1033959 - Remove use of wmemchr

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Hannes Verschore [:h4writer]]

Even though wmemchr used to be an improvement, it looks like we have even better performance with the unrolledMatcher. Possibly by the changes made, the unrolledMatcher is now compiled even better (removal of a branch). So it even defeats memchr/wmemchr in windows.

For linux memchr is still faster. The memchr provided by linux is 10x faster than the windows variant.

Comment 1

[Hannes Verschore [:h4writer]]

Attachment: strmatch_speed

Use the unrolledmatcher for windows and osx (clang isn't the issue. It are the libraries provided on osx for memchr)

Comment 2

[Hannes Verschore [:h4writer]]

On a testcase this gives an improvement from 400ms to 320ms

Comment 3

[Luke Wagner [:luke]]

Comment on attachment 8450038 [details] [diff] [review]
strmatch_speed

Review of attachment 8450038 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsstr.cpp
@@ +1174,2 @@
>      /*
> +     * Performance under memchr is horrible in OSX. Windows is better,

I'd say "Performance of memchr is..."

Comment 4

[Hannes Verschore [:h4writer]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a9ffbd31b81c

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a9ffbd31b81c

Comment 6

[Hannes Verschore [:h4writer]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/54a623fe6768

Comment 7

[Tim Taubert [:ttaubert] (inactive)]

https://hg.mozilla.org/mozilla-central/rev/54a623fe6768