PACTH: Enable Intel AVX

Jan Beulich
Mon Apr 7 14:25:00 GMT 2008

>>> "H.J. Lu" <> 07.04.08 15:03 >>>
>On Mon, Apr 07, 2008 at 09:29:30AM +0100, Jan Beulich wrote:
>> There's one oversight in the gas part: parse_real_register() shouldn't
>> return ymm registers without the according feature(s) enabled.
>Here is a patch to check it.

It only adds a check for the positive cases, not for the negative
ones (i.e. where ymmX does *not* reference a register)...

>> Besides that I wonder whether AES and PCLMUL really imply MMX
>> support? I don't think CPUID documentation allows any such
>> implication.
>AES/PCLMUL don't support MMX. But they do imply at least SSE2.

That would be what I expected (and you could thus imply SSE, too,
I would think). My concern is that with MMX becoming all the less
supported over time, it may be desirable to have a way to entirely
disable MMX support, and in that case .avx could needlessly turn it
(back) on.


More information about the Binutils mailing list