Bug 26957 – RegExp::match to be optimized

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Yong Li]

Created attachment 32246 [details]
the patch

RegExp::match() should use inline Vector rather than OwnArrayPtr for better performance.

The patch is attached.

It gives more than 20% boost to sunspider string-unpack-code test, and gives about 8% boost to sunspider overall result.

[George Staikos]

Note: the performance test was done on ARM with bytecode (no JIT), no FPU.

[Gavin Sherlock]

Possibly a dupe of Bug 26316

[Maciej Stachowiak]

Comment on attachment 32246 [details]
the patch

Thanks for the patch! Looks like a good fix. On Mac OS X, this gives a 5% speedup on string-unpack code and smaller speedups on the other string tests, for an overall benefit of around 0.3% to SunSpider overall (some tests seem to slow down a bit but this appears to be noise). This patch doesn't handle the YARR case, I went ahead and filled that in and will post the revised patch in addition to commiting it. Also this patch was missing a ChangeLog.

[Yong Li]

Thanks, Maciej. Should I create a changelog by myself?

The reason why we get more speedup is probably that our memory allocation is much slower than Mac OS.

[Maciej Stachowiak]

For future patches, please include a ChangeLog entry (the prepare-ChangeLog script can fill in a lot of it for you). For this patch, I wrote a ChangeLog entry for you and checked it in (along with a few code changes to work with YARR) as <http://trac.webkit.org/changeset/45545>.

If the allocator on your platform is that slow, you may want to look into whether it's possible to use FastMalloc.

[Yong Li]

(In reply to comment #5)
> For future patches, please include a ChangeLog entry (the prepare-ChangeLog
> script can fill in a lot of it for you). For this patch, I wrote a ChangeLog
> entry for you and checked it in (along with a few code changes to work with
> YARR) as <http://trac.webkit.org/changeset/45545>.
> If the allocator on your platform is that slow, you may want to look into
> whether it's possible to use FastMalloc.

Thanks a lot. prepare-ChangeLog never works for me (in GIT BASH). I have to do it manually.

[Gavin Barraclough]

*** Bug 26368 has been marked as a duplicate of this bug. ***