unsigned X86TTI::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
const APInt &Imm, Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
switch (IID) {
default: return TCC_Free;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
if ((Idx == 1) && Imm.getBitWidth() <= 64 && isInt<32>(Imm.getSExtValue()))
return TCC_Free;
break;
case Intrinsic::experimental_stackmap:
if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TCC_Free;
break;
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TCC_Free;
break;
}
return X86TTI::getIntImmCost(Imm, Ty);
}
int PPCTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
if (DisablePPCConstHoist)
return BaseT::getIntImmCost(Imm, Ty);
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
if (Imm == 0)
return TTI::TCC_Free;
if (Imm.getBitWidth() <= 64) {
if (isInt<16>(Imm.getSExtValue()))
return TTI::TCC_Basic;
if (isInt<32>(Imm.getSExtValue())) {
// A constant that can be materialized using lis.
if ((Imm.getZExtValue() & 0xFFFF) == 0)
return TTI::TCC_Basic;
return 2 * TTI::TCC_Basic;
}
}
return 4 * TTI::TCC_Basic;
}
int X86TTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// There is no cost model for constants with a bit size of 0. Return TCC_Free
// here, so that constant hoisting will ignore this constant.
if (BitSize == 0)
return TTI::TCC_Free;
switch (IID) {
default:
return TTI::TCC_Free;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
if ((Idx == 1) && Imm.getBitWidth() <= 64 && isInt<32>(Imm.getSExtValue()))
return TTI::TCC_Free;
break;
case Intrinsic::experimental_stackmap:
if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TTI::TCC_Free;
break;
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TTI::TCC_Free;
break;
}
return X86TTIImpl::getIntImmCost(Imm, Ty);
}
Value *StraightLineStrengthReduce::emitBump(const Candidate &Basis,
const Candidate &C,
IRBuilder<> &Builder,
const DataLayout *DL,
bool &BumpWithUglyGEP) {
APInt Idx = C.Index->getValue(), BasisIdx = Basis.Index->getValue();
unifyBitWidth(Idx, BasisIdx);
APInt IndexOffset = Idx - BasisIdx;
BumpWithUglyGEP = false;
if (Basis.CandidateKind == Candidate::GEP) {
APInt ElementSize(
IndexOffset.getBitWidth(),
DL->getTypeAllocSize(
cast<GetElementPtrInst>(Basis.Ins)->getType()->getElementType()));
APInt Q, R;
APInt::sdivrem(IndexOffset, ElementSize, Q, R);
if (R.getSExtValue() == 0)
IndexOffset = Q;
else
BumpWithUglyGEP = true;
}
// Compute Bump = C - Basis = (i' - i) * S.
// Common case 1: if (i' - i) is 1, Bump = S.
if (IndexOffset.getSExtValue() == 1)
return C.Stride;
// Common case 2: if (i' - i) is -1, Bump = -S.
if (IndexOffset.getSExtValue() == -1)
return Builder.CreateNeg(C.Stride);
// Otherwise, Bump = (i' - i) * sext/trunc(S). Note that (i' - i) and S may
// have different bit widths.
IntegerType *DeltaType =
IntegerType::get(Basis.Ins->getContext(), IndexOffset.getBitWidth());
Value *ExtendedStride = Builder.CreateSExtOrTrunc(C.Stride, DeltaType);
if (IndexOffset.isPowerOf2()) {
// If (i' - i) is a power of 2, Bump = sext/trunc(S) << log(i' - i).
ConstantInt *Exponent = ConstantInt::get(DeltaType, IndexOffset.logBase2());
return Builder.CreateShl(ExtendedStride, Exponent);
}
if ((-IndexOffset).isPowerOf2()) {
// If (i - i') is a power of 2, Bump = -sext/trunc(S) << log(i' - i).
