Bug 190799 – [ESNext][BigInt] Implement support for "**"

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Caio Lima]

...

[Caio Lima]

Created attachment 358509 [details]
WIP - Patch

[Caio Lima]

Created attachment 358518 [details]
WIP - Patch

[Caio Lima]

Created attachment 358520 [details]
Benchmarks

This Patch is perf neutral in x86_64

[Caio Lima]

Created attachment 360053 [details]
Patch

[Caio Lima]

Created attachment 360103 [details]
Patch

[Caio Lima]

Created attachment 360105 [details]
Benchmarks

The patch is perf neutral.

[Caio Lima]

Created attachment 362260 [details]
Patch

[Caio Lima]

Created attachment 364248 [details]
Patch

[Caio Lima]

Created attachment 369121 [details]
Patch

[Saam Barati]

Comment on attachment 369121 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=369121&action=review

LGTM overall. Some bugs I believe that are similar to ValueMod patch

> Source/JavaScriptCore/dfg/DFGAbstractInterpreterInlines.h:862
> +        if (childX && childY && childX.isNumber() && childY.isNumber()) {
> +            if (!node->isBinaryUseKind(BigIntUse))
> +                didFoldClobberWorld();
> +            setConstant(node, jsDoubleNumber(operationMathPow(childX.asNumber(), childY.asNumber())));
> +            break;
> +        }

same comment as I had in ValueMod patch. I believe you need to ensure that the types of X and Y are valid given BigInt.

> Source/JavaScriptCore/dfg/DFGOperations.cpp:545
> +    if (WTF::holds_alternative<JSBigInt*>(leftNumeric) || WTF::holds_alternative<JSBigInt*>(rightNumeric)) {
> +        if (WTF::holds_alternative<JSBigInt*>(leftNumeric) && WTF::holds_alternative<JSBigInt*>(rightNumeric))
> +            RELEASE_AND_RETURN(scope, JSValue::encode(JSBigInt::exponentiate(exec, WTF::get<JSBigInt*>(leftNumeric), WTF::get<JSBigInt*>(rightNumeric))));

Not related to this patch, but maybe we can make this type of pattern easier to read since it's so common in all the operations. Maybe this can be a followup. But if we had something like:
bool doBigIntMath(variant a, variant b, doMath lambda, throw lambda or throw message)

It returns false if both are *not* BigInt. Returns true otherwise. Throws an exception if 1 of 2 are big int, otherwise, returns lambda

[Caio Lima]

Created attachment 370657 [details]
Patch

[Caio Lima]

Created attachment 370705 [details]
Patch

[Saam Barati]

Comment on attachment 370705 [details]
Patch

r=me

[Caio Lima]

Created attachment 371195 [details]
Patch

[Caio Lima]

(In reply to Saam Barati from comment #13)
> Comment on attachment 370705 [details]
> Patch
> 
> r=me

Thx for the review!

[WebKit Commit Bot]

Comment on attachment 371195 [details]
Patch

Clearing flags on attachment: 371195

Committed r246041: <https://trac.webkit.org/changeset/246041>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/51361986>