in benchmarks/JetStream2/ARES-6/Air/allocate_stack.js [27:252]
function allocateStack(code)
{
if (code.frameSize)
throw new Error("Frame size already determined");
function attemptAssignment(slot, offsetFromFP, otherSlots)
{
if (offsetFromFP > 0)
throw new Error("Expect negative offset");
offsetFromFP = -roundUpToMultipleOf(slot.alignment, -offsetFromFP);
for (let otherSlot of otherSlots) {
if (!otherSlot.offsetFromFP)
continue;
let overlap = rangesOverlap(
offsetFromFP,
offsetFromFP + slot.byteSize,
otherSlot.offsetFromFP,
otherSlot.offsetFromFP + otherSlot.byteSize);
if (overlap)
return false;
}
slot.setOffsetFromFP(offsetFromFP);
return true;
}
function assign(slot, otherSlots)
{
if (attemptAssignment(slot, -slot.byteSize, otherSlots))
return;
for (let otherSlot of otherSlots) {
if (!otherSlot.offsetFromFP)
continue;
if (attemptAssignment(slot, otherSlot.offsetFromFP - slot.byteSize, otherSlots))
return;
}
throw new Error("Assignment failed");
}
// Allocate all of the escaped slots in order. This is kind of a crazy algorithm to allow for
// the possibility of stack slots being assigned frame offsets before we even get here.
let assignedEscapedStackSlots = [];
let escapedStackSlotsWorklist = [];
for (let slot of code.stackSlots) {
if (slot.isLocked) {
if (slot.offsetFromFP)
assignedEscapedStackSlots.push(slot);
else
escapedStackSlotsWorklist.push(slot);
} else {
if (slot.offsetFromFP)
throw new Error("Offset already assigned");
}
}
// This is a fairly espensive loop, but it's OK because we'll usually only have a handful of
// escaped stack slots.
while (escapedStackSlotsWorklist.length) {
let slot = escapedStackSlotsWorklist.pop();
assign(slot, assignedEscapedStackSlots);
assignedEscapedStackSlots.push(slot);
}
// Now we handle the spill slots.
let liveness = new Liveness(StackSlot, code);
let interference = new Map();
for (let slot of code.stackSlots)
interference.set(slot, new Set());
let slots = [];
for (let block of code) {
let localCalc = liveness.localCalc(block);
function interfere(instIndex)
{
Inst.forEachDef(
StackSlot, block.get(instIndex), block.get(instIndex + 1),
(slot, role, type, width) => {
if (!slot.isSpill)
return;
for (let otherSlot of localCalc.liveSet) {
interference.get(slot).add(otherSlot);
interference.get(otherSlot).add(slot);
}
});
}
for (let instIndex = block.size; instIndex--;) {
// Kill dead stores. For simplicity we say that a store is killable if it has only late
// defs and those late defs are to things that are dead right now. We only do that
// because that's the only kind of dead stack store we will see here.
let inst = block.at(instIndex);
if (!inst.hasNonArgEffects) {
let ok = true;
inst.forEachArg((arg, role, type, width) => {
if (Arg.isEarlyDef(role)) {
ok = false;
return;
}
if (!Arg.isLateDef(role))
return;
if (!arg.isStack) {
ok = false;
return;
}
let slot = arg.stackSlot;
if (!slot.isSpill) {
ok = false;
return;
}
if (localCalc.liveSet.has(slot)) {
ok = false;
return;
}
});
if (ok)
inst.clear();
}
interfere(instIndex);
localCalc.execute(instIndex);
}
interfere(-1);
removeAllMatching(block.insts, inst => inst.opcode == Nop);
}
// Now we assign stack locations. At its heart this algorithm is just first-fit. For each
// StackSlot we just want to find the offsetFromFP that is closest to zero while ensuring no
// overlap with other StackSlots that this overlaps with.
for (let slot of code.stackSlots) {
if (slot.offsetFromFP)
continue;
assign(slot, assignedEscapedStackSlots.concat(Array.from(interference.get(slot))));
}
// Figure out how much stack we're using for stack slots.
let frameSizeForStackSlots = 0;
for (let slot of code.stackSlots) {
frameSizeForStackSlots = Math.max(
frameSizeForStackSlots,
-slot.offsetFromFP);
}
frameSizeForStackSlots = roundUpToMultipleOf(stackAlignmentBytes, frameSizeForStackSlots);
// No we need to deduce how much argument area we need.
for (let block of code) {
for (let inst of block) {
for (let arg of inst.args) {
if (arg.isCallArg) {
// For now, we assume that we use 8 bytes of the call arg. But that's not
// such an awesome assumption.
// FIXME: https://bugs.webkit.org/show_bug.cgi?id=150454
if (arg.offset < 0)
throw new Error("Did not expect negative offset for callArg");
code.requestCallArgAreaSize(arg.offset + 8);
}
}
}
}
code.setFrameSize(frameSizeForStackSlots + code.callArgAreaSize);
// Finally transform the code to use Addrs instead of StackSlots. This is a lossless
// transformation since we can search the StackSlots array to figure out which StackSlot any
// offset-from-FP refers to.
// FIXME: This may produce addresses that aren't valid if we end up with a ginormous stack frame.
// We would have to scavenge for temporaries if this happened. Fortunately, this case will be
// extremely rare so we can do crazy things when it arises.
// https://bugs.webkit.org/show_bug.cgi?id=152530
let insertionSet = new InsertionSet();
for (let block of code) {
for (let instIndex = 0; instIndex < block.size; ++instIndex) {
let inst = block.at(instIndex);
inst.forEachArg((arg, role, type, width) => {
function stackAddr(offset)
{
return Arg.createStackAddr(offset, code.frameSize, width);
}
switch (arg.kind) {
case Arg.Stack: {
let slot = arg.stackSlot;
if (Arg.isZDef(role)
&& slot.isSpill
&& slot.byteSize > Arg.bytes(width)) {
// Currently we only handle this simple case because it's the only one
// that arises: ZDef's are only 32-bit right now. So, when we hit these
// assertions it means that we need to implement those other kinds of
// zero fills.
if (slot.byteSize != 8) {
throw new Error(
`Bad spill slot size for ZDef: ${slot.byteSize}, width is ${width}`);
}
if (width != 32)
throw new Error("Bad width for ZDef");
insertionSet.insert(
instIndex + 1,
new Inst(
StoreZero32,
[stackAddr(arg.offset + 4 + slot.offsetFromFP)]));
}
return stackAddr(arg.offset + slot.offsetFromFP);
}
case Arg.CallArg:
return stackAddr(arg.offset - code.frameSize);
default:
break;
}
});
}
insertionSet.execute(block.insts);
}
}