Stefan Schulze Frielinghaus via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> This is a follow up to commit 5c9669a0e6c respectively discussion
> https://gcc.gnu.org/pipermail/gcc-patches/2020-June/549132.html
>
> In case that an alignment constraint is less than the size of a
> corresponding scalar type, ensure that we advance at least by one
> iteration. For example, on s390x we have for a long double an alignment
> constraint of 8 bytes whereas the size is 16 bytes. Therefore,
> TARGET_ALIGN / DR_SIZE equals zero resulting in an infinite loop which
> can be reproduced by the following MWE:
>
> extern long double *a;
> extern double *b;
> void fun(void) {
> for (int i = 0; i < 42; i++)
> a[i] = b[i];
> }
>
> Increasing the number of peelings in each iteration at least by one
> fixes the issue for me. Any comments?
>
> Bootstrapped and regtested on s390x.
>
> gcc/ChangeLog:
>
> * tree-vect-data-refs.c (vect_enhance_data_refs_alignment):
> Ensure that loop variable npeel_tmp advances in each iteration.
OK, thanks. (For the record, I wondered whether changing the
/ to CEIL (…) would be better, but given that the alignment must
be a power of 2, I guess there's no practical difference.)
Richard
> ---
> gcc/tree-vect-data-refs.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c
> index e35a215e042..a78ae61d1b0 100644
> --- a/gcc/tree-vect-data-refs.c
> +++ b/gcc/tree-vect-data-refs.c
> @@ -1779,7 +1779,7 @@ vect_enhance_data_refs_alignment (loop_vec_info
> loop_vinfo)
> {
> vect_peeling_hash_insert (&peeling_htab, loop_vinfo,
> dr_info, npeel_tmp);
> - npeel_tmp += target_align / dr_size;
> + npeel_tmp += MAX (1, target_align / dr_size);
> }
>
> one_misalignment_known = true;
