[FFmpeg-devel] [PATCH] lavc/aacenc_utils: unroll abs_pow34_v loop

[Ganesh Ajjanagadde]

On Tue, Mar 22, 2016 at 11:52 AM, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
> On Tue, Mar 22, 2016 at 7:37 PM, Ganesh Ajjanagadde <gajjanag at gmail.com> wrote:
>> On Tue, Mar 22, 2016 at 11:30 AM, Rostislav Pehlivanov
>> <atomnuker at gmail.com> wrote:
>>> On 22 March 2016 at 18:14, Ganesh Ajjanagadde <gajjanag at gmail.com> wrote:
>>>>
>>>>
>>>> Per doc/optimization.txt, aac is a widely used codec, so even a 0.1%
>>>> improvement in aac is fair game for optimizations, assuming it is a
>>>> small code change. Of course, one can debate whether this is small or
>>>> not. I view it as simple and clean, others may disagree.
>>>
>>>
>>> Nope, I still doubt that that 0.1% is a definite performance improvement.
>>
>> Then change doc/optimization.txt.
>
> This particular doc doesn't give you a blanket argument. Specifically
> if the maintainer objects on account of complexity, you should honor
> that - the doc even says as much (ie. only "clean and simple" being
> justified)

It does not. And of course I honor a maintainer's wishes. If he
refuses to accept a performance improvement, so be it.

I just want it clear that from my view this is still ridiculous given
FFmpeg's track record on these sorts of things in the past. I also
find it ironic that there are objections to this on the lines of "what
about some (unspecified) platform?", when bccc81dfa was accepted with
no problems. On a broken libm, or a libm with a slow expf (see
lavu/libm, sometimes it falls back to exp, likely in at least some
MSVC version), this is an obvious performance regression.

>
> - Hendrik
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel