void SIRegisterInfo::spillSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
const DebugLoc &DL = MI->getDebugLoc();
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned SuperReg = MI->getOperand(0).getReg();
bool IsKill = MI->getOperand(0).isKill();
// SubReg carries the "Kill" flag when SubReg == SuperReg.
unsigned SubKillState = getKillRegState((NumSubRegs == 1) && IsKill);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, getSubRegFromChannel(i));
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
if (SuperReg == AMDGPU::M0) {
assert(NumSubRegs == 1);
unsigned CopyM0
= MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), CopyM0)
.addReg(SuperReg, getKillRegState(IsKill));
// The real spill now kills the temp copy.
SubReg = SuperReg = CopyM0;
IsKill = true;
}
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_WRITELANE_B32),
Spill.VGPR)
.addReg(SubReg, getKillRegState(IsKill))
.addImm(Spill.Lane);
// FIXME: Since this spills to another register instead of an actual
// frame index, we should delete the frame index when all references to
// it are fixed.
} else {
// Spill SGPR to a frame index.
// FIXME we should use S_STORE_DWORD here for VI.
MachineInstrBuilder Mov
= BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addReg(SubReg, SubKillState);
// There could be undef components of a spilled super register.
// TODO: Can we detect this and skip the spill?
if (NumSubRegs > 1) {
// The last implicit use of the SuperReg carries the "Kill" flag.
unsigned SuperKillState = 0;
if (i + 1 == e)
SuperKillState |= getKillRegState(IsKill);
Mov.addReg(SuperReg, RegState::Implicit | SuperKillState);
}
unsigned Size = FrameInfo.getObjectSize(Index);
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
Size, Align);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_SAVE))
.addReg(TmpReg, RegState::Kill) // src
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srrsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
}
}
MI->eraseFromParent();
MFI->addToSpilledSGPRs(NumSubRegs);
}
void SIRegisterInfo::restoreSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
const DebugLoc &DL = MI->getDebugLoc();
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned SuperReg = MI->getOperand(0).getReg();
// m0 is not allowed as with readlane/writelane, so a temporary SGPR and
// extra copy is needed.
bool IsM0 = (SuperReg == AMDGPU::M0);
if (IsM0) {
assert(NumSubRegs == 1);
SuperReg = MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
}
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, getSubRegFromChannel(i));
SIMachineFunctionInfo::SpilledReg Spill
= MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
} else {
// Restore SGPR from a stack slot.
// FIXME: We should use S_LOAD_DWORD here for VI.
unsigned Align = FrameInfo.getObjectAlignment(Index);
unsigned Size = FrameInfo.getObjectSize(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index);
MachineMemOperand *MMO = MF->getMachineMemOperand(
PtrInfo, MachineMemOperand::MOLoad, Size, Align);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_RESTORE), TmpReg)
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
BuildMI(*MBB, MI, DL,
TII->get(AMDGPU::V_READFIRSTLANE_B32), SubReg)
.addReg(TmpReg, RegState::Kill)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
}
}
if (IsM0 && SuperReg != AMDGPU::M0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
.addReg(SuperReg);
}
MI->eraseFromParent();
}
void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
int SPAdj, unsigned FIOperandNum,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo *FrameInfo = MF->getFrameInfo();
const SIInstrInfo *TII =
static_cast<const SIInstrInfo *>(MF->getSubtarget().getInstrInfo());
DebugLoc DL = MI->getDebugLoc();
MachineOperand &FIOp = MI->getOperand(FIOperandNum);
int Index = MI->getOperand(FIOperandNum).getIndex();
switch (MI->getOpcode()) {
// SGPR register spill
case AMDGPU::SI_SPILL_S512_SAVE:
case AMDGPU::SI_SPILL_S256_SAVE:
case AMDGPU::SI_SPILL_S128_SAVE:
case AMDGPU::SI_SPILL_S64_SAVE:
case AMDGPU::SI_SPILL_S32_SAVE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.VGPR == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling SGPR");
}
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_WRITELANE_B32),
Spill.VGPR)
.addReg(SubReg)
.addImm(Spill.Lane);
}
MI->eraseFromParent();
break;
}
// SGPR register restore
case AMDGPU::SI_SPILL_S512_RESTORE:
case AMDGPU::SI_SPILL_S256_RESTORE:
case AMDGPU::SI_SPILL_S128_RESTORE:
case AMDGPU::SI_SPILL_S64_RESTORE:
case AMDGPU::SI_SPILL_S32_RESTORE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.VGPR == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling SGPR");
}
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
}
// TODO: only do this when it is needed
switch (MF->getSubtarget<AMDGPUSubtarget>().getGeneration()) {
case AMDGPUSubtarget::SOUTHERN_ISLANDS:
// "VALU writes SGPR" -> "SMRD reads that SGPR" needs "S_NOP 3" on SI
TII->insertNOPs(MI, 3);
break;
case AMDGPUSubtarget::SEA_ISLANDS:
break;
default: // VOLCANIC_ISLANDS and later
// "VALU writes SGPR -> VMEM reads that SGPR" needs "S_NOP 4" on VI
// and later. This also applies to VALUs which write VCC, but we're
// unlikely to see VMEM use VCC.
