bmalloc: speed up the lock slow path

[Geoffrey Garen]

bmalloc: speed up the lock slow path

[Geoffrey Garen]

Created attachment 286629 [details]
Patch

[WebKit Commit Bot]

Attachment 286629 [details] did not pass style-queue:


ERROR: Source/bmalloc/bmalloc/StaticMutex.cpp:26:  Found other header before WebCore config.h. Should be: config.h, primary header, blank line, and then alphabetically sorted.  [build/include_order] [4]
ERROR: Source/bmalloc/bmalloc/StaticMutex.cpp:40:  More than one command on the same line  [whitespace/newline] [4]
Total errors found: 2 in 6 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Filip Pizlo]

Comment on attachment 286629 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=286629&action=review

> Source/bmalloc/ChangeLog:26
> +        To correct these problems, this patch adds a little bit of spinning to
> +        the bmalloc lock slow path.

Nice.

> Source/bmalloc/bmalloc/StaticMutex.cpp:40
> +        auto clear = makeScopeExit([&] { m_isSpinning.clear(); });

LOL.  I can't make up my mind if this is awesome or confusing.

> Source/bmalloc/bmalloc/ThreadSwitch.h:38
> +    swtch();

What is this?

[Geoffrey Garen]

> > Source/bmalloc/bmalloc/StaticMutex.cpp:40
> > +        auto clear = makeScopeExit([&] { m_isSpinning.clear(); });
> 
> LOL.  I can't make up my mind if this is awesome or confusing.

I kinda like it -- similar to the defer keyword in Go and Swift. The other option I considered was a "TestAndSetScope" class, which would clear() in its destructor if it had set in its constructor.

> > Source/bmalloc/bmalloc/ThreadSwitch.h:38
> > +    swtch();
> 
> What is this?

Documented here: https://www.gnu.org/software/hurd/gnumach-doc/Hand_002dOff-Scheduling.html.

My reading of the implementation is that it just runs the "find another thread to run" algorithm, without changing your priority or power management or anything.

[Geoffrey Garen]

Committed r204744: <http://trac.webkit.org/changeset/204744>

[Yusuke Suzuki]

Committed r204749: <http://trac.webkit.org/changeset/204749>

[Ryosuke Niwa]

This appears to be ~6% progression on Dromaeo DOM Core bench on MacBookAirs.

[Geoffrey Garen]

On Speedometer, I also see a ~4% progression on MacBook Air and a ~2% progression on MacBook Pro.

[Geoffrey Garen]

I can't seem to reproduce a Dromaeo DOM Core or Speedometer speedup locally on my MacBook Air. (I see a fraction of a % speedup or no change. I didn't compute a confident interval.)

[Gavin Barraclough]

Rolled out in https://trac.webkit.org/changeset/205022

[Geoffrey Garen]

Committed r205210: <http://trac.webkit.org/changeset/205210>