[FFmpeg-devel] [PATCH] tests/gapless: add gapless aac tests

Marton Balint cus at passwd.hu
Sun Mar 6 11:52:13 CET 2016


On Fri, 4 Mar 2016, Michael Niedermayer wrote:

> On Fri, Mar 04, 2016 at 10:46:33PM +0100, Marton Balint wrote:
>>
>> On Thu, 3 Mar 2016, Michael Niedermayer wrote:
>>
>>> On Thu, Mar 03, 2016 at 02:27:52AM +0100, Marton Balint wrote:
>>>> Signed-off-by: Marton Balint <cus at passwd.hu>
>>>> ---
>>>> tests/fate/gapless.mak     | 3 +++
>>>> tests/ref/fate/gapless-aac | 5 +++++
>>>> 2 files changed, 8 insertions(+)
>>>> create mode 100644 tests/ref/fate/gapless-aac
>>>
>>> seems to fail on x86-32
>>>
>>> --- ffmpeg/tests/ref/fate/gapless-aac  2016-03-03 03:06:35.306048679 +0100
>>> +++ tests/data/fate/gapless-aac 2016-03-03 03:11:56.286055441 +0100
>>> @@ -1,5 +1,5 @@
>>> 9459e7dc74af1b451eb890686f04f262 *tests/data/fate/gapless-aac.out-1
>>> -d3c3c4ea121b3f3b8a346a168d2fec0e
>>> +d00bed4d4a83ce1addb92c075b2fcaaf
>>> 9459e7dc74af1b451eb890686f04f262 *tests/data/fate/gapless-aac.out-2
>>> -d3c3c4ea121b3f3b8a346a168d2fec0e
>>> +d00bed4d4a83ce1addb92c075b2fcaaf
>>> 63dd86b78c8fbd22a99bf88583256bfe *tests/data/fate/gapless-aac.out-3
>>> Test gapless-aac failed. Look at tests/data/fate/gapless-aac.err for details.
>>> make: *** [fate-gapless-aac] Error 1
>>
>> Hmm, it seems there is a tiny difference in the 145th sample of the
>> first decoded frame. I thought this is not supposed to happen with
>> the fixed point decoder. Anybody has any ideas?
>
> in absence of better ideas
> 1:pepper the code with printf()
>  run on both x86-32 and 64
>  and diff to find te first difference
> if that explains the problem then you are done
> else goto 1
>

Okay, it seems that there are some architecture-dependant float-double 
rounding errors at compile time and the sinewin tables are generated at 
runtime, so as far as I see this will no be bitexact on all platforms...

I think I'll just come up with another test which does not involve the 
actual data, only the frame metadata, that should be enough for checking 
gapless results.

Regards,
Marton


More information about the ffmpeg-devel mailing list