TII->insertNOPs(MI, 4);
}
MI->eraseFromParent();
break;
}
// VGPR register spill
case AMDGPU::SI_SPILL_V512_SAVE:
case AMDGPU::SI_SPILL_V256_SAVE:
case AMDGPU::SI_SPILL_V128_SAVE:
case AMDGPU::SI_SPILL_V96_SAVE:
case AMDGPU::SI_SPILL_V64_SAVE:
case AMDGPU::SI_SPILL_V32_SAVE:
buildScratchLoadStore(MI, AMDGPU::BUFFER_STORE_DWORD_OFFSET,
TII->getNamedOperand(*MI, AMDGPU::OpName::src)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
FrameInfo->getObjectOffset(Index), RS);
MI->eraseFromParent();
break;
case AMDGPU::SI_SPILL_V32_RESTORE:
case AMDGPU::SI_SPILL_V64_RESTORE:
case AMDGPU::SI_SPILL_V96_RESTORE:
case AMDGPU::SI_SPILL_V128_RESTORE:
case AMDGPU::SI_SPILL_V256_RESTORE:
case AMDGPU::SI_SPILL_V512_RESTORE: {
buildScratchLoadStore(MI, AMDGPU::BUFFER_LOAD_DWORD_OFFSET,
TII->getNamedOperand(*MI, AMDGPU::OpName::dst)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
FrameInfo->getObjectOffset(Index), RS);
MI->eraseFromParent();
break;
}
default: {
int64_t Offset = FrameInfo->getObjectOffset(Index);
FIOp.ChangeToImmediate(Offset);
if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, SPAdj);
BuildMI(*MBB, MI, MI->getDebugLoc(),
TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addImm(Offset);
FIOp.ChangeToRegister(TmpReg, false, false, true);
}
}
}
}
bool SILowerControlFlow::runOnMachineFunction(MachineFunction &MF) {
TII = static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
TRI =
static_cast<const SIRegisterInfo *>(MF.getSubtarget().getRegisterInfo());
SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
bool HaveKill = false;
bool NeedWQM = false;
bool NeedFlat = false;
unsigned Depth = 0;
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 (TII->isWQM(MI) || TII->isDS(MI))
NeedWQM = true;
// Flat uses m0 in case it needs to access LDS.
if (TII->isFLAT(MI))
NeedFlat = true;
switch (MI.getOpcode()) {
default: break;
case AMDGPU::SI_IF:
++Depth;
If(MI);
break;
case AMDGPU::SI_ELSE:
Else(MI);
break;
case AMDGPU::SI_BREAK:
Break(MI);
break;
case AMDGPU::SI_IF_BREAK:
IfBreak(MI);
break;
case AMDGPU::SI_ELSE_BREAK:
ElseBreak(MI);
break;
case AMDGPU::SI_LOOP:
++Depth;
Loop(MI);
break;
case AMDGPU::SI_END_CF:
if (--Depth == 0 && HaveKill) {
SkipIfDead(MI);
HaveKill = false;
}
EndCf(MI);
break;
case AMDGPU::SI_KILL:
if (Depth == 0)
SkipIfDead(MI);
else
HaveKill = true;
Kill(MI);
break;
case AMDGPU::S_BRANCH:
Branch(MI);
break;
case AMDGPU::SI_INDIRECT_SRC_V1:
case AMDGPU::SI_INDIRECT_SRC_V2:
case AMDGPU::SI_INDIRECT_SRC_V4:
case AMDGPU::SI_INDIRECT_SRC_V8:
case AMDGPU::SI_INDIRECT_SRC_V16:
IndirectSrc(MI);
break;
case AMDGPU::SI_INDIRECT_DST_V1:
case AMDGPU::SI_INDIRECT_DST_V2:
case AMDGPU::SI_INDIRECT_DST_V4:
case AMDGPU::SI_INDIRECT_DST_V8:
case AMDGPU::SI_INDIRECT_DST_V16:
IndirectDst(MI);
break;
}
}
}
if (NeedWQM && MFI->getShaderType() == ShaderType::PIXEL) {
MachineBasicBlock &MBB = MF.front();
BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
AMDGPU::EXEC).addReg(AMDGPU::EXEC);
}
if (NeedFlat && MFI->IsKernel) {
// TODO: What to use with function calls?
// We will need to Initialize the flat scratch register pair.
if (NeedFlat)
MFI->setHasFlatInstructions(true);
}
return true;
}
void SIRegisterInfo::restoreSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
const DebugLoc &DL = MI->getDebugLoc();
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned SuperReg = MI->getOperand(0).getReg();
bool SpillToSMEM = ST.hasScalarStores() && EnableSpillSGPRToSMEM;
assert(SuperReg != AMDGPU::M0 && "m0 should never spill");
unsigned OffsetReg = AMDGPU::M0;
unsigned M0CopyReg = AMDGPU::NoRegister;
if (SpillToSMEM) {
if (RS->isRegUsed(AMDGPU::M0)) {
M0CopyReg = MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), M0CopyReg)
.addReg(AMDGPU::M0);
}
}
// SubReg carries the "Kill" flag when SubReg == SuperReg.
int64_t FrOffset = FrameInfo.getObjectOffset(Index);
const unsigned EltSize = 4;
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, getSubRegFromChannel(i));
if (SpillToSMEM) {
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
EltSize, MinAlign(Align, EltSize * i));
// Add i * 4 offset
int64_t Offset = ST.getWavefrontSize() * (FrOffset + 4 * i);
if (Offset != 0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_ADD_U32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg())
.addImm(Offset);
} else {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg());
}
auto MIB =
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_BUFFER_LOAD_DWORD_SGPR), SubReg)
.addReg(MFI->getScratchRSrcReg()) // sbase
.addReg(OffsetReg, RegState::Kill) // soff
.addImm(0) // glc
.addMemOperand(MMO);
if (NumSubRegs > 1)
MIB.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
continue;
}
SIMachineFunctionInfo::SpilledReg Spill
= MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
auto MIB =
BuildMI(*MBB, MI, DL, TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane);
if (NumSubRegs > 1)
MIB.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
} else {
// Restore SGPR from a stack slot.
// FIXME: We should use S_LOAD_DWORD here for VI.
