vector shift regression on sparc

[Richard Guenther]

On Sun, Oct 30, 2011 at 5:38 AM, David Miller <davem@davemloft.net> wrote:
>
> gcc.dg/pr48616.c segfaults on sparc as of a day or two ago
>
> vectorizable_shift() crashes because op1_vectype is NULL and
> we hit this code path:
>
>  /* Vector shifted by vector.  */
>  if (!scalar_shift_arg)
>    {
>      optab = optab_for_tree_code (code, vectype, optab_vector);
>      if (vect_print_dump_info (REPORT_DETAILS))
>        fprintf (vect_dump, "vector/vector shift/rotate found.");
> =>    if (TYPE_MODE (op1_vectype) != TYPE_MODE (vectype))
>
> dt[1] is vect_external_def and slp_node is non-NULL.
>
> Indeed, when the 'dt' arg to vect_is_simple_use_1() is
> vect_external_def *vectype will be set to NULL.
>
> And scalar_shift_arg gets set to false because of this loop
> check:
>
>      if (slp_node)
>        {
>          VEC (gimple, heap) *stmts = SLP_TREE_SCALAR_STMTS (slp_node);
>          gimple slpstmt;
>
>          FOR_EACH_VEC_ELT (gimple, stmts, k, slpstmt)
>            if (!operand_equal_p (gimple_assign_rhs2 (slpstmt), op1, 0))
>              scalar_shift_arg = false;
>        }
>
> Indeed, this crashing test was added by this change:

Yeah, I thought testing the scalar modes would have been more natural.

> 2011-10-28  Jakub Jelinek  <jakub@redhat.com>
>
>       * tree-vect-stmts.c (vectorizable_shift): Give up if op1 has different
>       vector mode from vectype's mode.
>