VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

mirandir · 2013-07-21 07:46:02

This doesn't work for me with the ATI Catalyst driver (13.6 beta) and a ATI Mobility Radeon HD5470, I have only graphical glitches with Flash videos.

i-rinat · 2013-07-21 08:13:10

mirandir wrote:

I have only graphical glitches with Flash videos.

Do they look like http://i.imgur.com/HAMeu5z.png ? I saw that kind of glitches too after a couple of crashes. I don't think I can do anything about it.

WonderWoofy · 2013-07-21 14:29:07

Thanks i-rinat for creating this, and VladikSS for finding it and throwing it in the AUR! It seems to be working quite nicely on my machine (Intel Core i5-3210M w/ HD4000).

Neato!

mirandir · 2013-07-24 19:41:57

i-rinat wrote:

mirandir wrote:
I have only graphical glitches with Flash videos.
Do they look like http://i.imgur.com/HAMeu5z.png ? I saw that kind of glitches too after a couple of crashes. I don't think I can do anything about it.

No, it looks like

In fact, the flash element doesn't stay in the right place in the web page. It's really sad, because the video seems to be accelerated !

[EDIT] Hum, fullscreen doesn't work too, it looks like your screenshot.

Last edited by mirandir (2013-07-24 20:09:46)

i-rinat · 2013-07-24 20:10:13

mirandir wrote:

No, it looks like http://pix.toile-libre.org/upload/thumb/1374694559.png
In fact, the flash element doesn't stay in the right place in the web page. It's really sad, because the video seems to be accelerated !

Do they appear after page scrolling (like here: https://www.youtube.com/watch?v=lHIOIbcwOqA&t=22 ) ?

The technique browser uses to draw plugin content is way too complicated for me to understand. I can't even determine in which part of the chain bug is.

mirandir · 2013-07-24 20:47:14

i-rinat wrote:

Do they appear after page scrolling (like here: https://www.youtube.com/watch?v=lHIOIbcwOqA&t=22 ) ?

Yes, exactly.

i-rinat · 2013-07-25 08:56:45

mirandir wrote:

Yes, exactly.

I have only intel chip on my laptop, so it's unlikely this bug will be fixed soon.

If you can use AMD open-source driver, I advise you to try it. I heard they added both power management and video acceleration in recent versions. So you'll be able to accelerate video decoding without libvdpau-va-gl. Complexity behind proprietary AMD driver stack and lack of testing essentially makes libvdpau-va-gl intel-only solution.

89c51 · 2013-07-25 14:21:57

i-rinat wrote:

If you can use AMD open-source driver, I advise you to try it. I heard they added both power management and video acceleration in recent versions. So you'll be able to accelerate video decoding without libvdpau-va-gl. Complexity behind proprietary AMD driver stack and lack of testing essentially makes libvdpau-va-gl intel-only solution.

Any info on how to do that (3.10 video acceleration for AMD)?? Do you need anything special or it is supposed to just work.

i-rinat · 2013-07-25 14:30:20

89c51 wrote:

Do you need anything special or it is supposed to just work.

linux 3.10 and mesa 9.2 (unreleased yet, use development version) should be enough. Then you can verify VDPAU presence with vdpauinfo tool.

WonderWoofy · 2013-07-25 14:34:44

89c51 wrote:

Any info on how to do that (3.10 video acceleration for AMD)?? Do you need anything special or it is supposed to just work.

I don't think this stuff was included in 3.10. I think that it is in the 3.11 rc's though. At least I know the power management is not in 3.10. I don't know about the acceleration specifically. But I know there has been a flurry of activity for the open-source radeon drivers of late... and the activity was recent enough to have not made it to 3.10.

Since I run btrfs, I have been using the 3.11 kernels with much success (I just use the vanilla sources from Linus' git tree). In the AUR, there is a nice package for it as 'linux-git'. The PKGBUILD will build all three packages, like the regular 'linux' package, and provides some very good commenting and layout so that you can tweak it to your needs pretty easily.

89c51 · 2013-07-25 14:48:35

3.11 will bring us power management (and stop my fan roaring ). Video acceleration is part of 3.10 (see link below). But i have no idea if you have to built mesa with any special parameters or install libraries or whatever.

http://www.bit-tech.net/news/bits/2013/ … nux-3-10/1

WonderWoofy · 2013-07-25 15:00:47

89c51 wrote:

3.11 will bring us power management (and stop my fan roaring ). Video acceleration is part of 3.10 (see link below). But i have no idea if you have to built mesa with any special parameters or install libraries or whatever.