ConstantInt *Exponent =
ConstantInt::get(DeltaType, (-IndexOffset).logBase2());
return Builder.CreateNeg(Builder.CreateShl(ExtendedStride, Exponent));
}
Constant *Delta = ConstantInt::get(DeltaType, IndexOffset);
return Builder.CreateMul(ExtendedStride, Delta);
}
int SystemZTTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// There is no cost model for constants with a bit size of 0. Return TCC_Free
// here, so that constant hoisting will ignore this constant.
if (BitSize == 0)
return TTI::TCC_Free;
// No cost model for operations on integers larger than 64 bit implemented yet.
if (BitSize > 64)
return TTI::TCC_Free;
switch (IID) {
default:
return TTI::TCC_Free;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
// These get expanded to include a normal addition/subtraction.
if (Idx == 1 && Imm.getBitWidth() <= 64) {
if (isUInt<32>(Imm.getZExtValue()))
return TTI::TCC_Free;
if (isUInt<32>(-Imm.getSExtValue()))
return TTI::TCC_Free;
}
break;
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
// These get expanded to include a normal multiplication.
if (Idx == 1 && Imm.getBitWidth() <= 64) {
if (isInt<32>(Imm.getSExtValue()))
return TTI::TCC_Free;
}
break;
case Intrinsic::experimental_stackmap:
if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TTI::TCC_Free;
break;
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TTI::TCC_Free;
break;
}
return SystemZTTIImpl::getIntImmCost(Imm, Ty);
}
int X86TTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
// Never hoist constants larger than 128bit, because this might lead to
// incorrect code generation or assertions in codegen.
// Fixme: Create a cost model for types larger than i128 once the codegen
// issues have been fixed.
if (BitSize > 128)
return TTI::TCC_Free;
if (Imm == 0)
return TTI::TCC_Free;
// Sign-extend all constants to a multiple of 64-bit.
APInt ImmVal = Imm;
if (BitSize & 0x3f)
ImmVal = Imm.sext((BitSize + 63) & ~0x3fU);
// Split the constant into 64-bit chunks and calculate the cost for each
// chunk.
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(1, Cost);
}
int ARMTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned Bits = Ty->getPrimitiveSizeInBits();
if (Bits == 0 || Imm.getActiveBits() >= 64)
return 4;
int64_t SImmVal = Imm.getSExtValue();
uint64_t ZImmVal = Imm.getZExtValue();
if (!ST->isThumb()) {
if ((SImmVal >= 0 && SImmVal < 65536) ||
(ARM_AM::getSOImmVal(ZImmVal) != -1) ||
(ARM_AM::getSOImmVal(~ZImmVal) != -1))
return 1;
return ST->hasV6T2Ops() ? 2 : 3;
}
if (ST->isThumb2()) {
if ((SImmVal >= 0 && SImmVal < 65536) ||
(ARM_AM::getT2SOImmVal(ZImmVal) != -1) ||
(ARM_AM::getT2SOImmVal(~ZImmVal) != -1))
return 1;
return ST->hasV6T2Ops() ? 2 : 3;
}
// Thumb1, any i8 imm cost 1.
if (Bits == 8 || (SImmVal >= 0 && SImmVal < 256))
return 1;
if ((~SImmVal < 256) || ARM_AM::isThumbImmShiftedVal(ZImmVal))
return 2;
// Load from constantpool.
return 3;
}
/// \brief Finds and combines constants that can be easily rematerialized with
/// an add from a common base constant.
void ConstantHoisting::FindBaseConstants() {
// Sort the constants by value and type. This invalidates the mapping.
std::sort(ConstantMap.begin(), ConstantMap.end(), ConstantMapLessThan);
// Simple linear scan through the sorted constant map for viable merge
// candidates.
ConstantMapType::iterator MinValItr = ConstantMap.begin();
for (ConstantMapType::iterator I = llvm::next(ConstantMap.begin()),
E = ConstantMap.end(); I != E; ++I) {
if (MinValItr->first->getType() == I->first->getType()) {
// Check if the constant is in range of an add with immediate.
