Kodi Community Forum
Initial native support for DXVA2 in SVN - Time to say goodbye to your firstborns - Printable Version

+- Kodi Community Forum (https://forum.kodi.tv)
+-- Forum: Support (https://forum.kodi.tv/forumdisplay.php?fid=33)
+--- Forum: General Support (https://forum.kodi.tv/forumdisplay.php?fid=111)
+---- Forum: Windows (https://forum.kodi.tv/forumdisplay.php?fid=59)
+---- Thread: Initial native support for DXVA2 in SVN - Time to say goodbye to your firstborns (/showthread.php?tid=69306)



- SlaveUnit - 2010-05-02

Intel is worse than ATI. I think in fact it's non working?


- elupus - 2010-05-02

Right, intel is not working. I'm just about to try it again on poulsbo to see if it has improved (just got black screen before)


- SlaveUnit - 2010-05-02

elupus,
Im sure you already know most of this but here is a read on on intel if anyone else wants to look at it.

The PDF on this page entitled "Using H.264/AVC DirectX* Video Acceleration with the IntelĀ® G45/GM45 Express Chipsets"


- SlaveUnit - 2010-05-02

elupus,
Any changes made with that misreported buffer issue from the past week (Transformers sample file)? I see in r29717 that with an nvidia card the file still has the same jumping and pixelation effect but now has flashing green every few frames. Just curious.


- elupus - 2010-05-02

SlaveUnit.. eh, you sure? should have been fixed in r29709. Could you post a debug log for me?

<edit> never mind, it commit the wrong thing, will fix </edit>


- steelman1991 - 2010-05-02

elupus Wrote:SlaveUnit.. eh, you sure? should have been fixed in r29709. Could you post a debug log for me?

Reported already one page back with debug log No


- elupus - 2010-05-02

SlaveUnit:
fixed in r29730


- elupus - 2010-05-02

The intel stuff needs to be fixed inside ffmpeg by somebody that has the hardware.


- SlaveUnit - 2010-05-03

elupus Wrote:SlaveUnit:
fixed in r29730

Thanks very much sir!!


- bb10 - 2010-05-03

elupus Wrote:CrystalP,

Hmm.. that sounds odd. Sounds like it's not swapping front and back buffer at the right time. I suppose it could be that It's such a new feature in windows that the drivers might not be uptodate on it yet. Maybe should be made optional.

Maybe increasing backbuffers would solve the issue. Maybe the DWM queuing too, but i really don't get how that would work with flipex.

With flipex, DWM is supposed to be using the same surface as the app for rendering to screen, ie not copying it to it's own buffer. If we are running with only 1 backbuffer, how the heck should DWM be able to queue up frames for later presentation? The docs doesn't mention it asfar as i've been able to tell.

Not sure if this helps:
Quote:BackBufferCount

(Windows 7 Only)

When SwapEffect is set to the new D3DSWAPEFFECT_FLIPEX swap chain effect type, the back buffer count should be equal or greater than 2, to prevent an application performance penalty as a result of waiting on the previous Present buffer to be released by DWM.

When the application also uses Present Statistics associated with D3DSWAPEFFECT_FLIPEX, the back buffer count is suggested to be 2 to 4.

Using D3DSWAPEFFECT_FLIPEX on Windows Vista or previous operating system versions will return fail from CreateDeviceEx.

More here: http://msdn.microsoft.com/en-us/library/ee890072%28VS.85%29.aspx


- CrystalP - 2010-05-03

bb10, you're right and XBMC specifies 1 back buffer frame. So who knows.

Increasing the backbuffers further may be a good thing too. It's recommended for multimedia applications, since they typically don't need the low latency. I don't understand that part of XBMC yet, but it may have side-effects on the audio-video synchronization, so it has to be done carefully.
With 4 back buffers + front buffer, the last decoded frame will show after 4 flips, and that's about 160ms at 24Hz refresh rate (the worst case). You're free to try and report though!


- bluenote - 2010-05-03

From r29697 to 29717 my video playback improved (less green flashes and choppy playback on certain HD mkv's) but it broke "true fullscreen" for me completely.

However, Im a total XBMC newb, having just stuck my toe in the water from mediaportal, so although I'm reading as much and as fast as I can its possible I just did something dumb.

My card is an nvidia GT 210 if it makes a diff.

Thanks for any advice or pointers, I'll be trying tonights build if it shows up although it sounds like Craig? at sshcs might be having trouble compiling atm.

Oops, one quick side question that just occurred to me -- I barely understand this REPO SVN stuff, and I've read the posts about the merge, but I've seen a couple posts here and there suggesting that in fact it can? does? work in these nightly unofficial builds. Is it possible? Can someone point me in the right direction if so? It doesnt seem to appear, I've tried installing manually to the best of my ability so far, but nothing seems to happen.

Anyways, in closing, this is an *awesome* piece of software, and even on its first day, the WAF is super high compared to other products I've used. So thank you all for a great great experience.

thanks for any pointers or comments.


- CrystalP - 2010-05-03

Writing about the backbuffers made me curious and it turns out that without FLIPEX there's not much effect (if any) for the max allowed number of buffers (30), but with FLIPEX there is an a/v sync + GUI response delay of about 1 second at 24Hz, as could be expected.

Additional information about the FLIPEX GUI artifact: it is more pronounced with more back buffers, but I'm not sure I'm still seeing the same type of artifact. There seem to be a 'going back in time' problem. At 24Hz, 4 back buffers and fake fullscreen, the RSS ticker on the home page doesn't go smoothly right to left, sometimes it briefly jumps back to the right, and then picks up where it left off. So it seems that you can see past images. The same effect can be seen on screen transitions, when moving the selection, etc.
Oddly, true fullscreen doesn't show these artifacts, only fake fullscreen.

It looks like somehow XBMC doesn't draw/hand off frames at the same speed Windows shows them.

Sorry for the off-topic follow-up but there doesn't seem to be a better place for this at the moment.


- elupus - 2010-05-03

Ok, that sound logical that one should use atleast 2 backbuffers. But sad that that doesn't fix your issue CrystalP


- ashlar - 2010-05-03

One thing I don't understand: do the current problems with DXVA2 implementation for ATI come from ffmpeg or it's something specific to XBMC?

I ask because in the former case I'd stop having high expectations of seeing this "fixed" in time for the next stable release.

Thanks! Smile