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO = MF->getMachineMemOperand(PtrInfo,
MachineMemOperand::MOLoad, EltSize,
MinAlign(Align, EltSize * i));
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_RESTORE), TmpReg)
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
auto MIB =
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), SubReg)
.addReg(TmpReg, RegState::Kill);
if (NumSubRegs > 1)
MIB.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
}
}
if (M0CopyReg != AMDGPU::NoRegister) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), AMDGPU::M0)
.addReg(M0CopyReg, RegState::Kill);
}
MI->eraseFromParent();
}
void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
int SPAdj, unsigned FIOperandNum,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
DebugLoc DL = MI->getDebugLoc();
MachineOperand &FIOp = MI->getOperand(FIOperandNum);
int Index = MI->getOperand(FIOperandNum).getIndex();
switch (MI->getOpcode()) {
// SGPR register spill
case AMDGPU::SI_SPILL_S512_SAVE:
case AMDGPU::SI_SPILL_S256_SAVE:
case AMDGPU::SI_SPILL_S128_SAVE:
case AMDGPU::SI_SPILL_S64_SAVE:
case AMDGPU::SI_SPILL_S32_SAVE: {
spillSGPR(MI, Index, RS);
break;
}
// SGPR register restore
case AMDGPU::SI_SPILL_S512_RESTORE:
case AMDGPU::SI_SPILL_S256_RESTORE:
case AMDGPU::SI_SPILL_S128_RESTORE:
case AMDGPU::SI_SPILL_S64_RESTORE:
case AMDGPU::SI_SPILL_S32_RESTORE: {
restoreSGPR(MI, Index, RS);
break;
}
// VGPR register spill
case AMDGPU::SI_SPILL_V512_SAVE:
case AMDGPU::SI_SPILL_V256_SAVE:
case AMDGPU::SI_SPILL_V128_SAVE:
case AMDGPU::SI_SPILL_V96_SAVE:
case AMDGPU::SI_SPILL_V64_SAVE:
case AMDGPU::SI_SPILL_V32_SAVE: {
const MachineOperand *VData = TII->getNamedOperand(*MI,
AMDGPU::OpName::vdata);
buildSpillLoadStore(MI, AMDGPU::BUFFER_STORE_DWORD_OFFSET,
Index,
VData->getReg(), VData->isKill(),
TII->getNamedOperand(*MI, AMDGPU::OpName::srsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::soffset)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(),
*MI->memoperands_begin(),
RS);
MFI->addToSpilledVGPRs(getNumSubRegsForSpillOp(MI->getOpcode()));
MI->eraseFromParent();
break;
}
case AMDGPU::SI_SPILL_V32_RESTORE:
case AMDGPU::SI_SPILL_V64_RESTORE:
case AMDGPU::SI_SPILL_V96_RESTORE:
case AMDGPU::SI_SPILL_V128_RESTORE:
case AMDGPU::SI_SPILL_V256_RESTORE:
case AMDGPU::SI_SPILL_V512_RESTORE: {
const MachineOperand *VData = TII->getNamedOperand(*MI,
AMDGPU::OpName::vdata);
buildSpillLoadStore(MI, AMDGPU::BUFFER_LOAD_DWORD_OFFSET,
Index,
VData->getReg(), VData->isKill(),
TII->getNamedOperand(*MI, AMDGPU::OpName::srsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::soffset)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(),
*MI->memoperands_begin(),
RS);
MI->eraseFromParent();
break;
}
default: {
if (TII->isMUBUF(*MI)) {
// Disable offen so we don't need a 0 vgpr base.
assert(static_cast<int>(FIOperandNum) ==
AMDGPU::getNamedOperandIdx(MI->getOpcode(),
AMDGPU::OpName::vaddr));
int64_t Offset = FrameInfo.getObjectOffset(Index);
int64_t OldImm
= TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm();
int64_t NewOffset = OldImm + Offset;
if (isUInt<12>(NewOffset) &&
buildMUBUFOffsetLoadStore(TII, FrameInfo, MI, Index, NewOffset)) {
MI->eraseFromParent();
break;
}
}
int64_t Offset = FrameInfo.getObjectOffset(Index);
FIOp.ChangeToImmediate(Offset);
if (!TII->isImmOperandLegal(*MI, FIOperandNum, FIOp)) {
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
BuildMI(*MBB, MI, MI->getDebugLoc(),
TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addImm(Offset);
FIOp.ChangeToRegister(TmpReg, false, false, true);
}
}
}
}
void SIRegisterInfo::restoreSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
const DebugLoc &DL = MI->getDebugLoc();
unsigned SuperReg = MI->getOperand(0).getReg();
bool SpillToSMEM = ST.hasScalarStores() && EnableSpillSGPRToSMEM;
assert(SuperReg != AMDGPU::M0 && "m0 should never spill");
unsigned OffsetReg = AMDGPU::M0;
unsigned M0CopyReg = AMDGPU::NoRegister;
if (SpillToSMEM) {
if (RS->isRegUsed(AMDGPU::M0)) {
M0CopyReg = MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), M0CopyReg)
.addReg(AMDGPU::M0);
}
}
unsigned EltSize = 4;
unsigned ScalarLoadOp;
const TargetRegisterClass *RC = getPhysRegClass(SuperReg);
if (SpillToSMEM && isSGPRClass(RC)) {
// XXX - if private_element_size is larger than 4 it might be useful to be
// able to spill wider vmem spills.
std::tie(EltSize, ScalarLoadOp) = getSpillEltSize(RC->getSize(), false);
}
ArrayRef<int16_t> SplitParts = getRegSplitParts(RC, EltSize);
unsigned NumSubRegs = SplitParts.empty() ? 1 : SplitParts.size();
// SubReg carries the "Kill" flag when SubReg == SuperReg.
int64_t FrOffset = FrameInfo.getObjectOffset(Index);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, SplitParts[i]);
if (SpillToSMEM) {
// FIXME: Size may be > 4 but extra bytes wasted.
