-//===-- AArch64TargetTransformInfo.cpp - AArch64 specific TTI pass --------===//
+//===-- AArch64TargetTransformInfo.cpp - AArch64 specific TTI -------------===//
//
// The LLVM Compiler Infrastructure
//
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
-/// \file
-/// This file implements a TargetTransformInfo analysis pass specific to the
-/// AArch64 target machine. It uses the target's detailed information to provide
-/// more precise answers to certain TTI queries, while letting the target
-/// independent and default TTI implementations handle the rest.
-///
-//===----------------------------------------------------------------------===//
-#include "AArch64.h"
-#include "AArch64TargetMachine.h"
+#include "AArch64TargetTransformInfo.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Analysis/LoopInfo.h"
#include "llvm/CodeGen/BasicTTIImpl.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/CostTable.h"
#define DEBUG_TYPE "aarch64tti"
-namespace {
-
-class AArch64TTIImpl : public BasicTTIImplBase<AArch64TTIImpl> {
- typedef BasicTTIImplBase<AArch64TTIImpl> BaseT;
- typedef TargetTransformInfo TTI;
-
- const AArch64Subtarget *ST;
- const AArch64TargetLowering *TLI;
-
- /// Estimate the overhead of scalarizing an instruction. Insert and Extract
- /// are set if the result needs to be inserted and/or extracted from vectors.
- unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract);
-
- enum MemIntrinsicType {
- VECTOR_LDST_TWO_ELEMENTS,
- VECTOR_LDST_THREE_ELEMENTS,
- VECTOR_LDST_FOUR_ELEMENTS
- };
-
-public:
- explicit AArch64TTIImpl(const AArch64TargetMachine *TM = nullptr)
- : BaseT(TM), ST(TM ? TM->getSubtargetImpl() : nullptr),
- TLI(ST ? ST->getTargetLowering() : nullptr) {}
-
- // Provide value semantics. MSVC requires that we spell all of these out.
- AArch64TTIImpl(const AArch64TTIImpl &Arg)
- : BaseT(static_cast<const BaseT &>(Arg)), ST(Arg.ST), TLI(Arg.TLI) {}
- AArch64TTIImpl(AArch64TTIImpl &&Arg)
- : BaseT(std::move(static_cast<BaseT &>(Arg))), ST(std::move(Arg.ST)),
- TLI(std::move(Arg.TLI)) {}
- AArch64TTIImpl &operator=(const AArch64TTIImpl &RHS) {
- BaseT::operator=(static_cast<const BaseT &>(RHS));
- ST = RHS.ST;
- TLI = RHS.TLI;
- return *this;
- }
- AArch64TTIImpl &operator=(AArch64TTIImpl &&RHS) {
- BaseT::operator=(std::move(static_cast<BaseT &>(RHS)));
- ST = std::move(RHS.ST);
- TLI = std::move(RHS.TLI);
- return *this;
- }
-
- /// \name Scalar TTI Implementations
- /// @{
-
- using BaseT::getIntImmCost;
- unsigned getIntImmCost(int64_t Val);
- unsigned getIntImmCost(const APInt &Imm, Type *Ty);
- unsigned getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
- Type *Ty);
- unsigned getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
- Type *Ty);
- TTI::PopcntSupportKind getPopcntSupport(unsigned TyWidth);
-
- /// @}
-
- /// \name Vector TTI Implementations
- /// @{
-
- unsigned getNumberOfRegisters(bool Vector) {
- if (Vector) {
- if (ST->hasNEON())
- return 32;
- return 0;
- }
- return 31;
- }
-
- unsigned getRegisterBitWidth(bool Vector) {
- if (Vector) {
- if (ST->hasNEON())
- return 128;
- return 0;
- }
- return 64;
- }
-
- unsigned getMaxInterleaveFactor();
-
- unsigned getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src);
-
- unsigned getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index);
-
- unsigned getArithmeticInstrCost(
- unsigned Opcode, Type *Ty,
