[Bug tree-optimization/18694] [4.0 regression] loop miscompilation at -O1 (-ftree-ch)
------- Additional Comments From kazu at cs dot umass dot edu 2004-12-09 03:35 -------
Created an attachment (id=7710)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7710&action=view)
a patch
This patch is fixes the bug but maybe a bit too conservative.
If thread_across_edge gets a comparison "op0 code op1",
and op0 and op1 are defined by PHI nodes at e->dest,
we could still replace op0 and op1 with the PHI arguments
before we look up the condition.
With my patch, if the condition involves an SSA_NAME
that is also used as a PHI argument at e->dest,
I would guess that the lookup is very likely to fail.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18694
------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.
Reply to: