Re: [PATCH v2 09/12] target/riscv/cpu.h: use 'vlenb' in vext_get_vlmax()

[Daniel Henrique Barboza]

On 1/15/24 20:14, Richard Henderson wrote:
> On 1/16/24 09:25, Daniel Henrique Barboza wrote:
> > Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
> > ---
> >   target/riscv/cpu.h | 6 +++++-
> >   1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> > index 11df226a00..7304e478c2 100644
> > --- a/target/riscv/cpu.h
> > +++ b/target/riscv/cpu.h
> > @@ -692,7 +692,11 @@ static inline uint32_t vext_get_vlmax(RISCVCPU *cpu, 
> > target_ulong vtype)
> >   {
> >       uint8_t sew = FIELD_EX64(vtype, VTYPE, VSEW);
> >       int8_t lmul = sextract32(FIELD_EX64(vtype, VTYPE, VLMUL), 0, 3);
> > -    return cpu->cfg.vlen >> (sew + 3 - lmul);
> > +    /*
> > +     * vlmax = vlen >> (sew + 3 - lmul). With vlenb,
> > +     * 3 less shifts: vlenb >> (sew + 3 - 3 - lmul)
> > +     */
> > +    return cpu->cfg.vlenb >> (sew - lmul);
> >   }
>
> I take it back -- this doesn't work without the + 3:
>
>    sew = 0
>    lmul = 3
>
>     vlenb >> (0 - 3)
>   = vlenb >> -3
>
> Need
>
>    vlen = vlenb << 3
>    vlmax = vlen >> (0 + 3 - 3)
>          = vlen >> 0
>          = vlen

I got tricked by the 'sew' variable name there. This formula is using vsew.

I'll fix and re-send. Thanks,


Daniel


> r~