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
EltSize, MinAlign(Align, EltSize * i));
// Add i * 4 offset
int64_t Offset = (ST.getWavefrontSize() * FrOffset) + (EltSize * i);
if (Offset != 0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_ADD_U32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg())
.addImm(Offset);
} else {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg());
}
auto MIB =
BuildMI(*MBB, MI, DL, TII->get(ScalarLoadOp), SubReg)
.addReg(MFI->getScratchRSrcReg()) // sbase
.addReg(OffsetReg, RegState::Kill) // soff
.addImm(0) // glc
.addMemOperand(MMO);
if (NumSubRegs > 1)
MIB.addReg(SuperReg, RegState::ImplicitDefine);
continue;
}
SIMachineFunctionInfo::SpilledReg Spill
= MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
auto MIB =
BuildMI(*MBB, MI, DL, TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane);
if (NumSubRegs > 1)
MIB.addReg(SuperReg, RegState::ImplicitDefine);
} else {
// Restore SGPR from a stack slot.
// FIXME: We should use S_LOAD_DWORD here for VI.
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO = MF->getMachineMemOperand(PtrInfo,
MachineMemOperand::MOLoad, EltSize,
MinAlign(Align, EltSize * i));
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_RESTORE), TmpReg)
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
auto MIB =
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), SubReg)
.addReg(TmpReg, RegState::Kill);
if (NumSubRegs > 1)
MIB.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
}
}
if (M0CopyReg != AMDGPU::NoRegister) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), AMDGPU::M0)
.addReg(M0CopyReg, RegState::Kill);
}
MI->eraseFromParent();
}
void SIRegisterInfo::spillSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineBasicBlock *MBB = MI->getParent();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned SuperReg = MI->getOperand(0).getReg();
bool IsKill = MI->getOperand(0).isKill();
const DebugLoc &DL = MI->getDebugLoc();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
bool SpillToSMEM = ST.hasScalarStores() && EnableSpillSGPRToSMEM;
assert(SuperReg != AMDGPU::M0 && "m0 should never spill");
const unsigned EltSize = 4;
unsigned OffsetReg = AMDGPU::M0;
unsigned M0CopyReg = AMDGPU::NoRegister;
if (SpillToSMEM) {
if (RS->isRegUsed(AMDGPU::M0)) {
M0CopyReg = MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), M0CopyReg)
.addReg(AMDGPU::M0);
}
}
// SubReg carries the "Kill" flag when SubReg == SuperReg.
unsigned SubKillState = getKillRegState((NumSubRegs == 1) && IsKill);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, getSubRegFromChannel(i));
if (SpillToSMEM) {
int64_t FrOffset = FrameInfo.getObjectOffset(Index);
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
EltSize, MinAlign(Align, EltSize * i));
// Add i * 4 wave offset.
//
// SMEM instructions only support a single offset, so increment the wave
// offset.
int64_t Offset = ST.getWavefrontSize() * (FrOffset + 4 * i);
if (Offset != 0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_ADD_U32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg())
.addImm(Offset);
} else {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg());
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_BUFFER_STORE_DWORD_SGPR))
.addReg(SubReg, getKillRegState(IsKill)) // sdata
.addReg(MFI->getScratchRSrcReg()) // sbase
.addReg(OffsetReg, RegState::Kill) // soff
.addImm(0) // glc
.addMemOperand(MMO);
continue;
}
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_WRITELANE_B32),
Spill.VGPR)
.addReg(SubReg, getKillRegState(IsKill))
.addImm(Spill.Lane);
// FIXME: Since this spills to another register instead of an actual
// frame index, we should delete the frame index when all references to
// it are fixed.
} else {
// Spill SGPR to a frame index.
// TODO: Should VI try to spill to VGPR and then spill to SMEM?
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
// TODO: Should VI try to spill to VGPR and then spill to SMEM?
MachineInstrBuilder Mov
= BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addReg(SubReg, SubKillState);
// There could be undef components of a spilled super register.
// TODO: Can we detect this and skip the spill?
if (NumSubRegs > 1) {
// The last implicit use of the SuperReg carries the "Kill" flag.
unsigned SuperKillState = 0;
if (i + 1 == e)
SuperKillState |= getKillRegState(IsKill);
Mov.addReg(SuperReg, RegState::Implicit | SuperKillState);
}
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
EltSize, MinAlign(Align, EltSize * i));
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_SAVE))
.addReg(TmpReg, RegState::Kill) // src
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srrsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
}
}
if (M0CopyReg != AMDGPU::NoRegister) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), AMDGPU::M0)
.addReg(M0CopyReg, RegState::Kill);
}
MI->eraseFromParent();
MFI->addToSpilledSGPRs(NumSubRegs);
}
void SIRegisterInfo::spillSGPR(MachineBasicBlock::iterator MI,
int Index,
RegScavenger *RS) const {
MachineBasicBlock *MBB = MI->getParent();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
unsigned SuperReg = MI->getOperand(0).getReg();
bool IsKill = MI->getOperand(0).isKill();
const DebugLoc &DL = MI->getDebugLoc();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo &FrameInfo = MF->getFrameInfo();
bool SpillToSMEM = ST.hasScalarStores() && EnableSpillSGPRToSMEM;
assert(SuperReg != AMDGPU::M0 && "m0 should never spill");
unsigned OffsetReg = AMDGPU::M0;
unsigned M0CopyReg = AMDGPU::NoRegister;
if (SpillToSMEM) {
if (RS->isRegUsed(AMDGPU::M0)) {
M0CopyReg = MRI.createVirtualRegister(&AMDGPU::SReg_32_XM0RegClass);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), M0CopyReg)
.addReg(AMDGPU::M0);
}
}
unsigned ScalarStoreOp;
unsigned EltSize = 4;
const TargetRegisterClass *RC = getPhysRegClass(SuperReg);
if (SpillToSMEM && isSGPRClass(RC)) {
// XXX - if private_element_size is larger than 4 it might be useful to be
// able to spill wider vmem spills.
std::tie(EltSize, ScalarStoreOp) = getSpillEltSize(RC->getSize(), true);
}
ArrayRef<int16_t> SplitParts = getRegSplitParts(RC, EltSize);
unsigned NumSubRegs = SplitParts.empty() ? 1 : SplitParts.size();
// SubReg carries the "Kill" flag when SubReg == SuperReg.
unsigned SubKillState = getKillRegState((NumSubRegs == 1) && IsKill);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = NumSubRegs == 1 ?
