[Bug rtl-optimization/31360] [4.2 Regression] RTL loop invariant is not aggressive enough

[mmitchel at gcc dot gnu dot org]

------- Comment #27 from mmitchel at gcc dot gnu dot org  2007-06-23 20:45 -------
Richard G., by:

"I guess its hard to find a fix that does not regress elsewhere, so I'm
inclined
to WONTFIX this on the 4.2 branch."

are you referring to functional problems or to Richard E's comments regarding
code size?  The followup PR 31676 patch looks very small; couldn't we backport
this patch plus that one to 4.2, if we wanted to do so?

As for the code size regression, Richard E., have you had a chance to identify
a specific CSiBE file that expands, so that Zdenek can look at that?

Meanwhile, I would prefer to leave this regression open.  I really do feel that
it's a severe optimization regression.  After all, this is basically a variant
on bzero.

Thanks,

-- Mark


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31360