Hi,
this patch makes vect_analyze_loop_2 to give up on loops with low max iteration
counts (instead of only considering estimated_stmt_executions_int). This
change was eaerlier approved by Richi in stage4 and reverted.
I got wrong the testm instead == -1 there was != -1 and thus the regressions.
Noticed that only now.
Bootstrapped/regtested x86_64-linux, comitted.
Honza
Index: ChangeLog
===================================================================
--- ChangeLog (revision 236477)
+++ ChangeLog (working copy)
@@ -1,3 +1,8 @@
+2016-05-19 Jan Hubicka <hubi...@ucw.cz>
+
+ * tree-vect-loop.c (vect_analyze_loop_2): Use also
+ max_loop_iterations_int.
+
2016-05-19 Marek Polacek <pola...@redhat.com>
PR tree-optimization/71031
Index: tree-vect-loop.c
===================================================================
--- tree-vect-loop.c (revision 236450)
+++ tree-vect-loop.c (working copy)
@@ -2065,6 +2065,8 @@ start_over:
estimated_niter
= estimated_stmt_executions_int (LOOP_VINFO_LOOP (loop_vinfo));
+ if (estimated_niter == -1)
+ estimated_niter = max_niter;
if (estimated_niter != -1
&& ((unsigned HOST_WIDE_INT) estimated_niter
<= MAX (th, (unsigned)min_profitable_estimate)))
