void handleCalleeSaves(Code& code)
{
PhaseScope phaseScope(code, "handleCalleeSaves");
RegisterSet usedCalleeSaves;
for (BasicBlock* block : code) {
for (Inst& inst : *block) {
inst.forEachTmpFast(
[&] (Tmp& tmp) {
// At first we just record all used regs.
usedCalleeSaves.set(tmp.reg());
});
if (inst.hasSpecial())
usedCalleeSaves.merge(inst.extraClobberedRegs());
}
}
// Now we filter to really get the callee saves.
usedCalleeSaves.filter(RegisterSet::calleeSaveRegisters());
usedCalleeSaves.exclude(RegisterSet::stackRegisters()); // We don't need to save FP here.
if (!usedCalleeSaves.numberOfSetRegisters())
return;
code.calleeSaveRegisters() = RegisterAtOffsetList(usedCalleeSaves);
size_t byteSize = 0;
for (const RegisterAtOffset& entry : code.calleeSaveRegisters())
byteSize = std::max(static_cast<size_t>(-entry.offset()), byteSize);
StackSlot* savesArea = code.addStackSlot(byteSize, StackSlotKind::Locked);
// This is a bit weird since we could have already pinned a different stack slot to this
// area. Also, our runtime does not require us to pin the saves area. Maybe we shouldn't pin it?
savesArea->setOffsetFromFP(-byteSize);
auto argFor = [&] (const RegisterAtOffset& entry) -> Arg {
return Arg::stack(savesArea, entry.offset() + byteSize);
};
InsertionSet insertionSet(code);
// First insert saving code in the prologue.
for (const RegisterAtOffset& entry : code.calleeSaveRegisters()) {
insertionSet.insert(
0, entry.reg().isGPR() ? Move : MoveDouble, code[0]->at(0).origin,
Tmp(entry.reg()), argFor(entry));
}
insertionSet.execute(code[0]);
// Now insert restore code at epilogues.
for (BasicBlock* block : code) {
Inst& last = block->last();
if (!isReturn(last.opcode))
continue;
for (const RegisterAtOffset& entry : code.calleeSaveRegisters()) {
insertionSet.insert(
block->size() - 1, entry.reg().isGPR() ? Move : MoveDouble, last.origin,
argFor(entry), Tmp(entry.reg()));
}
insertionSet.execute(block);
}
}