SuperReg : getSubReg(SuperReg, SplitParts[i]);
if (SpillToSMEM) {
int64_t FrOffset = FrameInfo.getObjectOffset(Index);
// The allocated memory size is really the wavefront size * the frame
// index size. The widest register class is 64 bytes, so a 4-byte scratch
// allocation is enough to spill this in a single stack object.
//
// FIXME: Frame size/offsets are computed earlier than this, so the extra
// space is still unnecessarily allocated.
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
EltSize, MinAlign(Align, EltSize * i));
// SMEM instructions only support a single offset, so increment the wave
// offset.
int64_t Offset = (ST.getWavefrontSize() * FrOffset) + (EltSize * i);
if (Offset != 0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_ADD_U32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg())
.addImm(Offset);
} else {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), OffsetReg)
.addReg(MFI->getScratchWaveOffsetReg());
}
BuildMI(*MBB, MI, DL, TII->get(ScalarStoreOp))
.addReg(SubReg, getKillRegState(IsKill)) // sdata
.addReg(MFI->getScratchRSrcReg()) // sbase
.addReg(OffsetReg, RegState::Kill) // soff
.addImm(0) // glc
.addMemOperand(MMO);
continue;
}
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_WRITELANE_B32),
Spill.VGPR)
.addReg(SubReg, getKillRegState(IsKill))
.addImm(Spill.Lane);
// FIXME: Since this spills to another register instead of an actual
// frame index, we should delete the frame index when all references to
// it are fixed.
} else {
// Spill SGPR to a frame index.
// TODO: Should VI try to spill to VGPR and then spill to SMEM?
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
// TODO: Should VI try to spill to VGPR and then spill to SMEM?
MachineInstrBuilder Mov
= BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addReg(SubReg, SubKillState);
// There could be undef components of a spilled super register.
// TODO: Can we detect this and skip the spill?
if (NumSubRegs > 1) {
// The last implicit use of the SuperReg carries the "Kill" flag.
unsigned SuperKillState = 0;
if (i + 1 == e)
SuperKillState |= getKillRegState(IsKill);
Mov.addReg(SuperReg, RegState::Implicit | SuperKillState);
}
unsigned Align = FrameInfo.getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index, EltSize * i);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
EltSize, MinAlign(Align, EltSize * i));
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_SAVE))
.addReg(TmpReg, RegState::Kill) // src
.addFrameIndex(Index) // vaddr
.addReg(MFI->getScratchRSrcReg()) // srrsrc
.addReg(MFI->getScratchWaveOffsetReg()) // soffset
.addImm(i * 4) // offset
.addMemOperand(MMO);
}
}
if (M0CopyReg != AMDGPU::NoRegister) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::COPY), AMDGPU::M0)
.addReg(M0CopyReg, RegState::Kill);
}
MI->eraseFromParent();
MFI->addToSpilledSGPRs(NumSubRegs);
}
bool SILowerControlFlowPass::runOnMachineFunction(MachineFunction &MF) {
TII = static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
TRI =
static_cast<const SIRegisterInfo *>(MF.getSubtarget().getRegisterInfo());
SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
bool HaveKill = false;
bool NeedM0 = false;
bool NeedWQM = false;
bool NeedFlat = false;
unsigned Depth = 0;
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 (TII->isDS(MI.getOpcode())) {
NeedM0 = true;
NeedWQM = true;
}
// Flat uses m0 in case it needs to access LDS.
if (TII->isFLAT(MI.getOpcode())) {
NeedM0 = true;
NeedFlat = true;
}
switch (MI.getOpcode()) {
default: break;
case AMDGPU::SI_IF:
++Depth;
If(MI);
break;
case AMDGPU::SI_ELSE:
Else(MI);
break;
case AMDGPU::SI_BREAK:
Break(MI);
break;
case AMDGPU::SI_IF_BREAK:
IfBreak(MI);
break;
case AMDGPU::SI_ELSE_BREAK:
ElseBreak(MI);
break;
case AMDGPU::SI_LOOP:
++Depth;
Loop(MI);
break;
case AMDGPU::SI_END_CF:
if (--Depth == 0 && HaveKill) {
SkipIfDead(MI);
HaveKill = false;
}
EndCf(MI);
break;
case AMDGPU::SI_KILL:
if (Depth == 0)
SkipIfDead(MI);
else
HaveKill = true;
Kill(MI);
break;
case AMDGPU::S_BRANCH:
Branch(MI);
break;
case AMDGPU::SI_INDIRECT_SRC:
IndirectSrc(MI);
break;
case AMDGPU::SI_INDIRECT_DST_V1:
case AMDGPU::SI_INDIRECT_DST_V2:
case AMDGPU::SI_INDIRECT_DST_V4:
case AMDGPU::SI_INDIRECT_DST_V8:
case AMDGPU::SI_INDIRECT_DST_V16:
IndirectDst(MI);
break;
case AMDGPU::V_INTERP_P1_F32:
case AMDGPU::V_INTERP_P2_F32:
case AMDGPU::V_INTERP_MOV_F32:
NeedWQM = true;
break;
}
}
}
if (NeedM0) {
MachineBasicBlock &MBB = MF.front();
// Initialize M0 to a value that won't cause LDS access to be discarded
// due to offset clamping
InitM0ForLDS(MBB.getFirstNonPHI());
}
if (NeedWQM && MFI->getShaderType() == ShaderType::PIXEL) {
MachineBasicBlock &MBB = MF.front();
BuildMI(MBB, MBB.getFirstNonPHI(), DebugLoc(), TII->get(AMDGPU::S_WQM_B64),
AMDGPU::EXEC).addReg(AMDGPU::EXEC);
}
// FIXME: This seems inappropriate to do here.
if (NeedFlat && MFI->IsKernel) {
// Insert the prologue initializing the SGPRs pointing to the scratch space
// for flat accesses.
const MachineFrameInfo *FrameInfo = MF.getFrameInfo();
// TODO: What to use with function calls?
