// Find a scratch register that we can use at the start of the prologue to
// re-align the stack pointer. We avoid using callee-save registers since they
// may appear to be free when this is called from canUseAsPrologue (during
// shrink wrapping), but then no longer be free when this is called from
// emitPrologue.
//
// FIXME: This is a bit conservative, since in the above case we could use one
// of the callee-save registers as a scratch temp to re-align the stack pointer,
// but we would then have to make sure that we were in fact saving at least one
// callee-save register in the prologue, which is additional complexity that
// doesn't seem worth the benefit.
static unsigned findScratchNonCalleeSaveRegister(MachineBasicBlock *MBB) {
MachineFunction *MF = MBB->getParent();
// If MBB is an entry block, use X9 as the scratch register
if (&MF->front() == MBB)
return AArch64::X9;
RegScavenger RS;
RS.enterBasicBlock(*MBB);
// Prefer X9 since it was historically used for the prologue scratch reg.
if (!RS.isRegUsed(AArch64::X9))
return AArch64::X9;
// Find a free non callee-save reg.
const AArch64Subtarget &Subtarget = MF->getSubtarget<AArch64Subtarget>();
const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo();
const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(MF);
BitVector CalleeSaveRegs(RegInfo->getNumRegs());
for (unsigned i = 0; CSRegs[i]; ++i)
CalleeSaveRegs.set(CSRegs[i]);
BitVector Available = RS.getRegsAvailable(&AArch64::GPR64RegClass);
for (int AvailReg = Available.find_first(); AvailReg != -1;
AvailReg = Available.find_next(AvailReg))
if (!CalleeSaveRegs.test(AvailReg))
return AvailReg;
return AArch64::NoRegister;
}
int AArch64A57FPLoadBalancing::scavengeRegister(Chain *G, Color C,
MachineBasicBlock &MBB) {
RegScavenger RS;
RS.enterBasicBlock(&MBB);
RS.forward(MachineBasicBlock::iterator(G->getStart()));
// Can we find an appropriate register that is available throughout the life
// of the chain?
unsigned RegClassID = G->getStart()->getDesc().OpInfo[0].RegClass;
BitVector AvailableRegs = RS.getRegsAvailable(TRI->getRegClass(RegClassID));
for (MachineBasicBlock::iterator I = G->getStart(), E = G->getEnd();
I != E; ++I) {
RS.forward(I);
AvailableRegs &= RS.getRegsAvailable(TRI->getRegClass(RegClassID));
// Remove any registers clobbered by a regmask.
for (auto J : I->operands()) {
if (J.isRegMask())
AvailableRegs.clearBitsNotInMask(J.getRegMask());
}
}
// Make sure we allocate in-order, to get the cheapest registers first.
auto Ord = RCI.getOrder(TRI->getRegClass(RegClassID));
for (auto Reg : Ord) {
if (!AvailableRegs[Reg])
continue;
if ((C == Color::Even && (Reg % 2) == 0) ||
(C == Color::Odd && (Reg % 2) == 1))
return Reg;
}
return -1;
}
/// fixupLoopInsts - For Hexagon, if the loop label is to far from the
/// loop instruction then we need to set the LC0 and SA0 registers
/// explicitly instead of using LOOP(start,count). This function
/// checks the distance, and generates register assignments if needed.
///
/// This function makes two passes over the basic blocks. The first
/// pass computes the offset of the basic block from the start.
/// The second pass checks all the loop instructions.
bool HexagonFixupHwLoops::fixupLoopInstrs(MachineFunction &MF) {
// Offset of the current instruction from the start.
unsigned InstOffset = 0;
// Map for each basic block to it's first instruction.
DenseMap<MachineBasicBlock*, unsigned> BlockToInstOffset;
// First pass - compute the offset of each basic block.
for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
MBB != MBBe; ++MBB) {
BlockToInstOffset[MBB] = InstOffset;
InstOffset += (MBB->size() * 4);
}
// Second pass - check each loop instruction to see if it needs to
// be converted.
InstOffset = 0;
bool Changed = false;
RegScavenger RS;
// Loop over all the basic blocks.
for (MachineFunction::iterator MBB = MF.begin(), MBBe = MF.end();
MBB != MBBe; ++MBB) {
InstOffset = BlockToInstOffset[MBB];
RS.enterBasicBlock(MBB);
// Loop over all the instructions.
MachineBasicBlock::iterator MIE = MBB->end();
MachineBasicBlock::iterator MII = MBB->begin();
while (MII != MIE) {
if (isHardwareLoop(MII)) {
RS.forward(MII);
assert(MII->getOperand(0).isMBB() &&
"Expect a basic block as loop operand");
int diff = InstOffset - BlockToInstOffset[MII->getOperand(0).getMBB()];
diff = (diff > 0 ? diff : -diff);
if ((unsigned)diff > MAX_LOOP_DISTANCE) {
// Convert to explicity setting LC0 and SA0.
convertLoopInstr(MF, MII, RS);
MII = MBB->erase(MII);
Changed = true;
} else {
++MII;
}
} else {
++MII;
}
InstOffset += 4;
}
}
return Changed;
}
int AArch64A57FPLoadBalancing::scavengeRegister(Chain *G, Color C,
MachineBasicBlock &MBB) {
RegScavenger RS;
RS.enterBasicBlock(&MBB);
RS.forward(MachineBasicBlock::iterator(G->getStart()));
// Can we find an appropriate register that is available throughout the life
// of the chain?
