ExpectedElt != static_cast<unsigned>(M[i]))
return false;
}
return true;
}
static bool isVEXTMask(ArrayRef<int> M, EVT VT,
bool &ReverseVEXT, unsigned &Imm) {
unsigned NumElts = VT.getVectorNumElements();
ReverseVEXT = false;
if (M[0] < 0)
return false;
Imm = M[0];
unsigned ExpectedElt = Imm;
for (unsigned i = 1; i < NumElts; ++i) {
ExpectedElt += 1;
if (ExpectedElt == NumElts * 2) {
ExpectedElt = 0;
ReverseVEXT = true;
}
if (M[i] < 0) continue; if (ExpectedElt != static_cast<unsigned>(M[i]))
return false;
}
if (ReverseVEXT)
Imm -= NumElts;
return true;
}
static bool isVREVMask(ArrayRef<int> M, EVT VT, unsigned BlockSize) {
assert((BlockSize==16 || BlockSize==32 || BlockSize==64) &&
"Only possible block sizes for VREV are: 16, 32, 64");
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
unsigned BlockElts = M[0] + 1;
if (M[0] < 0)
BlockElts = BlockSize / EltSz;
if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz)
return false;
for (unsigned i = 0; i < NumElts; ++i) {
if (M[i] < 0) continue; if ((unsigned) M[i] != (i - i%BlockElts) + (BlockElts - 1 - i%BlockElts))
return false;
}
return true;
}
static bool isVTBLMask(ArrayRef<int> M, EVT VT) {
return VT == MVT::v8i8 && M.size() == 8;
}
static bool isVTRNMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
WhichResult = (M[0] == 0 ? 0 : 1);
for (unsigned i = 0; i < NumElts; i += 2) {
if ((M[i] >= 0 && (unsigned) M[i] != i + WhichResult) ||
(M[i+1] >= 0 && (unsigned) M[i+1] != i + NumElts + WhichResult))
return false;
}
return true;
}
static bool isVTRN_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
WhichResult = (M[0] == 0 ? 0 : 1);
for (unsigned i = 0; i < NumElts; i += 2) {
if ((M[i] >= 0 && (unsigned) M[i] != i + WhichResult) ||
(M[i+1] >= 0 && (unsigned) M[i+1] != i + WhichResult))
return false;
}
return true;
}
static bool isVUZPMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
WhichResult = (M[0] == 0 ? 0 : 1);
for (unsigned i = 0; i != NumElts; ++i) {
if (M[i] < 0) continue; if ((unsigned) M[i] != 2 * i + WhichResult)
return false;
}
if (VT.is64BitVector() && EltSz == 32)
return false;
return true;
}
static bool isVUZP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned Half = VT.getVectorNumElements() / 2;
WhichResult = (M[0] == 0 ? 0 : 1);
for (unsigned j = 0; j != 2; ++j) {
unsigned Idx = WhichResult;
for (unsigned i = 0; i != Half; ++i) {
int MIdx = M[i + j * Half];
if (MIdx >= 0 && (unsigned) MIdx != Idx)
return false;
Idx += 2;
}
}
if (VT.is64BitVector() && EltSz == 32)
return false;
return true;
}
static bool isVZIPMask(ArrayRef<int> M, EVT VT, unsigned &WhichResult) {
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
WhichResult = (M[0] == 0 ? 0 : 1);
unsigned Idx = WhichResult * NumElts / 2;
for (unsigned i = 0; i != NumElts; i += 2) {
if ((M[i] >= 0 && (unsigned) M[i] != Idx) ||
(M[i+1] >= 0 && (unsigned) M[i+1] != Idx + NumElts))
return false;
Idx += 1;
}
if (VT.is64BitVector() && EltSz == 32)
return false;
return true;
}
static bool isVZIP_v_undef_Mask(ArrayRef<int> M, EVT VT, unsigned &WhichResult){
unsigned EltSz = VT.getVectorElementType().getSizeInBits();
if (EltSz == 64)
return false;
unsigned NumElts = VT.getVectorNumElements();
WhichResult = (M[0] == 0 ? 0 : 1);
unsigned Idx = WhichResult * NumElts / 2;
for (unsigned i = 0; i != NumElts; i += 2) {
if ((M[i] >= 0 && (unsigned) M[i] != Idx) ||
(M[i+1] >= 0 && (unsigned) M[i+1] != Idx))
return false;
Idx += 1;
}
if (VT.is64BitVector() && EltSz == 32)
return false;
return true;
}
static bool isReverseMask(ArrayRef<int> M, EVT VT) {
unsigned NumElts = VT.getVectorNumElements();
if (NumElts != M.size())
return false;
for (unsigned i = 0; i != NumElts; ++i)
if (M[i] >= 0 && M[i] != (int) (NumElts - 1 - i))
return false;
return true;
}
static SDValue IsSingleInstrConstant(SDValue N, SelectionDAG &DAG,
const ARMSubtarget *ST, SDLoc dl) {
uint64_t Val;
if (!isa<ConstantSDNode>(N))
return SDValue();
Val = cast<ConstantSDNode>(N)->getZExtValue();
if (ST->isThumb1Only()) {
if (Val <= 255 || ~Val <= 255)
return DAG.getConstant(Val, MVT::i32);
} else {
if (ARM_AM::getSOImmVal(Val) != -1 || ARM_AM::getSOImmVal(~Val) != -1)
return DAG.getConstant(Val, MVT::i32);
}
return SDValue();
}
SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *ST) const {
BuildVectorSDNode *BVN = cast<BuildVectorSDNode>(Op.getNode());
SDLoc dl(Op);
EVT VT = Op.getValueType();
APInt SplatBits, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
if (SplatBitSize <= 64) {
EVT VmovVT;
SDValue Val = isNEONModifiedImm(SplatBits.getZExtValue(),
SplatUndef.getZExtValue(), SplatBitSize,
DAG, VmovVT, VT.is128BitVector(),
VMOVModImm);
if (Val.getNode()) {
SDValue Vmov = DAG.getNode(ARMISD::VMOVIMM, dl, VmovVT, Val);
return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
}
uint64_t NegatedImm = (~SplatBits).getZExtValue();
Val = isNEONModifiedImm(NegatedImm,
SplatUndef.getZExtValue(), SplatBitSize,
DAG, VmovVT, VT.is128BitVector(),
VMVNModImm);
if (Val.getNode()) {
SDValue Vmov = DAG.getNode(ARMISD::VMVNIMM, dl, VmovVT, Val);
return DAG.getNode(ISD::BITCAST, dl, VT, Vmov);
}
if ((VT == MVT::v2f32 || VT == MVT::v4f32) && SplatBitSize == 32) {
int ImmVal = ARM_AM::getFP32Imm(SplatBits);
if (ImmVal != -1) {
SDValue Val = DAG.getTargetConstant(ImmVal, MVT::i32);
return DAG.getNode(ARMISD::VMOVFPIMM, dl, VT, Val);
}
}
}
}
unsigned NumElts = VT.getVectorNumElements();
bool isOnlyLowElement = true;
bool usesOnlyOneValue = true;
bool hasDominantValue = false;
bool isConstant = true;
DenseMap<SDValue, unsigned> ValueCounts;
SDValue Value;
for (unsigned i = 0; i < NumElts; ++i) {
SDValue V = Op.getOperand(i);
if (V.getOpcode() == ISD::UNDEF)
continue;
if (i > 0)
isOnlyLowElement = false;
if (!isa<ConstantFPSDNode>(V) && !isa<ConstantSDNode>(V))
isConstant = false;
ValueCounts.insert(std::make_pair(V, 0));
unsigned &Count = ValueCounts[V];
if (++Count > (NumElts / 2)) {
hasDominantValue = true;
Value = V;
}
}
if (ValueCounts.size() != 1)
usesOnlyOneValue = false;
if (!Value.getNode() && ValueCounts.size() > 0)
Value = ValueCounts.begin()->first;
if (ValueCounts.size() == 0)
return DAG.getUNDEF(VT);
if (isOnlyLowElement && !ISD::isNormalLoad(Value.getNode()))
return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value);
unsigned EltSize = VT.getVectorElementType().getSizeInBits();
if (hasDominantValue && EltSize <= 32) {
if (!isConstant) {
SDValue N;
if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
isa<ConstantSDNode>(Value->getOperand(1))) {
if (VT != Value->getOperand(0).getValueType()) {
ConstantSDNode *constIndex;
constIndex = dyn_cast<ConstantSDNode>(Value->getOperand(1));
assert(constIndex && "The index is not a constant!");
unsigned index = constIndex->getAPIntValue().getLimitedValue() %
VT.getVectorNumElements();
N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DAG.getUNDEF(VT),
Value, DAG.getConstant(index, MVT::i32)),
DAG.getConstant(index, MVT::i32));
} else
N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
Value->getOperand(0), Value->getOperand(1));
} else
N = DAG.getNode(ARMISD::VDUP, dl, VT, Value);
if (!usesOnlyOneValue) {
for (unsigned I = 0; I < NumElts; ++I) {
if (Op.getOperand(I) == Value)
continue;
SmallVector<SDValue, 3> Ops;
Ops.push_back(N);
Ops.push_back(Op.getOperand(I));
Ops.push_back(DAG.getConstant(I, MVT::i32));
N = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Ops);
}
}
return N;
}
if (VT.getVectorElementType().isFloatingPoint()) {
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0; i < NumElts; ++i)
Ops.push_back(DAG.getNode(ISD::BITCAST, dl, MVT::i32,
Op.getOperand(i)));
EVT VecVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumElts);
SDValue Val = DAG.getNode(ISD::BUILD_VECTOR, dl, VecVT, Ops);
Val = LowerBUILD_VECTOR(Val, DAG, ST);
if (Val.getNode())
return DAG.getNode(ISD::BITCAST, dl, VT, Val);
}
if (usesOnlyOneValue) {
SDValue Val = IsSingleInstrConstant(Value, DAG, ST, dl);
if (isConstant && Val.getNode())
return DAG.getNode(ARMISD::VDUP, dl, VT, Val);
}
}
if (isConstant)
return SDValue();
if (NumElts >= 4) {
SDValue shuffle = ReconstructShuffle(Op, DAG);
if (shuffle != SDValue())
return shuffle;
}
if (EltSize >= 32) {
EVT EltVT = EVT::getFloatingPointVT(EltSize);
EVT VecVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts);
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0; i < NumElts; ++i)
Ops.push_back(DAG.getNode(ISD::BITCAST, dl, EltVT, Op.getOperand(i)));
SDValue Val = DAG.getNode(ARMISD::BUILD_VECTOR, dl, VecVT, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, Val);
}
if (!isConstant && !usesOnlyOneValue) {
SDValue Vec = DAG.getUNDEF(VT);
for (unsigned i = 0 ; i < NumElts; ++i) {
SDValue V = Op.getOperand(i);
if (V.getOpcode() == ISD::UNDEF)
continue;
SDValue LaneIdx = DAG.getConstant(i, MVT::i32);
Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, Vec, V, LaneIdx);
}
return Vec;
}
return SDValue();
}
SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
EVT VT = Op.getValueType();
unsigned NumElts = VT.getVectorNumElements();
SmallVector<SDValue, 2> SourceVecs;
SmallVector<unsigned, 2> MinElts;
SmallVector<unsigned, 2> MaxElts;
for (unsigned i = 0; i < NumElts; ++i) {
SDValue V = Op.getOperand(i);
if (V.getOpcode() == ISD::UNDEF)
continue;
else if (V.getOpcode() != ISD::EXTRACT_VECTOR_ELT) {
return SDValue();
} else if (V.getOperand(0).getValueType().getVectorElementType() !=
VT.getVectorElementType()) {
return SDValue();
}
SDValue SourceVec = V.getOperand(0);
if (!isa<ConstantSDNode>(V.getOperand(1)))
return SDValue();
unsigned EltNo = cast<ConstantSDNode>(V.getOperand(1))->getZExtValue();
bool FoundSource = false;
for (unsigned j = 0; j < SourceVecs.size(); ++j) {
if (SourceVecs[j] == SourceVec) {
if (MinElts[j] > EltNo)
MinElts[j] = EltNo;
if (MaxElts[j] < EltNo)
MaxElts[j] = EltNo;
FoundSource = true;
break;
}
}
if (!FoundSource) {
SourceVecs.push_back(SourceVec);
MinElts.push_back(EltNo);
MaxElts.push_back(EltNo);
}
}
if (SourceVecs.size() > 2)
return SDValue();
SDValue ShuffleSrcs[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT) };
int VEXTOffsets[2] = {0, 0};
for (unsigned i = 0; i < SourceVecs.size(); ++i) {
if (SourceVecs[i].getValueType() == VT) {
ShuffleSrcs[i] = SourceVecs[i];
VEXTOffsets[i] = 0;
continue;
} else if (SourceVecs[i].getValueType().getVectorNumElements() < NumElts) {
return SDValue();
}
assert(SourceVecs[i].getValueType().getVectorNumElements() == 2*NumElts &&
"unexpected vector sizes in ReconstructShuffle");
if (MaxElts[i] - MinElts[i] >= NumElts) {
return SDValue();
}
if (MinElts[i] >= NumElts) {
VEXTOffsets[i] = NumElts;
ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
SourceVecs[i],
DAG.getIntPtrConstant(NumElts));
} else if (MaxElts[i] < NumElts) {
VEXTOffsets[i] = 0;
ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
SourceVecs[i],
DAG.getIntPtrConstant(0));
} else {
VEXTOffsets[i] = MinElts[i];
SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
SourceVecs[i],
DAG.getIntPtrConstant(0));
SDValue VEXTSrc2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT,
SourceVecs[i],
DAG.getIntPtrConstant(NumElts));
ShuffleSrcs[i] = DAG.getNode(ARMISD::VEXT, dl, VT, VEXTSrc1, VEXTSrc2,
DAG.getConstant(VEXTOffsets[i], MVT::i32));
}
}
SmallVector<int, 8> Mask;
for (unsigned i = 0; i < NumElts; ++i) {
SDValue Entry = Op.getOperand(i);
if (Entry.getOpcode() == ISD::UNDEF) {
Mask.push_back(-1);
continue;
}
SDValue ExtractVec = Entry.getOperand(0);
int ExtractElt = cast<ConstantSDNode>(Op.getOperand(i)
.getOperand(1))->getSExtValue();
if (ExtractVec == SourceVecs[0]) {
Mask.push_back(ExtractElt - VEXTOffsets[0]);
} else {
Mask.push_back(ExtractElt + NumElts - VEXTOffsets[1]);
}
}
if (isShuffleMaskLegal(Mask, VT))
return DAG.getVectorShuffle(VT, dl, ShuffleSrcs[0], ShuffleSrcs[1],
&Mask[0]);
return SDValue();
}
bool
ARMTargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &M,
EVT VT) const {
if (VT.getVectorNumElements() == 4 &&
(VT.is128BitVector() || VT.is64BitVector())) {
unsigned PFIndexes[4];
for (unsigned i = 0; i != 4; ++i) {
if (M[i] < 0)
PFIndexes[i] = 8;
else
PFIndexes[i] = M[i];
}
unsigned PFTableIndex =
PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
unsigned Cost = (PFEntry >> 30);
if (Cost <= 4)
return true;
}
bool ReverseVEXT;
unsigned Imm, WhichResult;
unsigned EltSize = VT.getVectorElementType().getSizeInBits();
return (EltSize >= 32 ||
ShuffleVectorSDNode::isSplatMask(&M[0], VT) ||
isVREVMask(M, VT, 64) ||
isVREVMask(M, VT, 32) ||
isVREVMask(M, VT, 16) ||
isVEXTMask(M, VT, ReverseVEXT, Imm) ||
isVTBLMask(M, VT) ||
isVTRNMask(M, VT, WhichResult) ||
isVUZPMask(M, VT, WhichResult) ||
isVZIPMask(M, VT, WhichResult) ||
isVTRN_v_undef_Mask(M, VT, WhichResult) ||
isVUZP_v_undef_Mask(M, VT, WhichResult) ||
isVZIP_v_undef_Mask(M, VT, WhichResult) ||
((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(M, VT)));
}
static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
SDValue RHS, SelectionDAG &DAG,
SDLoc dl) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
enum {
OP_COPY = 0, OP_VREV,
OP_VDUP0,
OP_VDUP1,
OP_VDUP2,
OP_VDUP3,
OP_VEXT1,
OP_VEXT2,
OP_VEXT3,
OP_VUZPL, OP_VUZPR, OP_VZIPL, OP_VZIPR, OP_VTRNL, OP_VTRNR };
if (OpNum == OP_COPY) {
if (LHSID == (1*9+2)*9+3) return LHS;
assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
return RHS;
}
SDValue OpLHS, OpRHS;
OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG, dl);
OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG, dl);
EVT VT = OpLHS.getValueType();
switch (OpNum) {
default: llvm_unreachable("Unknown shuffle opcode!");
case OP_VREV:
if (VT.getVectorElementType() == MVT::i32 ||
VT.getVectorElementType() == MVT::f32)
return DAG.getNode(ARMISD::VREV64, dl, VT, OpLHS);
if (VT.getVectorElementType() == MVT::i16)
return DAG.getNode(ARMISD::VREV32, dl, VT, OpLHS);
assert(VT.getVectorElementType() == MVT::i8);
return DAG.getNode(ARMISD::VREV16, dl, VT, OpLHS);
case OP_VDUP0:
case OP_VDUP1:
case OP_VDUP2:
case OP_VDUP3:
return DAG.getNode(ARMISD::VDUPLANE, dl, VT,
OpLHS, DAG.getConstant(OpNum-OP_VDUP0, MVT::i32));
case OP_VEXT1:
case OP_VEXT2:
case OP_VEXT3:
return DAG.getNode(ARMISD::VEXT, dl, VT,
OpLHS, OpRHS,
DAG.getConstant(OpNum-OP_VEXT1+1, MVT::i32));
case OP_VUZPL:
case OP_VUZPR:
return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
OpLHS, OpRHS).getValue(OpNum-OP_VUZPL);
case OP_VZIPL:
case OP_VZIPR:
return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
OpLHS, OpRHS).getValue(OpNum-OP_VZIPL);
case OP_VTRNL:
case OP_VTRNR:
return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
OpLHS, OpRHS).getValue(OpNum-OP_VTRNL);
}
}
static SDValue LowerVECTOR_SHUFFLEv8i8(SDValue Op,
ArrayRef<int> ShuffleMask,
SelectionDAG &DAG) {
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
SDLoc DL(Op);
SmallVector<SDValue, 8> VTBLMask;
for (ArrayRef<int>::iterator
I = ShuffleMask.begin(), E = ShuffleMask.end(); I != E; ++I)
VTBLMask.push_back(DAG.getConstant(*I, MVT::i32));
if (V2.getNode()->getOpcode() == ISD::UNDEF)
return DAG.getNode(ARMISD::VTBL1, DL, MVT::v8i8, V1,
DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v8i8, VTBLMask));
return DAG.getNode(ARMISD::VTBL2, DL, MVT::v8i8, V1, V2,
DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v8i8, VTBLMask));
}
static SDValue LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(SDValue Op,
SelectionDAG &DAG) {
SDLoc DL(Op);
SDValue OpLHS = Op.getOperand(0);
EVT VT = OpLHS.getValueType();
assert((VT == MVT::v8i16 || VT == MVT::v16i8) &&
"Expect an v8i16/v16i8 type");
OpLHS = DAG.getNode(ARMISD::VREV64, DL, VT, OpLHS);
unsigned ExtractNum = (VT == MVT::v16i8) ? 8 : 4;
return DAG.getNode(ARMISD::VEXT, DL, VT, OpLHS, OpLHS,
DAG.getConstant(ExtractNum, MVT::i32));
}
static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
SDLoc dl(Op);
EVT VT = Op.getValueType();
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());
ArrayRef<int> ShuffleMask = SVN->getMask();
unsigned EltSize = VT.getVectorElementType().getSizeInBits();
if (EltSize <= 32) {
if (ShuffleVectorSDNode::isSplatMask(&ShuffleMask[0], VT)) {
int Lane = SVN->getSplatIndex();
if (Lane == -1) Lane = 0;
if (Lane == 0 && V1.getOpcode() == ISD::SCALAR_TO_VECTOR) {
return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
}
if (Lane == 0 && V1.getOpcode() == ISD::BUILD_VECTOR &&
!isa<ConstantSDNode>(V1.getOperand(0))) {
bool IsScalarToVector = true;
for (unsigned i = 1, e = V1.getNumOperands(); i != e; ++i)
if (V1.getOperand(i).getOpcode() != ISD::UNDEF) {
IsScalarToVector = false;
break;
}
if (IsScalarToVector)
return DAG.getNode(ARMISD::VDUP, dl, VT, V1.getOperand(0));
}
return DAG.getNode(ARMISD::VDUPLANE, dl, VT, V1,
DAG.getConstant(Lane, MVT::i32));
}
bool ReverseVEXT;
unsigned Imm;
if (isVEXTMask(ShuffleMask, VT, ReverseVEXT, Imm)) {
if (ReverseVEXT)
std::swap(V1, V2);
return DAG.getNode(ARMISD::VEXT, dl, VT, V1, V2,
DAG.getConstant(Imm, MVT::i32));
}
if (isVREVMask(ShuffleMask, VT, 64))
return DAG.getNode(ARMISD::VREV64, dl, VT, V1);
if (isVREVMask(ShuffleMask, VT, 32))
return DAG.getNode(ARMISD::VREV32, dl, VT, V1);
if (isVREVMask(ShuffleMask, VT, 16))
return DAG.getNode(ARMISD::VREV16, dl, VT, V1);
if (V2->getOpcode() == ISD::UNDEF &&
isSingletonVEXTMask(ShuffleMask, VT, Imm)) {
return DAG.getNode(ARMISD::VEXT, dl, VT, V1, V1,
DAG.getConstant(Imm, MVT::i32));
}
unsigned WhichResult;
if (isVTRNMask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
V1, V2).getValue(WhichResult);
if (isVUZPMask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
V1, V2).getValue(WhichResult);
