Bug 196387 – Teach Call ICs how to call Wasm

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Saam Barati]

...

[Saam Barati]

Created attachment 366440 [details]
WIP

almost 2x faster on richards-wasm

I still need to make the code correct in the common case, e.g, teach the unwinder how to unwind past these frames and read callee saves, and restore wasm context

[Saam Barati]

Created attachment 366455 [details]
WIP

[Saam Barati]

Created attachment 366457 [details]
WIP

[Saam Barati]

Created attachment 366532 [details]
WIP

Wasm tests appear to pass

[Saam Barati]

I now need to make polymorphic calls work.

[Saam Barati]

Created attachment 366541 [details]
WIP

now with polymorphic ICs

[Saam Barati]

Created attachment 366571 [details]
WIP

Almost done. Just need to clean up a few things.

[Saam Barati]

Created attachment 366667 [details]
patch

[EWS Watchlist]

Attachment 366667 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/wasm/js/WebAssemblyFunctionHeapCellType.cpp:32:  Alphabetical sorting problem.  [build/include_order] [4]
Total errors found: 1 in 34 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Filip Pizlo]

Comment on attachment 366667 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=366667&action=review

> Source/JavaScriptCore/jit/Repatch.cpp:1151
> -            codePtr = variant.executable()->generatedJITCodeForCall()->addressForCall(ArityCheckNotRequired);
> +
> +#if ENABLE(WEBASSEMBLY)
> +            if (callLinkInfo.specializationKind() == CodeForCall) {
> +                if (JSFunction* function = variant.function()) {
> +                    if (auto* wasmFunction = jsDynamicCast<WebAssemblyFunction*>(vm, function))
> +                        codePtr = wasmFunction->jsCallEntrypoint();
> +                }
> +            }
> +#endif
> +
> +            if (!codePtr)
> +                codePtr = variant.executable()->generatedJITCodeForCall()->addressForCall(ArityCheckNotRequired);

Seems like we write this code or some variant of it in multiple places.

Helper?

[Saam Barati]

Comment on attachment 366667 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=366667&action=review

>> Source/JavaScriptCore/jit/Repatch.cpp:1151
>> +                codePtr = variant.executable()->generatedJITCodeForCall()->addressForCall(ArityCheckNotRequired);
> 
> Seems like we write this code or some variant of it in multiple places.
> 
> Helper?

Thanks for the review. Will fix.

[Saam Barati]

Created attachment 366674 [details]
patch for landing

[Saam Barati]

Created attachment 366730 [details]
patch for landing

With windows build fix.

[Saam Barati]

This is also a 12% speedup on https://pspdfkit.com/webassembly-benchmark/

[WebKit Commit Bot]

Comment on attachment 366730 [details]
patch for landing

Clearing flags on attachment: 366730

Committed r243886: <https://trac.webkit.org/changeset/243886>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/49616723>

[Yusuke Suzuki]

Committed r243898: <https://trac.webkit.org/changeset/243898>

[WebKit Commit Bot]

Re-opened since this is blocked by bug 196624

[Saam Barati]

Will land cloop build fix momentarily

[Saam Barati]

CLoop build fix:
http://trac.webkit.org/changeset/243907/webkit