Neato! I wasn't aware it was already in there. I found this about it from phoronix (the most amazing "news" publication ever )

89c51 · 2013-07-25 15:09:32

Yeah saw that. I just don't know if it needs something arch specific.

mods should we get this discussion somewhere else?

Gusar · 2013-07-25 15:18:16

You don't need anything special, Arch has been building VDPAU in mesa for some time now. Just use mesa-git from AUR until 9.2 is released. And some useful bit of info, the 9.2 release date is August 22nd.

89c51 · 2013-07-25 15:29:23

Thanks. I ll wait for mesa 9.2 then.

Oh. And what about that ldconfig stuff??

Last edited by 89c51 (2013-07-25 16:31:23)

blablubb1234 · 2013-07-27 12:42:45

ValdikSS wrote:

For AMD, you need also https://aur.archlinux.org/packages/xvba-video/
To use it, just add
#!/bin/sh
export VDPAU_DRIVER=va_gl
into /etc/profile.d/vdpau_vaapi.sh and reboot or relogin
So now you can use VDPAU in mplayer2 or flash and it will decode video with VAAPI.

I tried this, but I cannot get it to work in Firefox 22.0 with Shockwave Flash 11.2.202.297 Flash Videos on Youtube (f.e.) crash in fullscreen mode (blank screen, sound for some more seconds, then black screen with horizontal white stripes). I can't even switch to a tty afterwards to recover the system... have to do a hard reset when this happens. On other sites flash plugin crashes immediately, see here.

libva and xvba-video is installed and HW decoding works in mplayer-vaapi. I use the catalsyt-total-hd234k driver which works fine. glxinfo shows direct rendering is active, so is vainfo:

vainfo 
libva info: VA-API version 0.34.0
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib/dri/fglrx_drv_video.so
libva info: Found init function __vaDriverInit_0_32
libva info: va_openDriver() returns 0
vainfo: VA-API version: 0.34 (libva 1.2.1)
vainfo: Driver version: Splitted-Desktop Systems XvBA backend for VA-API - 0.7.8
vainfo: Supported profile and entrypoints
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileVC1Advanced            :	VAEntrypointVLD

Any suggestions what I could try? I do understand correctly, that I need to create the file /etc/profile.d/vdpau_vaapi.sh and paste

#!/bin/sh
export VDPAU_DRIVER=va_gl

into it?

Best regards.

Last edited by blablubb1234 (2013-07-27 14:00:48)

stqn · 2013-07-27 14:50:02

blablubb1234 wrote:

I tried this, but I cannot get it to work in Firefox 22.0 with Shockwave Flash 11.2.202.297 Flash Videos on Youtube (f.e.) crash in fullscreen mode

I’ve also had a crash when playing a fullscreen youtube video, but it generally works fine otherwise. (On Intel HD Graphics.)

I’ve noticed that sometimes the last row or column of pixels in a video is stretched towards the border of the video area. I can’t say if this happens without vdpau.

Here’s an example (first video on youtube’s homepage, I didn’t have to search for a long time !).

blablubb1234 wrote:

I do understand correctly, that I need to create the file /etc/profile.d/vdpau_vaapi.sh

You can also put:

export VDPAU_DRIVER=va_gl

in your .xinitrc. But what you described sounds ok (Remember to chmod +x the file though.)

Last edited by stqn (2013-07-27 14:55:16)

i-rinat · 2013-07-27 16:00:55

blablubb1234 wrote:

I tried this, but I cannot get it to work in Firefox 22.0 with Shockwave Flash 11.2.202.297 Flash Videos on Youtube (f.e.) crash in fullscreen mode (blank screen, sound for some more seconds, then black screen with horizontal white stripes). I can't even switch to a tty afterwards to recover the system... have to do a hard reset when this happens. On other sites flash plugin crashes immediately, see here.

