On Wed, 20 Mar 2024, Hao Liu OS wrote:
> Hi Richard,
>
> As mentioned in the comments of PR114322 (which has been fixed by PR114151
> r14-9540-ge0e9499a), this patch is to cover the case.
>
> Bootstrapped and regression tested on aarch64-linux-gnu, OK for trunk?
>
> gcc/testsuite/ChangeLog:
>
> PR tree-optimization/114322
> * gcc.dg/vect/pr114322.c: New testcase.
>
> diff --git a/gcc/testsuite/gcc.dg/vect/pr114322.c
> b/gcc/testsuite/gcc.dg/vect/pr114322.c
> new file mode 100644
> index 00000000000..7c3a4ffe260
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/vect/pr114322.c
> @@ -0,0 +1,18 @@
> +/* PR tree-optimization/114322 */
> +/* { dg-do compile } */
> +
> +int
> +foo (short *A, int x, int stride)
> +{
> + int sum = 0;
> +
> + if (stride > 1)
> + {
> + for (int i = 0; i < 1024; ++i)
> + sum += A[(i + x) * stride];
> + }
> +
> + return sum;
> +}
> +
> +/* { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" { target
> aarch64*-*-* } } } */
I don't like the target selector very much - if you make 'A'
a 'int *', the strided access should be the important part,
vect_int should match the requirements - the strided access should
end up with VMAT_ELEMENTWISE at worst (possibly gather on aarch64?),
and VMAT_ELEMENTWISE doesn't need any alignment.
So - OK with using { target vect_int } instead.
Richard.
