On Fri, Jul 31, 2020 at 4:35 AM Richard Henderson <
richard.henderson@linaro.org> wrote:

> On 7/22/20 2:16 AM, frank.chang@sifive.com wrote:
> > +    static gen_helper_gvec_3_ptr * const fns[6][4] = {
> > +        {
> > +            NULL, gen_helper_vzext_vf2_h,
> > +            gen_helper_vzext_vf2_w, gen_helper_vzext_vf2_d
> > +        },
> > +        {
> > +            NULL, NULL,
> > +            gen_helper_vzext_vf4_w, gen_helper_vzext_vf4_d,
> > +        },
> > +        {
> > +            NULL, NULL,
> > +            NULL, gen_helper_vzext_vf8_d
> > +        },
> > +        {
> > +            NULL, gen_helper_vsext_vf2_h,
> > +            gen_helper_vsext_vf2_w, gen_helper_vsext_vf2_d
> > +        },
> > +        {
> > +            NULL, NULL,
> > +            gen_helper_vsext_vf4_w, gen_helper_vsext_vf4_d,
> > +        },
> > +        {
> > +            NULL, NULL,
> > +            NULL, gen_helper_vsext_vf8_d
> > +        }
> > +    };
>
> I don't understand why there aren't more functions in this table.  As far
> as I
> can see, the only NULLs should be at [*][0].
>
>
> r~
>

As source EEW has to be 1/2, 1/4, 1/8 of SEW and the source EEW must be
a supported width (Section 12.3).

Shouldn't it impossible to have the cases, e.g.
vzext.vf4 with SEW = 16, i.e. EEW = SEW / 4 = 4 bits
vzext.vf8 with SEW = 16, i.e. EEW = SEW / 8 = 2 bits
vzext.vf8 with SEW = 32, i.e. EEW = SEW / 8 = 4 bits

Frank Chang