[PATCH] Fix incorrect loop exit edge probability [PR103270]

Richard Biener richard.guenther@gmail.com
Tue Nov 23 09:42:02 GMT 2021


On Tue, Nov 23, 2021 at 6:52 AM Xionghu Luo <luoxhu@linux.ibm.com> wrote:
>
> r12-4526 cancelled jump thread path rotates loop. It exposes a issue in
> profile-estimate when predict_extra_loop_exits, outer loop's exit edge
> is marked as inner loop's extra loop exit and set with incorrect
> prediction, then a hot inner loop will become cold loop finally through
> optimizations, this patch ignores the EDGE_DFS_BACK edge when searching
> extra exit edges to avoid unexpected predict_edge.

Not sure how outer vs. inner loop exit correlates with EDGE_DFS_BACK,
I have expected a check based on which loop is exited by the edge instead?
A backedge should never be an exit, no?

Note that the profile pass does not yet mark backedges so EDGE_DFS_BACK
settings are unreliable.

Richard.

>
> gcc/ChangeLog:
>
>         PR middle-end/103270
>         * predict.c (predict_extra_loop_exits): Ignore EDGE_DFS_BACK edge.
>
> gcc/ChangeLog:
>
>         PR middle-end/103270
>         * predict.c (predict_extra_loop_exits): New.
> ---
>  gcc/predict.c | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/gcc/predict.c b/gcc/predict.c
> index 68b11135680..1ae8ccff72c 100644
> --- a/gcc/predict.c
> +++ b/gcc/predict.c
> @@ -1910,6 +1910,10 @@ predict_extra_loop_exits (edge exit_edge)
>         continue;
>        if ((check_value_one ^ integer_onep (val)) == 1)
>         continue;
> +#if 0
> +      if (e->flags & EDGE_DFS_BACK)
> +       continue;
> +#endif
>        if (EDGE_COUNT (e->src->succs) != 1)
>         {
>           predict_paths_leading_to_edge (e, PRED_LOOP_EXTRA_EXIT, NOT_TAKEN);
> --
> 2.25.1
>


More information about the Gcc-patches mailing list