if (isVZIPMask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
V1, V2).getValue(WhichResult);
if (isVTRN_v_undef_Mask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VTRN, dl, DAG.getVTList(VT, VT),
V1, V1).getValue(WhichResult);
if (isVUZP_v_undef_Mask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VUZP, dl, DAG.getVTList(VT, VT),
V1, V1).getValue(WhichResult);
if (isVZIP_v_undef_Mask(ShuffleMask, VT, WhichResult))
return DAG.getNode(ARMISD::VZIP, dl, DAG.getVTList(VT, VT),
V1, V1).getValue(WhichResult);
}
unsigned NumElts = VT.getVectorNumElements();
if (NumElts == 4) {
unsigned PFIndexes[4];
for (unsigned i = 0; i != 4; ++i) {
if (ShuffleMask[i] < 0)
PFIndexes[i] = 8;
else
PFIndexes[i] = ShuffleMask[i];
}
unsigned PFTableIndex =
PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
unsigned Cost = (PFEntry >> 30);
if (Cost <= 4)
return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
}
if (EltSize >= 32) {
EVT EltVT = EVT::getFloatingPointVT(EltSize);
EVT VecVT = EVT::getVectorVT(*DAG.getContext(), EltVT, NumElts);
V1 = DAG.getNode(ISD::BITCAST, dl, VecVT, V1);
V2 = DAG.getNode(ISD::BITCAST, dl, VecVT, V2);
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0; i < NumElts; ++i) {
if (ShuffleMask[i] < 0)
Ops.push_back(DAG.getUNDEF(EltVT));
else
Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT,
ShuffleMask[i] < (int)NumElts ? V1 : V2,
DAG.getConstant(ShuffleMask[i] & (NumElts-1),
MVT::i32)));
}
SDValue Val = DAG.getNode(ARMISD::BUILD_VECTOR, dl, VecVT, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, Val);
}
if ((VT == MVT::v8i16 || VT == MVT::v16i8) && isReverseMask(ShuffleMask, VT))
return LowerReverse_VECTOR_SHUFFLEv16i8_v8i16(Op, DAG);
if (VT == MVT::v8i8) {
SDValue NewOp = LowerVECTOR_SHUFFLEv8i8(Op, ShuffleMask, DAG);
if (NewOp.getNode())
return NewOp;
}
return SDValue();
}
static SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
SDValue Lane = Op.getOperand(2);
if (!isa<ConstantSDNode>(Lane))
return SDValue();
return Op;
}
static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
SDValue Lane = Op.getOperand(1);
if (!isa<ConstantSDNode>(Lane))
return SDValue();
SDValue Vec = Op.getOperand(0);
if (Op.getValueType() == MVT::i32 &&
Vec.getValueType().getVectorElementType().getSizeInBits() < 32) {
SDLoc dl(Op);
return DAG.getNode(ARMISD::VGETLANEu, dl, MVT::i32, Vec, Lane);
}
return Op;
}
static SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
assert(Op.getValueType().is128BitVector() && Op.getNumOperands() == 2 &&
"unexpected CONCAT_VECTORS");
SDLoc dl(Op);
SDValue Val = DAG.getUNDEF(MVT::v2f64);
SDValue Op0 = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
if (Op0.getOpcode() != ISD::UNDEF)
Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Val,
DAG.getNode(ISD::BITCAST, dl, MVT::f64, Op0),
DAG.getIntPtrConstant(0));
if (Op1.getOpcode() != ISD::UNDEF)
Val = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v2f64, Val,
DAG.getNode(ISD::BITCAST, dl, MVT::f64, Op1),
DAG.getIntPtrConstant(1));
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Val);
}
static bool isExtendedBUILD_VECTOR(SDNode *N, SelectionDAG &DAG,
bool isSigned) {
EVT VT = N->getValueType(0);
if (VT == MVT::v2i64 && N->getOpcode() == ISD::BITCAST) {
SDNode *BVN = N->getOperand(0).getNode();
if (BVN->getValueType(0) != MVT::v4i32 ||
BVN->getOpcode() != ISD::BUILD_VECTOR)
return false;
unsigned LoElt = DAG.getTargetLoweringInfo().isBigEndian() ? 1 : 0;
unsigned HiElt = 1 - LoElt;
ConstantSDNode *Lo0 = dyn_cast<ConstantSDNode>(BVN->getOperand(LoElt));
ConstantSDNode *Hi0 = dyn_cast<ConstantSDNode>(BVN->getOperand(HiElt));
ConstantSDNode *Lo1 = dyn_cast<ConstantSDNode>(BVN->getOperand(LoElt+2));
ConstantSDNode *Hi1 = dyn_cast<ConstantSDNode>(BVN->getOperand(HiElt+2));
if (!Lo0 || !Hi0 || !Lo1 || !Hi1)
return false;
if (isSigned) {
if (Hi0->getSExtValue() == Lo0->getSExtValue() >> 32 &&
Hi1->getSExtValue() == Lo1->getSExtValue() >> 32)
return true;
} else {
if (Hi0->isNullValue() && Hi1->isNullValue())
return true;
}
return false;
}
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 bool isSignExtended(SDNode *N, SelectionDAG &DAG) {
if (N->getOpcode() == ISD::SIGN_EXTEND || ISD::isSEXTLoad(N))
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 || ISD::isZEXTLoad(N))
return true;
if (isExtendedBUILD_VECTOR(N, DAG, false))
return true;
return false;
}
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 AddRequiredExtensionForVMULL(SDValue N, SelectionDAG &DAG,
const EVT &OrigTy,
const EVT &ExtTy,
unsigned ExtOpcode) {
assert(ExtTy.is128BitVector() && "Unexpected extension size");
if (OrigTy.getSizeInBits() >= 64)
return N;
EVT NewVT = getExtensionTo64Bits(OrigTy);
return DAG.getNode(ExtOpcode, SDLoc(N), NewVT, N);
}
static SDValue SkipLoadExtensionForVMULL(LoadSDNode *LD, SelectionDAG& DAG) {
EVT ExtendedTy = getExtensionTo64Bits(LD->getMemoryVT());
if (ExtendedTy == LD->getMemoryVT())
return DAG.getLoad(LD->getMemoryVT(), SDLoc(LD), LD->getChain(),
LD->getBasePtr(), LD->getPointerInfo(), LD->isVolatile(),
LD->isNonTemporal(), LD->isInvariant(),
LD->getAlignment());
return DAG.getExtLoad(LD->getExtensionType(), SDLoc(LD), ExtendedTy,
LD->getChain(), LD->getBasePtr(), LD->getPointerInfo(),
LD->getMemoryVT(), LD->isVolatile(), LD->isInvariant(),
LD->isNonTemporal(), LD->getAlignment());
}
static SDValue SkipExtensionForVMULL(SDNode *N, SelectionDAG &DAG) {
if (N->getOpcode() == ISD::SIGN_EXTEND || N->getOpcode() == ISD::ZERO_EXTEND)
return AddRequiredExtensionForVMULL(N->getOperand(0), DAG,
N->getOperand(0)->getValueType(0),
N->getValueType(0),
N->getOpcode());
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N))
return SkipLoadExtensionForVMULL(LD, DAG);
if (N->getOpcode() == ISD::BITCAST) {
SDNode *BVN = N->getOperand(0).getNode();
assert(BVN->getOpcode() == ISD::BUILD_VECTOR &&
BVN->getValueType(0) == MVT::v4i32 && "expected v4i32 BUILD_VECTOR");
unsigned LowElt = DAG.getTargetLoweringInfo().isBigEndian() ? 1 : 0;
return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), MVT::v2i32,
BVN->getOperand(LowElt), BVN->getOperand(LowElt+2));
}
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();
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 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) {
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 = ARMISD::VMULLs;
else {
bool isN0ZExt = isZeroExtended(N0, DAG);
bool isN1ZExt = isZeroExtended(N1, DAG);
if (isN0ZExt && isN1ZExt)
NewOpc = ARMISD::VMULLu;
else if (isN1SExt || isN1ZExt) {
if (isN1SExt && isAddSubSExt(N0, DAG)) {
NewOpc = ARMISD::VMULLs;
isMLA = true;
} else if (isN1ZExt && isAddSubZExt(N0, DAG)) {
NewOpc = ARMISD::VMULLu;
isMLA = true;
} else if (isN0ZExt && isAddSubZExt(N1, DAG)) {
std::swap(N0, N1);
NewOpc = ARMISD::VMULLu;
isMLA = true;
}
}
if (!NewOpc) {
if (VT == MVT::v2i64)
return SDValue();
else
return Op;
}
}
SDLoc DL(Op);
SDValue Op0;
SDValue Op1 = SkipExtensionForVMULL(N1, DAG);
if (!isMLA) {
Op0 = SkipExtensionForVMULL(N0, DAG);
assert(Op0.getValueType().is64BitVector() &&
Op1.getValueType().is64BitVector() &&
"unexpected types for extended operands to VMULL");
return DAG.getNode(NewOpc, DL, VT, Op0, Op1);
}
SDValue N00 = SkipExtensionForVMULL(N0->getOperand(0).getNode(), DAG);
SDValue N01 = SkipExtensionForVMULL(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));
}
static SDValue
LowerSDIV_v4i8(SDValue X, SDValue Y, SDLoc dl, SelectionDAG &DAG) {
X = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, X);
Y = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, Y);
X = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, X);
Y = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, Y);
Y = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), Y);
X = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, X, Y);
X = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, X);
Y = DAG.getConstant(0xb000, MVT::i32);
Y = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Y, Y, Y, Y);
X = DAG.getNode(ISD::ADD, dl, MVT::v4i32, X, Y);
X = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, X);
X = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, X);
X = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, X);
return X;
}
static SDValue
LowerSDIV_v4i16(SDValue N0, SDValue N1, SDLoc dl, SelectionDAG &DAG) {
SDValue N2;
N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, N0);
N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v4i32, N1);
N0 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N0);
N1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N1);
N2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), N1);
N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
N1, N2);
N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
N0 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N0, N2);
N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, N0);
N1 = DAG.getConstant(0x89, MVT::i32);
N1 = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, N1, N1, N1, N1);
N0 = DAG.getNode(ISD::ADD, dl, MVT::v4i32, N0, N1);
N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, N0);
N0 = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, N0);
N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, N0);
return N0;
}
static SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getValueType();
assert((VT == MVT::v4i16 || VT == MVT::v8i8) &&
"unexpected type for custom-lowering ISD::SDIV");
SDLoc dl(Op);
SDValue N0 = Op.getOperand(0);
SDValue N1 = Op.getOperand(1);
SDValue N2, N3;
if (VT == MVT::v8i8) {
N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i16, N0);
N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::v8i16, N1);
N2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
DAG.getIntPtrConstant(4));
N3 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
DAG.getIntPtrConstant(4));
N0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
DAG.getIntPtrConstant(0));
N1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
DAG.getIntPtrConstant(0));
N0 = LowerSDIV_v4i8(N0, N1, dl, DAG); N2 = LowerSDIV_v4i8(N2, N3, dl, DAG);
N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i16, N0, N2);
N0 = LowerCONCAT_VECTORS(N0, DAG);
N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v8i8, N0);
return N0;
}
return LowerSDIV_v4i16(N0, N1, dl, DAG);
}
static SDValue LowerUDIV(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getValueType();
assert((VT == MVT::v4i16 || VT == MVT::v8i8) &&
"unexpected type for custom-lowering ISD::UDIV");
SDLoc dl(Op);
SDValue N0 = Op.getOperand(0);
SDValue N1 = Op.getOperand(1);
SDValue N2, N3;
if (VT == MVT::v8i8) {
N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v8i16, N0);
N1 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v8i16, N1);
N2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
DAG.getIntPtrConstant(4));
N3 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
DAG.getIntPtrConstant(4));
N0 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N0,
DAG.getIntPtrConstant(0));
N1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v4i16, N1,
DAG.getIntPtrConstant(0));
N0 = LowerSDIV_v4i16(N0, N1, dl, DAG); N2 = LowerSDIV_v4i16(N2, N3, dl, DAG);
N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i16, N0, N2);
N0 = LowerCONCAT_VECTORS(N0, DAG);
N0 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v8i8,
DAG.getConstant(Intrinsic::arm_neon_vqmovnsu, MVT::i32),
N0);
return N0;
}
N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v4i32, N0);
N1 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::v4i32, N1);
N0 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N0);
SDValue BN1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::v4f32, N1);
N2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecpe, MVT::i32), BN1);
N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
BN1, N2);
N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
N1 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, MVT::v4f32,
DAG.getConstant(Intrinsic::arm_neon_vrecps, MVT::i32),
BN1, N2);
N2 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N1, N2);
N0 = DAG.getNode(ISD::FMUL, dl, MVT::v4f32, N0, N2);
N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, N0);
N1 = DAG.getConstant(2, MVT::i32);
N1 = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, N1, N1, N1, N1);
N0 = DAG.getNode(ISD::ADD, dl, MVT::v4i32, N0, N1);
N0 = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, N0);
N0 = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::v4i32, N0);
N0 = DAG.getNode(ISD::TRUNCATE, dl, MVT::v4i16, N0);
return N0;
}
static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) {
EVT VT = Op.getNode()->getValueType(0);
SDVTList VTs = DAG.getVTList(VT, MVT::i32);
unsigned Opc;
bool ExtraOp = false;
switch (Op.getOpcode()) {
default: llvm_unreachable("Invalid code");
case ISD::ADDC: Opc = ARMISD::ADDC; break;
case ISD::ADDE: Opc = ARMISD::ADDE; ExtraOp = true; break;
case ISD::SUBC: Opc = ARMISD::SUBC; break;
case ISD::SUBE: Opc = ARMISD::SUBE; ExtraOp = true; break;
}
if (!ExtraOp)
return DAG.getNode(Opc, SDLoc(Op), VTs, Op.getOperand(0),
Op.getOperand(1));
return DAG.getNode(Opc, SDLoc(Op), VTs, Op.getOperand(0),
Op.getOperand(1), Op.getOperand(2));
}
SDValue ARMTargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
assert(Subtarget->isTargetDarwin());
SDLoc dl(Op);
SDValue Arg = Op.getOperand(0);
EVT ArgVT = Arg.getValueType();
Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
Type *RetTy = StructType::get(ArgTy, ArgTy, nullptr);
ArgListTy Args;
bool ShouldUseSRet = Subtarget->isAPCS_ABI();
SDValue SRet;
if (ShouldUseSRet) {
const uint64_t ByteSize = TLI.getDataLayout()->getTypeAllocSize(RetTy);
const unsigned StackAlign =
TLI.getDataLayout()->getPrefTypeAlignment(RetTy);
int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
SRet = DAG.getFrameIndex(FrameIdx, TLI.getPointerTy());
ArgListEntry Entry;
Entry.Node = SRet;
Entry.Ty = RetTy->getPointerTo();
Entry.isSExt = false;
Entry.isZExt = false;
Entry.isSRet = true;
Args.push_back(Entry);
RetTy = Type::getVoidTy(*DAG.getContext());
}
ArgListEntry Entry;
Entry.Node = Arg;
Entry.Ty = ArgTy;
Entry.isSExt = false;
Entry.isZExt = false;
Args.push_back(Entry);
const char *LibcallName = (ArgVT == MVT::f64)
? "__sincos_stret" : "__sincosf_stret";
RTLIB::Libcall LC =
(ArgVT == MVT::f64) ? RTLIB::SINCOS_F64 : RTLIB::SINCOS_F32;
CallingConv::ID CC = getLibcallCallingConv(LC);
SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(DAG.getEntryNode())
.setCallee(CC, RetTy, Callee, std::move(Args), 0)
.setDiscardResult(ShouldUseSRet);
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
if (!ShouldUseSRet)
return CallResult.first;
SDValue LoadSin = DAG.getLoad(ArgVT, dl, CallResult.second, SRet,
MachinePointerInfo(), false, false, false, 0);
SDValue Add = DAG.getNode(ISD::ADD, dl, getPointerTy(), SRet,
DAG.getIntPtrConstant(ArgVT.getStoreSize()));
SDValue LoadCos = DAG.getLoad(ArgVT, dl, LoadSin.getValue(1), Add,
MachinePointerInfo(), false, false, false, 0);
SDVTList Tys = DAG.getVTList(ArgVT, ArgVT);
return DAG.getNode(ISD::MERGE_VALUES, dl, Tys,
LoadSin.getValue(0), LoadCos.getValue(0));
}
static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) {
if (cast<AtomicSDNode>(Op)->getOrdering() <= Monotonic)
return Op;
return SDValue();
}
static void ReplaceREADCYCLECOUNTER(SDNode *N,
SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG,
const ARMSubtarget *Subtarget) {
SDLoc DL(N);
SDValue Cycles32, OutChain;
if (Subtarget->hasPerfMon()) {
SDValue Ops[] = { N->getOperand(0), DAG.getConstant(Intrinsic::arm_mrc, MVT::i32),
DAG.getConstant(15, MVT::i32),
DAG.getConstant(0, MVT::i32),
DAG.getConstant(9, MVT::i32),
DAG.getConstant(13, MVT::i32),
DAG.getConstant(0, MVT::i32)
};
Cycles32 = DAG.getNode(ISD::INTRINSIC_W_CHAIN, DL,
DAG.getVTList(MVT::i32, MVT::Other), Ops);
OutChain = Cycles32.getValue(1);
} else {
Cycles32 = DAG.getConstant(0, MVT::i32);
OutChain = DAG.getEntryNode();
}
SDValue Cycles64 = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64,
Cycles32, DAG.getConstant(0, MVT::i32));
Results.push_back(Cycles64);
Results.push_back(OutChain);
}
SDValue ARMTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
switch (Op.getOpcode()) {
default: llvm_unreachable("Don't know how to custom lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
case ISD::GlobalAddress:
switch (Subtarget->getTargetTriple().getObjectFormat()) {
default: llvm_unreachable("unknown object format");
case Triple::COFF:
return LowerGlobalAddressWindows(Op, DAG);
case Triple::ELF:
return LowerGlobalAddressELF(Op, DAG);
case Triple::MachO:
return LowerGlobalAddressDarwin(Op, DAG);
}
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::SELECT: return LowerSELECT(Op, DAG);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::BR_CC: return LowerBR_CC(Op, DAG);
case ISD::BR_JT: return LowerBR_JT(Op, DAG);
case ISD::VASTART: return LowerVASTART(Op, DAG);
case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, DAG, Subtarget);
case ISD::PREFETCH: return LowerPREFETCH(Op, DAG, Subtarget);
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP: return LowerINT_TO_FP(Op, DAG);
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT: return LowerFP_TO_INT(Op, DAG);
case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG);
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::GLOBAL_OFFSET_TABLE: return LowerGLOBAL_OFFSET_TABLE(Op, DAG);
case ISD::EH_SJLJ_SETJMP: return LowerEH_SJLJ_SETJMP(Op, DAG);
