Hi,

On Tue, May 15, 2012 at 4:21 PM, Jason Garrett-Glaser <[email protected]> wrote:
> On Tue, May 15, 2012 at 4:01 PM, Ronald S. Bultje <[email protected]> wrote:
>> On Tue, May 15, 2012 at 10:31 AM, Diego Biurrun <[email protected]> wrote:
>>> On Mon, May 14, 2012 at 12:11:09PM +0100, Måns Rullgård wrote:
>>>> Diego Biurrun <[email protected]> writes:
>>>>
>>>> > On Fri, May 11, 2012 at 12:01:23PM -0700, Ronald S. Bultje wrote:
>>>> >> On Fri, May 11, 2012 at 10:23 AM, Jason Garrett-Glaser <[email protected]> 
>>>> >> wrote:
>>>> >> > I dislike this; just make the function mmx2.
>>>> >>
>>>> >> I'm supporting this sentiment, at least somewhat, as in "I'd prefer to
>>>> >> not add too many hacks for prehistoric cpus". Are there mmx-only cpus
>>>> >> (possibly from Intel) that do support cmov? Or does cmov support imply
>>>> >> sse/mmx2?
>>>> >
>>>> > cmov support implies nothing of the sort.  It appeared with the i686
>>>> > architecture pioneered by the Pentium Pro, which did not even have MMX.
>>>>
>>>> And on the flip side, the original Pentium MMX did not have cmov.  I
>>>> don't think we particularly care about either of those CPUs though.
>>>
>>> So what's a suitable way forward if the original patch is deemed too
>>> ugly?  Ifdeffing out all the functions with HAVE_CMOV does not sound
>>> better to me, so ideas welcome...
>>
>> I guess I'm OK with the original patch, the HAVE_CMOV needs to go
>> somewhere and if it goes in x86util.asm, the ugliness is at least in
>> one specific place. So go ahead and commit that. Jason can you live
>> with that?
>
> Why don't we just make it mmx2?  All mmx2 CPUs have cmov.  I don't
> like adding new variables for the purpose of supporting 15-20-year-old
> systems.

Diego, can you produce a smaller patch / final code by marking these as mmx2?

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to