Yeah, unfortunately there is nothing can be done. In addition to some oddity in fglrx driver itself, xvba-video have workarounds for older fglrx bugs which now results in crashes. You may try to patch xvba-video:

diff -ur xvba-video-0.8.0/src/xvba_driver.c xvba-video-0.8.0-n/src/xvba_driver.c
--- xvba-video-0.8.0/src/xvba_driver.c	2011-06-14 15:07:13.000000000 +0400
+++ xvba-video-0.8.0-n/src/xvba_driver.c	2013-04-19 15:28:00.000000000 +0400
@@ -125,6 +125,8 @@
     }
 
     xvba_gate_exit();
+    if (driver_data->x11_dpy_local)
+        XCloseDisplay(driver_data->x11_dpy_local);
 }
 
 // vaInitialize

This will fix one reason of crashes. But those you described first (where you have to use hard reset) are related to fglrx driver itself and cannot be fixed by user-mode library.

In short, libvdpau-va-gl with proprietary AMD driver results to high system instability.

i-rinat · 2013-07-27 16:04:20

stqn wrote:

I’ve also had a crash when playing a fullscreen youtube video, but it generally works fine otherwise. (On Intel HD Graphics.)

If you'll find a way to crash which you can reproduce, drop me an e-mail, or open issue at github, please.

stqn wrote:

I’ve noticed that sometimes the last row or column of pixels in a video is stretched towards the border of the video area. I can’t say if this happens without vdpau.

That's a bug, and it should not be hard to fix. Added to TODO list.

Gusar · 2013-07-30 13:41:46

I was looking at the va_gl repo today... Did I get this right - after shader-based colorspace conversion is implemented in va_gl, I'll have hardware accelerated flash on a GMA950? Woohoo is what I can say to that! I realize it'll be presentation only, not decoding, but with the poor Atom N270 CPU, even moving just presentation to the GPU might be enough to make flash bearable on my netbook.

i-rinat · 2013-07-31 11:10:40

Gusar wrote:

I was looking at the va_gl repo today... Did I get this right - after shader-based colorspace conversion is implemented in va_gl, I'll have hardware accelerated flash on a GMA950? Woohoo is what I can say to that! I realize it'll be presentation only, not decoding, but with the poor Atom N270 CPU, even moving just presentation to the GPU might be enough to make flash bearable on my netbook.

Don't know if it becomes bearable. Last time I run it on Atom N270 netbook with GMA950 it was 2 to 3 times slower than acceptable.

There is a room for improvement, though. YV12 images 2.67 times smaller than RGBA ones, so their upload should become faster. Asynchronous texture upload can increase performance too. It's also worth to try libavcodec to emulate hardware decoding. All that left is to implement it.

WonderWoofy · 2013-07-31 16:53:04

You know, because the AUR package uses the git source (which I imagine is the only one available ATM), it should really be libvpdau-va-gl-git.

WonderWoofy · 2013-08-06 06:28:10

I like this project, and it has worked quite well for me, so I emailed the greatest Linux publication of all time, phoronix and it was just covered.
http://www.phoronix.com/scan.php?page=n … px=MTQyOTM

i-rinat · 2013-08-06 07:36:45

WonderWoofy wrote:

I like this project, and it has worked quite well for me, so I emailed the greatest Linux publication of all time, phoronix and it was just covered.

Thanks for spreading the word!

Thaodan · 2013-08-06 12:15:05

As far as I know all GMA GPU aren't even from Intel.

Arch Linux

#26 2013-07-21 07:46:02

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#27 2013-07-21 08:13:10

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#28 2013-07-21 14:29:07

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#29 2013-07-24 19:41:57

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#30 2013-07-24 20:10:13

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#31 2013-07-24 20:47:14

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#32 2013-07-25 08:56:45

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#33 2013-07-25 14:21:57

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#34 2013-07-25 14:30:20

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#35 2013-07-25 14:34:44

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#36 2013-07-25 14:48:35

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#37 2013-07-25 15:00:47

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#38 2013-07-25 15:09:32

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#39 2013-07-25 15:18:16

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#40 2013-07-25 15:29:23

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#41 2013-07-27 12:42:45

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#42 2013-07-27 14:50:02

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#43 2013-07-27 16:00:55

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#44 2013-07-27 16:04:20

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#45 2013-07-30 13:41:46

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#46 2013-07-31 11:10:40

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#47 2013-07-31 16:53:04

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#48 2013-08-06 06:28:10

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#49 2013-08-06 07:36:45

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

#50 2013-08-06 12:15:05

Re: VDPAU-VAAPI backend / Hardware accel on Intel/AMD in flash and other

Board footer