case ISD::EH_SJLJ_LONGJMP: return LowerEH_SJLJ_LONGJMP(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG,
Subtarget);
case ISD::BITCAST: return ExpandBITCAST(Op.getNode(), DAG);
case ISD::SHL:
case ISD::SRL:
case ISD::SRA: return LowerShift(Op.getNode(), DAG, Subtarget);
case ISD::SHL_PARTS: return LowerShiftLeftParts(Op, DAG);
case ISD::SRL_PARTS:
case ISD::SRA_PARTS: return LowerShiftRightParts(Op, DAG);
case ISD::CTTZ: return LowerCTTZ(Op.getNode(), DAG, Subtarget);
case ISD::CTPOP: return LowerCTPOP(Op.getNode(), DAG, Subtarget);
case ISD::SETCC: return LowerVSETCC(Op, DAG);
case ISD::ConstantFP: return LowerConstantFP(Op, DAG, Subtarget);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG, Subtarget);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT: return LowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::CONCAT_VECTORS: return LowerCONCAT_VECTORS(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
case ISD::MUL: return LowerMUL(Op, DAG);
case ISD::SDIV: return LowerSDIV(Op, DAG);
case ISD::UDIV: return LowerUDIV(Op, DAG);
case ISD::ADDC:
case ISD::ADDE:
case ISD::SUBC:
case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
case ISD::SADDO:
case ISD::UADDO:
case ISD::SSUBO:
case ISD::USUBO:
return LowerXALUO(Op, DAG);
case ISD::ATOMIC_LOAD:
case ISD::ATOMIC_STORE: return LowerAtomicLoadStore(Op, DAG);
case ISD::FSINCOS: return LowerFSINCOS(Op, DAG);
case ISD::SDIVREM:
case ISD::UDIVREM: return LowerDivRem(Op, DAG);
case ISD::DYNAMIC_STACKALLOC:
if (Subtarget->getTargetTriple().isWindowsItaniumEnvironment())
return LowerDYNAMIC_STACKALLOC(Op, DAG);
llvm_unreachable("Don't know how to custom lower this!");
case ISD::FP_ROUND: return LowerFP_ROUND(Op, DAG);
case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
}
}
void ARMTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue>&Results,
SelectionDAG &DAG) const {
SDValue Res;
switch (N->getOpcode()) {
default:
llvm_unreachable("Don't know how to custom expand this!");
case ISD::BITCAST:
Res = ExpandBITCAST(N, DAG);
break;
case ISD::SRL:
case ISD::SRA:
Res = Expand64BitShift(N, DAG, Subtarget);
break;
case ISD::READCYCLECOUNTER:
ReplaceREADCYCLECOUNTER(N, Results, DAG, Subtarget);
return;
}
if (Res.getNode())
Results.push_back(Res);
}
void ARMTargetLowering::
SetupEntryBlockForSjLj(MachineInstr *MI, MachineBasicBlock *MBB,
MachineBasicBlock *DispatchBB, int FI) const {
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
MachineConstantPool *MCP = MF->getConstantPool();
ARMFunctionInfo *AFI = MF->getInfo<ARMFunctionInfo>();
const Function *F = MF->getFunction();
bool isThumb = Subtarget->isThumb();
bool isThumb2 = Subtarget->isThumb2();
unsigned PCLabelId = AFI->createPICLabelUId();
unsigned PCAdj = (isThumb || isThumb2) ? 4 : 8;
ARMConstantPoolValue *CPV =
ARMConstantPoolMBB::Create(F->getContext(), DispatchBB, PCLabelId, PCAdj);
unsigned CPI = MCP->getConstantPoolIndex(CPV, 4);
const TargetRegisterClass *TRC = isThumb ? &ARM::tGPRRegClass
: &ARM::GPRRegClass;
MachineMemOperand *CPMMO =
MF->getMachineMemOperand(MachinePointerInfo::getConstantPool(),
MachineMemOperand::MOLoad, 4, 4);
MachineMemOperand *FIMMOSt =
MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
MachineMemOperand::MOStore, 4, 4);
if (isThumb2) {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2LDRpci), NewVReg1)
.addConstantPoolIndex(CPI)
.addMemOperand(CPMMO));
unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
AddDefaultCC(
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2ORRri), NewVReg2)
.addReg(NewVReg1, RegState::Kill)
.addImm(0x01)));
unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg3)
.addReg(NewVReg2, RegState::Kill)
.addImm(PCLabelId);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::t2STRi12))
.addReg(NewVReg3, RegState::Kill)
.addFrameIndex(FI)
.addImm(36) .addMemOperand(FIMMOSt));
} else if (isThumb) {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tLDRpci), NewVReg1)
.addConstantPoolIndex(CPI)
.addMemOperand(CPMMO));
unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
BuildMI(*MBB, MI, dl, TII->get(ARM::tPICADD), NewVReg2)
.addReg(NewVReg1, RegState::Kill)
.addImm(PCLabelId);
unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tMOVi8), NewVReg3)
.addReg(ARM::CPSR, RegState::Define)
.addImm(1));
unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tORR), NewVReg4)
.addReg(ARM::CPSR, RegState::Define)
.addReg(NewVReg2, RegState::Kill)
.addReg(NewVReg3, RegState::Kill));
unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
BuildMI(*MBB, MI, dl, TII->get(ARM::tADDframe), NewVReg5)
.addFrameIndex(FI)
.addImm(36); AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::tSTRi))
.addReg(NewVReg4, RegState::Kill)
.addReg(NewVReg5, RegState::Kill)
.addImm(0)
.addMemOperand(FIMMOSt));
} else {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::LDRi12), NewVReg1)
.addConstantPoolIndex(CPI)
.addImm(0)
.addMemOperand(CPMMO));
unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::PICADD), NewVReg2)
.addReg(NewVReg1, RegState::Kill)
.addImm(PCLabelId));
AddDefaultPred(BuildMI(*MBB, MI, dl, TII->get(ARM::STRi12))
.addReg(NewVReg2, RegState::Kill)
.addFrameIndex(FI)
.addImm(36) .addMemOperand(FIMMOSt));
}
}
MachineBasicBlock *ARMTargetLowering::
EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
MachineFrameInfo *MFI = MF->getFrameInfo();
int FI = MFI->getFunctionContextIndex();
const TargetRegisterClass *TRC = Subtarget->isThumb() ? &ARM::tGPRRegClass
: &ARM::GPRnopcRegClass;
DenseMap<unsigned, SmallVector<MachineBasicBlock*, 2> > CallSiteNumToLPad;
unsigned MaxCSNum = 0;
MachineModuleInfo &MMI = MF->getMMI();
for (MachineFunction::iterator BB = MF->begin(), E = MF->end(); BB != E;
++BB) {
if (!BB->isLandingPad()) continue;
for (MachineBasicBlock::iterator
II = BB->begin(), IE = BB->end(); II != IE; ++II) {
if (!II->isEHLabel()) continue;
MCSymbol *Sym = II->getOperand(0).getMCSymbol();
if (!MMI.hasCallSiteLandingPad(Sym)) continue;
SmallVectorImpl<unsigned> &CallSiteIdxs = MMI.getCallSiteLandingPad(Sym);
for (SmallVectorImpl<unsigned>::iterator
CSI = CallSiteIdxs.begin(), CSE = CallSiteIdxs.end();
CSI != CSE; ++CSI) {
CallSiteNumToLPad[*CSI].push_back(BB);
MaxCSNum = std::max(MaxCSNum, *CSI);
}
break;
}
}
std::vector<MachineBasicBlock*> LPadList;
SmallPtrSet<MachineBasicBlock*, 64> InvokeBBs;
LPadList.reserve(CallSiteNumToLPad.size());
for (unsigned I = 1; I <= MaxCSNum; ++I) {
SmallVectorImpl<MachineBasicBlock*> &MBBList = CallSiteNumToLPad[I];
for (SmallVectorImpl<MachineBasicBlock*>::iterator
II = MBBList.begin(), IE = MBBList.end(); II != IE; ++II) {
LPadList.push_back(*II);
InvokeBBs.insert((*II)->pred_begin(), (*II)->pred_end());
}
}
assert(!LPadList.empty() &&
"No landing pad destinations for the dispatch jump table!");
MachineJumpTableInfo *JTI =
MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline);
unsigned MJTI = JTI->createJumpTableIndex(LPadList);
Reloc::Model RelocM = getTargetMachine().getRelocationModel();
MachineBasicBlock *DispatchBB = MF->CreateMachineBasicBlock();
DispatchBB->setIsLandingPad();
MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
unsigned trap_opcode;
if (Subtarget->isThumb())
trap_opcode = ARM::tTRAP;
else
trap_opcode = Subtarget->useNaClTrap() ? ARM::TRAPNaCl : ARM::TRAP;
BuildMI(TrapBB, dl, TII->get(trap_opcode));
DispatchBB->addSuccessor(TrapBB);
MachineBasicBlock *DispContBB = MF->CreateMachineBasicBlock();
DispatchBB->addSuccessor(DispContBB);
MF->insert(MF->end(), DispatchBB);
MF->insert(MF->end(), DispContBB);
MF->insert(MF->end(), TrapBB);
SetupEntryBlockForSjLj(MI, MBB, DispatchBB, FI);
MachineMemOperand *FIMMOLd =
MF->getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
MachineMemOperand::MOLoad |
MachineMemOperand::MOVolatile, 4, 4);
MachineInstrBuilder MIB;
MIB = BuildMI(DispatchBB, dl, TII->get(ARM::Int_eh_sjlj_dispatchsetup));
const ARMBaseInstrInfo *AII = static_cast<const ARMBaseInstrInfo*>(TII);
const ARMBaseRegisterInfo &RI = AII->getRegisterInfo();
MIB.addRegMask(RI.getNoPreservedMask());
unsigned NumLPads = LPadList.size();
if (Subtarget->isThumb2()) {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2LDRi12), NewVReg1)
.addFrameIndex(FI)
.addImm(4)
.addMemOperand(FIMMOLd));
if (NumLPads < 256) {
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPri))
.addReg(NewVReg1)
.addImm(LPadList.size()));
} else {
unsigned VReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVi16), VReg1)
.addImm(NumLPads & 0xFFFF));
unsigned VReg2 = VReg1;
if ((NumLPads & 0xFFFF0000) != 0) {
VReg2 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2MOVTi16), VReg2)
.addReg(VReg1)
.addImm(NumLPads >> 16));
}
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::t2CMPrr))
.addReg(NewVReg1)
.addReg(VReg2));
}
BuildMI(DispatchBB, dl, TII->get(ARM::t2Bcc))
.addMBB(TrapBB)
.addImm(ARMCC::HI)
.addReg(ARM::CPSR);
unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::t2LEApcrelJT),NewVReg3)
.addJumpTableIndex(MJTI));
unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
AddDefaultCC(
AddDefaultPred(
BuildMI(DispContBB, dl, TII->get(ARM::t2ADDrs), NewVReg4)
.addReg(NewVReg3, RegState::Kill)
.addReg(NewVReg1)
.addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
BuildMI(DispContBB, dl, TII->get(ARM::t2BR_JT))
.addReg(NewVReg4, RegState::Kill)
.addReg(NewVReg1)
.addJumpTableIndex(MJTI);
} else if (Subtarget->isThumb()) {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRspi), NewVReg1)
.addFrameIndex(FI)
.addImm(1)
.addMemOperand(FIMMOLd));
if (NumLPads < 256) {
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPi8))
.addReg(NewVReg1)
.addImm(NumLPads));
} else {
MachineConstantPool *ConstantPool = MF->getConstantPool();
Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
const Constant *C = ConstantInt::get(Int32Ty, NumLPads);
unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
if (Align == 0)
Align = getDataLayout()->getTypeAllocSize(C->getType());
unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);
unsigned VReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tLDRpci))
.addReg(VReg1, RegState::Define)
.addConstantPoolIndex(Idx));
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::tCMPr))
.addReg(NewVReg1)
.addReg(VReg1));
}
BuildMI(DispatchBB, dl, TII->get(ARM::tBcc))
.addMBB(TrapBB)
.addImm(ARMCC::HI)
.addReg(ARM::CPSR);
unsigned NewVReg2 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLSLri), NewVReg2)
.addReg(ARM::CPSR, RegState::Define)
.addReg(NewVReg1)
.addImm(2));
unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLEApcrelJT), NewVReg3)
.addJumpTableIndex(MJTI));
unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg4)
.addReg(ARM::CPSR, RegState::Define)
.addReg(NewVReg2, RegState::Kill)
.addReg(NewVReg3));
MachineMemOperand *JTMMOLd =
MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
MachineMemOperand::MOLoad, 4, 4);
unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tLDRi), NewVReg5)
.addReg(NewVReg4, RegState::Kill)
.addImm(0)
.addMemOperand(JTMMOLd));
unsigned NewVReg6 = NewVReg5;
if (RelocM == Reloc::PIC_) {
NewVReg6 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
.addReg(ARM::CPSR, RegState::Define)
.addReg(NewVReg5, RegState::Kill)
.addReg(NewVReg3));
}
BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr))
.addReg(NewVReg6, RegState::Kill)
.addJumpTableIndex(MJTI);
} else {
unsigned NewVReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRi12), NewVReg1)
.addFrameIndex(FI)
.addImm(4)
.addMemOperand(FIMMOLd));
if (NumLPads < 256) {
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPri))
.addReg(NewVReg1)
.addImm(NumLPads));
} else if (Subtarget->hasV6T2Ops() && isUInt<16>(NumLPads)) {
unsigned VReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVi16), VReg1)
.addImm(NumLPads & 0xFFFF));
unsigned VReg2 = VReg1;
if ((NumLPads & 0xFFFF0000) != 0) {
VReg2 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::MOVTi16), VReg2)
.addReg(VReg1)
.addImm(NumLPads >> 16));
}
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
.addReg(NewVReg1)
.addReg(VReg2));
} else {
MachineConstantPool *ConstantPool = MF->getConstantPool();
Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
const Constant *C = ConstantInt::get(Int32Ty, NumLPads);
unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
if (Align == 0)
Align = getDataLayout()->getTypeAllocSize(C->getType());
unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);
unsigned VReg1 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::LDRcp))
.addReg(VReg1, RegState::Define)
.addConstantPoolIndex(Idx)
.addImm(0));
AddDefaultPred(BuildMI(DispatchBB, dl, TII->get(ARM::CMPrr))
.addReg(NewVReg1)
.addReg(VReg1, RegState::Kill));
}
BuildMI(DispatchBB, dl, TII->get(ARM::Bcc))
.addMBB(TrapBB)
.addImm(ARMCC::HI)
.addReg(ARM::CPSR);
unsigned NewVReg3 = MRI->createVirtualRegister(TRC);
AddDefaultCC(
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::MOVsi), NewVReg3)
.addReg(NewVReg1)
.addImm(ARM_AM::getSORegOpc(ARM_AM::lsl, 2))));
unsigned NewVReg4 = MRI->createVirtualRegister(TRC);
AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::LEApcrelJT), NewVReg4)
.addJumpTableIndex(MJTI));
MachineMemOperand *JTMMOLd =
MF->getMachineMemOperand(MachinePointerInfo::getJumpTable(),
MachineMemOperand::MOLoad, 4, 4);
unsigned NewVReg5 = MRI->createVirtualRegister(TRC);
AddDefaultPred(
BuildMI(DispContBB, dl, TII->get(ARM::LDRrs), NewVReg5)
.addReg(NewVReg3, RegState::Kill)
.addReg(NewVReg4)
.addImm(0)
.addMemOperand(JTMMOLd));
if (RelocM == Reloc::PIC_) {
BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
.addReg(NewVReg5, RegState::Kill)
.addReg(NewVReg4)
.addJumpTableIndex(MJTI);
} else {
BuildMI(DispContBB, dl, TII->get(ARM::BR_JTr))
.addReg(NewVReg5, RegState::Kill)
.addJumpTableIndex(MJTI);
}
}
SmallPtrSet<MachineBasicBlock*, 8> SeenMBBs;
for (std::vector<MachineBasicBlock*>::iterator
I = LPadList.begin(), E = LPadList.end(); I != E; ++I) {
MachineBasicBlock *CurMBB = *I;
if (SeenMBBs.insert(CurMBB).second)
DispContBB->addSuccessor(CurMBB);
}
const MCPhysReg *SavedRegs = RI.getCalleeSavedRegs(MF);
SmallVector<MachineBasicBlock*, 64> MBBLPads;
for (MachineBasicBlock *BB : InvokeBBs) {
SmallVector<MachineBasicBlock*, 4> Successors(BB->succ_begin(),
BB->succ_end());
while (!Successors.empty()) {
MachineBasicBlock *SMBB = Successors.pop_back_val();
if (SMBB->isLandingPad()) {
BB->removeSuccessor(SMBB);
MBBLPads.push_back(SMBB);
}
}
BB->addSuccessor(DispatchBB);
for (MachineBasicBlock::reverse_iterator
II = BB->rbegin(), IE = BB->rend(); II != IE; ++II) {
if (!II->isCall()) continue;
DenseMap<unsigned, bool> DefRegs;
for (MachineInstr::mop_iterator
OI = II->operands_begin(), OE = II->operands_end();
OI != OE; ++OI) {
if (!OI->isReg()) continue;
DefRegs[OI->getReg()] = true;
}
MachineInstrBuilder MIB(*MF, &*II);
for (unsigned i = 0; SavedRegs[i] != 0; ++i) {
unsigned Reg = SavedRegs[i];
if (Subtarget->isThumb2() &&
!ARM::tGPRRegClass.contains(Reg) &&
!ARM::hGPRRegClass.contains(Reg))
continue;
if (Subtarget->isThumb1Only() && !ARM::tGPRRegClass.contains(Reg))
continue;
if (!Subtarget->isThumb() && !ARM::GPRRegClass.contains(Reg))
continue;
if (!DefRegs[Reg])
MIB.addReg(Reg, RegState::ImplicitDefine | RegState::Dead);
}
break;
}
}
for (SmallVectorImpl<MachineBasicBlock*>::iterator
I = MBBLPads.begin(), E = MBBLPads.end(); I != E; ++I)
(*I)->setIsLandingPad(false);
MI->eraseFromParent();
return MBB;
}
static
MachineBasicBlock *OtherSucc(MachineBasicBlock *MBB, MachineBasicBlock *Succ) {
for (MachineBasicBlock::succ_iterator I = MBB->succ_begin(),
E = MBB->succ_end(); I != E; ++I)
if (*I != Succ)
return *I;
llvm_unreachable("Expecting a BB with two successors!");
}
static unsigned getLdOpcode(unsigned LdSize, bool IsThumb1, bool IsThumb2) {
if (LdSize >= 8)
return LdSize == 16 ? ARM::VLD1q32wb_fixed
: LdSize == 8 ? ARM::VLD1d32wb_fixed : 0;
if (IsThumb1)
return LdSize == 4 ? ARM::tLDRi
: LdSize == 2 ? ARM::tLDRHi
: LdSize == 1 ? ARM::tLDRBi : 0;
if (IsThumb2)
return LdSize == 4 ? ARM::t2LDR_POST
: LdSize == 2 ? ARM::t2LDRH_POST
: LdSize == 1 ? ARM::t2LDRB_POST : 0;
return LdSize == 4 ? ARM::LDR_POST_IMM
: LdSize == 2 ? ARM::LDRH_POST
: LdSize == 1 ? ARM::LDRB_POST_IMM : 0;
}
static unsigned getStOpcode(unsigned StSize, bool IsThumb1, bool IsThumb2) {
if (StSize >= 8)
return StSize == 16 ? ARM::VST1q32wb_fixed
: StSize == 8 ? ARM::VST1d32wb_fixed : 0;
if (IsThumb1)
return StSize == 4 ? ARM::tSTRi
: StSize == 2 ? ARM::tSTRHi
: StSize == 1 ? ARM::tSTRBi : 0;
if (IsThumb2)
return StSize == 4 ? ARM::t2STR_POST
: StSize == 2 ? ARM::t2STRH_POST
: StSize == 1 ? ARM::t2STRB_POST : 0;
return StSize == 4 ? ARM::STR_POST_IMM
: StSize == 2 ? ARM::STRH_POST
: StSize == 1 ? ARM::STRB_POST_IMM : 0;
}
static void emitPostLd(MachineBasicBlock *BB, MachineInstr *Pos,
const TargetInstrInfo *TII, DebugLoc dl,
unsigned LdSize, unsigned Data, unsigned AddrIn,
unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
unsigned LdOpc = getLdOpcode(LdSize, IsThumb1, IsThumb2);
assert(LdOpc != 0 && "Should have a load opcode");
if (LdSize >= 8) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
.addReg(AddrOut, RegState::Define).addReg(AddrIn)
.addImm(0));
} else if (IsThumb1) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
.addReg(AddrIn).addImm(0));
MachineInstrBuilder MIB =
BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
MIB = AddDefaultT1CC(MIB);
MIB.addReg(AddrIn).addImm(LdSize);
AddDefaultPred(MIB);
} else if (IsThumb2) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
