[Webkit-unassigned] [Bug 174232] Reimplement bmalloc::log2 for MSVC

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=174232

--- Comment #5 from Daewoong Jang <daewoong.jang at navercorp.com> ---
(In reply to Geoffrey Garen from comment #4)
> I'm curious: Do you have a working version of bmalloc on Windows? If so,
> have you measured its performance?
> 
> It would be helpful to see the full design of the finished work, and to know
> its performance.
> 

Of course I have a working version of bmalloc on Windows in my working tree and you can see its performance gain here: https://bugs.webkit.org/show_bug.cgi?id=143310#c28
It's old but bmalloc on Windows gives definite 10%+ performance gain on JavaScript benchmarks run on Windows.

Never mind the cluttered patch you could see on the page above. It won't be cluttered like that in the finished work and won't be upstreamed until you're happy with the change :)

> Some of the decisions you've made seem to be compromises that harm
> performance.
> 
> To avoid muddying up the code, I think it's preferable to emulate missing
> functions on Windows. So, in this case, you could define __builtin_clzl only
> on windows, and use a recursive function or a tree of branches to compute it.

It could be done in that way for this small case, but how about other POSIX functions like pthread/mman APIs? I am thinking of putting an abstraction layer instead of emulation. Will it be OK for you?

Thank you for your comment.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20170708/d4625732/attachment.html>