[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 5/5] x86emul: disable FPU/MMX/SIMD insn emulation when !HVM



On 03.04.2020 00:18, Andrew Cooper wrote:
> On 20/12/2019 16:01, Andrew Cooper wrote:
>>> Suggested-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>>> ---
>>> I'll be happy to take suggestions allowing to avoid -Wno-unused-label.
>> I think I'm going to need a little while to figure out how this works.
> 
> So, after having had an evening playing with this, things get massively
> simpler when NO_MMX is folded with NO_SIMD.
> 
> MMX is a SIMD technology, and I can't see a compelling reason to control
> their inclusion separately.  We're either going to want everything or
> nothing.

I disagree - while MMX is a form of SIMD, what SIMD here means is
anything using the XMM register file and its extensions. Iirc
AMD once considered dropping MMX, and if I'm not mistaken early
Phi's didn't support MMX nor FPU. Hence I view a mode not
allowing MMX but allowing SIMD as a viable one to support.

> The attached incremental works for me without a single out-of-place
> label.  There is some further cleanup which can be done such as not
> making the CASE_ macros conditional.

Well, if we were to follow your alternative model - perhaps.
What I dislike though is something like the last hunk (an #ifdef
around a construct which can already abstract away things, and
which is specifically intended to avoid some #ifdef-ary).

>  (OTOH, the compile error from
> might be helpful to keep in some form).

There looks to be a word missing here, which puts me into trouble
understanding what you mean.

Jan



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.