[onomatopellan]

Those UVD+ users with problems, are you using the last Catalyst?

I have some artifacts problems but I can't help because I'm sticked to Catalyst 9.2. To make long story short, any superior catalyst freeze my computer because WDDM1.1 uses Directx10 instead directx9 and there is some problem with my mobo and dx10.

[CrystalP]

Montellese, since it worked fine and then broke, we need to find the change(s) that broke the ATI 3xxx series. Can you narrow down the range of revisions that could have done this?
I hear there's a site with unofficial builds that go back further than the official nightlies
If you can build, you could even find the exact revision that did it and it would be really easy to fix then.

[Vorm]

I know it doesnt help the problem but I made up my mind this week that I was going to ditch ATI altogether and grab a GT240.

Should be cheap enough and will hopefully solve all these problems.

Saying that, anyone experiencing artefacting with a GT240??

[0wing]

Montellese, since it worked fine and then broke, we need to find the change(s) that broke the ATI 3xxx series. Can you narrow down the range of revisions that could have done this?
I hear there's a site with unofficial builds that go back further than the official nightlies
If you can build, you could even find the exact revision that did it and it would be really easy to fix then.

I tried reverting to 31903, the first dharma branch build, same problem there.

Debug log: http://pastebin.com/2KSksLvK

I'm currently testing trunk builds. 27376, where DXVA2 was introduced, gave me a green screen when using DXVA. Updating to a more recent rev now.

UPDATE: I'm having trouble building on revs between 29000-31000, don't know what's wrong. Tried 29000,31005 and 31500, neither worked.

[CrystalP]

Off the top of my head that's most likely the libraries downloaded by DownloadBuildDeps.bat. They changed a few times, so re-run the script.
There were also a few broken builds here and there.
The how-to-compile Wiki page was kept reasonably up to date, you can look at the instructions matching the revision you're trying to build.
Thanks for the effort!

[0wing]

Before rev. 30179 all I get is a green screen with DXVA. Rev. 30179 and above have artifacts.

I've compiled rev. 30178 and 30179 so others can verify this.

30178: http://www.multiupload.com/N2DT2XCD0K
30179: http://www.multiupload.com/3N7FU63F0B

The skin isn't working properly, I didn't bother fixing it since it's for testing only.

[CrystalP]

0wing, did it work properly for you at some point?
I asked to track a revision because Montellese stated that it worked fine around r31961. If it didn't, there is no use.
You can also try the latest VLC with hw accel - they use a similar method for DXVA and if it works for VLC, there's a pretty good chance the problem is on our side.

[0wing]

Nope, I just recently started using XBMC and at first I used the DSPlayer branch which worked but it crashed a lot for me.

Then i switched to Dharma and it works perfectly on my htpc with a Radeon HD4350 but on my computer, with the HD3870, it never has.

VLC DXVA doesn't work for me. It's even worse than XBMC, since some files play just fine with XBMC. Nothing works in VLC.

When I first switched to Dharma I thought it was working for some time but I had just played the "right" files. Maybe it's the same with Montellese.

Btw is the DXVA method used in XBMC very different to the one used in Boxee? Because DXVA works for me in Boxee.

[onomatopellan]

It seems your builds don't have the skin textures. Maybe you had a problem with TexturePacker while building? If so you can try to replace TexturePacker.exe on "xbmc\tools\TexturePacker" with this one: http://www.mediafire.com/?iytl20r4lfzgq05

Also always use the portable mode ("xbmc.exe -p") when testing so many builds.

Then i switched to Dharma and it works perfectly on my htpc with a Radeon HD4350 but on my computer, with the HD3870, it never has.

I also have some problems with my HD3450 that has UVD+ like yours. I think is a problem that the ffmpeg team has to resolve. In fact, the difference between rev. 30178 and 30179 that you posted is an ffmpeg update. I'm pretty sure problems like ours will be also resolved with an ffmpeg update.

VLC DXVA doesn't work for me. It's even worse than XBMC, since some files play just fine with XBMC. Nothing works in VLC.

VLC DXVA needs the very last Catalyst from ATI so I'm afraid legacy ATI cards will never have dxva on VLC.

Btw is the DXVA method used in XBMC very different to the one used in Boxee? Because DXVA works for me in Boxee.

Boxee uses a different method more alike MPC-HC. You should read Elis post from this thread http://forum.xbmc.org/showthread.php?tid=69306

[CrystalP]

The problem is most likely in ffmpeg, which does the actual DXVA decoding. To take this further, you'd need to go over there and search for information, report a bug, ...

[Montellese]

Hey guys, sorry for not getting back with more information but my ISP decided to f**k up my internet access for a while so I couldn't follow the thread.

VLC DXVA doesn't work for me. It's even worse than XBMC, since some files play just fine with XBMC. Nothing works in VLC.

VLC doesn't seem to use DXVA when I play files with it so I don't know if it would work or not. I'm using the latest Catalyst drivers (10.8) and VLC mentions that you need at least version 10.7 but I couldn't figure out why it doesn't use DXVA.

When I first switched to Dharma I thought it was working for some time but I had just played the "right" files. Maybe it's the same with Montellese.

