Bug 170406 – Inst::forEachArg could compile to more compact code

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

[Filip Pizlo]

Patch forthcoming.

[Filip Pizlo]

Created attachment 306084 [details]
the patch

[JF Bastien]

Comment on attachment 306084 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=306084&action=review

lgtm

> Source/JavaScriptCore/ChangeLog:21
> +        shrinks by 2.7 MB. This is a 15% reduction in JavaScriptCore binary size.

DAMN

> Source/JavaScriptCore/b3/air/opcode_generator.rb:-1143
> -# This is a hack for JSAir. It's a joke.

No more joke?

[Filip Pizlo]

(In reply to JF Bastien from comment #2)
> Comment on attachment 306084 [details]
> the patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=306084&action=review
> 
> lgtm
> 
> > Source/JavaScriptCore/ChangeLog:21
> > +        shrinks by 2.7 MB. This is a 15% reduction in JavaScriptCore binary size.
> 
> DAMN

Yeah!  We were ALWAYS_INLINE'ing a ~1600 LoC autogenerated nested switch lambda monster in >20 places.  Kudos to all C++ compilers that compiled this code without ICE'ing.

> 
> > Source/JavaScriptCore/b3/air/opcode_generator.rb:-1143
> > -# This is a hack for JSAir. It's a joke.
> 
> No more joke?

We needed that code to generate the ARES-6 "Air" benchmark.  It's already been generated.  So we don't really need to have this code anymore.

[Filip Pizlo]

It's possible, but unlikely, that this patch will show up as a perf regression or progression. We should roll it out if it regresses anything.

[Filip Pizlo]

Landed in https://trac.webkit.org/changeset/214827/webkit