Richard Henderson <richard.hender...@linaro.org> writes:
> On 7/10/19 8:32 PM, Jan Bobek wrote:
>> On 7/3/19 5:49 PM, Richard Henderson wrote:
>>> On 7/1/19 6:35 AM, Jan Bobek wrote:
>>>> +MOVQ MMX 00001111 011 d 1110 !emit { rex(w => 1);
>>>> modrm(mod => MOD_DIRECT, rm => ~REG_ESP); }
>>>> +MOVQ_mem MMX 00001111 011 d 1110 !emit { rex(w => 1);
>>>> modrm(mod => ~MOD_DIRECT); mem(size => 8); }
>>>
>>> Oh, note that there are only 8 mmx registers, so the respective rex.{r,b}
>>> bit
>>> can't be set.
>>
>> Actually, my CPU chewed it without choking even when the bits were
>> set, but it will taken care of in v3.
>
> That's interesting data.
>
> I wonder if it's worth retaining this as a feature in order to check qemu's
> implementation?
We could be some time, c.f. BlackHat 2017
https://www.youtube.com/watch?v=KrksBdWcZgQ
I suspect if we set https://github.com/xoreaxeaxeax/sandsifter on QEMU
we might find a few breakages.
>
>
> r~
--
Alex Bennée