APInt Diff = I->first->getValue() - MinValItr->first->getValue();
if ((Diff.getBitWidth() <= 64) &&
TTI->isLegalAddImmediate(Diff.getSExtValue()))
continue;
}
// We either have now a different constant type or the constant is not in
// range of an add with immediate anymore.
FindAndMakeBaseConstant(MinValItr, I);
// Start a new base constant search.
MinValItr = I;
}
// Finalize the last base constant search.
FindAndMakeBaseConstant(MinValItr, ConstantMap.end());
}
int SystemZTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// There is no cost model for constants with a bit size of 0. Return TCC_Free
// here, so that constant hoisting will ignore this constant.
if (BitSize == 0)
return TTI::TCC_Free;
// No cost model for operations on integers larger than 64 bit implemented yet.
if (BitSize > 64)
return TTI::TCC_Free;
if (Imm == 0)
return TTI::TCC_Free;
if (Imm.getBitWidth() <= 64) {
// Constants loaded via lgfi.
if (isInt<32>(Imm.getSExtValue()))
return TTI::TCC_Basic;
// Constants loaded via llilf.
if (isUInt<32>(Imm.getZExtValue()))
return TTI::TCC_Basic;
// Constants loaded via llihf:
if ((Imm.getZExtValue() & 0xffffffff) == 0)
return TTI::TCC_Basic;
return 2 * TTI::TCC_Basic;
}
return 4 * TTI::TCC_Basic;
}
unsigned ARMTTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned Bits = Ty->getPrimitiveSizeInBits();
if (Bits == 0 || Bits > 32)
return 4;
int32_t SImmVal = Imm.getSExtValue();
uint32_t ZImmVal = Imm.getZExtValue();
if (!ST->isThumb()) {
if ((SImmVal >= 0 && SImmVal < 65536) ||
(ARM_AM::getSOImmVal(ZImmVal) != -1) ||
(ARM_AM::getSOImmVal(~ZImmVal) != -1))
return 1;
return ST->hasV6T2Ops() ? 2 : 3;
} else if (ST->isThumb2()) {
if ((SImmVal >= 0 && SImmVal < 65536) ||
(ARM_AM::getT2SOImmVal(ZImmVal) != -1) ||
(ARM_AM::getT2SOImmVal(~ZImmVal) != -1))
return 1;
return ST->hasV6T2Ops() ? 2 : 3;
} else /*Thumb1*/ {
if (SImmVal >= 0 && SImmVal < 256)
return 1;
if ((~ZImmVal < 256) || ARM_AM::isThumbImmShiftedVal(ZImmVal))
return 2;
// Load from constantpool.
return 3;
}
return 2;
}
unsigned X86TTI::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
unsigned ImmIdx = ~0U;
switch (Opcode) {
default: return TCC_Free;
case Instruction::GetElementPtr:
if (Idx == 0)
return 2 * TCC_Basic;
return TCC_Free;
case Instruction::Store:
ImmIdx = 0;
break;
case Instruction::Add:
case Instruction::Sub:
case Instruction::Mul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::ICmp:
ImmIdx = 1;
break;
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::IntToPtr:
case Instruction::PtrToInt:
case Instruction::BitCast:
case Instruction::PHI:
case Instruction::Call:
case Instruction::Select:
case Instruction::Ret:
case Instruction::Load:
break;
}
if ((Idx == ImmIdx) &&
Imm.getBitWidth() <= 64 && isInt<32>(Imm.getSExtValue()))
return TCC_Free;
return X86TTI::getIntImmCost(Imm, Ty);
}
/// We do not support symbolic projections yet, only 32-bit unsigned integers.
bool swift::getIntegerIndex(SILValue IndexVal, unsigned &IndexConst) {
if (auto *IndexLiteral = dyn_cast<IntegerLiteralInst>(IndexVal)) {
APInt ConstInt = IndexLiteral->getValue();
// IntegerLiterals are signed.
