RE: [PATCH] x86: fold special-operand insn attributes into a single enum

["Jiang, Haochen" <haochen.jiang@intel.com>]

> -----Original Message-----
> From: H.J. Lu <hjl.tools@gmail.com>
> Sent: Saturday, November 12, 2022 3:49 AM
> To: Beulich, Jan <JBeulich@suse.com>; Kong, Lingling
> <lingling.kong@intel.com>; Jiang, Haochen <haochen.jiang@intel.com>; Cui,
> Lili <lili.cui@intel.com>
> Cc: Binutils <binutils@sourceware.org>
> Subject: Re: [PATCH] x86: fold special-operand insn attributes into a single
> enum
> 
> On Fri, Nov 11, 2022 at 12:22 AM Jan Beulich <jbeulich@suse.com> wrote:
> >
> > On 11.11.2022 09:00, Jan Beulich via Binutils wrote:
> > > On 10.11.2022 18:38, H.J. Lu wrote:
> > >> On Thu, Nov 10, 2022 at 5:45 AM Jan Beulich <jbeulich@suse.com>
> wrote:
> > >>>
> > >>> Attributes which aren't used together in any single insn template
> > >>> can be converted from individual booleans to a single enum, as was
> > >>> done for a few other attributes before. This is more space
> > >>> efficient. Collect together all attributes which express special
> > >>> operand constraints (and which fit the criteria for folding).
> > >>
> > >> These assumptions may not be all true for future new instructions.
> > >
> > > It seems pretty unlikely to me that any of these assumptions would break.
> > > If one does, converting back to an individual bit is an easy option.
> > > I think this minor risk is far outweighed by the benefits of the
> > > change. We really have still some ways to go to reach reasonable
> > > data representation within the generated opcode table (and of course
> > > also in the source table, but that's an orthogonal direction of
> > > work, where I also have further changes pending).
> >
> > I should probably expand: For most of these I don't expect new uses to
> > appear at all. Note also that I left alone e.g. ImmExt, for having a
> > bigger risk of a new conflicting use appearing.
> >
> > If you have concerns for a specific attribute, please voice it that
> > way. I'm certainly willing to re-consider for individual attributes
> > (albeit as said in the earlier reply, the way back for any individual
> > one is easy); I'm not going to accept a blanket "no".
> >
> > Jan
> 
> Lingling, Lili, Haochen, do you have any comments?

As far as I see, I go with Jan. If anything break in the future, we could convert
back to individual bits.

Another benefit is that if we have another special-operand insn attrbutes,
we could also add to this and do not need an extra bit since only 0-8 used. 

Haochen

> 
> Thanks.
> 
> --
> H.J.