unsigned RegClassID = G->getStart()->getDesc().OpInfo[0].RegClass;
BitVector AvailableRegs = RS.getRegsAvailable(TRI->getRegClass(RegClassID));
for (MachineBasicBlock::iterator I = G->getStart(), E = G->getEnd();
I != E; ++I) {
RS.forward(I);
AvailableRegs &= RS.getRegsAvailable(TRI->getRegClass(RegClassID));
// Remove any registers clobbered by a regmask or any def register that is
// immediately dead.
for (auto J : I->operands()) {
if (J.isRegMask())
AvailableRegs.clearBitsNotInMask(J.getRegMask());
if (J.isReg() && J.isDef()) {
MCRegAliasIterator AI(J.getReg(), TRI, /*IncludeSelf=*/true);
if (J.isDead())
for (; AI.isValid(); ++AI)
AvailableRegs.reset(*AI);
#ifndef NDEBUG
else
for (; AI.isValid(); ++AI)
assert(!AvailableRegs[*AI] &&
"Non-dead def should have been removed by now!");
#endif
}
}
}
// Make sure we allocate in-order, to get the cheapest registers first.
auto Ord = RCI.getOrder(TRI->getRegClass(RegClassID));
for (auto Reg : Ord) {
if (!AvailableRegs[Reg])
continue;
if ((C == Color::Even && (Reg % 2) == 0) ||
(C == Color::Odd && (Reg % 2) == 1))
return Reg;
}
return -1;
}
/// This function generates the sequence of instructions needed to get the
/// result of adding register REG and immediate IMM.
unsigned Mips16InstrInfo::loadImmediate(unsigned FrameReg, int64_t Imm,
MachineBasicBlock &MBB,
MachineBasicBlock::iterator II,
const DebugLoc &DL,
unsigned &NewImm) const {
//
// given original instruction is:
// Instr rx, T[offset] where offset is too big.
//
// lo = offset & 0xFFFF
// hi = ((offset >> 16) + (lo >> 15)) & 0xFFFF;
//
// let T = temporary register
// li T, hi
// shl T, 16
// add T, Rx, T
//
RegScavenger rs;
int32_t lo = Imm & 0xFFFF;
NewImm = lo;
int Reg =0;
int SpReg = 0;
rs.enterBasicBlock(MBB);
rs.forward(II);
//
// We need to know which registers can be used, in the case where there
// are not enough free registers. We exclude all registers that
// are used in the instruction that we are helping.
// // Consider all allocatable registers in the register class initially
BitVector Candidates =
RI.getAllocatableSet
(*II->getParent()->getParent(), &Mips::CPU16RegsRegClass);
// Exclude all the registers being used by the instruction.
for (unsigned i = 0, e = II->getNumOperands(); i != e; ++i) {
MachineOperand &MO = II->getOperand(i);
if (MO.isReg() && MO.getReg() != 0 && !MO.isDef() &&
!TargetRegisterInfo::isVirtualRegister(MO.getReg()))
Candidates.reset(MO.getReg());
}
// If the same register was used and defined in an instruction, then
// it will not be in the list of candidates.
//
// we need to analyze the instruction that we are helping.
// we need to know if it defines register x but register x is not
// present as an operand of the instruction. this tells
// whether the register is live before the instruction. if it's not
// then we don't need to save it in case there are no free registers.
int DefReg = 0;
for (unsigned i = 0, e = II->getNumOperands(); i != e; ++i) {
MachineOperand &MO = II->getOperand(i);
if (MO.isReg() && MO.isDef()) {
DefReg = MO.getReg();
break;
}
}
BitVector Available = rs.getRegsAvailable(&Mips::CPU16RegsRegClass);
Available &= Candidates;
//
// we use T0 for the first register, if we need to save something away.
// we use T1 for the second register, if we need to save something away.
//
unsigned FirstRegSaved =0, SecondRegSaved=0;
unsigned FirstRegSavedTo = 0, SecondRegSavedTo = 0;
Reg = Available.find_first();
if (Reg == -1) {
Reg = Candidates.find_first();
Candidates.reset(Reg);
if (DefReg != Reg) {
FirstRegSaved = Reg;
FirstRegSavedTo = Mips::T0;
copyPhysReg(MBB, II, DL, FirstRegSavedTo, FirstRegSaved, true);
}
}
else
Available.reset(Reg);
BuildMI(MBB, II, DL, get(Mips::LwConstant32), Reg).addImm(Imm).addImm(-1);
NewImm = 0;
if (FrameReg == Mips::SP) {
SpReg = Available.find_first();
if (SpReg == -1) {
SpReg = Candidates.find_first();
// Candidates.reset(SpReg); // not really needed
if (DefReg!= SpReg) {
SecondRegSaved = SpReg;
SecondRegSavedTo = Mips::T1;
}
if (SecondRegSaved)
copyPhysReg(MBB, II, DL, SecondRegSavedTo, SecondRegSaved, true);
}
else
Available.reset(SpReg);
copyPhysReg(MBB, II, DL, SpReg, Mips::SP, false);
BuildMI(MBB, II, DL, get(Mips:: AdduRxRyRz16), Reg).addReg(SpReg, RegState::Kill)
.addReg(Reg);
}
else
BuildMI(MBB, II, DL, get(Mips:: AdduRxRyRz16), Reg).addReg(FrameReg)
.addReg(Reg, RegState::Kill);
if (FirstRegSaved || SecondRegSaved) {
II = std::next(II);
if (FirstRegSaved)
copyPhysReg(MBB, II, DL, FirstRegSaved, FirstRegSavedTo, true);
if (SecondRegSaved)
copyPhysReg(MBB, II, DL, SecondRegSaved, SecondRegSavedTo, true);
}
return Reg;
}