if (ConstInt.isIntN(32) && ConstInt.isNonNegative()) {
IndexConst = (unsigned)ConstInt.getSExtValue();
return true;
}
}
return false;
}
static void EmitAPInt(SmallVectorImpl<uint64_t> &Vals,
unsigned &Code, unsigned &AbbrevToUse, const APInt &Val) {
if (Val.getBitWidth() <= 64) {
uint64_t V = Val.getSExtValue();
emitSignedInt64(Vals, V);
Code = naclbitc::CST_CODE_INTEGER;
AbbrevToUse =
Val == 0 ? CONSTANTS_INTEGER_ZERO_ABBREV : CONSTANTS_INTEGER_ABBREV;
} else {
report_fatal_error("Wide integers are not supported");
}
}
unsigned X86TTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
if (Imm.getBitWidth() <= 64 &&
(isInt<32>(Imm.getSExtValue()) || isUInt<32>(Imm.getZExtValue())))
return TCC_Basic;
else
return 2 * TCC_Basic;
}
unsigned AArch64TTI::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
const APInt &Imm, Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// There is no cost model for constants with a bit size of 0. Return TCC_Free
// here, so that constant hoisting will ignore this constant.
if (BitSize == 0)
return TCC_Free;
switch (IID) {
default:
return TCC_Free;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
if (Idx == 1) {
unsigned NumConstants = (BitSize + 63) / 64;
unsigned Cost = AArch64TTI::getIntImmCost(Imm, Ty);
return (Cost <= NumConstants * TCC_Basic)
? static_cast<unsigned>(TCC_Free) : Cost;
}
break;
case Intrinsic::experimental_stackmap:
if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TCC_Free;
break;
case Intrinsic::experimental_patchpoint_void:
case Intrinsic::experimental_patchpoint_i64:
if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
return TCC_Free;
break;
}
return AArch64TTI::getIntImmCost(Imm, Ty);
}
unsigned X86TTI::getIntImmCost(unsigned Opcode, const APInt &Imm,
Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
switch (Opcode) {
case Instruction::Add:
case Instruction::Sub:
case Instruction::Mul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::ICmp:
if (Imm.getBitWidth() <= 64 && isInt<32>(Imm.getSExtValue()))
return TCC_Free;
else
return X86TTI::getIntImmCost(Imm, Ty);
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::IntToPtr:
case Instruction::PtrToInt:
case Instruction::BitCast:
case Instruction::Call:
case Instruction::Select:
case Instruction::Ret:
case Instruction::Load:
case Instruction::Store:
return X86TTI::getIntImmCost(Imm, Ty);
}
return TargetTransformInfo::getIntImmCost(Opcode, Imm, Ty);
}
unsigned ARM64TTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
int64_t Val = Imm.getSExtValue();
if (Val == 0 || ARM64_AM::isLogicalImmediate(Val, BitSize))
return 1;
if ((int64_t)Val < 0)
Val = ~Val;
if (BitSize == 32)
Val &= (1LL << 32) - 1;
unsigned LZ = countLeadingZeros((uint64_t)Val);
unsigned Shift = (63 - LZ) / 16;
// MOVZ is free so return true for one or fewer MOVK.
return (Shift == 0) ? 1 : Shift;
}
int ARMTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
Type *Ty) {
// Division by a constant can be turned into multiplication, but only if we
// know it's constant. So it's not so much that the immediate is cheap (it's
// not), but that the alternative is worse.
// FIXME: this is probably unneeded with GlobalISel.
if ((Opcode == Instruction::SDiv || Opcode == Instruction::UDiv ||
Opcode == Instruction::SRem || Opcode == Instruction::URem) &&
Idx == 1)
return 0;
if (Opcode == Instruction::And) {
// UXTB/UXTH
if (Imm == 255 || Imm == 65535)
return 0;
// Conversion to BIC is free, and means we can use ~Imm instead.
return std::min(getIntImmCost(Imm, Ty), getIntImmCost(~Imm, Ty));
}
if (Opcode == Instruction::Add)
// Conversion to SUB is free, and means we can use -Imm instead.
