https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116290
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
We're finding one more BIV with -g due to
@@ -78,14 +78,7 @@
(scalar = _32)
(get_scalar_evolution
(scalar = _32)
- (scalar_evolution = _32))
-)
-(analyze_scalar_evolution
- (loop_nb = 1)
- (scalar = _32)
-(get_scalar_evolution
- (scalar = _32)
- (scalar_evolution = _32))
+ (scalar_evolution = {_28, +, 1}_1))
)
that's actually a stale cache - calling scev_reset () at the start of IVOPTs
fixes the compare-debug issue.
-fdisable-tree-ldist avoids the issue.
Oddly enough there's no IL difference between ivcanon (before ldist) and
slp1. IVCANON calls scev_reset () unconditionally at the end.
Ah, found it and testing fix.
diff --git a/gcc/tree-loop-distribution.cc b/gcc/tree-loop-distribution.cc
index f0430ede2f4..d7013ba5f8d 100644
--- a/gcc/tree-loop-distribution.cc
+++ b/gcc/tree-loop-distribution.cc
@@ -3565,8 +3565,8 @@ determine_reduction_stmt_1 (const loop_p loop, const
basic_block *bbs)
}
}
- for (gimple_stmt_iterator bsi = gsi_start_bb (bb); !gsi_end_p (bsi);
- gsi_next_nondebug (&bsi), ++ninsns)
+ for (gimple_stmt_iterator bsi = gsi_start_nondebug_bb (bb);
+ !gsi_end_p (bsi); gsi_next_nondebug (&bsi), ++ninsns)
{
/* Bail out early for loops which are unlikely to match. */
if (ninsns > 16)
