Re: [PATCH] x86: drop vex_encoding_vex2 enumerator

[Jan Beulich <jbeulich at suse dot com>]

On 15.01.2020 23:22, H.J. Lu wrote:
> On Tue, Jan 14, 2020 at 11:42 PM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> Documentation clearly says "prefer {2,3}-byte VEX prefix for VEX
>> instruction", as opposed to "encode with EVEX prefix" for {evex}. Hence
>> there not being a way to VEX-encode an insn should not be an error (not
> 
> We can't prefer any VEX prefix on non AVX instructions:
> 
> {vex} inc %rax
> 
> should be an error.

How is this in line with what the documentation states? (I
assume you mean {vex2} in your example; I could see a future
{vex} as being mandatory just like {evex} is. If {vex2} is
to be mandatory, what use is it for VEX-encodable insns? The
2-byte form will be preferred anyway.)

>> even a warning), and with this the separate enumerator becomes unneeded.
>> (Really I'm having trouble seeing what {vex2} would be useful for when
>> it's just a suggestion hint, not one allowing the programmer to mandate
>> the used encoding.)
> 
> Please keep it for now.   We are planning to use it in the future.

Then may I please ask that you update the documentation to
clarify which pseudo prefixes are hints (and silently ignored
when they cannot be fulfilled) and which ones are mandatory?

Thanks, Jan