return std::min(getIntImmCost(Imm, Ty), getIntImmCost(-Imm, Ty));
if (Opcode == Instruction::ICmp && Imm.isNegative() &&
Ty->getIntegerBitWidth() == 32) {
int64_t NegImm = -Imm.getSExtValue();
if (ST->isThumb2() && NegImm < 1<<12)
// icmp X, #-C -> cmn X, #C
return 0;
if (ST->isThumb() && NegImm < 1<<8)
// icmp X, #-C -> adds X, #C
return 0;
}
// xor a, -1 can always be folded to MVN
if (Opcode == Instruction::Xor && Imm.isAllOnesValue())
return 0;
return getIntImmCost(Imm, Ty);
}
/// \brief Calculate the cost of materializing the given constant.
int AArch64TTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
// Sign-extend all constants to a multiple of 64-bit.
APInt ImmVal = Imm;
if (BitSize & 0x3f)
ImmVal = Imm.sext((BitSize + 63) & ~0x3fU);
// Split the constant into 64-bit chunks and calculate the cost for each
// chunk.
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(1, Cost);
}
Expr::Expr(APInt Int)
: Expr_((long int)Int.getSExtValue()) {
}
bool SIFoldOperands::runOnMachineFunction(MachineFunction &MF) {
MachineRegisterInfo &MRI = MF.getRegInfo();
const SIInstrInfo *TII =
static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
const SIRegisterInfo &TRI = TII->getRegisterInfo();
for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
BI != BE; ++BI) {
MachineBasicBlock &MBB = *BI;
MachineBasicBlock::iterator I, Next;
for (I = MBB.begin(); I != MBB.end(); I = Next) {
Next = std::next(I);
MachineInstr &MI = *I;
if (!isSafeToFold(MI.getOpcode()))
continue;
unsigned OpSize = TII->getOpSize(MI, 1);
MachineOperand &OpToFold = MI.getOperand(1);
bool FoldingImm = OpToFold.isImm();
// FIXME: We could also be folding things like FrameIndexes and
// TargetIndexes.
if (!FoldingImm && !OpToFold.isReg())
continue;
// Folding immediates with more than one use will increase program size.
// FIXME: This will also reduce register usage, which may be better
// in some cases. A better heuristic is needed.
if (FoldingImm && !TII->isInlineConstant(OpToFold, OpSize) &&
!MRI.hasOneUse(MI.getOperand(0).getReg()))
continue;
// FIXME: Fold operands with subregs.
if (OpToFold.isReg() &&
(!TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()) ||
OpToFold.getSubReg()))
continue;
std::vector<FoldCandidate> FoldList;
for (MachineRegisterInfo::use_iterator
Use = MRI.use_begin(MI.getOperand(0).getReg()), E = MRI.use_end();
Use != E; ++Use) {
MachineInstr *UseMI = Use->getParent();
const MachineOperand &UseOp = UseMI->getOperand(Use.getOperandNo());
// FIXME: Fold operands with subregs.
if (UseOp.isReg() && ((UseOp.getSubReg() && OpToFold.isReg()) ||
UseOp.isImplicit())) {
continue;
}
APInt Imm;
if (FoldingImm) {
unsigned UseReg = UseOp.getReg();
const TargetRegisterClass *UseRC
= TargetRegisterInfo::isVirtualRegister(UseReg) ?
MRI.getRegClass(UseReg) :
TRI.getRegClass(UseReg);
Imm = APInt(64, OpToFold.getImm());
// Split 64-bit constants into 32-bits for folding.
if (UseOp.getSubReg()) {
if (UseRC->getSize() != 8)
continue;
if (UseOp.getSubReg() == AMDGPU::sub0) {
Imm = Imm.getLoBits(32);
} else {
assert(UseOp.getSubReg() == AMDGPU::sub1);
Imm = Imm.getHiBits(32);
}
}
// In order to fold immediates into copies, we need to change the
// copy to a MOV.
if (UseMI->getOpcode() == AMDGPU::COPY) {
unsigned DestReg = UseMI->getOperand(0).getReg();
const TargetRegisterClass *DestRC
= TargetRegisterInfo::isVirtualRegister(DestReg) ?