- TTI::OperandValueKind Opd1Info = TTI::OK_AnyValue,
- TTI::OperandValueKind Opd2Info = TTI::OK_AnyValue,
- TTI::OperandValueProperties Opd1PropInfo = TTI::OP_None,
- TTI::OperandValueProperties Opd2PropInfo = TTI::OP_None);
-
- unsigned getAddressComputationCost(Type *Ty, bool IsComplex);
-
- unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy);
-
- unsigned getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
- unsigned AddressSpace);
-
- unsigned getCostOfKeepingLiveOverCall(ArrayRef<Type *> Tys);
-
- void getUnrollingPreferences(const Function *F, Loop *L,
- TTI::UnrollingPreferences &UP);
-
- Value *getOrCreateResultFromMemIntrinsic(IntrinsicInst *Inst,
- Type *ExpectedType);
-
- bool getTgtMemIntrinsic(IntrinsicInst *Inst, MemIntrinsicInfo &Info);
-
- /// @}
-};
-
-} // end anonymous namespace
-
-ImmutablePass *
-llvm::createAArch64TargetTransformInfoPass(const AArch64TargetMachine *TM) {
- return new TargetTransformInfoWrapperPass(AArch64TTIImpl(TM));
-}
-
/// \brief Calculate the cost of materializing a 64-bit value. This helper
/// method might only calculate a fraction of a larger immediate. Therefore it
/// is valid to return a cost of ZERO.
-unsigned AArch64TTIImpl::getIntImmCost(int64_t Val) {
+int AArch64TTIImpl::getIntImmCost(int64_t Val) {
// Check if the immediate can be encoded within an instruction.
if (Val == 0 || AArch64_AM::isLogicalImmediate(Val, 64))
return 0;
}
/// \brief Calculate the cost of materializing the given constant.
-unsigned AArch64TTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
+int AArch64TTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// Split the constant into 64-bit chunks and calculate the cost for each
// chunk.
- unsigned Cost = 0;
+ int Cost = 0;
for (unsigned ShiftVal = 0; ShiftVal < BitSize; ShiftVal += 64) {
APInt Tmp = ImmVal.ashr(ShiftVal).sextOrTrunc(64);
int64_t Val = Tmp.getSExtValue();
Cost += getIntImmCost(Val);
}
// We need at least one instruction to materialze the constant.
- return std::max(1U, Cost);
+ return std::max(1, Cost);
}
-unsigned AArch64TTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx,
- const APInt &Imm, Type *Ty) {
+int AArch64TTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx,
+ const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
}
if (Idx == ImmIdx) {
- unsigned NumConstants = (BitSize + 63) / 64;
- unsigned Cost = AArch64TTIImpl::getIntImmCost(Imm, Ty);
+ int NumConstants = (BitSize + 63) / 64;
+ int Cost = AArch64TTIImpl::getIntImmCost(Imm, Ty);
return (Cost <= NumConstants * TTI::TCC_Basic)
- ? static_cast<unsigned>(TTI::TCC_Free)
+ ? static_cast<int>(TTI::TCC_Free)
: Cost;
}
return AArch64TTIImpl::getIntImmCost(Imm, Ty);
}
-unsigned AArch64TTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
- const APInt &Imm, Type *Ty) {
+int AArch64TTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
+ const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
if (Idx == 1) {
- unsigned NumConstants = (BitSize + 63) / 64;
- unsigned Cost = AArch64TTIImpl::getIntImmCost(Imm, Ty);
+ int NumConstants = (BitSize + 63) / 64;
+ int Cost = AArch64TTIImpl::getIntImmCost(Imm, Ty);
return (Cost <= NumConstants * TTI::TCC_Basic)
- ? static_cast<unsigned>(TTI::TCC_Free)
+ ? static_cast<int>(TTI::TCC_Free)
: Cost;
}
break;
return TTI::PSK_Software;
}
-unsigned AArch64TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
- Type *Src) {