.addReg(AddrOut, RegState::Define).addReg(AddrIn)
.addImm(LdSize));
} else { AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(LdOpc), Data)
.addReg(AddrOut, RegState::Define).addReg(AddrIn)
.addReg(0).addImm(LdSize));
}
}
static void emitPostSt(MachineBasicBlock *BB, MachineInstr *Pos,
const TargetInstrInfo *TII, DebugLoc dl,
unsigned StSize, unsigned Data, unsigned AddrIn,
unsigned AddrOut, bool IsThumb1, bool IsThumb2) {
unsigned StOpc = getStOpcode(StSize, IsThumb1, IsThumb2);
assert(StOpc != 0 && "Should have a store opcode");
if (StSize >= 8) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
.addReg(AddrIn).addImm(0).addReg(Data));
} else if (IsThumb1) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc)).addReg(Data)
.addReg(AddrIn).addImm(0));
MachineInstrBuilder MIB =
BuildMI(*BB, Pos, dl, TII->get(ARM::tADDi8), AddrOut);
MIB = AddDefaultT1CC(MIB);
MIB.addReg(AddrIn).addImm(StSize);
AddDefaultPred(MIB);
} else if (IsThumb2) {
AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
.addReg(Data).addReg(AddrIn).addImm(StSize));
} else { AddDefaultPred(BuildMI(*BB, Pos, dl, TII->get(StOpc), AddrOut)
.addReg(Data).addReg(AddrIn).addReg(0)
.addImm(StSize));
}
}
MachineBasicBlock *
ARMTargetLowering::EmitStructByval(MachineInstr *MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = BB;
++It;
unsigned dest = MI->getOperand(0).getReg();
unsigned src = MI->getOperand(1).getReg();
unsigned SizeVal = MI->getOperand(2).getImm();
unsigned Align = MI->getOperand(3).getImm();
DebugLoc dl = MI->getDebugLoc();
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
unsigned UnitSize = 0;
const TargetRegisterClass *TRC = nullptr;
const TargetRegisterClass *VecTRC = nullptr;
bool IsThumb1 = Subtarget->isThumb1Only();
bool IsThumb2 = Subtarget->isThumb2();
if (Align & 1) {
UnitSize = 1;
} else if (Align & 2) {
UnitSize = 2;
} else {
if (!MF->getFunction()->hasFnAttribute(Attribute::NoImplicitFloat) &&
Subtarget->hasNEON()) {
if ((Align % 16 == 0) && SizeVal >= 16)
UnitSize = 16;
else if ((Align % 8 == 0) && SizeVal >= 8)
UnitSize = 8;
}
if (UnitSize == 0)
UnitSize = 4;
}
bool IsNeon = UnitSize >= 8;
TRC = (IsThumb1 || IsThumb2) ? &ARM::tGPRRegClass : &ARM::GPRRegClass;
if (IsNeon)
VecTRC = UnitSize == 16 ? &ARM::DPairRegClass
: UnitSize == 8 ? &ARM::DPRRegClass
: nullptr;
unsigned BytesLeft = SizeVal % UnitSize;
unsigned LoopSize = SizeVal - BytesLeft;
if (SizeVal <= Subtarget->getMaxInlineSizeThreshold()) {
unsigned srcIn = src;
unsigned destIn = dest;
for (unsigned i = 0; i < LoopSize; i+=UnitSize) {
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
emitPostLd(BB, MI, TII, dl, UnitSize, scratch, srcIn, srcOut,
IsThumb1, IsThumb2);
emitPostSt(BB, MI, TII, dl, UnitSize, scratch, destIn, destOut,
IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
for (unsigned i = 0; i < BytesLeft; i++) {
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(TRC);
emitPostLd(BB, MI, TII, dl, 1, scratch, srcIn, srcOut,
IsThumb1, IsThumb2);
emitPostSt(BB, MI, TII, dl, 1, scratch, destIn, destOut,
IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
MI->eraseFromParent(); return BB;
}
MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MF->insert(It, loopMBB);
MF->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
unsigned varEnd = MRI.createVirtualRegister(TRC);
if (IsThumb2) {
unsigned Vtmp = varEnd;
if ((LoopSize & 0xFFFF0000) != 0)
Vtmp = MRI.createVirtualRegister(TRC);
AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVi16), Vtmp)
.addImm(LoopSize & 0xFFFF));
if ((LoopSize & 0xFFFF0000) != 0)
AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2MOVTi16), varEnd)
.addReg(Vtmp).addImm(LoopSize >> 16));
} else {
MachineConstantPool *ConstantPool = MF->getConstantPool();
Type *Int32Ty = Type::getInt32Ty(MF->getFunction()->getContext());
const Constant *C = ConstantInt::get(Int32Ty, LoopSize);
unsigned Align = getDataLayout()->getPrefTypeAlignment(Int32Ty);
if (Align == 0)
Align = getDataLayout()->getTypeAllocSize(C->getType());
unsigned Idx = ConstantPool->getConstantPoolIndex(C, Align);
if (IsThumb1)
AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::tLDRpci)).addReg(
varEnd, RegState::Define).addConstantPoolIndex(Idx));
else
AddDefaultPred(BuildMI(*BB, MI, dl, TII->get(ARM::LDRcp)).addReg(
varEnd, RegState::Define).addConstantPoolIndex(Idx).addImm(0));
}
BB->addSuccessor(loopMBB);
MachineBasicBlock *entryBB = BB;
BB = loopMBB;
unsigned varLoop = MRI.createVirtualRegister(TRC);
unsigned varPhi = MRI.createVirtualRegister(TRC);
unsigned srcLoop = MRI.createVirtualRegister(TRC);
unsigned srcPhi = MRI.createVirtualRegister(TRC);
unsigned destLoop = MRI.createVirtualRegister(TRC);
unsigned destPhi = MRI.createVirtualRegister(TRC);
BuildMI(*BB, BB->begin(), dl, TII->get(ARM::PHI), varPhi)
.addReg(varLoop).addMBB(loopMBB)
.addReg(varEnd).addMBB(entryBB);
BuildMI(BB, dl, TII->get(ARM::PHI), srcPhi)
.addReg(srcLoop).addMBB(loopMBB)
.addReg(src).addMBB(entryBB);
BuildMI(BB, dl, TII->get(ARM::PHI), destPhi)
.addReg(destLoop).addMBB(loopMBB)
.addReg(dest).addMBB(entryBB);
unsigned scratch = MRI.createVirtualRegister(IsNeon ? VecTRC : TRC);
emitPostLd(BB, BB->end(), TII, dl, UnitSize, scratch, srcPhi, srcLoop,
IsThumb1, IsThumb2);
emitPostSt(BB, BB->end(), TII, dl, UnitSize, scratch, destPhi, destLoop,
IsThumb1, IsThumb2);
if (IsThumb1) {
MachineInstrBuilder MIB =
BuildMI(*BB, BB->end(), dl, TII->get(ARM::tSUBi8), varLoop);
MIB = AddDefaultT1CC(MIB);
MIB.addReg(varPhi).addImm(UnitSize);
AddDefaultPred(MIB);
} else {
MachineInstrBuilder MIB =
BuildMI(*BB, BB->end(), dl,
TII->get(IsThumb2 ? ARM::t2SUBri : ARM::SUBri), varLoop);
AddDefaultCC(AddDefaultPred(MIB.addReg(varPhi).addImm(UnitSize)));
MIB->getOperand(5).setReg(ARM::CPSR);
MIB->getOperand(5).setIsDef(true);
}
BuildMI(*BB, BB->end(), dl,
TII->get(IsThumb1 ? ARM::tBcc : IsThumb2 ? ARM::t2Bcc : ARM::Bcc))
.addMBB(loopMBB).addImm(ARMCC::NE).addReg(ARM::CPSR);
BB->addSuccessor(loopMBB);
BB->addSuccessor(exitMBB);
BB = exitMBB;
MachineInstr *StartOfExit = exitMBB->begin();
unsigned srcIn = srcLoop;
unsigned destIn = destLoop;
for (unsigned i = 0; i < BytesLeft; i++) {
unsigned srcOut = MRI.createVirtualRegister(TRC);
unsigned destOut = MRI.createVirtualRegister(TRC);
unsigned scratch = MRI.createVirtualRegister(TRC);
emitPostLd(BB, StartOfExit, TII, dl, 1, scratch, srcIn, srcOut,
IsThumb1, IsThumb2);
emitPostSt(BB, StartOfExit, TII, dl, 1, scratch, destIn, destOut,
IsThumb1, IsThumb2);
srcIn = srcOut;
destIn = destOut;
}
MI->eraseFromParent(); return BB;
}
MachineBasicBlock *
ARMTargetLowering::EmitLowered__chkstk(MachineInstr *MI,
MachineBasicBlock *MBB) const {
const TargetMachine &TM = getTargetMachine();
const TargetInstrInfo &TII = *Subtarget->getInstrInfo();
DebugLoc DL = MI->getDebugLoc();
assert(Subtarget->isTargetWindows() &&
"__chkstk is only supported on Windows");
assert(Subtarget->isThumb2() && "Windows on ARM requires Thumb-2 mode");
switch (TM.getCodeModel()) {
case CodeModel::Small:
case CodeModel::Medium:
case CodeModel::Default:
case CodeModel::Kernel:
BuildMI(*MBB, MI, DL, TII.get(ARM::tBL))
.addImm((unsigned)ARMCC::AL).addReg(0)
.addExternalSymbol("__chkstk")
.addReg(ARM::R4, RegState::Implicit | RegState::Kill)
.addReg(ARM::R4, RegState::Implicit | RegState::Define)
.addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
break;
case CodeModel::Large:
case CodeModel::JITDefault: {
MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
unsigned Reg = MRI.createVirtualRegister(&ARM::rGPRRegClass);
BuildMI(*MBB, MI, DL, TII.get(ARM::t2MOVi32imm), Reg)
.addExternalSymbol("__chkstk");
BuildMI(*MBB, MI, DL, TII.get(ARM::tBLXr))
.addImm((unsigned)ARMCC::AL).addReg(0)
.addReg(Reg, RegState::Kill)
.addReg(ARM::R4, RegState::Implicit | RegState::Kill)
.addReg(ARM::R4, RegState::Implicit | RegState::Define)
.addReg(ARM::R12, RegState::Implicit | RegState::Define | RegState::Dead);
break;
}
}
AddDefaultCC(AddDefaultPred(BuildMI(*MBB, MI, DL, TII.get(ARM::t2SUBrr),
ARM::SP)
.addReg(ARM::SP).addReg(ARM::R4)));
MI->eraseFromParent();
return MBB;
}
MachineBasicBlock *
ARMTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = Subtarget->getInstrInfo();
DebugLoc dl = MI->getDebugLoc();
bool isThumb2 = Subtarget->isThumb2();
switch (MI->getOpcode()) {
default: {
MI->dump();
llvm_unreachable("Unexpected instr type to insert");
}
case ARM::t2STR_preidx:
MI->setDesc(TII->get(ARM::t2STR_PRE));
return BB;
case ARM::t2STRB_preidx:
MI->setDesc(TII->get(ARM::t2STRB_PRE));
return BB;
case ARM::t2STRH_preidx:
MI->setDesc(TII->get(ARM::t2STRH_PRE));
return BB;
case ARM::STRi_preidx:
case ARM::STRBi_preidx: {
unsigned NewOpc = MI->getOpcode() == ARM::STRi_preidx ?
ARM::STR_PRE_IMM : ARM::STRB_PRE_IMM;
unsigned Offset = MI->getOperand(4).getImm();
bool isSub = ARM_AM::getAM2Op(Offset) == ARM_AM::sub;
Offset = ARM_AM::getAM2Offset(Offset);
if (isSub)
Offset = -Offset;
MachineMemOperand *MMO = *MI->memoperands_begin();
BuildMI(*BB, MI, dl, TII->get(NewOpc))
.addOperand(MI->getOperand(0)) .addOperand(MI->getOperand(1)) .addOperand(MI->getOperand(2)) .addImm(Offset) .addOperand(MI->getOperand(5)) .addOperand(MI->getOperand(6))
.addMemOperand(MMO);
MI->eraseFromParent();
return BB;
}
case ARM::STRr_preidx:
case ARM::STRBr_preidx:
case ARM::STRH_preidx: {
unsigned NewOpc;
switch (MI->getOpcode()) {
default: llvm_unreachable("unexpected opcode!");
case ARM::STRr_preidx: NewOpc = ARM::STR_PRE_REG; break;
case ARM::STRBr_preidx: NewOpc = ARM::STRB_PRE_REG; break;
case ARM::STRH_preidx: NewOpc = ARM::STRH_PRE; break;
}
MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(NewOpc));
for (unsigned i = 0; i < MI->getNumOperands(); ++i)
MIB.addOperand(MI->getOperand(i));
MI->eraseFromParent();
return BB;
}
case ARM::tMOVCCr_pseudo: {
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = BB;
++It;
MachineBasicBlock *thisMBB = BB;
MachineFunction *F = BB->getParent();
MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, copy0MBB);
F->insert(It, sinkMBB);
sinkMBB->splice(sinkMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
BuildMI(BB, dl, TII->get(ARM::tBcc)).addMBB(sinkMBB)
.addImm(MI->getOperand(3).getImm()).addReg(MI->getOperand(4).getReg());
BB = copy0MBB;
BB->addSuccessor(sinkMBB);
BB = sinkMBB;
BuildMI(*BB, BB->begin(), dl,
TII->get(ARM::PHI), MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
.addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
MI->eraseFromParent(); return BB;
}
case ARM::BCCi64:
case ARM::BCCZi64: {
BB->erase(std::next(MachineBasicBlock::iterator(MI)), BB->end());
bool RHSisZero = MI->getOpcode() == ARM::BCCZi64;
unsigned LHS1 = MI->getOperand(1).getReg();
unsigned LHS2 = MI->getOperand(2).getReg();
if (RHSisZero) {
AddDefaultPred(BuildMI(BB, dl,
TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(LHS1).addImm(0));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(LHS2).addImm(0)
.addImm(ARMCC::EQ).addReg(ARM::CPSR);
} else {
unsigned RHS1 = MI->getOperand(3).getReg();
unsigned RHS2 = MI->getOperand(4).getReg();
AddDefaultPred(BuildMI(BB, dl,
TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr))
.addReg(LHS1).addReg(RHS1));
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2CMPrr : ARM::CMPrr))
.addReg(LHS2).addReg(RHS2)
.addImm(ARMCC::EQ).addReg(ARM::CPSR);
}
MachineBasicBlock *destMBB = MI->getOperand(RHSisZero ? 3 : 5).getMBB();
MachineBasicBlock *exitMBB = OtherSucc(BB, destMBB);
if (MI->getOperand(0).getImm() == ARMCC::NE)
std::swap(destMBB, exitMBB);
BuildMI(BB, dl, TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc))
.addMBB(destMBB).addImm(ARMCC::EQ).addReg(ARM::CPSR);
if (isThumb2)
AddDefaultPred(BuildMI(BB, dl, TII->get(ARM::t2B)).addMBB(exitMBB));
else
BuildMI(BB, dl, TII->get(ARM::B)) .addMBB(exitMBB);
MI->eraseFromParent(); return BB;
}
case ARM::Int_eh_sjlj_setjmp:
case ARM::Int_eh_sjlj_setjmp_nofp:
case ARM::tInt_eh_sjlj_setjmp:
case ARM::t2Int_eh_sjlj_setjmp:
case ARM::t2Int_eh_sjlj_setjmp_nofp:
EmitSjLjDispatchBlock(MI, BB);
return BB;
case ARM::ABS:
case ARM::t2ABS: {
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator BBI = BB;
++BBI;
MachineFunction *Fn = BB->getParent();
MachineBasicBlock *RSBBB = Fn->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *SinkBB = Fn->CreateMachineBasicBlock(LLVM_BB);
Fn->insert(BBI, RSBBB);
Fn->insert(BBI, SinkBB);
unsigned int ABSSrcReg = MI->getOperand(1).getReg();
unsigned int ABSDstReg = MI->getOperand(0).getReg();
bool isThumb2 = Subtarget->isThumb2();
MachineRegisterInfo &MRI = Fn->getRegInfo();
unsigned NewRsbDstReg =
MRI.createVirtualRegister(isThumb2 ? &ARM::rGPRRegClass : &ARM::GPRRegClass);
SinkBB->splice(SinkBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
SinkBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(RSBBB);
BB->addSuccessor(SinkBB);
RSBBB->addSuccessor(SinkBB);
AddDefaultPred(BuildMI(BB, dl,
TII->get(isThumb2 ? ARM::t2CMPri : ARM::CMPri))
.addReg(ABSSrcReg).addImm(0));
BuildMI(BB, dl,
TII->get(isThumb2 ? ARM::t2Bcc : ARM::Bcc)).addMBB(SinkBB)
.addImm(ARMCC::getOppositeCondition(ARMCC::MI)).addReg(ARM::CPSR);
BuildMI(*RSBBB, RSBBB->begin(), dl,
TII->get(isThumb2 ? ARM::t2RSBri : ARM::RSBri), NewRsbDstReg)
.addReg(ABSSrcReg, RegState::Kill)
.addImm(0).addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
BuildMI(*SinkBB, SinkBB->begin(), dl,
TII->get(ARM::PHI), ABSDstReg)
.addReg(NewRsbDstReg).addMBB(RSBBB)
.addReg(ABSSrcReg).addMBB(BB);
MI->eraseFromParent();
return SinkBB;
}
case ARM::COPY_STRUCT_BYVAL_I32:
++NumLoopByVals;
return EmitStructByval(MI, BB);
case ARM::WIN__CHKSTK:
return EmitLowered__chkstk(MI, BB);
}
}
void ARMTargetLowering::AdjustInstrPostInstrSelection(MachineInstr *MI,
SDNode *Node) const {
const MCInstrDesc *MCID = &MI->getDesc();
unsigned NewOpc = convertAddSubFlagsOpcode(MI->getOpcode());
if (NewOpc) {
const ARMBaseInstrInfo *TII = Subtarget->getInstrInfo();
MCID = &TII->get(NewOpc);
assert(MCID->getNumOperands() == MI->getDesc().getNumOperands() + 1 &&
"converted opcode should be the same except for cc_out");
MI->setDesc(*MCID);
MI->addOperand(MachineOperand::CreateReg(0, true));
}
unsigned ccOutIdx = MCID->getNumOperands() - 1;
if (!MI->hasOptionalDef() || !MCID->OpInfo[ccOutIdx].isOptionalDef()) {
assert(!NewOpc && "Optional cc_out operand required");
return;
}
bool definesCPSR = false;
bool deadCPSR = false;
for (unsigned i = MCID->getNumOperands(), e = MI->getNumOperands();
i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isDef() && MO.getReg() == ARM::CPSR) {
definesCPSR = true;
if (MO.isDead())
deadCPSR = true;
MI->RemoveOperand(i);
break;
}
}
if (!definesCPSR) {
assert(!NewOpc && "Optional cc_out operand required");
return;
}
assert(deadCPSR == !Node->hasAnyUseOfValue(1) && "inconsistent dead flag");
if (deadCPSR) {
assert(!MI->getOperand(ccOutIdx).getReg() &&
"expect uninitialized optional cc_out operand");
return;
}
MachineOperand &MO = MI->getOperand(ccOutIdx);
MO.setReg(ARM::CPSR);
MO.setIsDef(true);
}
static inline bool isZeroOrAllOnes(SDValue N, bool AllOnes) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(N);
if (!C)
return false;
return AllOnes ? C->isAllOnesValue() : C->isNullValue();
}
static bool isConditionalZeroOrAllOnes(SDNode *N, bool AllOnes,
SDValue &CC, bool &Invert,
SDValue &OtherOp,
SelectionDAG &DAG) {
switch (N->getOpcode()) {
default: return false;
case ISD::SELECT: {
CC = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue N2 = N->getOperand(2);
if (isZeroOrAllOnes(N1, AllOnes)) {
Invert = false;
OtherOp = N2;
return true;
}
if (isZeroOrAllOnes(N2, AllOnes)) {
Invert = true;
OtherOp = N1;
return true;
}
return false;
}
case ISD::ZERO_EXTEND:
if (AllOnes)
return false;
case ISD::SIGN_EXTEND: {
EVT VT = N->getValueType(0);
CC = N->getOperand(0);
if (CC.getValueType() != MVT::i1)
return false;
Invert = !AllOnes;
if (AllOnes)
OtherOp = DAG.getConstant(0, VT);
else if (N->getOpcode() == ISD::ZERO_EXTEND)
OtherOp = DAG.getConstant(1, VT);
else
OtherOp = DAG.getConstant(APInt::getAllOnesValue(VT.getSizeInBits()), VT);
return true;
}
}
}
static
SDValue combineSelectAndUse(SDNode *N, SDValue Slct, SDValue OtherOp,
TargetLowering::DAGCombinerInfo &DCI,
bool AllOnes = false) {
SelectionDAG &DAG = DCI.DAG;
EVT VT = N->getValueType(0);
SDValue NonConstantVal;
SDValue CCOp;
bool SwapSelectOps;
if (!isConditionalZeroOrAllOnes(Slct.getNode(), AllOnes, CCOp, SwapSelectOps,
NonConstantVal, DAG))
return SDValue();
SDValue TrueVal = OtherOp;
SDValue FalseVal = DAG.getNode(N->getOpcode(), SDLoc(N), VT,
OtherOp, NonConstantVal);
if (SwapSelectOps)
std::swap(TrueVal, FalseVal);
return DAG.getNode(ISD::SELECT, SDLoc(N), VT,
CCOp, TrueVal, FalseVal);
}
static
SDValue combineSelectAndUseCommutative(SDNode *N, bool AllOnes,
TargetLowering::DAGCombinerInfo &DCI) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N0.getNode()->hasOneUse()) {
SDValue Result = combineSelectAndUse(N, N0, N1, DCI, AllOnes);
if (Result.getNode())
return Result;
}
if (N1.getNode()->hasOneUse()) {
SDValue Result = combineSelectAndUse(N, N1, N0, DCI, AllOnes);
if (Result.getNode())
return Result;
}
return SDValue();
}
static SDValue AddCombineToVPADDL(SDNode *N, SDValue N0, SDValue N1,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
if (DCI.isBeforeLegalize() || !Subtarget->hasNEON()
|| N0.getOpcode() != ISD::BUILD_VECTOR
|| N1.getOpcode() != ISD::BUILD_VECTOR)
return SDValue();
EVT VT = N->getValueType(0);
if (!VT.isInteger() || VT.getVectorElementType() == MVT::i64)
return SDValue();