MRI.getRegClass(DestReg) :
TRI.getRegClass(DestReg);
unsigned MovOp = TII->getMovOpcode(DestRC);
if (MovOp == AMDGPU::COPY)
continue;
UseMI->setDesc(TII->get(MovOp));
}
}
const MCInstrDesc &UseDesc = UseMI->getDesc();
// Don't fold into target independent nodes. Target independent opcodes
// don't have defined register classes.
if (UseDesc.isVariadic() ||
UseDesc.OpInfo[Use.getOperandNo()].RegClass == -1)
continue;
if (FoldingImm) {
MachineOperand ImmOp = MachineOperand::CreateImm(Imm.getSExtValue());
tryAddToFoldList(FoldList, UseMI, Use.getOperandNo(), &ImmOp, TII);
continue;
}
tryAddToFoldList(FoldList, UseMI, Use.getOperandNo(), &OpToFold, TII);
// FIXME: We could try to change the instruction from 64-bit to 32-bit
// to enable more folding opportunites. The shrink operands pass
// already does this.
}
for (FoldCandidate &Fold : FoldList) {
if (updateOperand(Fold, TRI)) {
// Clear kill flags.
if (!Fold.isImm()) {
assert(Fold.OpToFold && Fold.OpToFold->isReg());
Fold.OpToFold->setIsKill(false);
}
DEBUG(dbgs() << "Folded source from " << MI << " into OpNo " <<
Fold.UseOpNo << " of " << *Fold.UseMI << '\n');
}
}
}
}
return false;
}
int SystemZTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx,
const APInt &Imm, Type *Ty) {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
// There is no cost model for constants with a bit size of 0. Return TCC_Free
// here, so that constant hoisting will ignore this constant.
if (BitSize == 0)
return TTI::TCC_Free;
// No cost model for operations on integers larger than 64 bit implemented yet.
if (BitSize > 64)
return TTI::TCC_Free;
switch (Opcode) {
default:
return TTI::TCC_Free;
case Instruction::GetElementPtr:
// Always hoist the base address of a GetElementPtr. This prevents the
// creation of new constants for every base constant that gets constant
// folded with the offset.
if (Idx == 0)
return 2 * TTI::TCC_Basic;
return TTI::TCC_Free;
case Instruction::Store:
if (Idx == 0 && Imm.getBitWidth() <= 64) {
// Any 8-bit immediate store can by implemented via mvi.
if (BitSize == 8)
return TTI::TCC_Free;
// 16-bit immediate values can be stored via mvhhi/mvhi/mvghi.
if (isInt<16>(Imm.getSExtValue()))
return TTI::TCC_Free;
}
break;
case Instruction::ICmp:
if (Idx == 1 && Imm.getBitWidth() <= 64) {
// Comparisons against signed 32-bit immediates implemented via cgfi.
if (isInt<32>(Imm.getSExtValue()))
return TTI::TCC_Free;
// Comparisons against unsigned 32-bit immediates implemented via clgfi.
if (isUInt<32>(Imm.getZExtValue()))
return TTI::TCC_Free;
}
break;
case Instruction::Add:
case Instruction::Sub:
if (Idx == 1 && Imm.getBitWidth() <= 64) {
// We use algfi/slgfi to add/subtract 32-bit unsigned immediates.
if (isUInt<32>(Imm.getZExtValue()))
return TTI::TCC_Free;
// Or their negation, by swapping addition vs. subtraction.
if (isUInt<32>(-Imm.getSExtValue()))
return TTI::TCC_Free;
}
break;
case Instruction::Mul:
if (Idx == 1 && Imm.getBitWidth() <= 64) {
// We use msgfi to multiply by 32-bit signed immediates.
if (isInt<32>(Imm.getSExtValue()))
return TTI::TCC_Free;
}
break;
case Instruction::Or:
case Instruction::Xor:
if (Idx == 1 && Imm.getBitWidth() <= 64) {
// Masks supported by oilf/xilf.
