Bug 100589 – ASSERT problem on MIPS

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

[Balazs Kilvady]

There is an
ASSERT(!value);
line at Source/JavaScriptCore/jit/JITOpcodes.cpp:1627 which works on most of the platforms but fails on MIPS as value == regT0 and regT0 is not the first register (int(regT0) != 0) on MIPS.

[Balazs Kilvady]

Created attachment 171088 [details]
ASSERT fix.

As value variable is const (const RegisterID value = regT0;) after quick look I would remove the ASSERT line but anyway I created a patch to make it work on MIPS in debug mode. Feel free to throw it out :)

[Balazs Kilvady]

Created attachment 173095 [details]
ASSERT fix.

[Oliver Hunt]

Comment on attachment 173095 [details]
ASSERT fix.

View in context: https://bugs.webkit.org/attachment.cgi?id=173095&action=review

> Source/JavaScriptCore/jit/JITOpcodes.cpp:1631
>              ASSERT(!value);

Can't this just be ASSERT(value == regT0) ?

[Balazs Kilvady]

Created attachment 173099 [details]
ASSERT fix.

Simple fix :)

[Balazs Kilvady]

(In reply to comment #3)
> (From update of attachment 173095 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=173095&action=review
> 
> > Source/JavaScriptCore/jit/JITOpcodes.cpp:1631
> >              ASSERT(!value);
> 
> Can't this just be ASSERT(value == regT0) ?
Yes, works, new patch added. Is this ASSERT useful for a const variable? Assembly code could hurt it?

[Oliver Hunt]

Comment on attachment 173099 [details]
ASSERT fix.

The assertion is intended to prevent anyone from trying to use a different register for value, as the profiling logic in the baseline jit always assumes regT0 contains the value and t1 the tag (if appropriate)

[WebKit Review Bot]

Comment on attachment 173099 [details]
ASSERT fix.

Clearing flags on attachment: 173099

Committed r133950: <http://trac.webkit.org/changeset/133950>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.