Fix PR23410
Sebastian Pop
sebastian.pop@cri.ensmp.fr
Thu Sep 1 15:09:00 GMT 2005
Hi,
This patch fixes PR23410. Tested on i686 and amd64, committed. I've
worked on more cleanups for fixing this PR, but I think that these
should wait till 4.2.
PR tree-optimization/23410
* tree-ssa-loop-niter.c (scev_probably_wraps_p): Check that the
sequence is not wrapping during the first step.
Index: tree-ssa-loop-niter.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-loop-niter.c,v
retrieving revision 2.40
diff -d -u -p -r2.40 tree-ssa-loop-niter.c
--- tree-ssa-loop-niter.c 23 Aug 2005 08:15:13 -0000 2.40
+++ tree-ssa-loop-niter.c 1 Sep 2005 09:00:07 -0000
@@ -1823,7 +1823,8 @@ scev_probably_wraps_p (tree type, tree b
struct nb_iter_bound *bound;
tree delta, step_abs;
tree unsigned_type, valid_niter;
- tree base_plus_step;
+ tree base_plus_step, bpsps;
+ int cps, cpsps;
/* FIXME: The following code will not be used anymore once
http://gcc.gnu.org/ml/gcc-patches/2005-06/msg02025.html is
@@ -1865,7 +1866,17 @@ scev_probably_wraps_p (tree type, tree b
*unknown_max = false;
base_plus_step = fold_build2 (PLUS_EXPR, type, base, step);
- switch (compare_trees (base_plus_step, base))
+ bpsps = fold_build2 (PLUS_EXPR, type, base_plus_step, step);
+ cps = compare_trees (base_plus_step, base);
+ cpsps = compare_trees (bpsps, base_plus_step);
+
+ /* Check that the sequence is not wrapping in the first step: it
+ should have the same monotonicity for the first two steps. See
+ PR23410. */
+ if (cps != cpsps)
+ return true;
+
+ switch (cps)
{
case -1:
{
More information about the Gcc-patches
mailing list