Unreviewed patch: Old loop optimizer clobbers frame pointer

[Björn Haase]

Roger Sayle wrote on Sonntag, 19. Juni 2005 00:50 :
> On Sat, 18 Jun 2005, Ulrich Weigand wrote:
> > http://gcc.gnu.org/ml/gcc-patches/2005-06/msg00157.html
>
> This is OK for mainline.  I agree that the risks of moving modifications
> to the frame pointer (especially in the presence of pseudos) outway the
> potential performance benefits.  If we ever identify any cases where this
> transformation is worthwhile, we can at that point investigate alternate
> approaches, such as recording the dependency between all pseudos and
> frame_pointer_rtx.  However, this fix seems preferable for now.
>
> Roger
> --
Hi,

I'd just like to draw your attention to PR21990 "reload absentmindly clobbers 
frame pointer". Is it possible that both problems have the same origin? I 
have tried Ulrich's patch, but it did not help for the test case of PR21990.

Yours,

Björn