Re: [PATCH i386][google]With -mtune=core2, avoid generating the slow unaligned vector load/store (issue5488054)

[Ira Rosen <IRAR at il dot ibm dot com>]

gcc-patches-owner@gcc.gnu.org wrote on 13/12/2011 04:05:57 AM:

> On core2, unaligned vector load/store using movdqu is a very slow
operation.
> Experiments show it is six times slower than movdqa (aligned) and this is
> irrespective of whether the resulting data happens to be aligned or not.
> For Corei7, there is no performance difference between the two and on
AMDs,
> movdqu is only about 10% slower.
>
> This patch does not vectorize loops that need to generate the slow
unaligned
> memory load/stores on core2.
>
>
>    Do not vectorize loops on Core2 that need to use unaligned
>    vector load/stores.
>    * tree-vect-stmts.c (is_slow_vect_unaligned_load_store): New function.
>    (vect_analyze_stmt): Check if the vectorizable load/store is slow.
>    * target.def (TARGET_SLOW_UNALIGNED_VECTOR_MEMOP): New target hook.
>    * doc/m.texi.in: Document new target hook:
>    TARGET_SLOW_UNALIGNED_VECTOR_MEMOP
>    * doc/m.texi: Regenerate.
>    * config/i386/i386.c (ix86_slow_unaligned_vector_memop): New function.
>    (TARGET_SLOW_UNALIGNED_VECTOR_MEMOP): New macro.
>


> @@ -5065,27 +5112,43 @@ vect_analyze_stmt (gimple stmt, bool
*need_to_vect
>     if (!bb_vinfo
>         && (STMT_VINFO_RELEVANT_P (stmt_info)
>             || STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def))
> +    {
>        ok = (vectorizable_type_promotion (stmt, NULL, NULL, NULL)
>              || vectorizable_type_demotion (stmt, NULL, NULL, NULL)
>              || vectorizable_conversion (stmt, NULL, NULL, NULL)
>              || vectorizable_shift (stmt, NULL, NULL, NULL)
>              || vectorizable_operation (stmt, NULL, NULL, NULL)
>              || vectorizable_assignment (stmt, NULL, NULL, NULL)
> -            || vectorizable_load (stmt, NULL, NULL, NULL, NULL)
>              || vectorizable_call (stmt, NULL, NULL)
> -            || vectorizable_store (stmt, NULL, NULL, NULL)
> -            || vectorizable_reduction (stmt, NULL, NULL, NULL)
> +       || vectorizable_reduction (stmt, NULL, NULL, NULL)
>              || vectorizable_condition (stmt, NULL, NULL, NULL, 0));
> +
> +      if (!ok)
> +   {
> +          ok = (vectorizable_load (stmt, NULL, NULL, NULL, NULL)
> +           || vectorizable_store (stmt, NULL, NULL, NULL));
> +
> +     if (ok && is_slow_vect_unaligned_load_store (stmt))
> +       ok = false;

Why not call is_slow_vect_unaligned_load_store from
vectorizable_load/store?

Ira


> +   }
> +    }
>      else
>        {
>          if (bb_vinfo)
> -          ok = (vectorizable_type_promotion (stmt, NULL, NULL, node)
> -                || vectorizable_type_demotion (stmt, NULL, NULL, node)
> -               || vectorizable_shift (stmt, NULL, NULL, node)
> -                || vectorizable_operation (stmt, NULL, NULL, node)
> -                || vectorizable_assignment (stmt, NULL, NULL, node)
> -                || vectorizable_load (stmt, NULL, NULL, node, NULL)
> -                || vectorizable_store (stmt, NULL, NULL, node));
> +     {
> +       ok = (vectorizable_type_promotion (stmt, NULL, NULL, node)
> +        || vectorizable_type_demotion (stmt, NULL, NULL, node)
> +        || vectorizable_shift (stmt, NULL, NULL, node)
> +                  || vectorizable_operation (stmt, NULL, NULL, node)
> +                  || vectorizable_assignment (stmt, NULL, NULL, node));
> +            if (!ok)
> +         {
> +                ok = (vectorizable_load (stmt, NULL, NULL, node, NULL)
> +            || vectorizable_store (stmt, NULL, NULL, node));
> +           if (ok && is_slow_vect_unaligned_load_store (stmt))
> +        ok = false;
> +         }
> +     }
>        }