Re: [PATCH] x86: allow suffix-less sign-extending movsb, movsw, and movsl

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

On Mon, Jul 4, 2016 at 1:07 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>> On 01.07.16 at 17:20, <jonas-devlists@watlock.be> wrote:
>
>> Jan Beulich wrote on Fri, 01 Jul 2016:
>>
>>>>>> On 01.07.16 at 16:24, <jonas-devlists@watlock.be> wrote:
>>>> Referring again to the above document, it says about movsb/movsw:
>>>> "movsb is not movsb{wlq}" and "movsw
>>>> is not movsw{lq}" (on p. 37). Those are the only mnemonics that are
>>>> singled out in this way.
>>>
>>> Well, the document referenced is a random one; it's way too new
>>> to be a canonical reference.
>>
>> I disagree that (originally) Sun documentation is a random reference
>> in the context of AT&T/System V UNIX. Yes, SVR4.2 i386 documentation
>> would be even better (if it mentions this issue), but I guess that
>> only exists as hard copy in someone's basement.
>>
>>> I do not understand what inconsistency you refer to here. The
>>> only inconsistency I can see is that one can't omit the suffixes
>>> from these three instructions, unlike any others with GPR
>>> operands.
>>
>> It is not consistent that all base mnemonics (i.e., without size
>> suffix) refer to individual opcodes (or groups of opcodes) as defined
>> in Intel's architecture manuals, except for movsb/w/l.
>
> I don't see what's wrong with this, when it's okay for the assembler
> to accept all kinds on non-AT&T syntax instructions in AT&T mode.
> Note how, for example, both movsx and movzx have specific AT&T
> entries despite these being Intel syntax mnemonics.

Those are for historical reasons.  We shouldn't add new ones when
they buy nothing for programmers.

-- 
H.J.