Re: [PATCH 3/5] x86: support AVX10.1/512

["H.J. Lu" <hjl.tools@gmail.com>]

On Mon, Aug 28, 2023 at 12:09 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 28.08.2023 08:59, Jiang, Haochen wrote:
> >> But of course the question remains on whether this is needed in the first
> >> place. Personally I'd favor making AES a prereq of VAES (and PCLMULQDQ one
> >> of VPCLMULQDQ); if we decided to go that route, I'd make that change a
> >
> > For PCLMULQDQ and VPCLMULQDQ, the imply should be ok. Actually, I have
> > added that in GCC in April, but I forgot to check Binutils at that time:
> >
> > https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=4246611d1915f1664c01f286dbeb946dd06e2a4d
> >
> > For AES/VAES, I am conservative on that at that time because not all the insts are
> > included in VAES:
> >
> > https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=24a8acc1662c37003a7b54814bf840019fec2190
> >
> > But I suppose the reason why AESIMC/AESKEYGENASSIST are not included might
> > be one key length is 128 bit while generating/transforming two at the same time
> > and putting them into ymm register is not quite needed.
>
> Plus VAES is going to be of less use without those two insns, so assuming
> AES to also be there if VAES is ought to be sufficiently "natural". Yet
> as said, it would be quite nice if the SDM could make such "connections"
> explicit.
>
> > I am both ok on whether to imply or not since it would be simple if implied.
>
> Okay, I'll tentatively add a new prereq patch then, but before posting v2
> I'd like to give H.J. the opportunity to voice an opinion.
>

I don't think we need to make AVX10 to imply AES and VPCLMULQDQ
in assembler.  The AES and PCLMULQDQ changes aren't needed.


-- 
H.J.