Bug 122861 – Baseline JIT should use the DFG GetById IC

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

[Filip Pizlo]

Patch forthcoming.

[Filip Pizlo]

Created attachment 214299 [details]
work in progress

Almost done.  Still testing it.

[Filip Pizlo]

Created attachment 214323 [details]
the patch

[Oliver Hunt]

Comment on attachment 214323 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=214323&action=review

wow, that is a lot of red

> Source/JavaScriptCore/jit/JITPropertyAccess32_64.cpp:473
>      BEGIN_UNINTERRUPTED_SEQUENCE(sequenceGetByIdHotPath);

Remove this -- you've removed the corresponding end.  Maybe worth just having an initial patch remove the *_UNINTERRUPTED_* macros -- assuming the patchable jump/label logic is handling everything correctly we no long reed to worry about inline constant pools breaking repatching.

[Build Bot]

Comment on attachment 214323 [details]
the patch

Attachment 214323 [details] did not pass win-ews (win):
Output: http://webkit-queues.appspot.com/results/3744094

[Filip Pizlo]

(In reply to comment #3)
> (From update of attachment 214323 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=214323&action=review
> 
> wow, that is a lot of red
> 
> > Source/JavaScriptCore/jit/JITPropertyAccess32_64.cpp:473
> >      BEGIN_UNINTERRUPTED_SEQUENCE(sequenceGetByIdHotPath);
> 
> Remove this -- you've removed the corresponding end.  Maybe worth just having an initial patch remove the *_UNINTERRUPTED_* macros -- assuming the patchable jump/label logic is handling everything correctly we no long reed to worry about inline constant pools breaking repatching.

And, I don't think we use inline constant pools on any of the platforms we care about...

[Filip Pizlo]

(In reply to comment #4)
> (From update of attachment 214323 [details])
> Attachment 214323 [details] did not pass win-ews (win):
> Output: http://webkit-queues.appspot.com/results/3744094

Oops, those look like easy 32-bit fixes.

[Filip Pizlo]

(In reply to comment #3)
> (From update of attachment 214323 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=214323&action=review
> 
> wow, that is a lot of red
> 
> > Source/JavaScriptCore/jit/JITPropertyAccess32_64.cpp:473
> >      BEGIN_UNINTERRUPTED_SEQUENCE(sequenceGetByIdHotPath);
> 
> Remove this -- you've removed the corresponding end.  Maybe worth just having an initial patch remove the *_UNINTERRUPTED_* macros -- assuming the patchable jump/label logic is handling everything correctly we no long reed to worry about inline constant pools breaking repatching.

I think after this lands, I'll do a couple of cycles of looking for stupidities to remove, including those macros.

[Filip Pizlo]

Created attachment 214328 [details]
patch for ews

[Filip Pizlo]

Landed in http://trac.webkit.org/changeset/157480

[Roger Fong]

Hi, I think the following failure started occurring on Windows after this patch:
http://build.webkit.org/results/Apple%20Win%207%20Release%20(Tests)/r157480%20(39294)/js/dictionary-prototype-caching-pretty-diff.html

does this look related to you? thx

[Filip Pizlo]

(In reply to comment #10)
> Hi, I think the following failure started occurring on Windows after this patch:
> http://build.webkit.org/results/Apple%20Win%207%20Release%20(Tests)/r157480%20(39294)/js/dictionary-prototype-caching-pretty-diff.html
> 
> does this look related to you? thx

See https://bugs.webkit.org/show_bug.cgi?id=122902

[Roger Fong]

(In reply to comment #11)
> (In reply to comment #10)
> > Hi, I think the following failure started occurring on Windows after this patch:
> > http://build.webkit.org/results/Apple%20Win%207%20Release%20(Tests)/r157480%20(39294)/js/dictionary-prototype-caching-pretty-diff.html
> > 
> > does this look related to you? thx
> 
> See https://bugs.webkit.org/show_bug.cgi?id=122902

Hmm this is a layout test failure and not a jscore test failure, still broken. I'll take a closer look and let you know what i find.