[afedchin]

Link: http://www.dvdloc8.com/clip.php?movieid=8419&clipid=2
MediaInfo: http://xbmclogs.com/show.php?id=175750
Debug log: http://xbmclogs.com/show.php?id=175751

Thanks. Please, can you provide debug log from Frodo 12.3 with same movie? Just for compare.

[kit500]

http://xbmclogs.com/show.php?id=176461

[afedchin]

http://xbmclogs.com/show.php?id=176461

Thanks. Can you try this build https://drive.google.com/file/d/0B42OdYa...hORmc/edit with same movie. Also needs debug log.

[kit500]

still no luck
http://xbmclogs.com/show.php?id=176587
(also creates xbmc_crashlog.dmp)

[afedchin]

I have only one idea. Problem is what dxva decoder initiated twice and for radeon HD it's fatal behavior. Why? I don't know. Need summon @FernetMenta or/and @elupus to this thread.

[nille02]

Beta 4 still broken.

[cg110]

Doing some debugging of this, it looks like the issue is somewhere in ffmpeg, and the DVXA implementation of getformat. IE this code:
https://github.com/xbmc/xbmc/blob/master...eg.cpp#L98

It appears that GetFormat is called twice from ffmpeg. I'm not sure where or why, as the windows build of ffmpeg doesn't have symbols in a useful form for visual studio.

I suspect something like this:
https://github.com/cg110/xbmc/commit/315...c49271f382

might workaround the issue, as it'll close the current dxva decoder before asking for another one. However, I've no idea how to make a release build for others to try, as I tend to do debug builds.

[afedchin]

I will build windows version with your workaround. It take a while.

[nille02]

Sounds plausible since the HD3XXX and below cards can only decode one stream at the same time.

[afedchin]

nille02, please try this build https://drive.google.com/file/d/0B42OdYa...FJS3M/edit

If it also doesn't work correctly, please provide a debug log.

[nille02]

Thank you very much, it works.

https://www.dropbox.com/s/vhdnmqbiefghpe....31.14.png

https://www.dropbox.com/s/ftudqzwks1rndl....34.50.png

[afedchin]

Awesome! I'll create a PR at github.com.

[cg110]

Note that there's also a follow up change I found by making opendecoder randomly fail:
https://github.com/cg110/xbmc/commit/2f7...300f11da9b

It's the likely cause of the access violations in the debug logs.

If opendecoder works once, then avctx has the hw hooks set, if subsequent calls fail the hooks are left in place, but the hardware context is null, and so it causes access violations.

[nille02]

@cg110 @afedchin i want to thank you both for your patches and everyone else who helped. I have now Build my own Beta 4 with the Patch from cg110's https://github.com/cg110/xbmc/commit/682...065fa85843 . This runs very well on my HD3200, HD4200, Mobility HD 3650, HD 3850,HD5770, HD7850, Intel HD 2500 and Nvidia GTX760 (i don't have more Cards to test, only a ATI xpress 200m, and a S3 Savage DDR but both don't have a video decoder and the last one has is not xbmc capable because of opengl 1.4).

This is my Build with the Gotham Final Source and cg110's Patch

XBMCSetup-20140504-0d1d803-Gotham.exe

[afedchin]

@nille02, can you try this build of xbmc.exe https://drive.google.com/file/d/0B42OdYa...M5a3c/edit with an other small workaround for your issue. just replace XBMC.exe from beta4 to this new one.