1448590 - Avoid useless use of SIMD-accelerated Latin1 to UTF-16 conversion in layout

Status: RESOLVED FIXED

(Core :: Layout, enhancement)

Description

[Henri Sivonen (:hsivonen) (temporarily away from Bugzilla)]

These uses are with strings that don't benefit from a function optimized for longer strings and I'd have to adjust the callers anyway for a new signature that no longer accepts zero-terminated strings.

Comment 1

[Henri Sivonen (:hsivonen) (temporarily away from Bugzilla)]

Attachment: Bug 1448590 - Avoid useless use of SIMD-accelerated Latin1 to UTF-16 conversion in layout. .

Review commit: https://reviewboard.mozilla.org/r/230944/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/230944/

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 8962086 [details]
Bug 1448590 - Avoid useless use of SIMD-accelerated Latin1 to UTF-16 conversion in layout. .

https://reviewboard.mozilla.org/r/230944/#review236732

Hmmm.  It feels weird that what was designed as an equivalent API with a different signature for when that signature made sense now has substantially different performance characteristics, and now needs to be avoided.  (That's more a comment as String module owner than Layout.)

From a layout perspective these changes are fine, though.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/7f3837852827
Avoid useless use of SIMD-accelerated Latin1 to UTF-16 conversion in layout. r=dbaron.

Comment 4

[Henri Sivonen (:hsivonen) (temporarily away from Bugzilla)]

(In reply to David Baron :dbaron: ⌚️UTC-7 from comment #2)
> Comment on attachment 8962086 [details]
> Bug 1448590 - Avoid useless use of SIMD-accelerated Latin1 to UTF-16
> conversion in layout. .
> 
> https://reviewboard.mozilla.org/r/230944/#review236732
> 
> Hmmm.  It feels weird that what was designed as an equivalent API with a
> different signature for when that signature made sense now has substantially
> different performance characteristics, and now needs to be avoided.  (That's
> more a comment as String module owner than Layout.)

Maybe putting SIMD rather than strlen in the summary was misleading. The SIMD vs. no SIMD difference and the strlen vs. no strlen difference was already there before I started changing stuff.

At least my removal of the bare-pointer-taking overloads that did strlen invisibly to the caller should make the performance characteristics more visible.

> From a layout perspective these changes are fine, though.

Thanks.

Comment 5

[Bogdan Tara[:bogdan_tara | bogdant]]

https://hg.mozilla.org/mozilla-central/rev/7f3837852827