[Bug tree-optimization/34016] [4.3 Regression] ICE in execute_todo with -ftree-loop-linear for 454.calculix

[jakub at gcc dot gnu dot org]

------- Comment #2 from jakub at gcc dot gnu dot org  2007-11-27 12:31 -------
Testcase in C:
/* PR tree-optimization/34016 */
/* { dg-do compile } */
/* { dg-options "-O2 -ftree-loop-linear" } */

extern void bar (double *);

void foo (void)
{
  double gr[36];
  int i, j;
  for (i = 0; i <= 5; i++)
    {
      for (j = 0; j <= 5; j++)
        gr[i + j * 6] = 0.0;
      if (i <= 2)
        gr[i + i * 6] = 1.0;
    }
  bar (gr);
}

As perfect_nestify will call mark_syms_for_renaming, I believe the right fix is
to update todo flags of this pass either using:
--- tree-ssa-loop.c.jj6 2007-09-17 22:25:50.000000000 +0200
+++ tree-ssa-loop.c     2007-11-27 13:18:41.000000000 +0100
@@ -272,6 +272,7 @@ struct tree_opt_pass pass_linear_transfo
   0,                                   /* properties_destroyed */
   0,                                   /* todo_flags_start */
   TODO_dump_func | TODO_verify_loops
+    | TODO_update_ssa_only_virtuals
     | TODO_ggc_collect,                        /* todo_flags_finish */
   0                                    /* letter */
 };

or by returning TODO_update_ssa_only_virtuals from tree_linear_transform if
there wasn't already a perfect nest.  As update_ssa is a nop if
!need_ssa_update_p, I believe just adding | TODO_update_ssa_only_virtuals
to pass_linear_transform.todo_flags_finish is better.  Will test.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34016