On Wed, Feb 22, 2023 at 12:03 AM Michael Collison <colli...@rivosinc.com> wrote:
>
> While working on autovectorizing for the RISCV port I encountered an
> issue where vect_do_peeling assumes that the vectorization factor is a
> compile-time constant. The vectorization is not a compile-time constant
> on RISCV.
>
> Tested on RISCV and x86_64-linux-gnu. Okay?
I wonder how you arrive at prologue peeling with a non-constant VF?
In any case it would probably be better to use constant_lower_bound (vf)
here? Also it looks wrong to apply this limit in case we are using
a fully masked main vector loop. But as said, the specific case of
non-constant VF and prologue peeling probably wasn't supposed to happen,
instead the prologue usually is applied via an offset to a fully masked loop?
Richard?
Thanks,
Richard.
> Michael
>
> gcc/
>
> * tree-vect-loop-manip.cc (vect_do_peeling): Verify
> that vectorization factor is a compile-time constant.
>
> ---
> gcc/tree-vect-loop-manip.cc | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc
> index 6aa3d2ed0bf..1ad1961c788 100644
> --- a/gcc/tree-vect-loop-manip.cc
> +++ b/gcc/tree-vect-loop-manip.cc
> @@ -2930,7 +2930,7 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree
> niters, tree nitersm1,
> niters = vect_build_loop_niters (loop_vinfo, &new_var_p);
> /* It's guaranteed that vector loop bound before vectorization is at
> least VF, so set range information for newly generated var. */
> - if (new_var_p)
> + if (new_var_p && vf.is_constant ())
> {
> value_range vr (type,
> wi::to_wide (build_int_cst (type, vf)),
> --
> 2.34.1
>