+int AArch64TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) {
int ISD = TLI->InstructionOpcodeToISD(Opcode);
assert(ISD && "Invalid opcode");
- EVT SrcTy = TLI->getValueType(Src);
- EVT DstTy = TLI->getValueType(Dst);
+ EVT SrcTy = TLI->getValueType(DL, Src);
+ EVT DstTy = TLI->getValueType(DL, Dst);
if (!SrcTy.isSimple() || !DstTy.isSimple())
return BaseT::getCastInstrCost(Opcode, Dst, Src);
- static const TypeConversionCostTblEntry<MVT> ConversionTbl[] = {
+ static const TypeConversionCostTblEntry
+ ConversionTbl[] = {
+ { ISD::TRUNCATE, MVT::v4i16, MVT::v4i32, 1 },
+ { ISD::TRUNCATE, MVT::v4i32, MVT::v4i64, 0 },
+ { ISD::TRUNCATE, MVT::v8i8, MVT::v8i32, 3 },
+ { ISD::TRUNCATE, MVT::v16i8, MVT::v16i32, 6 },
+
+ // The number of shll instructions for the extension.
+ { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
+ { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
+ { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i32, 2 },
+ { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i32, 2 },
+ { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
+ { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
+ { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i16, 2 },
+ { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i16, 2 },
+ { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
+ { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
+ { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
+ { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
+ { ISD::SIGN_EXTEND, MVT::v16i16, MVT::v16i8, 2 },
+ { ISD::ZERO_EXTEND, MVT::v16i16, MVT::v16i8, 2 },
+ { ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
+ { ISD::ZERO_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
+
// LowerVectorINT_TO_FP:
{ ISD::SINT_TO_FP, MVT::v2f32, MVT::v2i32, 1 },
{ ISD::SINT_TO_FP, MVT::v4f32, MVT::v4i32, 1 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i8, 3 },
{ ISD::UINT_TO_FP, MVT::v4f32, MVT::v4i16, 2 },
+ // Complex: to v8f32
+ { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i8, 10 },
+ { ISD::SINT_TO_FP, MVT::v8f32, MVT::v8i16, 4 },
+ { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i8, 10 },
+ { ISD::UINT_TO_FP, MVT::v8f32, MVT::v8i16, 4 },
+
+ // Complex: to v16f32
+ { ISD::SINT_TO_FP, MVT::v16f32, MVT::v16i8, 21 },
+ { ISD::UINT_TO_FP, MVT::v16f32, MVT::v16i8, 21 },
+
// Complex: to v2f64
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i8, 4 },
{ ISD::SINT_TO_FP, MVT::v2f64, MVT::v2i16, 4 },
{ ISD::FP_TO_UINT, MVT::v2i8, MVT::v2f64, 2 },
};
- int Idx = ConvertCostTableLookup<MVT>(
- ConversionTbl, array_lengthof(ConversionTbl), ISD, DstTy.getSimpleVT(),
- SrcTy.getSimpleVT());
- if (Idx != -1)
- return ConversionTbl[Idx].Cost;
+ if (const auto *Entry = ConvertCostTableLookup(ConversionTbl, ISD,
+ DstTy.getSimpleVT(),
+ SrcTy.getSimpleVT()))
+ return Entry->Cost;
return BaseT::getCastInstrCost(Opcode, Dst, Src);
}
-unsigned AArch64TTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
- unsigned Index) {
+int AArch64TTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
+ unsigned Index) {
assert(Val->isVectorTy() && "This must be a vector type");
if (Index != -1U) {
// Legalize the type.
- std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Val);
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Val);
// This type is legalized to a scalar type.
if (!LT.second.isVector())
}
// All other insert/extracts cost this much.