// FIXME: This is reporting stack size that is used in a scratch buffer
// rather than registers as well.
uint64_t StackSizeBytes = FrameInfo->getStackSize();
int IndirectBegin
= static_cast<const AMDGPUInstrInfo*>(TII)->getIndirectIndexBegin(MF);
// Convert register index to 256-byte unit.
uint64_t StackOffset = IndirectBegin < 0 ? 0 : (4 * IndirectBegin / 256);
assert((StackSizeBytes < 0xffff) && StackOffset < 0xffff &&
"Stack limits should be smaller than 16-bits");
// Initialize the flat scratch register pair.
// TODO: Can we use one s_mov_b64 here?
// Offset is in units of 256-bytes.
MachineBasicBlock &MBB = MF.front();
DebugLoc NoDL;
MachineBasicBlock::iterator Start = MBB.getFirstNonPHI();
const MCInstrDesc &SMovK = TII->get(AMDGPU::S_MOVK_I32);
BuildMI(MBB, Start, NoDL, SMovK, AMDGPU::FLAT_SCR_LO)
.addImm(StackOffset);
// Documentation says size is "per-thread scratch size in bytes"
BuildMI(MBB, Start, NoDL, SMovK, AMDGPU::FLAT_SCR_HI)
.addImm(StackSizeBytes);
}
return true;
}
void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
int SPAdj, unsigned FIOperandNum,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo *FrameInfo = MF->getFrameInfo();
const SISubtarget &ST = MF->getSubtarget<SISubtarget>();
const SIInstrInfo *TII = ST.getInstrInfo();
DebugLoc DL = MI->getDebugLoc();
MachineOperand &FIOp = MI->getOperand(FIOperandNum);
int Index = MI->getOperand(FIOperandNum).getIndex();
switch (MI->getOpcode()) {
// SGPR register spill
case AMDGPU::SI_SPILL_S512_SAVE:
case AMDGPU::SI_SPILL_S256_SAVE:
case AMDGPU::SI_SPILL_S128_SAVE:
case AMDGPU::SI_SPILL_S64_SAVE:
case AMDGPU::SI_SPILL_S32_SAVE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
unsigned SuperReg = MI->getOperand(0).getReg();
bool IsKill = MI->getOperand(0).isKill();
// SubReg carries the "Kill" flag when SubReg == SuperReg.
unsigned SubKillState = getKillRegState((NumSubRegs == 1) && IsKill);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(SuperReg,
&AMDGPU::SGPR_32RegClass, i);
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_WRITELANE_B32),
Spill.VGPR)
.addReg(SubReg, getKillRegState(IsKill))
.addImm(Spill.Lane);
// FIXME: Since this spills to another register instead of an actual
// frame index, we should delete the frame index when all references to
// it are fixed.
} else {
// Spill SGPR to a frame index.
// FIXME we should use S_STORE_DWORD here for VI.
MachineInstrBuilder Mov
= BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addReg(SubReg, SubKillState);
// There could be undef components of a spilled super register.
// TODO: Can we detect this and skip the spill?
if (NumSubRegs > 1) {
// The last implicit use of the SuperReg carries the "Kill" flag.
unsigned SuperKillState = 0;
if (i + 1 == e)
SuperKillState |= getKillRegState(IsKill);
Mov.addReg(SuperReg, RegState::Implicit | SuperKillState);
}
unsigned Size = FrameInfo->getObjectSize(Index);
unsigned Align = FrameInfo->getObjectAlignment(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index);
MachineMemOperand *MMO
= MF->getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
Size, Align);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_SAVE))
.addReg(TmpReg, RegState::Kill) // src
.addFrameIndex(Index) // frame_idx
.addReg(MFI->getScratchRSrcReg()) // scratch_rsrc
.addReg(MFI->getScratchWaveOffsetReg()) // scratch_offset
.addImm(i * 4) // offset
.addMemOperand(MMO);
}
}
MI->eraseFromParent();
break;
}
// SGPR register restore
case AMDGPU::SI_SPILL_S512_RESTORE:
case AMDGPU::SI_SPILL_S256_RESTORE:
case AMDGPU::SI_SPILL_S128_RESTORE:
case AMDGPU::SI_SPILL_S64_RESTORE:
case AMDGPU::SI_SPILL_S32_RESTORE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.hasReg()) {
BuildMI(*MBB, MI, DL,
TII->getMCOpcodeFromPseudo(AMDGPU::V_READLANE_B32),
SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
} else {
// Restore SGPR from a stack slot.
