[FFmpeg-devel] [PATCH v4 3/3] avcodec/adpcm: Fixes output from Westwood ADPCM.

[Zane van Iperen]

On 25/4/21 11:02 am, Andreas Rheinhardt wrote:
> Zane van Iperen:
>>
>>
>> On 25/4/21 9:40 am, Andreas Rheinhardt wrote:
>>
>>> This format is covered by FATE (namely the adpcm-ima-ws test) and so you
>>> need to update the checksums in your patch; of course you should verify
>>> that your change is correct. See https://ffmpeg.org/fate.html for more.
>>>
>>> - Andreas
>>> _______________________________________________
>>
>> The attached file here https://trac.ffmpeg.org/ticket/9198 decodes
>> incorrectly without the patch. I'd be inclined to say that it's correct.
>>
>> The issue isn't caught by FATE because it only tests a mono sample.
>>
> 
> If it weren't caught by FATE, the test wouldn't need an update.
> 

Interesting, FATE didn't catch it for me. Which test is it?

$ make fate-adpcm fate-acodec fate-westwood-aud SAMPLES=fate-suite/

All pass for me.

> Both the sample from the ticket as well as the fate-sample
> (vqa/cc-demo1-partial.vqa) are 22050 Hz, mono, s16; the fate sample is
> version 2, the new sample version 0. None of them test the version 3
> codepath, so the change to it is untested.

Yeah, don't know where stereo came from. I need to sleep more...

What I meant was see this. Top is before, bottom is after:

https://0x0.st/-mex.png

> I have tested a few files and the changes to the non-version 3 codepath
> seem to be correct; I have also found out that your commit
> c012f9b265e172de9c240c9dfab8665936fa3e83 made decoding
> https://samples.ffmpeg.org/game-formats/vqa/Tiberian Sun
> VQAs/startup.vqa (which is a version 3 format ima-ws file) crash.
> 

I've fixed the crash, see https://ffmpeg.org/pipermail/ffmpeg-devel/2021-April/279504.html