- return 2;
+ return 3;
}
-unsigned AArch64TTIImpl::getArithmeticInstrCost(
+int AArch64TTIImpl::getArithmeticInstrCost(
unsigned Opcode, Type *Ty, TTI::OperandValueKind Opd1Info,
TTI::OperandValueKind Opd2Info, TTI::OperandValueProperties Opd1PropInfo,
TTI::OperandValueProperties Opd2PropInfo) {
// Legalize the type.
- std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Ty);
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Ty);
int ISD = TLI->InstructionOpcodeToISD(Opcode);
// normally expanded to the sequence ADD + CMP + SELECT + SRA.
// The OperandValue properties many not be same as that of previous
// operation; conservatively assume OP_None.
- unsigned Cost =
- getArithmeticInstrCost(Instruction::Add, Ty, Opd1Info, Opd2Info,
- TargetTransformInfo::OP_None,
- TargetTransformInfo::OP_None);
+ int Cost = getArithmeticInstrCost(Instruction::Add, Ty, Opd1Info, Opd2Info,
+ TargetTransformInfo::OP_None,
+ TargetTransformInfo::OP_None);
Cost += getArithmeticInstrCost(Instruction::Sub, Ty, Opd1Info, Opd2Info,
TargetTransformInfo::OP_None,
TargetTransformInfo::OP_None);
}
}
-unsigned AArch64TTIImpl::getAddressComputationCost(Type *Ty, bool IsComplex) {
+int AArch64TTIImpl::getAddressComputationCost(Type *Ty, bool IsComplex) {
// Address computations in vectorized code with non-consecutive addresses will
// likely result in more instructions compared to scalar code where the
// computation can more often be merged into the index mode. The resulting
return 1;
}
-unsigned AArch64TTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
- Type *CondTy) {
+int AArch64TTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+ Type *CondTy) {
int ISD = TLI->InstructionOpcodeToISD(Opcode);
- // We don't lower vector selects well that are wider than the register width.
+ // We don't lower some vector selects well that are wider than the register
+ // width.
if (ValTy->isVectorTy() && ISD == ISD::SELECT) {
// We would need this many instructions to hide the scalarization happening.
- unsigned AmortizationCost = 20;
- static const TypeConversionCostTblEntry<MVT::SimpleValueType>
+ const int AmortizationCost = 20;
+ static const TypeConversionCostTblEntry
VectorSelectTbl[] = {
- { ISD::SELECT, MVT::v16i1, MVT::v16i16, 16 * AmortizationCost },
- { ISD::SELECT, MVT::v8i1, MVT::v8i32, 8 * AmortizationCost },
- { ISD::SELECT, MVT::v16i1, MVT::v16i32, 16 * AmortizationCost },
+ { ISD::SELECT, MVT::v16i1, MVT::v16i16, 16 },
+ { ISD::SELECT, MVT::v8i1, MVT::v8i32, 8 },
+ { ISD::SELECT, MVT::v16i1, MVT::v16i32, 16 },
{ ISD::SELECT, MVT::v4i1, MVT::v4i64, 4 * AmortizationCost },
{ ISD::SELECT, MVT::v8i1, MVT::v8i64, 8 * AmortizationCost },
{ ISD::SELECT, MVT::v16i1, MVT::v16i64, 16 * AmortizationCost }
};
- EVT SelCondTy = TLI->getValueType(CondTy);
- EVT SelValTy = TLI->getValueType(ValTy);
+ EVT SelCondTy = TLI->getValueType(DL, CondTy);
+ EVT SelValTy = TLI->getValueType(DL, ValTy);
if (SelCondTy.isSimple() && SelValTy.isSimple()) {
- int Idx =
- ConvertCostTableLookup(VectorSelectTbl, ISD, SelCondTy.getSimpleVT(),
- SelValTy.getSimpleVT());
- if (Idx != -1)
- return VectorSelectTbl[Idx].Cost;
+ if (const auto *Entry = ConvertCostTableLookup(VectorSelectTbl, ISD,
+ SelCondTy.getSimpleVT(),
+ SelValTy.getSimpleVT()))
+ return Entry->Cost;
}
}
return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy);
}
-unsigned AArch64TTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src,
- unsigned Alignment,
- unsigned AddressSpace) {
- std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+int AArch64TTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src,
+ unsigned Alignment, unsigned AddressSpace) {
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Src);
if (Opcode == Instruction::Store && Src->isVectorTy() && Alignment != 16 &&
Src->getVectorElementType()->isIntegerTy(64)) {
// practice on inlined memcpy code.
