Bug 241252

Summary:	Optimize HTMLTokenizer::bufferASCIICharacter()
Product:	WebKit	Reporter:	Chris Dumez <cdumez>
Component:	WebCore Misc.	Assignee:	Chris Dumez <cdumez>
Status:	RESOLVED FIXED
Severity:	Normal	CC:	ap, webkit-bug-importer
Priority:	P2	Keywords:	InRadar
Version:	WebKit Nightly Build
Hardware:	Unspecified
OS:	Unspecified

Description Chris Dumez 2022-06-02 20:08:12 PDT

Optimize HTMLTokenizer::bufferASCIICharacter() by appending several characters at once whenever possible.

Comment 1 Chris Dumez 2022-06-02 20:17:41 PDT

Pull request: https://github.com/WebKit/WebKit/pull/1265

Comment 2 EWS 2022-06-02 22:42:38 PDT

Committed r295171 (251255@main): <https://commits.webkit.org/251255@main>

Reviewed commits have been landed. Closing PR #1265 and removing active labels.

Comment 3 Radar WebKit Bug Importer 2022-06-02 22:43:13 PDT

<rdar://problem/94330783>

Comment 4 Alexey Proskuryakov 2022-06-02 23:40:01 PDT

Is this actually an optimization? Seems like loop un-unrolling, this has to be slower than before?

Comment 5 Chris Dumez 2022-06-03 07:38:30 PDT

(In reply to Alexey Proskuryakov from comment #4)
> Is this actually an optimization? Seems like loop un-unrolling, this has to
> be slower than before?

I started A/B testing to confirm. In this case, the length of the literal is known at build time and the function is inlined so I wouldn't expect this to regress. However, I believe that appending all the characters at once to the Vector is beneficial so that it can make appropriate decisions internally about resizing its buffer.

We'll see.