if (N0->getOperand(0)->getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return SDValue();
SDValue Vec = N0->getOperand(0)->getOperand(0);
SDNode *V = Vec.getNode();
unsigned nextIndex = 0;
for (unsigned i = 0, e = N0->getNumOperands(); i != e; ++i) {
if (N0->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT
&& N1->getOperand(i)->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
SDValue ExtVec0 = N0->getOperand(i);
SDValue ExtVec1 = N1->getOperand(i);
if (V != ExtVec0->getOperand(0).getNode() ||
V != ExtVec1->getOperand(0).getNode())
return SDValue();
ConstantSDNode *C0 = dyn_cast<ConstantSDNode>(ExtVec0->getOperand(1));
ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(ExtVec1->getOperand(1));
if (!C0 || !C1 || C0->getZExtValue() != nextIndex
|| C1->getZExtValue() != nextIndex+1)
return SDValue();
nextIndex+=2;
} else
return SDValue();
}
SelectionDAG &DAG = DCI.DAG;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SmallVector<SDValue, 8> Ops;
Ops.push_back(DAG.getConstant(Intrinsic::arm_neon_vpaddls,
TLI.getPointerTy()));
Ops.push_back(Vec);
MVT widenType;
unsigned numElem = VT.getVectorNumElements();
EVT inputLaneType = Vec.getValueType().getVectorElementType();
switch (inputLaneType.getSimpleVT().SimpleTy) {
case MVT::i8: widenType = MVT::getVectorVT(MVT::i16, numElem); break;
case MVT::i16: widenType = MVT::getVectorVT(MVT::i32, numElem); break;
case MVT::i32: widenType = MVT::getVectorVT(MVT::i64, numElem); break;
default:
llvm_unreachable("Invalid vector element type for padd optimization.");
}
SDValue tmp = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N), widenType, Ops);
unsigned ExtOp = VT.bitsGT(tmp.getValueType()) ? ISD::ANY_EXTEND : ISD::TRUNCATE;
return DAG.getNode(ExtOp, SDLoc(N), VT, tmp);
}
static SDValue findMUL_LOHI(SDValue V) {
if (V->getOpcode() == ISD::UMUL_LOHI ||
V->getOpcode() == ISD::SMUL_LOHI)
return V;
return SDValue();
}
static SDValue AddCombineTo64bitMLAL(SDNode *AddcNode,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
if (Subtarget->isThumb1Only()) return SDValue();
if (DCI.isBeforeLegalize()) return SDValue();
assert(AddcNode->getOpcode() == ISD::ADDC && "Expect an ADDC");
SDValue AddcOp0 = AddcNode->getOperand(0);
SDValue AddcOp1 = AddcNode->getOperand(1);
if (AddcOp0.getNode() == AddcOp1.getNode())
return SDValue();
assert(AddcNode->getNumValues() == 2 &&
AddcNode->getValueType(0) == MVT::i32 &&
"Expect ADDC with two result values. First: i32");
if (AddcNode->getValueType(1) != MVT::Glue)
return SDValue();
if (AddcOp0->getOpcode() != ISD::UMUL_LOHI &&
AddcOp0->getOpcode() != ISD::SMUL_LOHI &&
AddcOp1->getOpcode() != ISD::UMUL_LOHI &&
AddcOp1->getOpcode() != ISD::SMUL_LOHI)
return SDValue();
SDNode* AddeNode = AddcNode->getGluedUser();
if (!AddeNode)
return SDValue();
if (AddeNode->getOpcode() != ISD::ADDE)
return SDValue();
assert(AddeNode->getNumOperands() == 3 &&
AddeNode->getOperand(2).getValueType() == MVT::Glue &&
"ADDE node has the wrong inputs");
SDValue AddeOp0 = AddeNode->getOperand(0);
SDValue AddeOp1 = AddeNode->getOperand(1);
if (AddeOp0.getNode() == AddeOp1.getNode())
return SDValue();
bool IsLeftOperandMUL = false;
SDValue MULOp = findMUL_LOHI(AddeOp0);
if (MULOp == SDValue())
MULOp = findMUL_LOHI(AddeOp1);
else
IsLeftOperandMUL = true;
if (MULOp == SDValue())
return SDValue();
unsigned Opc = MULOp->getOpcode();
unsigned FinalOpc = (Opc == ISD::SMUL_LOHI) ? ARMISD::SMLAL : ARMISD::UMLAL;
SDValue* HiAdd = nullptr;
SDValue* LoMul = nullptr;
SDValue* LowAdd = nullptr;
if ((AddeOp0 != MULOp.getValue(1)) && (AddeOp1 != MULOp.getValue(1)))
return SDValue();
if (IsLeftOperandMUL)
HiAdd = &AddeOp1;
else
HiAdd = &AddeOp0;
if (AddcOp0 == MULOp.getValue(0)) {
LoMul = &AddcOp0;
LowAdd = &AddcOp1;
}
if (AddcOp1 == MULOp.getValue(0)) {
LoMul = &AddcOp1;
LowAdd = &AddcOp0;
}
if (!LoMul)
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SmallVector<SDValue, 8> Ops;
Ops.push_back(LoMul->getOperand(0));
Ops.push_back(LoMul->getOperand(1));
Ops.push_back(*LowAdd);
Ops.push_back(*HiAdd);
SDValue MLALNode = DAG.getNode(FinalOpc, SDLoc(AddcNode),
DAG.getVTList(MVT::i32, MVT::i32), Ops);
SDValue HiMLALResult(MLALNode.getNode(), 1);
DAG.ReplaceAllUsesOfValueWith(SDValue(AddeNode, 0), HiMLALResult);
SDValue LoMLALResult(MLALNode.getNode(), 0);
DAG.ReplaceAllUsesOfValueWith(SDValue(AddcNode, 0), LoMLALResult);
SDValue resNode(AddcNode, 0);
return resNode;
}
static SDValue PerformADDCCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
return AddCombineTo64bitMLAL(N, DCI, Subtarget);
}
static SDValue PerformADDCombineWithOperands(SDNode *N, SDValue N0, SDValue N1,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget){
SDValue Result = AddCombineToVPADDL(N, N0, N1, DCI, Subtarget);
if (Result.getNode())
return Result;
if (N0.getNode()->hasOneUse()) {
SDValue Result = combineSelectAndUse(N, N0, N1, DCI);
if (Result.getNode()) return Result;
}
return SDValue();
}
static SDValue PerformADDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue Result = PerformADDCombineWithOperands(N, N0, N1, DCI, Subtarget);
if (Result.getNode())
return Result;
return PerformADDCombineWithOperands(N, N1, N0, DCI, Subtarget);
}
static SDValue PerformSUBCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N1.getNode()->hasOneUse()) {
SDValue Result = combineSelectAndUse(N, N1, N0, DCI);
if (Result.getNode()) return Result;
}
return SDValue();
}
static SDValue PerformVMULCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
if (!Subtarget->hasVMLxForwarding())
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
unsigned Opcode = N0.getOpcode();
if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
Opcode != ISD::FADD && Opcode != ISD::FSUB) {
Opcode = N1.getOpcode();
if (Opcode != ISD::ADD && Opcode != ISD::SUB &&
Opcode != ISD::FADD && Opcode != ISD::FSUB)
return SDValue();
std::swap(N0, N1);
}
if (N0 == N1)
return SDValue();
EVT VT = N->getValueType(0);
SDLoc DL(N);
SDValue N00 = N0->getOperand(0);
SDValue N01 = N0->getOperand(1);
return DAG.getNode(Opcode, DL, VT,
DAG.getNode(ISD::MUL, DL, VT, N00, N1),
DAG.getNode(ISD::MUL, DL, VT, N01, N1));
}
static SDValue PerformMULCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (Subtarget->isThumb1Only())
return SDValue();
if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
return SDValue();
EVT VT = N->getValueType(0);
if (VT.is64BitVector() || VT.is128BitVector())
return PerformVMULCombine(N, DCI, Subtarget);
if (VT != MVT::i32)
return SDValue();
ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (!C)
return SDValue();
int64_t MulAmt = C->getSExtValue();
unsigned ShiftAmt = countTrailingZeros<uint64_t>(MulAmt);
ShiftAmt = ShiftAmt & (32 - 1);
SDValue V = N->getOperand(0);
SDLoc DL(N);
SDValue Res;
MulAmt >>= ShiftAmt;
if (MulAmt >= 0) {
if (isPowerOf2_32(MulAmt - 1)) {
Res = DAG.getNode(ISD::ADD, DL, VT,
V,
DAG.getNode(ISD::SHL, DL, VT,
V,
DAG.getConstant(Log2_32(MulAmt - 1),
MVT::i32)));
} else if (isPowerOf2_32(MulAmt + 1)) {
Res = DAG.getNode(ISD::SUB, DL, VT,
DAG.getNode(ISD::SHL, DL, VT,
V,
DAG.getConstant(Log2_32(MulAmt + 1),
MVT::i32)),
V);
} else
return SDValue();
} else {
uint64_t MulAmtAbs = -MulAmt;
if (isPowerOf2_32(MulAmtAbs + 1)) {
Res = DAG.getNode(ISD::SUB, DL, VT,
V,
DAG.getNode(ISD::SHL, DL, VT,
V,
DAG.getConstant(Log2_32(MulAmtAbs + 1),
MVT::i32)));
} else if (isPowerOf2_32(MulAmtAbs - 1)) {
Res = DAG.getNode(ISD::ADD, DL, VT,
V,
DAG.getNode(ISD::SHL, DL, VT,
V,
DAG.getConstant(Log2_32(MulAmtAbs-1),
MVT::i32)));
Res = DAG.getNode(ISD::SUB, DL, VT,
DAG.getConstant(0, MVT::i32),Res);
} else
return SDValue();
}
if (ShiftAmt != 0)
Res = DAG.getNode(ISD::SHL, DL, VT,
Res, DAG.getConstant(ShiftAmt, MVT::i32));
DCI.CombineTo(N, Res, false);
return SDValue();
}
static SDValue PerformANDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
SDLoc dl(N);
EVT VT = N->getValueType(0);
SelectionDAG &DAG = DCI.DAG;
if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
return SDValue();
APInt SplatBits, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (BVN &&
BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
if (SplatBitSize <= 64) {
EVT VbicVT;
SDValue Val = isNEONModifiedImm((~SplatBits).getZExtValue(),
SplatUndef.getZExtValue(), SplatBitSize,
DAG, VbicVT, VT.is128BitVector(),
OtherModImm);
if (Val.getNode()) {
SDValue Input =
DAG.getNode(ISD::BITCAST, dl, VbicVT, N->getOperand(0));
SDValue Vbic = DAG.getNode(ARMISD::VBICIMM, dl, VbicVT, Input, Val);
return DAG.getNode(ISD::BITCAST, dl, VT, Vbic);
}
}
}
if (!Subtarget->isThumb1Only()) {
SDValue Result = combineSelectAndUseCommutative(N, true, DCI);
if (Result.getNode())
return Result;
}
return SDValue();
}
static SDValue PerformORCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
SDLoc dl(N);
EVT VT = N->getValueType(0);
SelectionDAG &DAG = DCI.DAG;
if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
return SDValue();
APInt SplatBits, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (BVN && Subtarget->hasNEON() &&
BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
if (SplatBitSize <= 64) {
EVT VorrVT;
SDValue Val = isNEONModifiedImm(SplatBits.getZExtValue(),
SplatUndef.getZExtValue(), SplatBitSize,
DAG, VorrVT, VT.is128BitVector(),
OtherModImm);
if (Val.getNode()) {
SDValue Input =
DAG.getNode(ISD::BITCAST, dl, VorrVT, N->getOperand(0));
SDValue Vorr = DAG.getNode(ARMISD::VORRIMM, dl, VorrVT, Input, Val);
return DAG.getNode(ISD::BITCAST, dl, VT, Vorr);
}
}
}
if (!Subtarget->isThumb1Only()) {
SDValue Result = combineSelectAndUseCommutative(N, false, DCI);
if (Result.getNode())
return Result;
}
SDValue N0 = N->getOperand(0);
if (N0.getOpcode() != ISD::AND || !N0.hasOneUse())
return SDValue();
SDValue N1 = N->getOperand(1);
if (Subtarget->hasNEON() && N1.getOpcode() == ISD::AND && VT.isVector() &&
DAG.getTargetLoweringInfo().isTypeLegal(VT)) {
APInt SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
APInt SplatBits0, SplatBits1;
BuildVectorSDNode *BVN0 = dyn_cast<BuildVectorSDNode>(N0->getOperand(1));
BuildVectorSDNode *BVN1 = dyn_cast<BuildVectorSDNode>(N1->getOperand(1));
if (BVN0 && BVN0->isConstantSplat(SplatBits0, SplatUndef, SplatBitSize,
HasAnyUndefs) && !HasAnyUndefs) {
if (BVN1 && BVN1->isConstantSplat(SplatBits1, SplatUndef, SplatBitSize,
HasAnyUndefs) && !HasAnyUndefs) {
if (SplatBits0.getBitWidth() == SplatBits1.getBitWidth() &&
SplatBits0 == ~SplatBits1) {
EVT CanonicalVT = VT.is128BitVector() ? MVT::v4i32 : MVT::v2i32;
SDValue Result = DAG.getNode(ARMISD::VBSL, dl, CanonicalVT,
N0->getOperand(1),
N0->getOperand(0),
N1->getOperand(0));
return DAG.getNode(ISD::BITCAST, dl, VT, Result);
}
}
}
}
if (Subtarget->isThumb1Only() || !Subtarget->hasV6T2Ops())
return SDValue();
SDLoc DL(N);
if (VT != MVT::i32)
return SDValue();
SDValue N00 = N0.getOperand(0);
SDValue MaskOp = N0.getOperand(1);
ConstantSDNode *MaskC = dyn_cast<ConstantSDNode>(MaskOp);
if (!MaskC)
return SDValue();
unsigned Mask = MaskC->getZExtValue();
if (Mask == 0xffff)
return SDValue();
SDValue Res;
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
if (N1C) {
unsigned Val = N1C->getZExtValue();
if ((Val & ~Mask) != Val)
return SDValue();
if (ARM::isBitFieldInvertedMask(Mask)) {
Val >>= countTrailingZeros(~Mask);
Res = DAG.getNode(ARMISD::BFI, DL, VT, N00,
DAG.getConstant(Val, MVT::i32),
DAG.getConstant(Mask, MVT::i32));
DCI.CombineTo(N, Res, false);
return SDValue();
}
} else if (N1.getOpcode() == ISD::AND) {
ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1));
if (!N11C)
return SDValue();
unsigned Mask2 = N11C->getZExtValue();
if (ARM::isBitFieldInvertedMask(Mask) &&
(Mask == ~Mask2)) {
if (Subtarget->hasT2ExtractPack() &&
(Mask == 0xffff || Mask == 0xffff0000))
return SDValue();
unsigned amt = countTrailingZeros(Mask2);
Res = DAG.getNode(ISD::SRL, DL, VT, N1.getOperand(0),
DAG.getConstant(amt, MVT::i32));
Res = DAG.getNode(ARMISD::BFI, DL, VT, N00, Res,
DAG.getConstant(Mask, MVT::i32));
DCI.CombineTo(N, Res, false);
return SDValue();
} else if (ARM::isBitFieldInvertedMask(~Mask) &&
(~Mask == Mask2)) {
if (Subtarget->hasT2ExtractPack() &&
(Mask2 == 0xffff || Mask2 == 0xffff0000))
return SDValue();
unsigned lsb = countTrailingZeros(Mask);
Res = DAG.getNode(ISD::SRL, DL, VT, N00,
DAG.getConstant(lsb, MVT::i32));
Res = DAG.getNode(ARMISD::BFI, DL, VT, N1.getOperand(0), Res,
DAG.getConstant(Mask2, MVT::i32));
DCI.CombineTo(N, Res, false);
return SDValue();
}
}
if (DAG.MaskedValueIsZero(N1, MaskC->getAPIntValue()) &&
N00.getOpcode() == ISD::SHL && isa<ConstantSDNode>(N00.getOperand(1)) &&
ARM::isBitFieldInvertedMask(~Mask)) {
SDValue ShAmt = N00.getOperand(1);
unsigned ShAmtC = cast<ConstantSDNode>(ShAmt)->getZExtValue();
unsigned LSB = countTrailingZeros(Mask);
if (ShAmtC != LSB)
return SDValue();
Res = DAG.getNode(ARMISD::BFI, DL, VT, N1, N00.getOperand(0),
DAG.getConstant(~Mask, MVT::i32));
DCI.CombineTo(N, Res, false);
}
return SDValue();
}
static SDValue PerformXORCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
EVT VT = N->getValueType(0);
SelectionDAG &DAG = DCI.DAG;
if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
return SDValue();
if (!Subtarget->isThumb1Only()) {
SDValue Result = combineSelectAndUseCommutative(N, false, DCI);
if (Result.getNode())
return Result;
}
return SDValue();
}
static SDValue PerformBFICombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SDValue N1 = N->getOperand(1);
if (N1.getOpcode() == ISD::AND) {
ConstantSDNode *N11C = dyn_cast<ConstantSDNode>(N1.getOperand(1));
if (!N11C)
return SDValue();
unsigned InvMask = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue();
unsigned LSB = countTrailingZeros(~InvMask);
unsigned Width = (32 - countLeadingZeros(~InvMask)) - LSB;
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),
N->getOperand(0), N1.getOperand(0),
N->getOperand(2));
}
return SDValue();
}
static SDValue PerformVMOVRRDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SDValue InDouble = N->getOperand(0);
if (InDouble.getOpcode() == ARMISD::VMOVDRR && !Subtarget->isFPOnlySP())
return DCI.CombineTo(N, InDouble.getOperand(0), InDouble.getOperand(1));
SDNode *InNode = InDouble.getNode();
if (ISD::isNormalLoad(InNode) && InNode->hasOneUse() &&
InNode->getValueType(0) == MVT::f64 &&
InNode->getOperand(1).getOpcode() == ISD::FrameIndex &&
!cast<LoadSDNode>(InNode)->isVolatile()) {
LoadSDNode *LD = cast<LoadSDNode>(InNode);
SelectionDAG &DAG = DCI.DAG;
SDLoc DL(LD);
SDValue BasePtr = LD->getBasePtr();
SDValue NewLD1 = DAG.getLoad(MVT::i32, DL, LD->getChain(), BasePtr,
LD->getPointerInfo(), LD->isVolatile(),
LD->isNonTemporal(), LD->isInvariant(),
LD->getAlignment());
SDValue OffsetPtr = DAG.getNode(ISD::ADD, DL, MVT::i32, BasePtr,
DAG.getConstant(4, MVT::i32));
SDValue NewLD2 = DAG.getLoad(MVT::i32, DL, NewLD1.getValue(1), OffsetPtr,
LD->getPointerInfo(), LD->isVolatile(),
LD->isNonTemporal(), LD->isInvariant(),
std::min(4U, LD->getAlignment() / 2));
DAG.ReplaceAllUsesOfValueWith(SDValue(LD, 1), NewLD2.getValue(1));
if (DCI.DAG.getTargetLoweringInfo().isBigEndian())
std::swap (NewLD1, NewLD2);
SDValue Result = DCI.CombineTo(N, NewLD1, NewLD2);
return Result;
}
return SDValue();
}
static SDValue PerformVMOVDRRCombine(SDNode *N, SelectionDAG &DAG) {
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
if (Op0.getOpcode() == ISD::BITCAST)
Op0 = Op0.getOperand(0);
if (Op1.getOpcode() == ISD::BITCAST)
Op1 = Op1.getOperand(0);
if (Op0.getOpcode() == ARMISD::VMOVRRD &&
Op0.getNode() == Op1.getNode() &&
Op0.getResNo() == 0 && Op1.getResNo() == 1)
return DAG.getNode(ISD::BITCAST, SDLoc(N),
N->getValueType(0), Op0.getOperand(0));
return SDValue();
}
static bool hasNormalLoadOperand(SDNode *N) {
unsigned NumElts = N->getValueType(0).getVectorNumElements();
for (unsigned i = 0; i < NumElts; ++i) {
SDNode *Elt = N->getOperand(i).getNode();
if (ISD::isNormalLoad(Elt) && !cast<LoadSDNode>(Elt)->isVolatile())
return true;
}
return false;
}
static SDValue PerformBUILD_VECTORCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (N->getNumOperands() == 2) {
SDValue RV = PerformVMOVDRRCombine(N, DAG);
if (RV.getNode())
return RV;
}
EVT VT = N->getValueType(0);
if (VT.getVectorElementType() != MVT::i64 || !hasNormalLoadOperand(N))
return SDValue();
SDLoc dl(N);
SmallVector<SDValue, 8> Ops;
unsigned NumElts = VT.getVectorNumElements();
for (unsigned i = 0; i < NumElts; ++i) {
SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::f64, N->getOperand(i));
Ops.push_back(V);
DCI.AddToWorklist(V.getNode());
}
EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64, NumElts);
SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, FloatVT, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, BV);
}
static SDValue
PerformARMBUILD_VECTORCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
EVT VT = N->getValueType(0);
EVT EltVT = VT.getVectorElementType();
if (EltVT.getSizeInBits() != 32 || !N->hasOneUse())
return SDValue();
assert(EltVT == MVT::f32 && "Unexpected type!");
SDNode *Use = *N->use_begin();
if (Use->getOpcode() != ISD::BITCAST ||
Use->getValueType(0).isFloatingPoint())
return SDValue();
unsigned NumOfBitCastedElts = 0;
unsigned NumElts = VT.getVectorNumElements();
unsigned NumOfRelevantElts = NumElts;
for (unsigned Idx = 0; Idx < NumElts; ++Idx) {
SDValue Elt = N->getOperand(Idx);
if (Elt->getOpcode() == ISD::BITCAST) {
if (Elt->getOperand(0).getValueType() == MVT::i32)
++NumOfBitCastedElts;
} else if (Elt.getOpcode() == ISD::UNDEF || isa<ConstantSDNode>(Elt))
--NumOfRelevantElts;
}
if (NumOfBitCastedElts <= NumOfRelevantElts / 2)