// FIXME: We should use S_LOAD_DWORD here for VI.
unsigned Align = FrameInfo->getObjectAlignment(Index);
unsigned Size = FrameInfo->getObjectSize(Index);
MachinePointerInfo PtrInfo
= MachinePointerInfo::getFixedStack(*MF, Index);
MachineMemOperand *MMO = MF->getMachineMemOperand(
PtrInfo, MachineMemOperand::MOLoad, Size, Align);
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::SI_SPILL_V32_RESTORE), TmpReg)
.addFrameIndex(Index) // frame_idx
.addReg(MFI->getScratchRSrcReg()) // scratch_rsrc
.addReg(MFI->getScratchWaveOffsetReg()) // scratch_offset
.addImm(i * 4) // offset
.addMemOperand(MMO);
BuildMI(*MBB, MI, DL,
TII->get(AMDGPU::V_READFIRSTLANE_B32), SubReg)
.addReg(TmpReg, RegState::Kill)
.addReg(MI->getOperand(0).getReg(), RegState::ImplicitDefine);
}
}
MI->eraseFromParent();
break;
}
// VGPR register spill
case AMDGPU::SI_SPILL_V512_SAVE:
case AMDGPU::SI_SPILL_V256_SAVE:
case AMDGPU::SI_SPILL_V128_SAVE:
case AMDGPU::SI_SPILL_V96_SAVE:
case AMDGPU::SI_SPILL_V64_SAVE:
case AMDGPU::SI_SPILL_V32_SAVE:
buildScratchLoadStore(MI, AMDGPU::BUFFER_STORE_DWORD_OFFSET,
TII->getNamedOperand(*MI, AMDGPU::OpName::src),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
FrameInfo->getObjectOffset(Index) +
TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(), RS);
MI->eraseFromParent();
break;
case AMDGPU::SI_SPILL_V32_RESTORE:
case AMDGPU::SI_SPILL_V64_RESTORE:
case AMDGPU::SI_SPILL_V96_RESTORE:
case AMDGPU::SI_SPILL_V128_RESTORE:
case AMDGPU::SI_SPILL_V256_RESTORE:
case AMDGPU::SI_SPILL_V512_RESTORE: {
buildScratchLoadStore(MI, AMDGPU::BUFFER_LOAD_DWORD_OFFSET,
TII->getNamedOperand(*MI, AMDGPU::OpName::dst),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
FrameInfo->getObjectOffset(Index) +
TII->getNamedOperand(*MI, AMDGPU::OpName::offset)->getImm(), RS);
MI->eraseFromParent();
break;
}
default: {
int64_t Offset = FrameInfo->getObjectOffset(Index);
FIOp.ChangeToImmediate(Offset);
if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
BuildMI(*MBB, MI, MI->getDebugLoc(),
TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addImm(Offset);
FIOp.ChangeToRegister(TmpReg, false, false, true);
}
}
}
}
bool SILowerControlFlow::runOnMachineFunction(MachineFunction &MF) {
TII = static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
TRI =
static_cast<const SIRegisterInfo *>(MF.getSubtarget().getRegisterInfo());
SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
bool HaveKill = false;
bool NeedFlat = false;
unsigned Depth = 0;
for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
BI != BE; ++BI) {
MachineBasicBlock *EmptyMBBAtEnd = NULL;
MachineBasicBlock &MBB = *BI;
MachineBasicBlock::iterator I, Next;
bool ExecModified = false;
for (I = MBB.begin(); I != MBB.end(); I = Next) {
Next = std::next(I);
MachineInstr &MI = *I;
// Flat uses m0 in case it needs to access LDS.
if (TII->isFLAT(MI))
NeedFlat = true;
for (const auto &Def : I->defs()) {
if (Def.isReg() && Def.isDef() && Def.getReg() == AMDGPU::EXEC) {
ExecModified = true;
break;
}
}
switch (MI.getOpcode()) {
default: break;
case AMDGPU::SI_IF:
++Depth;
If(MI);
break;
case AMDGPU::SI_ELSE:
Else(MI, ExecModified);
break;
case AMDGPU::SI_BREAK:
Break(MI);
break;
case AMDGPU::SI_IF_BREAK:
IfBreak(MI);
break;
case AMDGPU::SI_ELSE_BREAK:
ElseBreak(MI);
break;
case AMDGPU::SI_LOOP:
++Depth;
Loop(MI);
break;
case AMDGPU::SI_END_CF:
if (--Depth == 0 && HaveKill) {
SkipIfDead(MI);
HaveKill = false;
}
EndCf(MI);
break;
case AMDGPU::SI_KILL:
if (Depth == 0)
SkipIfDead(MI);
else
HaveKill = true;
Kill(MI);
break;
case AMDGPU::S_BRANCH:
Branch(MI);
break;
case AMDGPU::SI_INDIRECT_SRC_V1:
case AMDGPU::SI_INDIRECT_SRC_V2:
case AMDGPU::SI_INDIRECT_SRC_V4:
case AMDGPU::SI_INDIRECT_SRC_V8:
case AMDGPU::SI_INDIRECT_SRC_V16:
IndirectSrc(MI);
break;
case AMDGPU::SI_INDIRECT_DST_V1:
case AMDGPU::SI_INDIRECT_DST_V2:
case AMDGPU::SI_INDIRECT_DST_V4:
case AMDGPU::SI_INDIRECT_DST_V8:
case AMDGPU::SI_INDIRECT_DST_V16:
IndirectDst(MI);
break;
case AMDGPU::S_ENDPGM: {
if (MF.getInfo<SIMachineFunctionInfo>()->returnsVoid())
break;
// Graphics shaders returning non-void shouldn't contain S_ENDPGM,
// because external bytecode will be appended at the end.
if (BI != --MF.end() || I != MBB.getFirstTerminator()) {
// S_ENDPGM is not the last instruction. Add an empty block at
// the end and jump there.
if (!EmptyMBBAtEnd) {
EmptyMBBAtEnd = MF.CreateMachineBasicBlock();
MF.insert(MF.end(), EmptyMBBAtEnd);
}
MBB.addSuccessor(EmptyMBBAtEnd);
BuildMI(*BI, I, MI.getDebugLoc(), TII->get(AMDGPU::S_BRANCH))
.addMBB(EmptyMBBAtEnd);
}
I->eraseFromParent();
break;
}
}
}
}
if (NeedFlat && MFI->IsKernel) {
// TODO: What to use with function calls?
