Re: [tuples] Robustify cleanup_omp_return

[Aldy Hernandez <aldyh at redhat dot com>]

On Fri, May 30, 2008 at 10:32:57AM -0400, Diego Novillo wrote:
> On Fri, May 30, 2008 at 10:26, Jakub Jelinek <jakub@redhat.com> wrote:
> > Hi!
> >
> > This function is called during cfg cleanup, so bb might be preceeded
> > by an empty useless block.  Given the cfg creation for
> > GIMPLE_OMP_SECTIONS_SWITCH, I'm convinced the final sections
> > GIMPLE_OMP_RETURN will be always preceeded by GIMPLE_OMP_SECTIONS_SWITCH,
> > so there is no need to search through empty bbs to find
> > GIMPLE_OMP_SECTIONS_SWITCH.
> 
> Hmm, Aldy had a similar patch yesterday and I asked him to verify that
> he wasn't papering over anything.  It looked strange to me that we
> need this in tuples but not in trunk.  Maybe we do, but I'd like to
> understand why.
> 
> Aldy, did you find out anything new about this?

Hey, I just woke up, let drink some coffee or something!

I actually didn't see anything wrong.  I just saw that tuples had a
slightly different BB layout due to our flattened GIMPLE_COND's, which
was causing BB merging to look a tad different from trunk to branch.  

In the tuples branch, we ended up merging different BBs, which gave
us an empy BB with a fallthru edge.  This was causing the problem Jakub
and I tried idedependently to fix.

I'll see if I can find a reason why we didn't merge the BB's differently
on tuples, but it may just be a different way of representing COND's.

Aldy