if (isUInt<32>(Imm.getZExtValue()))
return TTI::TCC_Free;
// Masks supported by oihf/xihf.
if ((Imm.getZExtValue() & 0xffffffff) == 0)
return TTI::TCC_Free;
}
break;
case Instruction::And:
if (Idx == 1 && Imm.getBitWidth() <= 64) {
// Any 32-bit AND operation can by implemented via nilf.
if (BitSize <= 32)
return TTI::TCC_Free;
// 64-bit masks supported by nilf.
if (isUInt<32>(~Imm.getZExtValue()))
return TTI::TCC_Free;
// 64-bit masks supported by nilh.
if ((Imm.getZExtValue() & 0xffffffff) == 0xffffffff)
return TTI::TCC_Free;
// Some 64-bit AND operations can be implemented via risbg.
const SystemZInstrInfo *TII = ST->getInstrInfo();
unsigned Start, End;
if (TII->isRxSBGMask(Imm.getZExtValue(), BitSize, Start, End))
return TTI::TCC_Free;
}
break;
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
// Always return TCC_Free for the shift value of a shift instruction.
if (Idx == 1)
return TTI::TCC_Free;
break;
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::IntToPtr:
case Instruction::PtrToInt:
case Instruction::BitCast:
case Instruction::PHI:
case Instruction::Call:
case Instruction::Select:
case Instruction::Ret:
case Instruction::Load:
break;
}
return SystemZTTIImpl::getIntImmCost(Imm, Ty);
}
static void foldOperand(MachineOperand &OpToFold, MachineInstr *UseMI,
unsigned UseOpIdx,
std::vector<FoldCandidate> &FoldList,
SmallVectorImpl<MachineInstr *> &CopiesToReplace,
const SIInstrInfo *TII, const SIRegisterInfo &TRI,
MachineRegisterInfo &MRI) {
const MachineOperand &UseOp = UseMI->getOperand(UseOpIdx);
// FIXME: Fold operands with subregs.
if (UseOp.isReg() && ((UseOp.getSubReg() && OpToFold.isReg()) ||
UseOp.isImplicit())) {
return;
}
bool FoldingImm = OpToFold.isImm();
APInt Imm;
if (FoldingImm) {
unsigned UseReg = UseOp.getReg();
const TargetRegisterClass *UseRC
= TargetRegisterInfo::isVirtualRegister(UseReg) ?
MRI.getRegClass(UseReg) :
TRI.getPhysRegClass(UseReg);
Imm = APInt(64, OpToFold.getImm());
const MCInstrDesc &FoldDesc = TII->get(OpToFold.getParent()->getOpcode());
const TargetRegisterClass *FoldRC =
TRI.getRegClass(FoldDesc.OpInfo[0].RegClass);
// Split 64-bit constants into 32-bits for folding.
if (FoldRC->getSize() == 8 && UseOp.getSubReg()) {
if (UseRC->getSize() != 8)
return;
if (UseOp.getSubReg() == AMDGPU::sub0) {
Imm = Imm.getLoBits(32);
} else {
assert(UseOp.getSubReg() == AMDGPU::sub1);
Imm = Imm.getHiBits(32);
}
}
// In order to fold immediates into copies, we need to change the
// copy to a MOV.
if (UseMI->getOpcode() == AMDGPU::COPY) {
unsigned DestReg = UseMI->getOperand(0).getReg();
const TargetRegisterClass *DestRC
= TargetRegisterInfo::isVirtualRegister(DestReg) ?
