// FIXME: Does this need to check IEEE bit on function?
bool SIFoldOperands::tryFoldOMod(MachineInstr &MI) {
const MachineOperand *RegOp;
int OMod;
std::tie(RegOp, OMod) = isOMod(MI);
if (OMod == SIOutMods::NONE || !RegOp->isReg() ||
RegOp->getSubReg() != AMDGPU::NoSubRegister ||
!hasOneNonDBGUseInst(*MRI, RegOp->getReg()))
return false;
MachineInstr *Def = MRI->getVRegDef(RegOp->getReg());
MachineOperand *DefOMod = TII->getNamedOperand(*Def, AMDGPU::OpName::omod);
if (!DefOMod || DefOMod->getImm() != SIOutMods::NONE)
return false;
// Clamp is applied after omod. If the source already has clamp set, don't
// fold it.
if (TII->hasModifiersSet(*Def, AMDGPU::OpName::clamp))
return false;
DEBUG(dbgs() << "Folding omod " << MI << " into " << *Def << '\n');
DefOMod->setImm(OMod);
MRI->replaceRegWith(MI.getOperand(0).getReg(), Def->getOperand(0).getReg());
MI.eraseFromParent();
return true;
}
void SIRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg,
int64_t Offset) const {
MachineBasicBlock *MBB = MI.getParent();
MachineFunction *MF = MBB->getParent();
const SISubtarget &Subtarget = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = Subtarget.getInstrInfo();
#ifndef NDEBUG
// FIXME: Is it possible to be storing a frame index to itself?
bool SeenFI = false;
for (const MachineOperand &MO: MI.operands()) {
if (MO.isFI()) {
if (SeenFI)
llvm_unreachable("should not see multiple frame indices");
SeenFI = true;
}
}
#endif
MachineOperand *FIOp = TII->getNamedOperand(MI, AMDGPU::OpName::vaddr);
assert(FIOp && FIOp->isFI() && "frame index must be address operand");
assert(TII->isMUBUF(MI));
MachineOperand *OffsetOp = TII->getNamedOperand(MI, AMDGPU::OpName::offset);
int64_t NewOffset = OffsetOp->getImm() + Offset;
assert(isUInt<12>(NewOffset) && "offset should be legal");
FIOp->ChangeToRegister(BaseReg, false);
OffsetOp->setImm(NewOffset);
}
void HexagonPeephole::ChangeOpInto(MachineOperand &Dst, MachineOperand &Src) {
assert (&Dst != &Src && "Cannot duplicate into itself");
switch (Dst.getType()) {
case MachineOperand::MO_Register:
if (Src.isReg()) {
Dst.setReg(Src.getReg());
} else if (Src.isImm()) {
Dst.ChangeToImmediate(Src.getImm());
} else {
llvm_unreachable("Unexpected src operand type");
}
break;
case MachineOperand::MO_Immediate:
if (Src.isImm()) {
Dst.setImm(Src.getImm());
} else if (Src.isReg()) {
Dst.ChangeToRegister(Src.getReg(), Src.isDef(), Src.isImplicit(),
Src.isKill(), Src.isDead(), Src.isUndef(),
Src.isDebug());
} else {
llvm_unreachable("Unexpected src operand type");
}
break;
default:
llvm_unreachable("Unexpected dst operand type");
break;
}
}
void SIRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg,
int64_t Offset) const {
MachineBasicBlock *MBB = MI.getParent();
MachineFunction *MF = MBB->getParent();
const SISubtarget &Subtarget = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = Subtarget.getInstrInfo();
#ifndef NDEBUG
// FIXME: Is it possible to be storing a frame index to itself?
bool SeenFI = false;
for (const MachineOperand &MO: MI.operands()) {
if (MO.isFI()) {
if (SeenFI)
llvm_unreachable("should not see multiple frame indices");
SeenFI = true;
}
}
#endif
MachineOperand *FIOp = TII->getNamedOperand(MI, AMDGPU::OpName::vaddr);
assert(FIOp && FIOp->isFI() && "frame index must be address operand");
assert(TII->isMUBUF(MI));
MachineOperand *OffsetOp = TII->getNamedOperand(MI, AMDGPU::OpName::offset);
int64_t NewOffset = OffsetOp->getImm() + Offset;
if (isUInt<12>(NewOffset)) {
// If we have a legal offset, fold it directly into the instruction.
FIOp->ChangeToRegister(BaseReg, false);
OffsetOp->setImm(NewOffset);
return;
}
// The offset is not legal, so we must insert an add of the offset.
MachineRegisterInfo &MRI = MF->getRegInfo();
unsigned NewReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
DebugLoc DL = MI.getDebugLoc();
assert(Offset != 0 && "Non-zero offset expected");
unsigned UnusedCarry = MRI.createVirtualRegister(&AMDGPU::SReg_64RegClass);
unsigned OffsetReg = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
// In the case the instruction already had an immediate offset, here only
// the requested new offset is added because we are leaving the original
// immediate in place.
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), OffsetReg)
.addImm(Offset);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_ADD_I32_e64), NewReg)
.addReg(UnusedCarry, RegState::Define | RegState::Dead)
.addReg(OffsetReg, RegState::Kill)
.addReg(BaseReg);
FIOp->ChangeToRegister(NewReg, false);
}
/// Add 4 to the displacement of operand MO.
static void fixDisp(MachineOperand &MO) {
switch (MO.getType()) {
default:
llvm_unreachable("Unhandled operand type.");
case MachineOperand::MO_Immediate:
MO.setImm(MO.getImm() + 4);
break;
case MachineOperand::MO_GlobalAddress:
case MachineOperand::MO_ConstantPoolIndex:
case MachineOperand::MO_BlockAddress:
case MachineOperand::MO_TargetIndex:
case MachineOperand::MO_ExternalSymbol:
MO.setOffset(MO.getOffset() + 4);
break;
}
}
bool SIFoldOperands::tryFoldClamp(MachineInstr &MI) {
const MachineOperand *ClampSrc = isClamp(MI);
if (!ClampSrc || !hasOneNonDBGUseInst(*MRI, ClampSrc->getReg()))
return false;
MachineInstr *Def = MRI->getVRegDef(ClampSrc->getReg());
if (!TII->hasFPClamp(*Def))
return false;
MachineOperand *DefClamp = TII->getNamedOperand(*Def, AMDGPU::OpName::clamp);
if (!DefClamp)
return false;
DEBUG(dbgs() << "Folding clamp " << *DefClamp << " into " << *Def << '\n');
// Clamp is applied after omod, so it is OK if omod is set.
DefClamp->setImm(1);
MRI->replaceRegWith(MI.getOperand(0).getReg(), Def->getOperand(0).getReg());
MI.eraseFromParent();
return true;
}