// We will need to Initialize the flat scratch register pair.
if (NeedFlat)
MFI->setHasFlatInstructions(true);
}
return true;
}
void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
int SPAdj, unsigned FIOperandNum,
RegScavenger *RS) const {
MachineFunction *MF = MI->getParent()->getParent();
MachineBasicBlock *MBB = MI->getParent();
SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
MachineFrameInfo *FrameInfo = MF->getFrameInfo();
const SIInstrInfo *TII = static_cast<const SIInstrInfo*>(ST.getInstrInfo());
DebugLoc DL = MI->getDebugLoc();
MachineOperand &FIOp = MI->getOperand(FIOperandNum);
int Index = MI->getOperand(FIOperandNum).getIndex();
switch (MI->getOpcode()) {
// SGPR register spill
case AMDGPU::SI_SPILL_S512_SAVE:
case AMDGPU::SI_SPILL_S256_SAVE:
case AMDGPU::SI_SPILL_S128_SAVE:
case AMDGPU::SI_SPILL_S64_SAVE:
case AMDGPU::SI_SPILL_S32_SAVE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.VGPR == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling SGPR");
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_WRITELANE_B32), Spill.VGPR)
.addReg(SubReg)
.addImm(Spill.Lane);
}
MI->eraseFromParent();
break;
}
// SGPR register restore
case AMDGPU::SI_SPILL_S512_RESTORE:
case AMDGPU::SI_SPILL_S256_RESTORE:
case AMDGPU::SI_SPILL_S128_RESTORE:
case AMDGPU::SI_SPILL_S64_RESTORE:
case AMDGPU::SI_SPILL_S32_RESTORE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
for (unsigned i = 0, e = NumSubRegs; i < e; ++i) {
unsigned SubReg = getPhysRegSubReg(MI->getOperand(0).getReg(),
&AMDGPU::SGPR_32RegClass, i);
bool isM0 = SubReg == AMDGPU::M0;
struct SIMachineFunctionInfo::SpilledReg Spill =
MFI->getSpilledReg(MF, Index, i);
if (Spill.VGPR == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling SGPR");
}
if (isM0) {
SubReg = RS->scavengeRegister(&AMDGPU::SGPR_32RegClass, MI, 0);
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::V_READLANE_B32), SubReg)
.addReg(Spill.VGPR)
.addImm(Spill.Lane);
if (isM0) {
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
.addReg(SubReg);
}
}
TII->insertNOPs(MI, 3);
MI->eraseFromParent();
break;
}
// VGPR register spill
case AMDGPU::SI_SPILL_V512_SAVE:
case AMDGPU::SI_SPILL_V256_SAVE:
case AMDGPU::SI_SPILL_V128_SAVE:
case AMDGPU::SI_SPILL_V96_SAVE:
case AMDGPU::SI_SPILL_V64_SAVE:
case AMDGPU::SI_SPILL_V32_SAVE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned SrcReg = MI->getOperand(0).getReg();
int64_t Offset = FrameInfo->getObjectOffset(Index);
unsigned Size = NumSubRegs * 4;
unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, 0);
for (unsigned i = 0, e = NumSubRegs; i != e; ++i) {
unsigned SubReg = NumSubRegs > 1 ?
getPhysRegSubReg(SrcReg, &AMDGPU::VGPR_32RegClass, i) :
SrcReg;
Offset += (i * 4);
MFI->LDSWaveSpillSize = std::max((unsigned)Offset + 4, (unsigned)MFI->LDSWaveSpillSize);
unsigned AddrReg = TII->calculateLDSSpillAddress(*MBB, MI, RS, TmpReg,
Offset, Size);
if (AddrReg == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling VGPRS");
AddrReg = AMDGPU::VGPR0;
}
// Store the value in LDS
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::DS_WRITE_B32))
.addImm(0) // gds
.addReg(AddrReg, RegState::Kill) // addr
.addReg(SubReg) // data0
.addImm(0); // offset
}
MI->eraseFromParent();
break;
}
case AMDGPU::SI_SPILL_V32_RESTORE:
case AMDGPU::SI_SPILL_V64_RESTORE:
case AMDGPU::SI_SPILL_V128_RESTORE:
case AMDGPU::SI_SPILL_V256_RESTORE:
case AMDGPU::SI_SPILL_V512_RESTORE: {
unsigned NumSubRegs = getNumSubRegsForSpillOp(MI->getOpcode());
unsigned DstReg = MI->getOperand(0).getReg();
int64_t Offset = FrameInfo->getObjectOffset(Index);
unsigned Size = NumSubRegs * 4;
unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, 0);
// FIXME: We could use DS_READ_B64 here to optimize for larger registers.
for (unsigned i = 0, e = NumSubRegs; i != e; ++i) {
unsigned SubReg = NumSubRegs > 1 ?
getPhysRegSubReg(DstReg, &AMDGPU::VGPR_32RegClass, i) :
DstReg;
Offset += (i * 4);
unsigned AddrReg = TII->calculateLDSSpillAddress(*MBB, MI, RS, TmpReg,
Offset, Size);
if (AddrReg == AMDGPU::NoRegister) {
LLVMContext &Ctx = MF->getFunction()->getContext();
Ctx.emitError("Ran out of VGPRs for spilling VGPRs");
AddrReg = AMDGPU::VGPR0;
}
BuildMI(*MBB, MI, DL, TII->get(AMDGPU::DS_READ_B32), SubReg)
.addImm(0) // gds
.addReg(AddrReg, RegState::Kill) // addr
.addImm(0); //offset
}
MI->eraseFromParent();
break;
}
default: {
int64_t Offset = FrameInfo->getObjectOffset(Index);
FIOp.ChangeToImmediate(Offset);
if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VReg_32RegClass, MI, SPAdj);
BuildMI(*MBB, MI, MI->getDebugLoc(),
TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
.addImm(Offset);
FIOp.ChangeToRegister(TmpReg, false);
}
}
}
}
