Enable Intel AVX512_FP16 instructions and test

[Jan Beulich]

On 05.08.2021 14:10, Jan Beulich via Binutils wrote:
> On 05.08.2021 13:54, Cui, Lili wrote:
>>
>>
>>> -----Original Message-----
>>> From: Jan Beulich <jbeulich@suse.com>
>>> Sent: Wednesday, August 4, 2021 11:33 PM
>>> To: Cui, Lili <lili.cui@intel.com>
>>> Cc: hjl.tools@gmail.com; binutils@sourceware.org
>>> Subject: Re: Enable Intel AVX512_FP16 instructions and test
>>>
>>> On 04.08.2021 16:40, Cui, Lili wrote:
>>>>> From: Jan Beulich <jbeulich@suse.com>
>>>>> Sent: Tuesday, August 3, 2021 10:23 PM
>>>>>
>>>>> On 30.07.2021 11:00, Cui, Lili wrote:
>>>>>> Rebased AVX512-FP16 patch and added error checking for some
>>>>>> instructions that require different destination and source registers
>>>>>
>>>>> I can't seem to be able to spot respective disassembler changes? (I
>>>>> have to also admit that the restriction is quite well hidden in the
>>>>> spec - there's no mention of it except in the "Additionally:"
>>>>> sections. I'd have expected such special cases to be mentioned in
>>>>> "Operation" as well.)
>>>>>
>>>> Add distinct register check for disassembler and also add test case for it,
>>> thanks.
>>>>
>>>>> I'm not sure a new insn attribute is warranted here (iirc you got
>>>>> away without for the AMX special restrictions), but I also don't
>>>>> really want to request that you redo this. What I would like to see
>>>>> improved though is the
>>>>> name: It would better express that it's the destination that needs to
>>>>> be distinct (unlike for the AMX insns, where all registers need to be
>>> distinct).
>>>>> Also the error message wording "destination and source registers must
>>>>> be distinct" is ambiguous (one may read it to mean the same as what
>>>>> the AMX requirement is). I'd suggest "destination must be distinct
>>>>> from source registers".
>>>>>
>>>> Done.
>>>
>>> Hmm, I see you've changed the message (which, seeing it again, is still
>>> ambiguous I'm afraid), but you've kept the field name.
>>>
>> How about ReqDistinctDest ?
> 
> Quite a bit better, yes. Perhaps even drop the "Req" part of it?
> 
>>> As to the message (I'm sorry for adjusting my prior suggestion), how about
>>> "destination must be distinct from source registers"?
>>>
>> I am confused here, do I need to make some changes ?
> 
> Well - use the new suggested text?

And just to clarify - ideally with these two remaining small items
addressed, I'm okay with the patch now; afaic no need for another
re-posting before you commit the result. I may subsequently try to
deal with the two items you indicated you wouldn't want to address.

Jan