[thespecialist]

I've found other WMV9 advanced profile, VC1, AP@L3, VBR video's and they played fine with DXVA, unlike that problematic video ( http://www.speedyshare.com/files/28692352/test.wmv ). Makes me think it might have nothing to do with the codec/profile ...

Besides, what makes it even more weird: there are parts in the movie that play perfectly all right with DXVA !

Been looking at the FFMPEG DXVA code, found a few things that looked suspicious at first, but turned out to be allright ... It sucks that the GPU doesn't supply more info about what's going wrong, it's kinda hard to debug like this ...

[thespecialist]

Been busy this weekend analyzing and hacking around. Pretty sure I've found the problem and in fact, I have solved 95% of the artifacts, but there's a bit more to it, to get the whole 100% good. Made some great progress today, I think I'll be able to solve the bug in FFMPEG related to this issue

Will keep you guys posted

[izst]

Looking forward to it

[thespecialist]

Looking forward to it

It seems there's a whole lot more to it. The guy who wrote the VC1 implementation did some stuff that kind of links the decoder to the header parser. This means that the software decoder works fine with the parser but can become problematic in some situations when you use another (hardware) decoder with the same parser. The guy who wrote it didn't know we were going to decode in hardware at some point in the future

I will work on it, but also got tons of other stuff to do and since this isn't an easy fix, don't expect anything too soon. But in 1-2 weeks I'll know a lot more

[thespecialist]

Yihaaaa Fixed it already

Luckily the problem wasn't as deep into the code as I had expected. Will need to change some stuff in my solution (it's too hackerish right now) and make sure I didn't break support for other VC1 stuff. But you can certainly expect something soon

*EDIT* Whoops, yeah I broke support for some stuff that worked before But not to worry, that's fixible

[thespecialist]

Ok, also fixed broken support, BUT there's another problem: FFMPEG hands out some dxva info that's not correct (according to MS specs) and I need that info to make the decision when to use the original FFMPEG code and when to use my fix (it depends on the videofile). So, now I'll need to fix that bug as well. Shouldn't be as difficult as the other bugs, but we'll see ...

Damn, it's quite amazing most of the VC1 video's seemed to work, there was quite a bit wrong in FFMPEG's dxva implementation for VC-1

[isidrogar]

Hi,

I have opened this thread regarding a fix I have made in ffmpeg for DXVA2 decoding with ATI cards. It's only related to H264 and needs more work for interlaced content but I think is very useful for now.

It will be very helpful some reports about which videos are still not working and the description of artifacts (screen capture will be great! ).

[thespecialist]

Hi,

I have opened this thread regarding a fix I have made in ffmpeg for DXVA2 decoding with ATI cards. It's only related to H264 and needs more work for interlaced content but I think is very useful for now.

It will be very helpful some reports about which videos are still not working and the description of artifacts (screen capture will be great! ).

Awesome ! I'll post my patch as well soon (it's for VC-1 only). Almost done, needs some additional work ...

[thespecialist]

First beta of my fix is ready ! Brave xbmc-ers can try it here: http://forum.xbmc.org/showthread.php?p=8...post818486

[isidrogar]

First beta of my fix is ready ! Brave xbmc-ers can try it here: http://forum.xbmc.org/showthread.php?p=8...post818486

Great!

I will try it ASAP...

[thespecialist]

Great!

I will try it ASAP...

Cool, would love to hear how it works for you. Going to clean up the code and go through everything one more time, before I release the source code patch, probably this weekend.