return SDValue();
SelectionDAG &DAG = DCI.DAG;
EVT VecVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumElts);
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (!TLI.isTypeLegal(VecVT))
return SDValue();
SDValue Vec = DAG.getUNDEF(VecVT);
SDLoc dl(N);
for (unsigned Idx = 0 ; Idx < NumElts; ++Idx) {
SDValue V = N->getOperand(Idx);
if (V.getOpcode() == ISD::UNDEF)
continue;
if (V.getOpcode() == ISD::BITCAST &&
V->getOperand(0).getValueType() == MVT::i32)
V = V.getOperand(0);
else {
V = DAG.getNode(ISD::BITCAST, SDLoc(V), MVT::i32, V);
DCI.AddToWorklist(V.getNode());
}
SDValue LaneIdx = DAG.getConstant(Idx, MVT::i32);
Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VecVT, Vec, V, LaneIdx);
}
Vec = DAG.getNode(ISD::BITCAST, dl, VT, Vec);
DCI.AddToWorklist(Vec.getNode());
return Vec;
}
static SDValue PerformInsertEltCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
EVT VT = N->getValueType(0);
SDNode *Elt = N->getOperand(1).getNode();
if (VT.getVectorElementType() != MVT::i64 ||
!ISD::isNormalLoad(Elt) || cast<LoadSDNode>(Elt)->isVolatile())
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
EVT FloatVT = EVT::getVectorVT(*DAG.getContext(), MVT::f64,
VT.getVectorNumElements());
SDValue Vec = DAG.getNode(ISD::BITCAST, dl, FloatVT, N->getOperand(0));
SDValue V = DAG.getNode(ISD::BITCAST, dl, MVT::f64, N->getOperand(1));
DCI.AddToWorklist(Vec.getNode());
DCI.AddToWorklist(V.getNode());
SDValue InsElt = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, FloatVT,
Vec, V, N->getOperand(2));
return DAG.getNode(ISD::BITCAST, dl, VT, InsElt);
}
static SDValue PerformVECTOR_SHUFFLECombine(SDNode *N, SelectionDAG &DAG) {
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
if (Op0.getOpcode() != ISD::CONCAT_VECTORS ||
Op1.getOpcode() != ISD::CONCAT_VECTORS ||
Op0.getNumOperands() != 2 ||
Op1.getNumOperands() != 2)
return SDValue();
SDValue Concat0Op1 = Op0.getOperand(1);
SDValue Concat1Op1 = Op1.getOperand(1);
if (Concat0Op1.getOpcode() != ISD::UNDEF ||
Concat1Op1.getOpcode() != ISD::UNDEF)
return SDValue();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
EVT VT = N->getValueType(0);
if (!TLI.isTypeLegal(VT) ||
!TLI.isTypeLegal(Concat0Op1.getValueType()) ||
!TLI.isTypeLegal(Concat1Op1.getValueType()))
return SDValue();
SDValue NewConcat = DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(N), VT,
Op0.getOperand(0), Op1.getOperand(0));
SmallVector<int, 16> NewMask;
unsigned NumElts = VT.getVectorNumElements();
unsigned HalfElts = NumElts/2;
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N);
for (unsigned n = 0; n < NumElts; ++n) {
int MaskElt = SVN->getMaskElt(n);
int NewElt = -1;
if (MaskElt < (int)HalfElts)
NewElt = MaskElt;
else if (MaskElt >= (int)NumElts && MaskElt < (int)(NumElts + HalfElts))
NewElt = HalfElts + MaskElt - NumElts;
NewMask.push_back(NewElt);
}
return DAG.getVectorShuffle(VT, SDLoc(N), NewConcat,
DAG.getUNDEF(VT), NewMask.data());
}
static SDValue CombineBaseUpdate(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
const bool isIntrinsic = (N->getOpcode() == ISD::INTRINSIC_VOID ||
N->getOpcode() == ISD::INTRINSIC_W_CHAIN);
const bool isStore = N->getOpcode() == ISD::STORE;
const unsigned AddrOpIdx = ((isIntrinsic || isStore) ? 2 : 1);
SDValue Addr = N->getOperand(AddrOpIdx);
MemSDNode *MemN = cast<MemSDNode>(N);
for (SDNode::use_iterator UI = Addr.getNode()->use_begin(),
UE = Addr.getNode()->use_end(); UI != UE; ++UI) {
SDNode *User = *UI;
if (User->getOpcode() != ISD::ADD ||
UI.getUse().getResNo() != Addr.getResNo())
continue;
if (User->isPredecessorOf(N) || N->isPredecessorOf(User))
continue;
bool isLoadOp = true;
bool isLaneOp = false;
unsigned NewOpc = 0;
unsigned NumVecs = 0;
if (isIntrinsic) {
unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
switch (IntNo) {
default: llvm_unreachable("unexpected intrinsic for Neon base update");
case Intrinsic::arm_neon_vld1: NewOpc = ARMISD::VLD1_UPD;
NumVecs = 1; break;
case Intrinsic::arm_neon_vld2: NewOpc = ARMISD::VLD2_UPD;
NumVecs = 2; break;
case Intrinsic::arm_neon_vld3: NewOpc = ARMISD::VLD3_UPD;
NumVecs = 3; break;
case Intrinsic::arm_neon_vld4: NewOpc = ARMISD::VLD4_UPD;
NumVecs = 4; break;
case Intrinsic::arm_neon_vld2lane: NewOpc = ARMISD::VLD2LN_UPD;
NumVecs = 2; isLaneOp = true; break;
case Intrinsic::arm_neon_vld3lane: NewOpc = ARMISD::VLD3LN_UPD;
NumVecs = 3; isLaneOp = true; break;
case Intrinsic::arm_neon_vld4lane: NewOpc = ARMISD::VLD4LN_UPD;
NumVecs = 4; isLaneOp = true; break;
case Intrinsic::arm_neon_vst1: NewOpc = ARMISD::VST1_UPD;
NumVecs = 1; isLoadOp = false; break;
case Intrinsic::arm_neon_vst2: NewOpc = ARMISD::VST2_UPD;
NumVecs = 2; isLoadOp = false; break;
case Intrinsic::arm_neon_vst3: NewOpc = ARMISD::VST3_UPD;
NumVecs = 3; isLoadOp = false; break;
case Intrinsic::arm_neon_vst4: NewOpc = ARMISD::VST4_UPD;
NumVecs = 4; isLoadOp = false; break;
case Intrinsic::arm_neon_vst2lane: NewOpc = ARMISD::VST2LN_UPD;
NumVecs = 2; isLoadOp = false; isLaneOp = true; break;
case Intrinsic::arm_neon_vst3lane: NewOpc = ARMISD::VST3LN_UPD;
NumVecs = 3; isLoadOp = false; isLaneOp = true; break;
case Intrinsic::arm_neon_vst4lane: NewOpc = ARMISD::VST4LN_UPD;
NumVecs = 4; isLoadOp = false; isLaneOp = true; break;
}
} else {
isLaneOp = true;
switch (N->getOpcode()) {
default: llvm_unreachable("unexpected opcode for Neon base update");
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; isLaneOp = false; isLoadOp = false; break;
}
}
EVT VecTy;
if (isLoadOp) {
VecTy = N->getValueType(0);
} else if (isIntrinsic) {
VecTy = N->getOperand(AddrOpIdx+1).getValueType();
} else {
assert(isStore && "Node has to be a load, a store, or an intrinsic!");
VecTy = N->getOperand(1).getValueType();
}
unsigned NumBytes = NumVecs * VecTy.getSizeInBits() / 8;
if (isLaneOp)
NumBytes /= VecTy.getVectorNumElements();
SDValue Inc = User->getOperand(User->getOperand(0) == Addr ? 1 : 0);
if (ConstantSDNode *CInc = dyn_cast<ConstantSDNode>(Inc.getNode())) {
uint64_t IncVal = CInc->getZExtValue();
if (IncVal != NumBytes)
continue;
} else if (NumBytes >= 3 * 16) {
continue;
}
EVT AlignedVecTy = VecTy;
unsigned Alignment = MemN->getAlignment();
if (isa<LSBaseSDNode>(N)) {
if (Alignment == 0)
Alignment = 1;
if (Alignment < VecTy.getScalarSizeInBits() / 8) {
MVT EltTy = MVT::getIntegerVT(Alignment * 8);
assert(NumVecs == 1 && "Unexpected multi-element generic load/store.");
assert(!isLaneOp && "Unexpected generic load/store lane.");
unsigned NumElts = NumBytes / (EltTy.getSizeInBits() / 8);
AlignedVecTy = MVT::getVectorVT(EltTy, NumElts);
}
Alignment = 1;
}
EVT Tys[6];
unsigned NumResultVecs = (isLoadOp ? NumVecs : 0);
unsigned n;
for (n = 0; n < NumResultVecs; ++n)
Tys[n] = AlignedVecTy;
Tys[n++] = MVT::i32;
Tys[n] = MVT::Other;
SDVTList SDTys = DAG.getVTList(makeArrayRef(Tys, NumResultVecs+2));
SmallVector<SDValue, 8> Ops;
Ops.push_back(N->getOperand(0)); Ops.push_back(N->getOperand(AddrOpIdx));
Ops.push_back(Inc);
if (StoreSDNode *StN = dyn_cast<StoreSDNode>(N)) {
Ops.push_back(StN->getValue());
} else {
for (unsigned i = AddrOpIdx + 1; i < N->getNumOperands() - 1; ++i)
Ops.push_back(N->getOperand(i));
}
Ops.push_back(DAG.getConstant(Alignment, MVT::i32));
if (AlignedVecTy != VecTy && N->getOpcode() == ISD::STORE) {
SDValue &StVal = Ops[Ops.size()-2];
StVal = DAG.getNode(ISD::BITCAST, SDLoc(N), AlignedVecTy, StVal);
}
SDValue UpdN = DAG.getMemIntrinsicNode(NewOpc, SDLoc(N), SDTys,
Ops, AlignedVecTy,
MemN->getMemOperand());
SmallVector<SDValue, 5> NewResults;
for (unsigned i = 0; i < NumResultVecs; ++i)
NewResults.push_back(SDValue(UpdN.getNode(), i));
if (AlignedVecTy != VecTy && N->getOpcode() == ISD::LOAD) {
SDValue &LdVal = NewResults[0];
LdVal = DAG.getNode(ISD::BITCAST, SDLoc(N), VecTy, LdVal);
}
NewResults.push_back(SDValue(UpdN.getNode(), NumResultVecs+1)); DCI.CombineTo(N, NewResults);
DCI.CombineTo(User, SDValue(UpdN.getNode(), NumResultVecs));
break;
}
return SDValue();
}
static SDValue PerformVLDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
return SDValue();
return CombineBaseUpdate(N, DCI);
}
static bool CombineVLDDUP(SDNode *N, TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
EVT VT = N->getValueType(0);
if (!VT.is64BitVector())
return false;
SDNode *VLD = N->getOperand(0).getNode();
if (VLD->getOpcode() != ISD::INTRINSIC_W_CHAIN)
return false;
unsigned NumVecs = 0;
unsigned NewOpc = 0;
unsigned IntNo = cast<ConstantSDNode>(VLD->getOperand(1))->getZExtValue();
if (IntNo == Intrinsic::arm_neon_vld2lane) {
NumVecs = 2;
NewOpc = ARMISD::VLD2DUP;
} else if (IntNo == Intrinsic::arm_neon_vld3lane) {
NumVecs = 3;
NewOpc = ARMISD::VLD3DUP;
} else if (IntNo == Intrinsic::arm_neon_vld4lane) {
NumVecs = 4;
NewOpc = ARMISD::VLD4DUP;
} else {
return false;
}
unsigned VLDLaneNo =
cast<ConstantSDNode>(VLD->getOperand(NumVecs+3))->getZExtValue();
for (SDNode::use_iterator UI = VLD->use_begin(), UE = VLD->use_end();
UI != UE; ++UI) {
if (UI.getUse().getResNo() == NumVecs)
continue;
SDNode *User = *UI;
if (User->getOpcode() != ARMISD::VDUPLANE ||
VLDLaneNo != cast<ConstantSDNode>(User->getOperand(1))->getZExtValue())
return false;
}
EVT Tys[5];
unsigned n;
for (n = 0; n < NumVecs; ++n)
Tys[n] = VT;
Tys[n] = MVT::Other;
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,
Ops, VLDMemInt->getMemoryVT(),
VLDMemInt->getMemOperand());
for (SDNode::use_iterator UI = VLD->use_begin(), UE = VLD->use_end();
UI != UE; ++UI) {
unsigned ResNo = UI.getUse().getResNo();
if (ResNo == NumVecs)
continue;
SDNode *User = *UI;
DCI.CombineTo(User, SDValue(VLDDup.getNode(), ResNo));
}
std::vector<SDValue> VLDDupResults;
for (unsigned n = 0; n < NumVecs; ++n)
VLDDupResults.push_back(SDValue(VLDDup.getNode(), n));
VLDDupResults.push_back(SDValue(VLDDup.getNode(), NumVecs));
DCI.CombineTo(VLD, VLDDupResults);
return true;
}
static SDValue PerformVDUPLANECombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SDValue Op = N->getOperand(0);
if (CombineVLDDUP(N, DCI))
return SDValue(N, 0);
while (Op.getOpcode() == ISD::BITCAST)
Op = Op.getOperand(0);
if (Op.getOpcode() != ARMISD::VMOVIMM && Op.getOpcode() != ARMISD::VMVNIMM)
return SDValue();
unsigned EltSize = Op.getValueType().getVectorElementType().getSizeInBits();
unsigned Imm = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
unsigned EltBits;
if (ARM_AM::decodeNEONModImm(Imm, EltBits) == 0)
EltSize = 8;
EVT VT = N->getValueType(0);
if (EltSize > VT.getVectorElementType().getSizeInBits())
return SDValue();
return DCI.DAG.getNode(ISD::BITCAST, SDLoc(N), VT, Op);
}
static SDValue PerformLOADCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
EVT VT = N->getValueType(0);
if (ISD::isNormalLoad(N) && VT.isVector() &&
DCI.DAG.getTargetLoweringInfo().isTypeLegal(VT))
return CombineBaseUpdate(N, DCI);
return SDValue();
}
static SDValue PerformSTORECombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
StoreSDNode *St = cast<StoreSDNode>(N);
if (St->isVolatile())
return SDValue();
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();
if (!isPowerOf2_32(NumElems * FromEltSz * ToEltSz)) return SDValue();
if (0 != (NumElems * FromEltSz) % ToEltSz) return SDValue();
unsigned SizeRatio = FromEltSz / ToEltSz;
assert(SizeRatio * NumElems * ToEltSz == VT.getSizeInBits());
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;
if (!TLI.isTypeLegal(WideVecVT)) return SDValue();
SDValue Shuff = DAG.getVectorShuffle(WideVecVT, DL, WideVec,
DAG.getUNDEF(WideVec.getValueType()),
ShuffleVec.data());
MVT StoreType = MVT::i8;
for (MVT Tp : MVT::integer_valuetypes()) {
if (TLI.isTypeLegal(Tp) && Tp.getSizeInBits() <= NumElems * ToEltSz)
StoreType = Tp;
}
if (!TLI.isTypeLegal(StoreType))
return SDValue();
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();
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();
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) {
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);
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 (ISD::isNormalStore(N) && VT.isVector() &&
DCI.DAG.getTargetLoweringInfo().isTypeLegal(VT))
return CombineBaseUpdate(N, DCI);
return SDValue();
}
static bool isConstVecPow2(SDValue ConstVec, bool isSigned, uint64_t &C)
{
integerPart cN;
integerPart c0 = 0;
for (unsigned I = 0, E = ConstVec.getValueType().getVectorNumElements();
I != E; I++) {
ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(ConstVec.getOperand(I));
if (!C)
return false;
bool isExact;
APFloat APF = C->getValueAPF();
if (APF.convertToInteger(&cN, 64, isSigned, APFloat::rmTowardZero, &isExact)
!= APFloat::opOK || !isExact)
return false;
c0 = (I == 0) ? cN : c0;
if (!isPowerOf2_64(cN) || c0 != cN || Log2_64(c0) < 1 || Log2_64(c0) > 32)
return false;
}
C = c0;
return true;
}
static SDValue PerformVCVTCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SelectionDAG &DAG = DCI.DAG;
SDValue Op = N->getOperand(0);
if (!Subtarget->hasNEON() || !Op.getValueType().isVector() ||
Op.getOpcode() != ISD::FMUL)
return SDValue();
uint64_t C;
SDValue N0 = Op->getOperand(0);
SDValue ConstVec = Op->getOperand(1);
bool isSigned = N->getOpcode() == ISD::FP_TO_SINT;
if (ConstVec.getOpcode() != ISD::BUILD_VECTOR ||
!isConstVecPow2(ConstVec, isSigned, C))
return SDValue();
MVT FloatTy = Op.getSimpleValueType().getVectorElementType();
MVT IntTy = N->getSimpleValueType(0).getVectorElementType();
unsigned NumLanes = Op.getValueType().getVectorNumElements();
if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32 ||
NumLanes > 4) {
return SDValue();
}
unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfp2fxs :
Intrinsic::arm_neon_vcvtfp2fxu;
SDValue FixConv = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N),
NumLanes == 2 ? MVT::v2i32 : MVT::v4i32,
DAG.getConstant(IntrinsicOpcode, MVT::i32), N0,
DAG.getConstant(Log2_64(C), MVT::i32));
if (IntTy.getSizeInBits() < FloatTy.getSizeInBits())
FixConv = DAG.getNode(ISD::TRUNCATE, SDLoc(N), N->getValueType(0), FixConv);
return FixConv;
}
static SDValue PerformVDIVCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const ARMSubtarget *Subtarget) {
SelectionDAG &DAG = DCI.DAG;
SDValue Op = N->getOperand(0);
unsigned OpOpcode = Op.getNode()->getOpcode();
if (!Subtarget->hasNEON() || !N->getValueType(0).isVector() ||
(OpOpcode != ISD::SINT_TO_FP && OpOpcode != ISD::UINT_TO_FP))
return SDValue();
uint64_t C;
SDValue ConstVec = N->getOperand(1);
bool isSigned = OpOpcode == ISD::SINT_TO_FP;
if (ConstVec.getOpcode() != ISD::BUILD_VECTOR ||
!isConstVecPow2(ConstVec, isSigned, C))
return SDValue();
MVT FloatTy = N->getSimpleValueType(0).getVectorElementType();
MVT IntTy = Op.getOperand(0).getSimpleValueType().getVectorElementType();
if (FloatTy.getSizeInBits() != 32 || IntTy.getSizeInBits() > 32) {
return SDValue();
}
SDValue ConvInput = Op.getOperand(0);
unsigned NumLanes = Op.getValueType().getVectorNumElements();
if (IntTy.getSizeInBits() < FloatTy.getSizeInBits())
ConvInput = DAG.getNode(isSigned ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND,
SDLoc(N), NumLanes == 2 ? MVT::v2i32 : MVT::v4i32,
ConvInput);
unsigned IntrinsicOpcode = isSigned ? Intrinsic::arm_neon_vcvtfxs2fp :
Intrinsic::arm_neon_vcvtfxu2fp;
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, SDLoc(N),
Op.getValueType(),
DAG.getConstant(IntrinsicOpcode, MVT::i32),
ConvInput, DAG.getConstant(Log2_64(C), MVT::i32));
}
static bool getVShiftImm(SDValue Op, unsigned ElementBits, int64_t &Cnt) {
while (Op.getOpcode() == ISD::BITCAST)
Op = Op.getOperand(0);
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
APInt SplatBits, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (! BVN || ! BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize,
HasAnyUndefs, ElementBits) ||
SplatBitSize > ElementBits)
return false;
Cnt = SplatBits.getSExtValue();
return true;
}
static bool isVShiftLImm(SDValue Op, EVT VT, bool isLong, int64_t &Cnt) {
assert(VT.isVector() && "vector shift count is not a vector type");
unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
if (! getVShiftImm(Op, ElementBits, Cnt))
return false;
return (Cnt >= 0 && (isLong ? Cnt-1 : Cnt) < ElementBits);
}
static bool isVShiftRImm(SDValue Op, EVT VT, bool isNarrow, bool isIntrinsic,
int64_t &Cnt) {
assert(VT.isVector() && "vector shift count is not a vector type");
unsigned ElementBits = VT.getVectorElementType().getSizeInBits();
if (! getVShiftImm(Op, ElementBits, Cnt))
return false;
if (isIntrinsic)
Cnt = -Cnt;
return (Cnt >= 1 && Cnt <= (isNarrow ? ElementBits/2 : ElementBits));
}
static SDValue PerformIntrinsicCombine(SDNode *N, SelectionDAG &DAG) {
unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
switch (IntNo) {
default:
break;
case Intrinsic::arm_neon_vshifts:
case Intrinsic::arm_neon_vshiftu:
case Intrinsic::arm_neon_vrshifts:
case Intrinsic::arm_neon_vrshiftu:
case Intrinsic::arm_neon_vrshiftn:
case Intrinsic::arm_neon_vqshifts:
case Intrinsic::arm_neon_vqshiftu:
case Intrinsic::arm_neon_vqshiftsu:
case Intrinsic::arm_neon_vqshiftns:
case Intrinsic::arm_neon_vqshiftnu:
case Intrinsic::arm_neon_vqshiftnsu:
case Intrinsic::arm_neon_vqrshiftns:
case Intrinsic::arm_neon_vqrshiftnu:
case Intrinsic::arm_neon_vqrshiftnsu: {
EVT VT = N->getOperand(1).getValueType();
int64_t Cnt;
unsigned VShiftOpc = 0;
switch (IntNo) {
case Intrinsic::arm_neon_vshifts:
case Intrinsic::arm_neon_vshiftu:
if (isVShiftLImm(N->getOperand(2), VT, false, Cnt)) {
VShiftOpc = ARMISD::VSHL;
break;
}
if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt)) {
VShiftOpc = (IntNo == Intrinsic::arm_neon_vshifts ?
