On Mon, May 21, 2012 at 3:59 PM, Jakub Jelinek <ja...@redhat.com> wrote:
> Hi!
>
> On the following testcase we ICE, because op_def_stmt isn't inside
> of the loop (phi is a degenerate phi referencing a phi result from before the
> loop) and stmts outside of the loop don't have uid set (or it can contain
> garbage).
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk/4.7?
Ok.
Thanks,
Richard.
> 2012-05-21 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/53409
> * tree-vect-loop.c (vect_analyze_loop_operations): Don't check
> vinfo_for_stmt (op_def_stmt) if op_def_stmt isn't inside loop.
>
> * gcc.c-torture/compile/pr53409.c: New test.
>
> --- gcc/tree-vect-loop.c.jj 2012-05-17 08:40:45.000000000 +0200
> +++ gcc/tree-vect-loop.c 2012-05-21 11:30:28.721333458 +0200
> @@ -1327,7 +1327,9 @@ vect_analyze_loop_operations (loop_vec_i
> return false;
>
> op_def_stmt = SSA_NAME_DEF_STMT (phi_op);
> - if (!op_def_stmt || !vinfo_for_stmt (op_def_stmt))
> + if (!op_def_stmt
> + || !flow_bb_inside_loop_p (loop, gimple_bb
> (op_def_stmt))
> + || !vinfo_for_stmt (op_def_stmt))
> return false;
>
> if (STMT_VINFO_RELEVANT (vinfo_for_stmt (op_def_stmt))
> --- gcc/testsuite/gcc.c-torture/compile/pr53409.c.jj 2012-05-21
> 11:33:23.621318856 +0200
> +++ gcc/testsuite/gcc.c-torture/compile/pr53409.c 2012-05-21
> 11:33:13.000000000 +0200
> @@ -0,0 +1,19 @@
> +/* PR tree-optimization/53409 */
> +
> +int a, c, d, e, f;
> +int b[0];
> +
> +int
> +main ()
> +{
> + if (f)
> + e = 0;
> + int g = d;
> + for (c = 0; c <= 1; c++)
> + {
> + for (a = 0; a <= 1; a = (char) a + 1)
> + b[c] = g;
> + a = 0;
> + }
> + return 0;
> +}
>
> Jakub