// We make v2i64 stores expensive so that we will only vectorize if there
// are 6 other instructions getting vectorized.
- unsigned AmortizationCost = 6;
+ int AmortizationCost = 6;
return LT.first * 2 * AmortizationCost;
}
return LT.first;
}
-unsigned AArch64TTIImpl::getCostOfKeepingLiveOverCall(ArrayRef<Type *> Tys) {
- unsigned Cost = 0;
+int AArch64TTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
+ unsigned Factor,
+ ArrayRef<unsigned> Indices,
+ unsigned Alignment,
+ unsigned AddressSpace) {
+ assert(Factor >= 2 && "Invalid interleave factor");
+ assert(isa<VectorType>(VecTy) && "Expect a vector type");
+
+ if (Factor <= TLI->getMaxSupportedInterleaveFactor()) {
+ unsigned NumElts = VecTy->getVectorNumElements();
+ Type *SubVecTy = VectorType::get(VecTy->getScalarType(), NumElts / Factor);
+ unsigned SubVecSize = DL.getTypeSizeInBits(SubVecTy);
+
+ // ldN/stN only support legal vector types of size 64 or 128 in bits.
+ if (NumElts % Factor == 0 && (SubVecSize == 64 || SubVecSize == 128))
+ return Factor;
+ }
+
+ return BaseT::getInterleavedMemoryOpCost(Opcode, VecTy, Factor, Indices,
+ Alignment, AddressSpace);
+}
+
+int AArch64TTIImpl::getCostOfKeepingLiveOverCall(ArrayRef<Type *> Tys) {
+ int Cost = 0;
for (auto *I : Tys) {
if (!I->isVectorTy())
continue;
return Cost;
}
-unsigned AArch64TTIImpl::getMaxInterleaveFactor() {
+unsigned AArch64TTIImpl::getMaxInterleaveFactor(unsigned VF) {
if (ST->isCortexA57())
return 4;
return 2;
}
-void AArch64TTIImpl::getUnrollingPreferences(const Function *F, Loop *L,
+void AArch64TTIImpl::getUnrollingPreferences(Loop *L,
TTI::UnrollingPreferences &UP) {
+ // Enable partial unrolling and runtime unrolling.
+ BaseT::getUnrollingPreferences(L, UP);
+
+ // For inner loop, it is more likely to be a hot one, and the runtime check
+ // can be promoted out from LICM pass, so the overhead is less, let's try
+ // a larger threshold to unroll more loops.
+ if (L->getLoopDepth() > 1)
+ UP.PartialThreshold *= 2;
+
// Disable partial & runtime unrolling on -Os.
UP.PartialOptSizeThreshold = 0;
}
case Intrinsic::aarch64_neon_ld4:
Info.ReadMem = true;
Info.WriteMem = false;
- Info.Vol = false;
+ Info.IsSimple = true;
Info.NumMemRefs = 1;
Info.PtrVal = Inst->getArgOperand(0);
break;
case Intrinsic::aarch64_neon_st4:
Info.ReadMem = false;
Info.WriteMem = true;
- Info.Vol = false;
+ Info.IsSimple = true;
Info.NumMemRefs = 1;
Info.PtrVal = Inst->getArgOperand(Inst->getNumArgOperands() - 1);
break;