[llvm-commits] [llvm-gcc-4.2] r55718 - /llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp

[Dale Johannesen]

On Sep 4, 2008, at 1:00 AMPDT, Duncan Sands wrote:

>> Does this mean we can stop using gcc's trivial inliner?
>
> IIRC, we are using the full gcc inliner because otherwise
> we get webkit performance regressions.  Also (again IIRC),
> the trivial gcc inliner does not inline inline-always
> functions!  So I originally set things up so that the full
> inliner was run (not the trivial one), but skipped everything
> except for inline-always functions; but the skipping part had
> to be undone due to webkit.

Right, see PR 2353.
I have a bug report calling for changing it back to the way Duncan had  
it (inline-always functions only).
It's not clear to me what the criteria should be for doing that.   
Certainly the webkit regression has to be
analyzed and fixed, but it's unlikely that's the only regression there  
will be....