That is possible. I mostly watch tv episodes (in SD quality) so I wouldn't notice. I watched a view HD movies and was awestruck by how fast the movies played using DXVA. I never really noticed the noise but maybe I was just too far a way from the screen. I only noticed the noise after I took a closer look at the image when I got the artefacts all over the screen.

Btw I found this in the VLC Wiki at http://wiki.videolan.org/VLC_GPU_Decoding:

For ATI GPUs, you NEED Catalyst 10.7, that is just out.
Then, you are required to use a GPU supporting Unified Video Decoder.
We believe you need a GPU supporting UVD2, like HD4xxx, 5xxx or 3200. One might have success with UVD+ GPU, like some HD 3xxx, but this isn't tested.

Really seems to be a non-XBMC-related issue.

[theccap10]

I wasn’t sure whether to start a new thread or add to this one but my issue is very likely the same as Montellese and 0wing so thought I might be able to cement the verdict one way or the other.

I have the same symptoms as described and, like Montellese, I’m sure this was working previously in the nightly build I was using prior to upgrading to Dharma Beta 2. But of course I have never attempted to use the mkv file that is currently causing me grief, so it may not have worked in the earlier release either.

I can play all manner of HD formats with DXVA2 enabled and it works beautifully no problems. It just some mkv files (and I can’t see any obvious format differences) that the pixilation occurs.

My build is:
Dharma Beta 2
Windows 7 Ultimate 32bit
Intel® Pentium® 4 CPU 3.00GHz
2.5 GB RAM
160GB SATA Hard Drive
Graphics card = ATI Radeon HD3850
Catalyst Drivers = v 10.9
DirectX runtime = June 2010
FFdshow = 2010-09-13 (rev. 3572)
Desktop Resolution: 1920x1080
Aero = off

Video settings within XBMC are:

Render method: Auto detect
Allow hardware acceleration (DXVA2): Enabled
Adjust display refresh rate to match video: Disabled
Sync playback to display: Disabled
A/V sync method (greyed out): Video clock (Resample audio)
Allow error in aspect ratio to minimize black bars: None
Display 4:3 videos as: Normal
Video post-processing: disabled

So far I have tried the usual stuff:

Tried VSYNC on and off
Post processing on and off
Changed adjust display refresh setting etc etc.

I should also mention that the CPU never goes above 15% when playing the MKV file in question and there is no stutter or performance glitches, only pixilation.

I am going to try uninstalling Dharma Beta2 and retry the nightly I used successfully for so long.

At the very least I would like to understand the subtle difference between the MKV files that work and the ones that don’t.

Here is my debug when opening the mkv file in question

http://pastebin.com/AcfiGtzV

As already stated It looks very likely that the issue is ffmpeg related but just wanted to add my two cents to see if that is consistent.

It’s been a while since I posted last but wanted to say Congrats to all involved in this project. By far and away the best OS project on the net IMHO.

[theccap10]

Just a quick update on this.

Last night I completely unistalled XBMC and used the earlier release I was on before upgrading to Dharma B2.

I had the same pixelation and artefacts as I do using Dharma... confirms everything already discussed in this thread.

I'll do a bit more research to identify the specific mkv format that doesn't work

in the meantime it looks like a new Graphics card or wait for and update from ffmpeg.

[Devlin]

Just chiming in to say that I'm suffering from the same pixelation noise that's been mentioned here. (only in XBMC, in this case Dharma Beta 2). The movies look fine under VLC and also on my HTPC using an older XBMC version.

Desktop (noisy) - Dharma Beta 2
Gfx: ATI HD5850 - using Catalyst 10.8.

HTPC (fine)
Asrock 330HT.

Anything else needed?

[theccap10]

After much Googling and experimentation I’m not sure I am any the wiser with respect to this artefacts issue except to say that it's most definitely not an XBMC issue.

http://forums.amd.com/game/messageview.c...TMP=Linear

http://imouto.my/watching-h264-videos-using-dxva/

The above links offer some interesting reading on the issue of DXVA offloading with ATI cards but ultimately there is no solution in there.

I’ve been experimenting lately to work out which files fail and why. For my system (and I have a HD3650 not 3850 as stated in my earlier post) I noted that the MKV files that play with no issues are typically files with low Reference Frames.

In one of the links above, one of the posters in that forum makes the observation that it’s not the resolution or the number of reference frames but a combination of these. He came up with the formula –

Resolution x * Resolution y * Reference frames < 10000000

I have found that to be exactly the case with my own experimentation. If the media I am playing is under 10000000 it works without pixilation, if higher it pixilates (is that a word?) or shows artefacts all over the place.

From all my reading and googling I’m not sure if this is a hardware limitation of the card/UVD+ or a software driver issue but either way looks like an ATI issue.

I’ve spent 12 months building my HTPC and experimenting with GPU offloading etc. I deliberately used an older spec machine to try and get some use/value out of an old system I had lying around . Ironically I’ve ended up spending more money on the case, cooling and quiet technology than I have saved by using old technology and an ATI graphics card.

Suffice to say I’m giving up on ATI and their drivers.

Goodbye old HTPC……. Hello Acer Revo