On Wed, Jun 15, 2016 at 10:50 AM, Richard Sandiford
<richard.sandif...@arm.com> wrote:
> vect_model_store_cost had:
>
> /* Costs of the stores. */
> if (STMT_VINFO_STRIDED_P (stmt_info)
> && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
> {
> /* N scalar stores plus extracting the elements. */
> inside_cost += record_stmt_cost (body_cost_vec,
> ncopies * TYPE_VECTOR_SUBPARTS
> (vectype),
> scalar_store, stmt_info, 0, vect_body);
>
> But non-SLP strided groups also use individual scalar stores rather than
> vector stores, so I think we should skip this only for SLP groups.
>
> The same applies to vect_model_load_cost.
>
> Tested on aarch64-linux-gnu and x86_64-linux-gnu. OK to install?
The && grouped_access is redundant, for slp_nodes that's always true.
I don't think we have something like strided groups in the interleaving case,
so the proposed fix (minus the && grouped_access_p) looks like a (good) no-op.
Ok with the suggested change,
> + if (STMT_VINFO_STRIDED_P (stmt_info) && !slp_node)
Thanks,
Richard.
> Thanks,
> Richard
>
>
> gcc/
> * tree-vect-stmts.c (vect_model_store_cost): For non-SLP
> strided groups, use the cost of N scalar accesses instead
> of ncopies vector accesses.
> (vect_model_load_cost): Likewise.
>
> diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
> index e90eeda..f883580 100644
> --- a/gcc/tree-vect-stmts.c
> +++ b/gcc/tree-vect-stmts.c
> @@ -926,8 +926,7 @@ vect_model_store_cost (stmt_vec_info stmt_info, int
> ncopies,
>
> tree vectype = STMT_VINFO_VECTYPE (stmt_info);
> /* Costs of the stores. */
> - if (STMT_VINFO_STRIDED_P (stmt_info)
> - && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
> + if (STMT_VINFO_STRIDED_P (stmt_info) && !(slp_node && grouped_access_p))
> {
> /* N scalar stores plus extracting the elements. */
> inside_cost += record_stmt_cost (body_cost_vec,
> @@ -1059,8 +1058,7 @@ vect_model_load_cost (stmt_vec_info stmt_info, int
> ncopies,
> }
>
> /* The loads themselves. */
> - if (STMT_VINFO_STRIDED_P (stmt_info)
> - && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
> + if (STMT_VINFO_STRIDED_P (stmt_info) && !(slp_node && grouped_access_p))
> {
> /* N scalar loads plus gathering them into a vector. */
> tree vectype = STMT_VINFO_VECTYPE (stmt_info);
>
