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;
}
void SIFoldOperands::foldOperand(
MachineOperand &OpToFold,
MachineInstr *UseMI,
unsigned UseOpIdx,
SmallVectorImpl<FoldCandidate> &FoldList,
SmallVectorImpl<MachineInstr *> &CopiesToReplace) const {
const MachineOperand &UseOp = UseMI->getOperand(UseOpIdx);
if (!isUseSafeToFold(TII, *UseMI, UseOp))
return;
// FIXME: Fold operands with subregs.
if (UseOp.isReg() && OpToFold.isReg()) {
if (UseOp.isImplicit() || UseOp.getSubReg() != AMDGPU::NoSubRegister)
return;
// Don't fold subregister extracts into tied operands, only if it is a full
// copy since a subregister use tied to a full register def doesn't really
// make sense. e.g. don't fold:
//
// %vreg1 = COPY %vreg0:sub1
// %vreg2<tied3> = V_MAC_{F16, F32} %vreg3, %vreg4, %vreg1<tied0>
//
// into
// %vreg2<tied3> = V_MAC_{F16, F32} %vreg3, %vreg4, %vreg0:sub1<tied0>
if (UseOp.isTied() && OpToFold.getSubReg() != AMDGPU::NoSubRegister)
return;
}
// 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->isRegSequence()) {
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);
}
return;
}
bool FoldingImm = OpToFold.isImm();
// In order to fold immediates into copies, we need to change the
// copy to a MOV.
if (FoldingImm && UseMI->isCopy()) {
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);
} else {
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) {
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;
}
const MCInstrDesc &FoldDesc = OpToFold.getParent()->getDesc();
const TargetRegisterClass *FoldRC =
TRI->getRegClass(FoldDesc.OpInfo[0].RegClass);
// Split 64-bit constants into 32-bits for folding.
if (UseOp.getSubReg() && AMDGPU::getRegBitWidth(FoldRC->getID()) == 64) {
unsigned UseReg = UseOp.getReg();
const TargetRegisterClass *UseRC
= TargetRegisterInfo::isVirtualRegister(UseReg) ?
MRI->getRegClass(UseReg) :
TRI->getPhysRegClass(UseReg);
if (AMDGPU::getRegBitWidth(UseRC->getID()) != 64)
return;
APInt Imm(64, OpToFold.getImm());
if (UseOp.getSubReg() == AMDGPU::sub0) {
Imm = Imm.getLoBits(32);
} else {
assert(UseOp.getSubReg() == AMDGPU::sub1);
Imm = Imm.getHiBits(32);
}
MachineOperand ImmOp = MachineOperand::CreateImm(Imm.getSExtValue());
tryAddToFoldList(FoldList, UseMI, UseOpIdx, &ImmOp, TII);
return;
}
tryAddToFoldList(FoldList, UseMI, UseOpIdx, &OpToFold, TII);
}