1502051 - Cleanups in BaselineScript and BaselineCompiler

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, enhancement, P2)

Description

[Ted Campbell [:tcampbell]]

I'd like to clean up a few things in BaselineScript around being more precise about how we use code labels and entry points.

Comment 1

[Ted Campbell [:tcampbell]]

Attachment: Bug 1502051 - Cleanup up BaselineScript initializers. r?djvj

- Use C++ member initialization to simplify constructors.
- Mark class as final since it has trailing allocations.
- Make constructor private.

Comment 2

[Ted Campbell [:tcampbell]]

Attachment: Bug 1502051 - Cleanup BaselineScript entry point names. r?jandem

These are very specialized entry points so instead name them after their
use cases to make code easier to read and keep in sync.

Depends on D9787

Comment 3

[Ted Campbell [:tcampbell]]

Attachment: Bug 1502051 - Always check slotInfo for BaselineScript::nativeCodeForPc. r?djvj

Make the slotInfo argument required and assert that stack is
synchronized. A native code address for baseline is only well-defined if
the registers are setup correctly. Existing code was fine, this just
asserts some previously un-documented things.

Depends on D9789

Comment 4

[Ted Campbell [:tcampbell]]

Attachment: Bug 1502051 - Cleanup BaselineScript::nativeCodeForPc. r?djvj

Add a maybeNativeCodeForPc that returns nullptr on optimized-out code.
This is useful later for generating mappings without having to first
determine 'dead code'. Also reorganize the logic a little.

Depends on D9790

Comment 5

[Ted Campbell [:tcampbell]]

Attachment: Bug 1502051 - BaselineCompiler should use JSScript::yieldAndAwaitOffsets. r?jandem

It currently generates the same table in a round-about way so we should
just use the same table for consistency. Note that the BaselineScript
still gets it's own yield/await offsets table that is tracked in native
address offsets.

Depends on D9799

Comment 6

[Ted Campbell [:tcampbell]]

Apparently I didn't run with --ion-eager. I need to fix a failing testcase

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1eb75409c3e9
Cleanup up BaselineScript initializers. r=djvj
https://hg.mozilla.org/integration/autoland/rev/8f8923703a09
Cleanup BaselineScript entry point names. r=jandem
https://hg.mozilla.org/integration/autoland/rev/5c6343e86e05
Always check slotInfo for BaselineScript::nativeCodeForPc. r=djvj
https://hg.mozilla.org/integration/autoland/rev/fdda51be24d2
Cleanup BaselineScript::nativeCodeForPc. r=djvj
https://hg.mozilla.org/integration/autoland/rev/b2d4c84560b1
BaselineCompiler should use JSScript::yieldAndAwaitOffsets. r=jandem

Comment 8

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/1eb75409c3e9
https://hg.mozilla.org/mozilla-central/rev/8f8923703a09
https://hg.mozilla.org/mozilla-central/rev/5c6343e86e05
https://hg.mozilla.org/mozilla-central/rev/fdda51be24d2
https://hg.mozilla.org/mozilla-central/rev/b2d4c84560b1