On February 10, 2019 12:16:12 AM GMT+01:00, Jakub Jelinek <ja...@redhat.com>
wrote:
>Hi!
>
>WHen tree_if_conversion is called from within the vectorizer (for
>masked
>epilogues), preds is NULL, so we shouldn't be pushing anything there,
>the caller doesn't care anyway.
Oops.
>Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
Richard.
>2019-02-09 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/89268
> * tree-if-conv.c (version_loop_for_if_conversion): Push to preds only
> if preds is non-NULL.
>
> * gcc.dg/vect/pr89268.c: New test.
>
>--- gcc/tree-if-conv.c.jj 2019-02-08 20:00:40.774827920 +0100
>+++ gcc/tree-if-conv.c 2019-02-09 17:35:36.995782510 +0100
>@@ -2760,7 +2760,8 @@ version_loop_for_if_conversion (struct l
> new_loop->force_vectorize = false;
> gsi = gsi_last_bb (cond_bb);
>gimple_call_set_arg (g, 1, build_int_cst (integer_type_node,
>new_loop->num));
>- preds->safe_push (g);
>+ if (preds)
>+ preds->safe_push (g);
> gsi_insert_before (&gsi, g, GSI_SAME_STMT);
> update_ssa (TODO_update_ssa);
> return new_loop;
>--- gcc/testsuite/gcc.dg/vect/pr89268.c.jj 2019-02-09
>17:53:15.099481437 +0100
>+++ gcc/testsuite/gcc.dg/vect/pr89268.c 2019-02-09 17:53:50.900896069
>+0100
>@@ -0,0 +1,7 @@
>+/* PR tree-optimization/89268 */
>+/* { dg-do compile } */
>+/* { dg-require-effective-target vect_condition } */
>+/* { dg-additional-options "-fno-trapping-math --param
>vect-epilogues-nomask=1" } */
>+/* { dg-additional-options "-mavx512ifma -mtune=intel" { target
>x86_64-*-* i?86-*-* } } */
>+
>+#include "pr79887.c"
>
> Jakub
