1765397 - Optimise Array.prototype.slice on arguments objects

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, enhancement)

Description

[André Bargull [:anba]]

A while ago I was checking when we fail to scalar replace the arguments object. I found two common code patterns:

• super(...arguments) (bug 1740737)
• Array.prototype.slice.call(arguments)

Optimising Array.prototype.slice.call(arguments) was blocked on bug 1738413, so now we can finally tackle this issue.

Comment 1

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 1: Add CacheIR support for Array.prototype.slice on arguments objects. r=iain!

The next patches will add Warp and scalar replacement support. The tests are based
on the existing arguments objects tests and cover the following cases:

1. begin and end index are both missing,
2. begin index is positive,
3. begin index is negative,
4. begin and end index are both present.

Comment 2

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 2: Transpile ArgumentsSliceResult for Warp. r=iain!

Transpile "ArgumentsSliceResult" similar to how "PackedArraySliceResult" is
optimised.

Depends on D144041

Comment 3

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 3: Scalar replace ArgumentsSlice on frame arguments. r=iain!

Adds support to scalar replace MArgumentsSlice for frame arguments. The next
patch will add support for inlined arguments.

Depends on D144042

Comment 4

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 4: Scalar replace ArgumentsSlice on inline arguments. r=iain!

Depends on D144043

Comment 5

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 5: Teach MMinMax that ArgumentsLength is never negative. r=iain!

This change ensures the MMinMax nodes added in part 3 can be folded away.

Depends on D144044

Comment 6

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 6: Inline CodeGenerator::emitRest(). r=iain!

There used to be multiple callers to emitRest() (bug 867471), but now
there's only a single caller, so we can inline it again.

Depends on D144047

Comment 7

[André Bargull [:anba]]

Attachment: Bug 1765397 - Part 7: Optimise slice when no end-index is present. r=iain!

This is the case which actually happens in practice and we should aim to
provide support for.

Depends on D144048

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/ba4b24512db8
Part 1: Add CacheIR support for Array.prototype.slice on arguments objects. r=iain
https://hg.mozilla.org/integration/autoland/rev/aeb3d90e0434
Part 2: Transpile ArgumentsSliceResult for Warp. r=iain
https://hg.mozilla.org/integration/autoland/rev/b037127d38f3
Part 3: Scalar replace ArgumentsSlice on frame arguments. r=iain
https://hg.mozilla.org/integration/autoland/rev/4676e18c3cf0
Part 4: Scalar replace ArgumentsSlice on inline arguments. r=iain
https://hg.mozilla.org/integration/autoland/rev/1e2bfa74a88e
Part 5: Teach MMinMax that ArgumentsLength is never negative. r=iain
https://hg.mozilla.org/integration/autoland/rev/257207e109b2
Part 6: Inline CodeGenerator::emitRest(). r=iain
https://hg.mozilla.org/integration/autoland/rev/201252c849f4
Part 7: Optimise slice when no end-index is present. r=iain

Comment 9

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/ba4b24512db8
https://hg.mozilla.org/mozilla-central/rev/aeb3d90e0434
https://hg.mozilla.org/mozilla-central/rev/b037127d38f3
https://hg.mozilla.org/mozilla-central/rev/4676e18c3cf0
https://hg.mozilla.org/mozilla-central/rev/1e2bfa74a88e
https://hg.mozilla.org/mozilla-central/rev/257207e109b2
https://hg.mozilla.org/mozilla-central/rev/201252c849f4