[Webkit-unassigned] [Bug 160799] Replace % by bitwise &
bugzilla-daemon at webkit.org
bugzilla-daemon at webkit.org
Mon Aug 15 23:45:43 PDT 2016
https://bugs.webkit.org/show_bug.cgi?id=160799
Darin Adler <darin at apple.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Attachment #285899|review? |review-
Flags| |
--- Comment #6 from Darin Adler <darin at apple.com> ---
Comment on attachment 285899
--> https://bugs.webkit.org/attachment.cgi?id=285899
Patch
View in context: https://bugs.webkit.org/attachment.cgi?id=285899&action=review
>> Source/WTF/ChangeLog:9
>> + If we make size of HashTable as power
>
> I tried to run TestWTF with "/usr/bin/time" to measure usr/sys/real times, but variation for
> different test runs made me conclude that it is not the reliable way to measure performance.
> This variation is probably because tests in TestWTF depends on scheduling order of different
> threads which might be different in each run.
>
> Is there any official tool/tests/process to measure performance in webkit?
>
> I am not sure if there will be a immediate measurable performance improvement with
> this single change.
>
> This change was done with the thought that "If same result can be achieved with lesser
> cpu cycle, we should use that approach". I am guessing division by "power of 2 number"
> done thru & uses lesser cpu cycle than division by some arbitrary number.
>
> I think these smaller improvements could slowly adds up to give us better performance
> over time.
Sorry, thatâs not how we do performance on this project. We donât make changes based on a guess that something might be measurably faster. Further, I think your guess is likely incorrect. We do indeed have many real speedups that result from a collection many smaller improvements, but it is almost always many smaller *measurable* ones. Not changes that we guess are an improvement but canât measure. There are lots of techniques to measure performance in WebKit but I doubt there is any that will be able to measure this. All the standard techniques, such as writing a targeted micro benchmark that uses a function in a tight loop. If making the table larger to avoid collisions is worth it, you might be able to prove that using collision statistics. But I donât think any of that is going to work here; this really isnât a worthwhile improvement. It makes the code harder to read and almost certainly does not provide a commensurate speedup.
I think you should choose a more valuable project for your first WebKit patch, perhaps a bug fix. Do you have any other areas of interest?
> Source/WTF/wtf/ParkingLot.cpp:220
> + int leadingZeros = size > 1 ? __builtin_clz(size - 1) : __builtin_clz(size);
Does not compile on Windows because __builtin_clz is not supported on that platform.
--
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20160816/6ebe9bd1/attachment.html>
More information about the webkit-unassigned
mailing list