[x264-devel] Rare macroblock errors in recent x264...

[Jeppe Øland]

On Wed, Jul 18, 2012 at 9:02 AM, Jason Garrett-Glaser <jason at x264.com> wrote:
>>>>>>> As Jason wrote you last month we need sample. Preferably short source
>>>>>>> sample with which we can reproduce your problem. If you can't provide
>>>>>>> source sample than at least provide short encoded sample with error so
>>>>>>> we can at least see it and analyse encoded stream (what caused error:
>>>>>>> high QP, wrong prediction, etc.)
>>>>>>
>>>>>> As for getting some source content that exhibits the problem, so far
>>>>>> it's been hard to nail down something small.
>>>>>> But yesterday I may have found something ...
>>>>>
>>>>> No idea if its the same problem as I'm seeing in the other files, but
>>>>> I managed to get it to crap out with a small sourcefile!
>>>>
>>>> I can't replicate the problem with the same version; it seems to
>>>> encode fine here.
>>>
>>> I have tried it on 3 different PCs now (Core2 and i7) and they all do it.
>>> Even with '--asm 0" it happens.
>>>
>>> Can you try with this one:
>>>     http://mirror01.x264.nl/x264/32bit/8bit_depth/revision2145/x264.exe
>>
>> Replicated, let me see if I can figure this one out.
>>
>> Jason
>
> This bug was already fixed in revision bcd41db.  Please use the latest
> versions of x264 to encode and you won't have problem with bugs that
> were previously fixed.

Thanks for checking it - but crap!
That puts me back in square 1 then :-(

I'm trying to find a problem that is happening in all the newer versions.
(You can see my r2197 file for an example).

It's extremely hard for me to get it reproduced ... I have only one
file that does it, and it takes like half a day to encode to where the
problem happens. So far, encoding a snippet has not been able to
reproduce it, and changing settings to faster encoding has not
reproduced it either.
I was trying to regression test against it, and found the r2145
problem that way ... too bad it wasn't the same problem!

I don't have an environment set up to compile Windows versions ... but
if somebody can provide me a 32-bit Windows r2146 build, I will see if
that reproduces the problem with the other file.

Regards,
-Jeppe