[FFmpeg-devel] [PATCH v2] aarch64/opusdsp: implement NEON accelerated postfilter and deemphasis

[Lynne]

Apr 6, 2019, 3:41 PM by ceffmpeg at gmail.com:

> 2019-04-06 16:08 GMT+02:00, Lynne <> dev at lynne.ee <mailto:dev at lynne.ee>> >:
>
>>
>>
>>
>> Apr 6, 2019, 2:42 PM by >> ceffmpeg at gmail.com <mailto:ceffmpeg at gmail.com>>> :
>>
>>> 2019-04-02 13:47 GMT+02:00, Lynne <> >>> dev at lynne.ee <mailto:dev at lynne.ee>>>>  <mailto:>>> dev at lynne.ee <mailto:dev at lynne.ee>>>> >>
>>> >:
>>>
>>>> 153372 UNITS in postfilter_c,   65536 runs,      0 skips
>>>> 73164 UNITS in postfilter_neon,   65536 runs,      0 skips -> 2.1x
>>>> speedup
>>>>
>>>> 80591 UNITS in deemphasis_c,  131072 runs,      0 skips
>>>> 43969 UNITS in deemphasis_neon,  131072 runs,      0 skips -> 1.83x
>>>> speedup
>>>>
>>>> Total decoder speedup: ~15% on a Raspberry Pi 3 (from
>>>> 28.1x to 33.5x realtime)
>>>>
>>>
>>> How can I reproduce this?
>>>
>
> Meaning:
> Where can I find the file you used?
>

Sorry for the delay. Here's a test file:
http://www.mediafire.com/file/awg5lfy4j2dv85k/testfile.opus/file <http://www.mediafire.com/file/awg5lfy4j2dv85k/testfile.opus/file>



>>> On something called "X-Gene" Linux, I see a 2,5% overall
>>> speedup (126x to 129x)
>>>
>>> The issue is that there is some probability that your patch will
>>> break mac and Windows compilation, so I would at least
>>> like to understand why I can not reproduce your speedup.
>>>
>>
>> I tested this on real hardware.
>>
>
> Is X-Gene not real hardware?
>

Didn't find much about it at all. If you don't get performance improvements I think someone with a more popular board should test it.


>> How can this break mac support at all when it doesn't
>> support aarch64?
>>
>
> Have you heard of iPhones?
>

Those use iOS.


>> And I can't find a windows aarch64 compiler either.
>>
>
> There is one (I don't know more).
>

I still don't know how you can suggest this patch will break any operating systems when you haven't tested either.