This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix rtl-opt/24626
- From: Andrew Pinski <pinskia at physics dot uc dot edu>
- To: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Sun, 11 Dec 2005 21:37:31 -0500
- Subject: [PATCH] Fix rtl-opt/24626
The problem here is that the change which Mostafa did to
rtl_verify_flow_info_1 did not make sense. It changed the check
to make sure that if we have a jump instruction and don't have an edge
which is a non fallthrough. This is true when we have a jump
instruction
which is also have some code associated with it (like a delay slot of
a branch instruction which hppa has).
The correct check is to make sure that the jump label is the first
instruction of the fallthrough edge as that would mean the jump
instruction
has only has one edge and the jump instruction goes to the same BB no
matter
what happens to the if the jump is taken or not.
This reverts part of Mostafa which causes us to ICE when the jump
instructions
goes to the same BB no matter if the jump is taken or not.
OK? Bootstrapped and tested on x86_64-linux-gnu and tested (by David)
on hppa-hpux
(or hppa-linux I don't know fully).
Thanks,
Andrew Pinski
ChangeLog:
* cfgrtl.c (rtl_verify_flow_info_1): Fix the case where the jump
instructions goes to only one BB.
Index: cfgrtl.c
===================================================================
--- cfgrtl.c (revision 108350)
+++ cfgrtl.c (working copy)
@@ -2046,7 +2046,7 @@
err = 1;
}
if (n_branch != 1 && any_condjump_p (BB_END (bb))
- && JUMP_LABEL (BB_END (bb)) == BB_HEAD (fallthru->dest))
+ && JUMP_LABEL (BB_END (bb)) != BB_HEAD (fallthru->dest))
{
error ("wrong amount of branch edges after conditional jump %i",
bb->index);
err = 1;