ARMISD::VSHRs : ARMISD::VSHRu);
break;
}
return SDValue();
case Intrinsic::arm_neon_vrshifts:
case Intrinsic::arm_neon_vrshiftu:
if (isVShiftRImm(N->getOperand(2), VT, false, true, Cnt))
break;
return SDValue();
case Intrinsic::arm_neon_vqshifts:
case Intrinsic::arm_neon_vqshiftu:
if (isVShiftLImm(N->getOperand(2), VT, false, Cnt))
break;
return SDValue();
case Intrinsic::arm_neon_vqshiftsu:
if (isVShiftLImm(N->getOperand(2), VT, false, Cnt))
break;
llvm_unreachable("invalid shift count for vqshlu intrinsic");
case Intrinsic::arm_neon_vrshiftn:
case Intrinsic::arm_neon_vqshiftns:
case Intrinsic::arm_neon_vqshiftnu:
case Intrinsic::arm_neon_vqshiftnsu:
case Intrinsic::arm_neon_vqrshiftns:
case Intrinsic::arm_neon_vqrshiftnu:
case Intrinsic::arm_neon_vqrshiftnsu:
if (isVShiftRImm(N->getOperand(2), VT, true, true, Cnt))
break;
llvm_unreachable("invalid shift count for narrowing vector shift "
"intrinsic");
default:
llvm_unreachable("unhandled vector shift");
}
switch (IntNo) {
case Intrinsic::arm_neon_vshifts:
case Intrinsic::arm_neon_vshiftu:
break;
case Intrinsic::arm_neon_vrshifts:
VShiftOpc = ARMISD::VRSHRs; break;
case Intrinsic::arm_neon_vrshiftu:
VShiftOpc = ARMISD::VRSHRu; break;
case Intrinsic::arm_neon_vrshiftn:
VShiftOpc = ARMISD::VRSHRN; break;
case Intrinsic::arm_neon_vqshifts:
VShiftOpc = ARMISD::VQSHLs; break;
case Intrinsic::arm_neon_vqshiftu:
VShiftOpc = ARMISD::VQSHLu; break;
case Intrinsic::arm_neon_vqshiftsu:
VShiftOpc = ARMISD::VQSHLsu; break;
case Intrinsic::arm_neon_vqshiftns:
VShiftOpc = ARMISD::VQSHRNs; break;
case Intrinsic::arm_neon_vqshiftnu:
VShiftOpc = ARMISD::VQSHRNu; break;
case Intrinsic::arm_neon_vqshiftnsu:
VShiftOpc = ARMISD::VQSHRNsu; break;
case Intrinsic::arm_neon_vqrshiftns:
VShiftOpc = ARMISD::VQRSHRNs; break;
case Intrinsic::arm_neon_vqrshiftnu:
VShiftOpc = ARMISD::VQRSHRNu; break;
case Intrinsic::arm_neon_vqrshiftnsu:
VShiftOpc = ARMISD::VQRSHRNsu; break;
}
return DAG.getNode(VShiftOpc, SDLoc(N), N->getValueType(0),
N->getOperand(1), DAG.getConstant(Cnt, MVT::i32));
}
case Intrinsic::arm_neon_vshiftins: {
EVT VT = N->getOperand(1).getValueType();
int64_t Cnt;
unsigned VShiftOpc = 0;
if (isVShiftLImm(N->getOperand(3), VT, false, Cnt))
VShiftOpc = ARMISD::VSLI;
else if (isVShiftRImm(N->getOperand(3), VT, false, true, Cnt))
VShiftOpc = ARMISD::VSRI;
else {
llvm_unreachable("invalid shift count for vsli/vsri intrinsic");
}
return DAG.getNode(VShiftOpc, SDLoc(N), N->getValueType(0),
N->getOperand(1), N->getOperand(2),
DAG.getConstant(Cnt, MVT::i32));
}
case Intrinsic::arm_neon_vqrshifts:
case Intrinsic::arm_neon_vqrshiftu:
break;
}
return SDValue();
}
static SDValue PerformShiftCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
EVT VT = N->getValueType(0);
if (N->getOpcode() == ISD::SRL && VT == MVT::i32 && ST->hasV6Ops()) {
SDValue N1 = N->getOperand(1);
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N1)) {
SDValue N0 = N->getOperand(0);
if (C->getZExtValue() == 16 && N0.getOpcode() == ISD::BSWAP &&
DAG.MaskedValueIsZero(N0.getOperand(0),
APInt::getHighBitsSet(32, 16)))
return DAG.getNode(ISD::ROTR, SDLoc(N), VT, N0, N1);
}
}
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (!VT.isVector() || !TLI.isTypeLegal(VT))
return SDValue();
assert(ST->hasNEON() && "unexpected vector shift");
int64_t Cnt;
switch (N->getOpcode()) {
default: llvm_unreachable("unexpected shift opcode");
case ISD::SHL:
if (isVShiftLImm(N->getOperand(1), VT, false, Cnt))
return DAG.getNode(ARMISD::VSHL, SDLoc(N), VT, N->getOperand(0),
DAG.getConstant(Cnt, MVT::i32));
break;
case ISD::SRA:
case ISD::SRL:
if (isVShiftRImm(N->getOperand(1), VT, false, false, Cnt)) {
unsigned VShiftOpc = (N->getOpcode() == ISD::SRA ?
ARMISD::VSHRs : ARMISD::VSHRu);
return DAG.getNode(VShiftOpc, SDLoc(N), VT, N->getOperand(0),
DAG.getConstant(Cnt, MVT::i32));
}
}
return SDValue();
}
static SDValue PerformExtendCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
SDValue N0 = N->getOperand(0);
if (ST->hasNEON() && N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
SDValue Vec = N0.getOperand(0);
SDValue Lane = N0.getOperand(1);
EVT VT = N->getValueType(0);
EVT EltVT = N0.getValueType();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (VT == MVT::i32 &&
(EltVT == MVT::i8 || EltVT == MVT::i16) &&
TLI.isTypeLegal(Vec.getValueType()) &&
isa<ConstantSDNode>(Lane)) {
unsigned Opc = 0;
switch (N->getOpcode()) {
default: llvm_unreachable("unexpected opcode");
case ISD::SIGN_EXTEND:
Opc = ARMISD::VGETLANEs;
break;
case ISD::ZERO_EXTEND:
case ISD::ANY_EXTEND:
Opc = ARMISD::VGETLANEu;
break;
}
return DAG.getNode(Opc, SDLoc(N), VT, Vec, Lane);
}
}
return SDValue();
}
static SDValue PerformSELECT_CCCombine(SDNode *N, SelectionDAG &DAG,
const ARMSubtarget *ST) {
if (!ST->hasNEON() || !ST->useNEONForSinglePrecisionFP() ||
N->getValueType(0) != MVT::f32)
return SDValue();
SDValue CondLHS = N->getOperand(0);
SDValue CondRHS = N->getOperand(1);
SDValue LHS = N->getOperand(2);
SDValue RHS = N->getOperand(3);
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(4))->get();
unsigned Opcode = 0;
bool IsReversed;
if (DAG.isEqualTo(LHS, CondLHS) && DAG.isEqualTo(RHS, CondRHS)) {
IsReversed = false; } else if (DAG.isEqualTo(LHS, CondRHS) && DAG.isEqualTo(RHS, CondLHS)) {
IsReversed = true ; } else {
return SDValue();
}
bool IsUnordered;
switch (CC) {
default: break;
case ISD::SETOLT:
case ISD::SETOLE:
case ISD::SETLT:
case ISD::SETLE:
case ISD::SETULT:
case ISD::SETULE:
IsUnordered = (CC == ISD::SETULT || CC == ISD::SETULE);
if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
break;
if ((CC == ISD::SETLE || CC == ISD::SETOLE || CC == ISD::SETULE) &&
!DAG.getTarget().Options.UnsafeFPMath &&
!(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
break;
Opcode = IsReversed ? ARMISD::FMAX : ARMISD::FMIN;
break;
case ISD::SETOGT:
case ISD::SETOGE:
case ISD::SETGT:
case ISD::SETGE:
case ISD::SETUGT:
case ISD::SETUGE:
IsUnordered = (CC == ISD::SETUGT || CC == ISD::SETUGE);
if (!DAG.isKnownNeverNaN(IsUnordered ? RHS : LHS))
break;
if ((CC == ISD::SETGE || CC == ISD::SETOGE || CC == ISD::SETUGE) &&
!DAG.getTarget().Options.UnsafeFPMath &&
!(DAG.isKnownNeverZero(LHS) || DAG.isKnownNeverZero(RHS)))
break;
Opcode = IsReversed ? ARMISD::FMIN : ARMISD::FMAX;
break;
}
if (!Opcode)
return SDValue();
return DAG.getNode(Opcode, SDLoc(N), N->getValueType(0), LHS, RHS);
}
SDValue
ARMTargetLowering::PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const {
SDValue Cmp = N->getOperand(4);
if (Cmp.getOpcode() != ARMISD::CMPZ)
return SDValue();
EVT VT = N->getValueType(0);
SDLoc dl(N);
SDValue LHS = Cmp.getOperand(0);
SDValue RHS = Cmp.getOperand(1);
SDValue FalseVal = N->getOperand(0);
SDValue TrueVal = N->getOperand(1);
SDValue ARMcc = N->getOperand(2);
ARMCC::CondCodes CC =
(ARMCC::CondCodes)cast<ConstantSDNode>(ARMcc)->getZExtValue();
SDValue Res;
if (CC == ARMCC::NE && FalseVal == RHS && FalseVal != LHS) {
Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, TrueVal, ARMcc,
N->getOperand(3), Cmp);
} else if (CC == ARMCC::EQ && TrueVal == RHS) {
SDValue ARMcc;
SDValue NewCmp = getARMCmp(LHS, RHS, ISD::SETNE, ARMcc, DAG, dl);
Res = DAG.getNode(ARMISD::CMOV, dl, VT, LHS, FalseVal, ARMcc,
N->getOperand(3), NewCmp);
}
if (Res.getNode()) {
APInt KnownZero, KnownOne;
DAG.computeKnownBits(SDValue(N,0), KnownZero, KnownOne);
if (KnownZero == 0xfffffffe)
Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
DAG.getValueType(MVT::i1));
else if (KnownZero == 0xffffff00)
Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
DAG.getValueType(MVT::i8));
else if (KnownZero == 0xffff0000)
Res = DAG.getNode(ISD::AssertZext, dl, MVT::i32, Res,
DAG.getValueType(MVT::i16));
}
return Res;
}
SDValue ARMTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
switch (N->getOpcode()) {
default: break;
case ISD::ADDC: return PerformADDCCombine(N, DCI, Subtarget);
case ISD::ADD: return PerformADDCombine(N, DCI, Subtarget);
case ISD::SUB: return PerformSUBCombine(N, DCI);
case ISD::MUL: return PerformMULCombine(N, DCI, Subtarget);
case ISD::OR: return PerformORCombine(N, DCI, Subtarget);
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, 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, 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::FP_TO_SINT:
case ISD::FP_TO_UINT: return PerformVCVTCombine(N, DCI, Subtarget);
case ISD::FDIV: return PerformVDIVCombine(N, DCI, Subtarget);
case ISD::INTRINSIC_WO_CHAIN: return PerformIntrinsicCombine(N, DCI.DAG);
case ISD::SHL:
case ISD::SRA:
case ISD::SRL: return PerformShiftCombine(N, DCI.DAG, Subtarget);
case ISD::SIGN_EXTEND:
case ISD::ZERO_EXTEND:
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 PerformVLDCombine(N, DCI);
case ARMISD::BUILD_VECTOR:
return PerformARMBUILD_VECTORCombine(N, DCI);
case ISD::INTRINSIC_VOID:
case ISD::INTRINSIC_W_CHAIN:
switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
case Intrinsic::arm_neon_vld1:
case Intrinsic::arm_neon_vld2:
case Intrinsic::arm_neon_vld3:
case Intrinsic::arm_neon_vld4:
case Intrinsic::arm_neon_vld2lane:
case Intrinsic::arm_neon_vld3lane:
case Intrinsic::arm_neon_vld4lane:
case Intrinsic::arm_neon_vst1:
case Intrinsic::arm_neon_vst2:
case Intrinsic::arm_neon_vst3:
case Intrinsic::arm_neon_vst4:
case Intrinsic::arm_neon_vst2lane:
case Intrinsic::arm_neon_vst3lane:
case Intrinsic::arm_neon_vst4lane:
return PerformVLDCombine(N, DCI);
default: break;
}
break;
}
return SDValue();
}
bool ARMTargetLowering::isDesirableToTransformToIntegerOp(unsigned Opc,
EVT VT) const {
return (VT == MVT::f32) && (Opc == ISD::LOAD || Opc == ISD::STORE);
}
bool ARMTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
unsigned,
unsigned,
bool *Fast) const {
bool AllowsUnaligned = Subtarget->allowsUnalignedMem();
switch (VT.getSimpleVT().SimpleTy) {
default:
return false;
case MVT::i8:
case MVT::i16:
case MVT::i32: {
if (AllowsUnaligned) {
if (Fast)
*Fast = Subtarget->hasV7Ops();
return true;
}
return false;
}
case MVT::f64:
case MVT::v2f64: {
if (Subtarget->hasNEON() && (AllowsUnaligned || isLittleEndian())) {
if (Fast)
*Fast = true;
return true;
}
return false;
}
}
}
static bool memOpAlign(unsigned DstAlign, unsigned SrcAlign,
unsigned AlignCheck) {
return ((SrcAlign == 0 || SrcAlign % AlignCheck == 0) &&
(DstAlign == 0 || DstAlign % AlignCheck == 0));
}
EVT ARMTargetLowering::getOptimalMemOpType(uint64_t Size,
unsigned DstAlign, unsigned SrcAlign,
bool IsMemset, bool ZeroMemset,
bool MemcpyStrSrc,
MachineFunction &MF) const {
const Function *F = MF.getFunction();
if ((!IsMemset || ZeroMemset) && Subtarget->hasNEON() &&
!F->hasFnAttribute(Attribute::NoImplicitFloat)) {
bool Fast;
if (Size >= 16 &&
(memOpAlign(SrcAlign, DstAlign, 16) ||
(allowsMisalignedMemoryAccesses(MVT::v2f64, 0, 1, &Fast) && Fast))) {
return MVT::v2f64;
} else if (Size >= 8 &&
(memOpAlign(SrcAlign, DstAlign, 8) ||
(allowsMisalignedMemoryAccesses(MVT::f64, 0, 1, &Fast) &&
Fast))) {
return MVT::f64;
}
}
if (Size >= 4)
return MVT::i32;
else if (Size >= 2)
return MVT::i16;
return MVT::Other;
}
bool ARMTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
if (Val.getOpcode() != ISD::LOAD)
return false;
EVT VT1 = Val.getValueType();
if (!VT1.isSimple() || !VT1.isInteger() ||
!VT2.isSimple() || !VT2.isInteger())
return false;
switch (VT1.getSimpleVT().SimpleTy) {
default: break;
case MVT::i1:
case MVT::i8:
case MVT::i16:
return true;
}
return false;
}
bool ARMTargetLowering::allowTruncateForTailCall(Type *Ty1, Type *Ty2) const {
if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
return false;
if (!isTypeLegal(EVT::getEVT(Ty1)))
return false;
assert(Ty1->getPrimitiveSizeInBits() <= 64 && "i128 is probably not a noop");
return true;
}
static bool isLegalT1AddressImmediate(int64_t V, EVT VT) {
if (V < 0)
return false;
unsigned Scale = 1;
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
break;
case MVT::i16:
Scale = 2;
break;
case MVT::i32:
Scale = 4;
break;
}
if ((V & (Scale - 1)) != 0)
return false;
V /= Scale;
return V == (V & ((1LL << 5) - 1));
}
static bool isLegalT2AddressImmediate(int64_t V, EVT VT,
const ARMSubtarget *Subtarget) {
bool isNeg = false;
if (V < 0) {
isNeg = true;
V = - V;
}
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
if (isNeg)
return V == (V & ((1LL << 8) - 1));
return V == (V & ((1LL << 12) - 1));
case MVT::f32:
case MVT::f64:
if (!Subtarget->hasVFP2())
return false;
if ((V & 3) != 0)
return false;
V >>= 2;
return V == (V & ((1LL << 8) - 1));
}
}
static bool isLegalAddressImmediate(int64_t V, EVT VT,
const ARMSubtarget *Subtarget) {
if (V == 0)
return true;
if (!VT.isSimple())
return false;
if (Subtarget->isThumb1Only())
return isLegalT1AddressImmediate(V, VT);
else if (Subtarget->isThumb2())
return isLegalT2AddressImmediate(V, VT, Subtarget);
if (V < 0)
V = - V;
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
case MVT::i32:
return V == (V & ((1LL << 12) - 1));
case MVT::i16:
return V == (V & ((1LL << 8) - 1));
case MVT::f32:
case MVT::f64:
if (!Subtarget->hasVFP2()) return false;
if ((V & 3) != 0)
return false;
V >>= 2;
return V == (V & ((1LL << 8) - 1));
}
}
bool ARMTargetLowering::isLegalT2ScaledAddressingMode(const AddrMode &AM,
EVT VT) const {
int Scale = AM.Scale;
if (Scale < 0)
return false;
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
if (Scale == 1)
return true;
Scale = Scale & ~1;
return Scale == 2 || Scale == 4 || Scale == 8;
case MVT::i64:
if (((unsigned)AM.HasBaseReg + Scale) <= 2)
return true;
return false;
case MVT::isVoid:
if (Scale & 1) return false;
return isPowerOf2_32(Scale);
}
}
bool ARMTargetLowering::isLegalAddressingMode(const AddrMode &AM,
Type *Ty) const {
EVT VT = getValueType(Ty, true);
if (!isLegalAddressImmediate(AM.BaseOffs, VT, Subtarget))
return false;
if (AM.BaseGV)
return false;
switch (AM.Scale) {
case 0: break;
case 1:
if (Subtarget->isThumb1Only())
return false;
default:
if (AM.BaseOffs)
return false;
if (!VT.isSimple())
return false;
if (Subtarget->isThumb2())
return isLegalT2ScaledAddressingMode(AM, VT);
int Scale = AM.Scale;
switch (VT.getSimpleVT().SimpleTy) {
default: return false;
case MVT::i1:
case MVT::i8:
case MVT::i32:
if (Scale < 0) Scale = -Scale;
if (Scale == 1)
return true;
return isPowerOf2_32(Scale & ~1);
case MVT::i16:
case MVT::i64:
if (((unsigned)AM.HasBaseReg + Scale) <= 2)
return true;
return false;
case MVT::isVoid:
if (Scale & 1) return false;
return isPowerOf2_32(Scale);
}
}
return true;
}
bool ARMTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
if (!Subtarget->isThumb())
return ARM_AM::getSOImmVal(llvm::abs64(Imm)) != -1;
if (Subtarget->isThumb2())
return ARM_AM::getT2SOImmVal(llvm::abs64(Imm)) != -1;
return Imm >= 0 && Imm <= 255;
}
bool ARMTargetLowering::isLegalAddImmediate(int64_t Imm) const {
int64_t AbsImm = llvm::abs64(Imm);
if (!Subtarget->isThumb())
return ARM_AM::getSOImmVal(AbsImm) != -1;
if (Subtarget->isThumb2())
return ARM_AM::getT2SOImmVal(AbsImm) != -1;
return AbsImm >= 0 && AbsImm <= 255;
}
static bool getARMIndexedAddressParts(SDNode *Ptr, EVT VT,
bool isSEXTLoad, SDValue &Base,
SDValue &Offset, bool &isInc,
SelectionDAG &DAG) {
if (Ptr->getOpcode() != ISD::ADD && Ptr->getOpcode() != ISD::SUB)
return false;
if (VT == MVT::i16 || ((VT == MVT::i8 || VT == MVT::i1) && isSEXTLoad)) {
Base = Ptr->getOperand(0);
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
int RHSC = (int)RHS->getZExtValue();
if (RHSC < 0 && RHSC > -256) {
assert(Ptr->getOpcode() == ISD::ADD);
isInc = false;
Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
return true;
}
}
isInc = (Ptr->getOpcode() == ISD::ADD);
Offset = Ptr->getOperand(1);
return true;
} else if (VT == MVT::i32 || VT == MVT::i8 || VT == MVT::i1) {
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
int RHSC = (int)RHS->getZExtValue();
if (RHSC < 0 && RHSC > -0x1000) {
assert(Ptr->getOpcode() == ISD::ADD);
isInc = false;
Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
Base = Ptr->getOperand(0);
return true;
}
}
if (Ptr->getOpcode() == ISD::ADD) {
isInc = true;
ARM_AM::ShiftOpc ShOpcVal=
ARM_AM::getShiftOpcForNode(Ptr->getOperand(0).getOpcode());
if (ShOpcVal != ARM_AM::no_shift) {
Base = Ptr->getOperand(1);
Offset = Ptr->getOperand(0);
} else {
Base = Ptr->getOperand(0);
Offset = Ptr->getOperand(1);
}
return true;
}
isInc = (Ptr->getOpcode() == ISD::ADD);
Base = Ptr->getOperand(0);
Offset = Ptr->getOperand(1);
return true;
}
return false;
}
static bool getT2IndexedAddressParts(SDNode *Ptr, EVT VT,
bool isSEXTLoad, SDValue &Base,
SDValue &Offset, bool &isInc,
SelectionDAG &DAG) {
if (Ptr->getOpcode() != ISD::ADD && Ptr->getOpcode() != ISD::SUB)
return false;
Base = Ptr->getOperand(0);
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Ptr->getOperand(1))) {
int RHSC = (int)RHS->getZExtValue();
if (RHSC < 0 && RHSC > -0x100) { assert(Ptr->getOpcode() == ISD::ADD);
isInc = false;
Offset = DAG.getConstant(-RHSC, RHS->getValueType(0));
return true;
} else if (RHSC > 0 && RHSC < 0x100) { isInc = Ptr->getOpcode() == ISD::ADD;
Offset = DAG.getConstant(RHSC, RHS->getValueType(0));
return true;
}
}
return false;
}
bool
ARMTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
if (Subtarget->isThumb1Only())
return false;
EVT VT;
SDValue Ptr;
bool isSEXTLoad = false;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
VT = LD->getMemoryVT();
isSEXTLoad = LD->getExtensionType() == ISD::SEXTLOAD;
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
Ptr = ST->getBasePtr();
VT = ST->getMemoryVT();
} else
return false;
bool isInc;
bool isLegal = false;
if (Subtarget->isThumb2())
isLegal = getT2IndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
Offset, isInc, DAG);
else
isLegal = getARMIndexedAddressParts(Ptr.getNode(), VT, isSEXTLoad, Base,
Offset, isInc, DAG);
if (!isLegal)
return false;
AM = isInc ? ISD::PRE_INC : ISD::PRE_DEC;
return true;
}
bool ARMTargetLowering::getPostIndexedAddressParts(SDNode *N, SDNode *Op,
SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
if (Subtarget->isThumb1Only())
return false;
EVT VT;
SDValue Ptr;
bool isSEXTLoad = false;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
VT = LD->getMemoryVT();
Ptr = LD->getBasePtr();
isSEXTLoad = LD->getExtensionType() == ISD::SEXTLOAD;
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
VT = ST->getMemoryVT();
Ptr = ST->getBasePtr();
} else
return false;
bool isInc;
bool isLegal = false;
if (Subtarget->isThumb2())
isLegal = getT2IndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
isInc, DAG);
else
isLegal = getARMIndexedAddressParts(Op, VT, isSEXTLoad, Base, Offset,
isInc, DAG);
if (!isLegal)
return false;
if (Ptr != Base) {
if (Ptr == Offset && Op->getOpcode() == ISD::ADD &&
!Subtarget->isThumb2())
std::swap(Base, Offset);
if (Ptr != Base)
return false;
}
AM = isInc ? ISD::POST_INC : ISD::POST_DEC;
return true;
}
void ARMTargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
APInt &KnownZero,
APInt &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
unsigned BitWidth = KnownOne.getBitWidth();
KnownZero = KnownOne = APInt(BitWidth, 0);
switch (Op.getOpcode()) {
default: break;
case ARMISD::ADDC:
case ARMISD::ADDE:
case ARMISD::SUBC:
case ARMISD::SUBE:
if (Op.getResNo() == 0)
break;
KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - 1);
break;
case ARMISD::CMOV: {
DAG.computeKnownBits(Op.getOperand(0), KnownZero, KnownOne, Depth+1);
if (KnownZero == 0 && KnownOne == 0) return;
APInt KnownZeroRHS, KnownOneRHS;
DAG.computeKnownBits(Op.getOperand(1), KnownZeroRHS, KnownOneRHS, Depth+1);
KnownZero &= KnownZeroRHS;
KnownOne &= KnownOneRHS;
return;
}
case ISD::INTRINSIC_W_CHAIN: {
ConstantSDNode *CN = cast<ConstantSDNode>(Op->getOperand(1));
Intrinsic::ID IntID = static_cast<Intrinsic::ID>(CN->getZExtValue());
switch (IntID) {
default: return;
case Intrinsic::arm_ldaex:
case Intrinsic::arm_ldrex: {
EVT VT = cast<MemIntrinsicSDNode>(Op)->getMemoryVT();
unsigned MemBits = VT.getScalarType().getSizeInBits();
KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - MemBits);
return;
}
}
}
}
}
bool ARMTargetLowering::ExpandInlineAsm(CallInst *CI) const {
if (!Subtarget->hasV6Ops())
return false;
InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
std::string AsmStr = IA->getAsmString();
SmallVector<StringRef, 4> AsmPieces;
SplitString(AsmStr, AsmPieces, ";\n");
switch (AsmPieces.size()) {
default: return false;
case 1:
AsmStr = AsmPieces[0];
AsmPieces.clear();
SplitString(AsmStr, AsmPieces, " \t,");
if (AsmPieces.size() == 3 &&
AsmPieces[0] == "rev" && AsmPieces[1] == "$0" && AsmPieces[2] == "$1" &&
IA->getConstraintString().compare(0, 4, "=l,l") == 0) {
IntegerType *Ty = dyn_cast<IntegerType>(CI->getType());
if (Ty && Ty->getBitWidth() == 32)
return IntrinsicLowering::LowerToByteSwap(CI);
}
break;
}
return false;
}
ARMTargetLowering::ConstraintType
ARMTargetLowering::getConstraintType(const std::string &Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default: break;
case 'l': return C_RegisterClass;
case 'w': return C_RegisterClass;
case 'h': return C_RegisterClass;
case 'x': return C_RegisterClass;
case 't': return C_RegisterClass;
case 'j': return C_Other; case 'Q': return C_Memory;
}
} else if (Constraint.size() == 2) {
switch (Constraint[0]) {
default: break;
case 'U': return C_Memory;
}
}
return TargetLowering::getConstraintType(Constraint);
}
TargetLowering::ConstraintWeight
ARMTargetLowering::getSingleConstraintMatchWeight(
AsmOperandInfo &info, const char *constraint) const {
ConstraintWeight weight = CW_Invalid;
Value *CallOperandVal = info.CallOperandVal;
if (!CallOperandVal)
return CW_Default;
Type *type = CallOperandVal->getType();
switch (*constraint) {
default:
weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
break;
case 'l':
if (type->isIntegerTy()) {
if (Subtarget->isThumb())
weight = CW_SpecificReg;
else
weight = CW_Register;
}
break;
case 'w':
if (type->isFloatingPointTy())
weight = CW_Register;
break;
}
return weight;
}
typedef std::pair<unsigned, const TargetRegisterClass*> RCPair;
RCPair
ARMTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
const std::string &Constraint,
MVT VT) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
case 'l': if (Subtarget->isThumb())
return RCPair(0U, &ARM::tGPRRegClass);
return RCPair(0U, &ARM::GPRRegClass);
case 'h': if (Subtarget->isThumb())
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)
break;
if (VT == MVT::f32)
return RCPair(0U, &ARM::SPRRegClass);
if (VT.getSizeInBits() == 64)
return RCPair(0U, &ARM::DPRRegClass);
if (VT.getSizeInBits() == 128)
return RCPair(0U, &ARM::QPRRegClass);
break;
case 'x':
if (VT == MVT::Other)
break;
if (VT == MVT::f32)
return RCPair(0U, &ARM::SPR_8RegClass);
if (VT.getSizeInBits() == 64)
return RCPair(0U, &ARM::DPR_8RegClass);
if (VT.getSizeInBits() == 128)
return RCPair(0U, &ARM::QPR_8RegClass);
break;
case 't':
if (VT == MVT::f32)
return RCPair(0U, &ARM::SPRRegClass);
break;
}
}
if (StringRef("{cc}").equals_lower(Constraint))
return std::make_pair(unsigned(ARM::CPSR), &ARM::CCRRegClass);
return TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
}
void ARMTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
std::string &Constraint,
std::vector<SDValue>&Ops,
SelectionDAG &DAG) const {
SDValue Result;
if (Constraint.length() != 1) return;
char ConstraintLetter = Constraint[0];
switch (ConstraintLetter) {
default: break;
case 'j':
case 'I': case 'J': case 'K': case 'L':
case 'M': case 'N': case 'O':
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
if (!C)
return;
int64_t CVal64 = C->getSExtValue();
int CVal = (int) CVal64;
if (CVal != CVal64)
return;
switch (ConstraintLetter) {
case 'j':
if (Subtarget->hasV6T2Ops())
if (CVal >= 0 && CVal <= 65535)
break;
return;
case 'I':
if (Subtarget->isThumb1Only()) {
if (CVal >= 0 && CVal <= 255)
break;
} else if (Subtarget->isThumb2()) {
if (ARM_AM::getT2SOImmVal(CVal) != -1)
break;
} else {
if (ARM_AM::getSOImmVal(CVal) != -1)
break;
}
return;
case 'J':
if (Subtarget->isThumb()) { if (CVal >= -255 && CVal <= -1)
break;
} else {
if (CVal >= -4095 && CVal <= 4095)
break;
}
return;
case 'K':
if (Subtarget->isThumb1Only()) {
if (CVal != 0 && ARM_AM::isThumbImmShiftedVal(CVal))
break;
} else if (Subtarget->isThumb2()) {
if (ARM_AM::getT2SOImmVal(~CVal) != -1)
break;
} else {
if (ARM_AM::getSOImmVal(~CVal) != -1)
break;
}
return;
case 'L':
if (Subtarget->isThumb1Only()) {
if (CVal >= -7 && CVal < 7)
break;
} else if (Subtarget->isThumb2()) {
if (ARM_AM::getT2SOImmVal(-CVal) != -1)
break;
} else {
if (ARM_AM::getSOImmVal(-CVal) != -1)
break;
}
return;
case 'M':
if (Subtarget->isThumb()) { if ((CVal >= 0 && CVal <= 1020) && ((CVal & 3) == 0))
break;
} else {
if ((CVal >= 0 && CVal <= 32) || ((CVal & (CVal - 1)) == 0))
break;
}
return;
case 'N':
if (Subtarget->isThumb()) { if (CVal >= 0 && CVal <= 31)
break;
}
return;
case 'O':
if (Subtarget->isThumb()) { if ((CVal >= -508 && CVal <= 508) && ((CVal & 3) == 0))
break;
}
return;
}
Result = DAG.getTargetConstant(CVal, Op.getValueType());
break;
}
if (Result.getNode()) {
Ops.push_back(Result);
return;
}
return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}
SDValue ARMTargetLowering::LowerDivRem(SDValue Op, SelectionDAG &DAG) const {
assert(Subtarget->isTargetAEABI() && "Register-based DivRem lowering only");
unsigned Opcode = Op->getOpcode();
assert((Opcode == ISD::SDIVREM || Opcode == ISD::UDIVREM) &&
"Invalid opcode for Div/Rem lowering");
bool isSigned = (Opcode == ISD::SDIVREM);
EVT VT = Op->getValueType(0);
Type *Ty = VT.getTypeForEVT(*DAG.getContext());
RTLIB::Libcall LC;
switch (VT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unexpected request for libcall!");
case MVT::i8: LC = isSigned ? RTLIB::SDIVREM_I8 : RTLIB::UDIVREM_I8; break;
case MVT::i16: LC = isSigned ? RTLIB::SDIVREM_I16 : RTLIB::UDIVREM_I16; break;
case MVT::i32: LC = isSigned ? RTLIB::SDIVREM_I32 : RTLIB::UDIVREM_I32; break;
case MVT::i64: LC = isSigned ? RTLIB::SDIVREM_I64 : RTLIB::UDIVREM_I64; break;
}
SDValue InChain = DAG.getEntryNode();
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
for (unsigned i = 0, e = Op->getNumOperands(); i != e; ++i) {
EVT ArgVT = Op->getOperand(i).getValueType();
Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
Entry.Node = Op->getOperand(i);
Entry.Ty = ArgTy;
Entry.isSExt = isSigned;
Entry.isZExt = !isSigned;
Args.push_back(Entry);
}
SDValue Callee = DAG.getExternalSymbol(getLibcallName(LC),
getPointerTy());
Type *RetTy = (Type*)StructType::get(Ty, Ty, nullptr);
SDLoc dl(Op);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(InChain)
.setCallee(getLibcallCallingConv(LC), RetTy, Callee, std::move(Args), 0)
.setInRegister().setSExtResult(isSigned).setZExtResult(!isSigned);
std::pair<SDValue, SDValue> CallInfo = LowerCallTo(CLI);
return CallInfo.first;
}
SDValue
ARMTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const {
assert(Subtarget->isTargetWindows() && "unsupported target platform");
SDLoc DL(Op);
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
SDValue Words = DAG.getNode(ISD::SRL, DL, MVT::i32, Size,
DAG.getConstant(2, MVT::i32));
SDValue Flag;
Chain = DAG.getCopyToReg(Chain, DL, ARM::R4, Words, Flag);
Flag = Chain.getValue(1);
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
Chain = DAG.getNode(ARMISD::WIN__CHKSTK, DL, NodeTys, Chain, Flag);
SDValue NewSP = DAG.getCopyFromReg(Chain, DL, ARM::SP, MVT::i32);
Chain = NewSP.getValue(1);
SDValue Ops[2] = { NewSP, Chain };
return DAG.getMergeValues(Ops, DL);
}
SDValue ARMTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
assert(Op.getValueType() == MVT::f64 && Subtarget->isFPOnlySP() &&
"Unexpected type for custom-lowering FP_EXTEND");
RTLIB::Libcall LC;
LC = RTLIB::getFPEXT(Op.getOperand(0).getValueType(), Op.getValueType());
SDValue SrcVal = Op.getOperand(0);
return makeLibCall(DAG, LC, Op.getValueType(), &SrcVal, 1,
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,
false, SDLoc(Op)).first;
}
bool
ARMTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
return false;
}
bool ARM::isBitFieldInvertedMask(unsigned v) {
if (v == 0xffffffff)
return false;
unsigned TO = CountTrailingOnes_32(v);
unsigned LO = CountLeadingOnes_32(v);
v = (v >> TO) << TO;
v = (v << LO) >> LO;
return v == 0;
}
bool ARMTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
if (!Subtarget->hasVFP3())
return false;
if (VT == MVT::f32)
return ARM_AM::getFP32Imm(Imm) != -1;
if (VT == MVT::f64 && !Subtarget->isFPOnlySP())
return ARM_AM::getFP64Imm(Imm) != -1;
return false;
}
bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
const CallInst &I,
unsigned Intrinsic) const {
switch (Intrinsic) {
case Intrinsic::arm_neon_vld1:
case Intrinsic::arm_neon_vld2:
case Intrinsic::arm_neon_vld3:
case Intrinsic::arm_neon_vld4:
case Intrinsic::arm_neon_vld2lane:
case Intrinsic::arm_neon_vld3lane:
case Intrinsic::arm_neon_vld4lane: {
Info.opc = ISD::INTRINSIC_W_CHAIN;
uint64_t NumElts = getDataLayout()->getTypeAllocSize(I.getType()) / 8;
Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
Info.vol = false; Info.readMem = true;
Info.writeMem = false;
return true;
}
case Intrinsic::arm_neon_vst1:
case Intrinsic::arm_neon_vst2:
case Intrinsic::arm_neon_vst3:
case Intrinsic::arm_neon_vst4:
case Intrinsic::arm_neon_vst2lane:
case Intrinsic::arm_neon_vst3lane:
case Intrinsic::arm_neon_vst4lane: {
Info.opc = ISD::INTRINSIC_VOID;
unsigned NumElts = 0;
for (unsigned ArgI = 1, ArgE = I.getNumArgOperands(); ArgI < ArgE; ++ArgI) {
Type *ArgTy = I.getArgOperand(ArgI)->getType();
if (!ArgTy->isVectorTy())
break;
NumElts += getDataLayout()->getTypeAllocSize(ArgTy) / 8;
}
Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
Info.vol = false; Info.readMem = false;
Info.writeMem = true;
return true;
}
case Intrinsic::arm_ldaex:
case Intrinsic::arm_ldrex: {
PointerType *PtrTy = cast<PointerType>(I.getArgOperand(0)->getType());
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = MVT::getVT(PtrTy->getElementType());
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Info.align = getDataLayout()->getABITypeAlignment(PtrTy->getElementType());
Info.vol = true;
Info.readMem = true;
Info.writeMem = false;
return true;
}
case Intrinsic::arm_stlex:
case Intrinsic::arm_strex: {
PointerType *PtrTy = cast<PointerType>(I.getArgOperand(1)->getType());
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = MVT::getVT(PtrTy->getElementType());
Info.ptrVal = I.getArgOperand(1);
Info.offset = 0;
Info.align = getDataLayout()->getABITypeAlignment(PtrTy->getElementType());
Info.vol = true;
Info.readMem = false;
Info.writeMem = true;
return true;
}
case Intrinsic::arm_stlexd:
case Intrinsic::arm_strexd: {
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = MVT::i64;
Info.ptrVal = I.getArgOperand(2);
Info.offset = 0;
Info.align = 8;
Info.vol = true;
Info.readMem = false;
Info.writeMem = true;
return true;
}
case Intrinsic::arm_ldaexd:
case Intrinsic::arm_ldrexd: {
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = MVT::i64;
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Info.align = 8;
Info.vol = true;
Info.readMem = true;
Info.writeMem = false;
return true;
}
default:
break;
}
return false;
}
bool ARMTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned Bits = Ty->getPrimitiveSizeInBits();
if (Bits == 0 || Bits > 32)
return false;
return true;
}
bool ARMTargetLowering::hasLoadLinkedStoreConditional() const { return true; }
Instruction* ARMTargetLowering::makeDMB(IRBuilder<> &Builder,
ARM_MB::MemBOpt Domain) const {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
if (!Subtarget->hasDataBarrier()) {
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 {
llvm_unreachable("makeDMB on a target so old that it has no barriers");
}
} else {
Function *DMB = llvm::Intrinsic::getDeclaration(M, Intrinsic::arm_dmb);
Domain = Subtarget->isMClass() ? ARM_MB::SY : Domain;
Constant *CDomain = Builder.getInt32(Domain);
return Builder.CreateCall(DMB, CDomain);
}
}
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; case SequentiallyConsistent:
if (!IsStore)
return nullptr;
case Release:
case AcquireRelease:
if (Subtarget->isSwift())
return makeDMB(Builder, ARM_MB::ISHST);
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; case Acquire:
case AcquireRelease:
case SequentiallyConsistent:
return makeDMB(Builder, ARM_MB::ISH);
}
llvm_unreachable("Unknown fence ordering in emitTrailingFence");
}
bool ARMTargetLowering::shouldExpandAtomicStoreInIR(StoreInst *SI) const {
unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
return (Size == 64) && !Subtarget->isMClass();
}
bool ARMTargetLowering::shouldExpandAtomicLoadInIR(LoadInst *LI) const {
unsigned Size = LI->getType()->getPrimitiveSizeInBits();
return (Size == 64) && !Subtarget->isMClass();
}
bool ARMTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
unsigned Size = AI->getType()->getPrimitiveSizeInBits();
return Size <= (Subtarget->isMClass() ? 32U : 64U);
}
bool ARMTargetLowering::useLoadStackGuardNode() const {
return Subtarget->isTargetMachO();
}
bool ARMTargetLowering::canCombineStoreAndExtract(Type *VectorTy, Value *Idx,
unsigned &Cost) const {
if (!Subtarget->hasNEON())
return false;
if (VectorTy->isFPOrFPVectorTy())
return false;
if (!isa<ConstantInt>(Idx))
return false;
assert(VectorTy->isVectorTy() && "VectorTy is not a vector type");
unsigned BitWidth = cast<VectorType>(VectorTy)->getBitWidth();
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 = isAtLeastAcquire(Ord);
if (ValTy->getPrimitiveSizeInBits() == 64) {
Intrinsic::ID Int =
IsAcquire ? Intrinsic::arm_ldaexd : Intrinsic::arm_ldrexd;
Function *Ldrex = llvm::Intrinsic::getDeclaration(M, Int);
Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
Value *LoHi = Builder.CreateCall(Ldrex, Addr, "lohi");
Value *Lo = Builder.CreateExtractValue(LoHi, 0, "lo");
Value *Hi = Builder.CreateExtractValue(LoHi, 1, "hi");
if (!Subtarget->isLittle())
std::swap (Lo, Hi);
Lo = Builder.CreateZExt(Lo, ValTy, "lo64");
Hi = Builder.CreateZExt(Hi, ValTy, "hi64");
return Builder.CreateOr(
Lo, Builder.CreateShl(Hi, ConstantInt::get(ValTy, 32)), "val64");
}
Type *Tys[] = { Addr->getType() };
Intrinsic::ID Int = IsAcquire ? Intrinsic::arm_ldaex : Intrinsic::arm_ldrex;
Function *Ldrex = llvm::Intrinsic::getDeclaration(M, Int, Tys);
return Builder.CreateTruncOrBitCast(
Builder.CreateCall(Ldrex, Addr),
cast<PointerType>(Addr->getType())->getElementType());
}
Value *ARMTargetLowering::emitStoreConditional(IRBuilder<> &Builder, Value *Val,
Value *Addr,
AtomicOrdering Ord) const {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
bool IsRelease = isAtLeastRelease(Ord);
if (Val->getType()->getPrimitiveSizeInBits() == 64) {
Intrinsic::ID Int =
IsRelease ? Intrinsic::arm_stlexd : Intrinsic::arm_strexd;
Function *Strex = Intrinsic::getDeclaration(M, Int);
Type *Int32Ty = Type::getInt32Ty(M->getContext());
Value *Lo = Builder.CreateTrunc(Val, Int32Ty, "lo");
Value *Hi = Builder.CreateTrunc(Builder.CreateLShr(Val, 32), Int32Ty, "hi");
if (!Subtarget->isLittle())
std::swap (Lo, Hi);
Addr = Builder.CreateBitCast(Addr, Type::getInt8PtrTy(M->getContext()));
return Builder.CreateCall3(Strex, Lo, Hi, Addr);
}
Intrinsic::ID Int = IsRelease ? Intrinsic::arm_stlex : Intrinsic::arm_strex;
Type *Tys[] = { Addr->getType() };
Function *Strex = Intrinsic::getDeclaration(M, Int, Tys);
return Builder.CreateCall2(
Strex, Builder.CreateZExtOrBitCast(
Val, Strex->getFunctionType()->getParamType(0)),
Addr);
}
enum HABaseType {
HA_UNKNOWN = 0,
HA_FLOAT,
HA_DOUBLE,
HA_VECT64,
HA_VECT128
};
static bool isHomogeneousAggregate(Type *Ty, HABaseType &Base,
uint64_t &Members) {
if (const StructType *ST = dyn_cast<StructType>(Ty)) {
for (unsigned i = 0; i < ST->getNumElements(); ++i) {
uint64_t SubMembers = 0;
if (!isHomogeneousAggregate(ST->getElementType(i), Base, SubMembers))
return false;
Members += SubMembers;
}
} else if (const ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
uint64_t SubMembers = 0;
if (!isHomogeneousAggregate(AT->getElementType(), Base, SubMembers))
return false;
Members += SubMembers * AT->getNumElements();
} else if (Ty->isFloatTy()) {
if (Base != HA_UNKNOWN && Base != HA_FLOAT)
return false;
Members = 1;
Base = HA_FLOAT;
} else if (Ty->isDoubleTy()) {
if (Base != HA_UNKNOWN && Base != HA_DOUBLE)
return false;
Members = 1;
Base = HA_DOUBLE;
} else if (const VectorType *VT = dyn_cast<VectorType>(Ty)) {
Members = 1;
switch (Base) {
case HA_FLOAT:
case HA_DOUBLE:
return false;
case HA_VECT64:
return VT->getBitWidth() == 64;
case HA_VECT128:
return VT->getBitWidth() == 128;
case HA_UNKNOWN:
switch (VT->getBitWidth()) {
case 64:
Base = HA_VECT64;
return true;
case 128:
Base = HA_VECT128;
return true;
default:
return false;
}
}
}
return (Members > 0 && Members <= 4);
}
bool ARMTargetLowering::functionArgumentNeedsConsecutiveRegisters(
Type *Ty, CallingConv::ID CallConv, bool isVarArg) const {
if (getEffectiveCallingConv(CallConv, isVarArg) !=
CallingConv::ARM_AAPCS_VFP)
return false;
HABaseType Base = HA_UNKNOWN;
uint64_t Members = 0;
bool IsHA = isHomogeneousAggregate(Ty, Base, Members);
DEBUG(dbgs() << "isHA: " << IsHA << " "; Ty->dump());
bool IsIntArray = Ty->isArrayTy() && Ty->getArrayElementType()->isIntegerTy();
return IsHA || IsIntArray;
}