[FFmpeg-devel] [PATCH 1/2] x86inc: Extend FMA_INSTR functionality

[Michael Niedermayer]

On Sat, Feb 15, 2014 at 08:29:53PM -0300, James Almer wrote:
> On 15/02/14 4:49 PM, Michael Niedermayer wrote:
> > On Sat, Feb 15, 2014 at 04:05:34PM -0300, James Almer wrote:
> >> On 14/02/14 10:46 PM, Michael Niedermayer wrote:
> >>> On Fri, Feb 14, 2014 at 11:50:15AM -0300, James Almer wrote:
> >>>> On 14/02/14 5:57 AM, Christophe Gisquet wrote:
> >>>>> 2014-02-13 James Almer <jamrial at gmail.com>:
> >>>>>> You're right, a fifth parameter is probably the proper way. See
> >>>>>> FMULADD_PS in x86util. It would allow actual non-destructive emulation
> >>>>>> of these XOP instructions if it's ever needed.
> >>>>>> It's not for now, but changing it will not hurt and it will probably have
> >>>>>> to be done at some point anyway.
> >>>>>
> >>>>> You can probably make it optional (haven't looked at FMULADD_PS), ie
> >>>>> make the macro use 4-8 arguments, and if %0 == 8, use %8, otherwise
> >>>>> use %2.
> >>>>>
> >>>>
> >>>> Someone else committed the opposite to x264, making the %1 = %2 * %3 + %1 case 
> >>>> unsupported instead, so I'm not sure at this point if this should be on x86inc 
> >>>> or added as a local macro on a given asm file like i originally did.
> >>>
> >>> <michaelni> Skyler_, should i revert 23a8c63452009df21b3f184936b343593d4ccb04 (x86inc: Extend FMA_INSTR functionality) and apply "Warn about not supported emulation of some XOP instructions. Also add pmacsdql emulation." from x264 ?
> >>> <michaelni> i think it would be best if a 5th argument, that is a temporary for emulation could be specified
> >>> <Skyler_> yeah, that might be a better solution than both, but in that case we should move it out of x86inc and into x86util and capitalize it
> >>> <Skyler_> because its semantics don't match the original instruction
> >>> <Skyler_> e.g. like PALIGNR vs palignr
> >>
> >> Sounds good to me.
> > 
> > agree, do you volunteer to implement it ?
> 
> Alright.
> 
> Should each XOP macro accept both four and five arguments, or only five?
> I ask because i can't seem to find a way to make it work with the former.
> What Christophe mentioned above does not work.
> 
> The latter is easy and i have it working, but of course requires providing 
> a value for the fifth argument even when it's not needed.

id guess supporting both would be nice but then again its probably
primarely going to be used when the 5th one is needed
anyway, this is more a question for x264devel

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Those who are best at talking, realize last or never when they are wrong.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140216/a2956b4c/attachment.asc>