MRI.getRegClass(DestReg) :
TRI.getPhysRegClass(DestReg);
unsigned MovOp = TII->getMovOpcode(DestRC);
if (MovOp == AMDGPU::COPY)
return;
UseMI->setDesc(TII->get(MovOp));
CopiesToReplace.push_back(UseMI);
}
}
// Special case for REG_SEQUENCE: We can't fold literals into
// REG_SEQUENCE instructions, so we have to fold them into the
// uses of REG_SEQUENCE.
if (UseMI->getOpcode() == AMDGPU::REG_SEQUENCE) {
unsigned RegSeqDstReg = UseMI->getOperand(0).getReg();
unsigned RegSeqDstSubReg = UseMI->getOperand(UseOpIdx + 1).getImm();
for (MachineRegisterInfo::use_iterator
RSUse = MRI.use_begin(RegSeqDstReg),
RSE = MRI.use_end(); RSUse != RSE; ++RSUse) {
MachineInstr *RSUseMI = RSUse->getParent();
if (RSUse->getSubReg() != RegSeqDstSubReg)
continue;
foldOperand(OpToFold, RSUseMI, RSUse.getOperandNo(), FoldList,
CopiesToReplace, TII, TRI, MRI);
}
return;
}
const MCInstrDesc &UseDesc = UseMI->getDesc();
// Don't fold into target independent nodes. Target independent opcodes
// don't have defined register classes.
if (UseDesc.isVariadic() ||
UseDesc.OpInfo[UseOpIdx].RegClass == -1)
return;
if (FoldingImm) {
MachineOperand ImmOp = MachineOperand::CreateImm(Imm.getSExtValue());
tryAddToFoldList(FoldList, UseMI, UseOpIdx, &ImmOp, TII);
return;
}
tryAddToFoldList(FoldList, UseMI, UseOpIdx, &OpToFold, TII);
// FIXME: We could try to change the instruction from 64-bit to 32-bit
// to enable more folding opportunites. The shrink operands pass
// already does this.
return;
}
int PPCTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
Type *Ty) {
if (DisablePPCConstHoist)
return BaseT::getIntImmCost(Opcode, Idx, Imm, Ty);
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
if (BitSize == 0)
return ~0U;
unsigned ImmIdx = ~0U;
bool ShiftedFree = false, RunFree = false, UnsignedFree = false,
ZeroFree = false;
switch (Opcode) {
default:
return TTI::TCC_Free;
case Instruction::GetElementPtr:
// Always hoist the base address of a GetElementPtr. This prevents the
// creation of new constants for every base constant that gets constant
// folded with the offset.
if (Idx == 0)
return 2 * TTI::TCC_Basic;
return TTI::TCC_Free;
case Instruction::And:
RunFree = true; // (for the rotate-and-mask instructions)
LLVM_FALLTHROUGH;
case Instruction::Add:
case Instruction::Or:
case Instruction::Xor:
ShiftedFree = true;
LLVM_FALLTHROUGH;
case Instruction::Sub:
case Instruction::Mul:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
ImmIdx = 1;
break;
case Instruction::ICmp:
UnsignedFree = true;
ImmIdx = 1;
// Zero comparisons can use record-form instructions.
LLVM_FALLTHROUGH;
case Instruction::Select:
ZeroFree = true;
break;
case Instruction::PHI:
case Instruction::Call:
case Instruction::Ret:
case Instruction::Load:
case Instruction::Store:
break;
}
if (ZeroFree && Imm == 0)
return TTI::TCC_Free;
if (Idx == ImmIdx && Imm.getBitWidth() <= 64) {
if (isInt<16>(Imm.getSExtValue()))
return TTI::TCC_Free;
if (RunFree) {
if (Imm.getBitWidth() <= 32 &&
(isShiftedMask_32(Imm.getZExtValue()) ||
isShiftedMask_32(~Imm.getZExtValue())))
return TTI::TCC_Free;
if (ST->isPPC64() &&
(isShiftedMask_64(Imm.getZExtValue()) ||
isShiftedMask_64(~Imm.getZExtValue())))
return TTI::TCC_Free;
}
if (UnsignedFree && isUInt<16>(Imm.getZExtValue()))
return TTI::TCC_Free;
if (ShiftedFree && (Imm.getZExtValue() & 0xFFFF) == 0)
return TTI::TCC_Free;
}
return PPCTTIImpl::getIntImmCost(Imm, Ty);
}