| View previous topic :: View next topic |
| Author |
Message |
ivan2k
n00b
Joined: 10 Sep 2010
Posts: 50
Location: Codognè (TV), Italy
|
 Posted: Fri Jan 04, 2013 8:35 pm Post subject: Radeon HD3200 - vdpau xvmc xvba uvd vaapi mesa gallium g3dvl |
 |
|
Hi all,
I have an amd64 box with a Radeon HD3200 on it, using open source radeon driver.
In the past I'm using an Nvidia GF 9600 with VDPAU support which offload decoding to GPU and reduce CPU usage, but now with this radeon I don't know what can I do.
Digging on the web, found nothing usefull, so: is there some good man explain to me exactly what I should activate for video decoding?
Or at least, how to get information of what my GPU support and don't support.
Thanks |
|
| Back to top |
|
 |
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Fri Jan 04, 2013 8:39 pm Post subject: |
|
|
| xv and xvmc at the moment. I've been reading about the vdpau galliun3d state tracker, but it isnt done yet and no driver supports it. When that gets done though all gallium3d drivers should be able to support it. Its just a matter of time now. I think they are in desperate need of coders that know how the graphics stack functions. |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Fri Jan 04, 2013 8:41 pm Post subject: |
|
|
I am amazed again and again about people's poor google-fu. How exactly did you dig? Cos it's not like this stuff is a big secret. Anyway: https://forums.gentoo.org/viewtopic-t-946360.html
@duby2291: The gallium vdpau tracker is very much finished. The problem is somewhere else - GPUs suck at decoding (so you won't see a shader based h264 decoder), and accessing the dedicated decoders would require either reverse-engineering or AMD releasing docs. And that's not a simple thing: https://forums.gentoo.org/viewtopic-t-946818.html |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Fri Jan 04, 2013 9:04 pm Post subject: |
|
|
| Gusar wrote: | I am amazed again and again about people's poor google-fu. How exactly did you dig? Cos it's not like this stuff is a big secret. Anyway: https://forums.gentoo.org/viewtopic-t-946360.html
@duby2291: The gallium vdpau tracker is very much finished. The problem is somewhere else - GPUs suck at decoding (so you won't see a shader based h264 decoder), and accessing the dedicated decoders would require either reverse-engineering or AMD releasing docs. And that's not a simple thing: https://forums.gentoo.org/viewtopic-t-946818.html |
My understanding is that AMD is not going to release documents. It won't happen. It has something to do with DRM, as in restrictions management. As far as shaders doing the work, that is kinda the whole point of gallium3d. If they didnt intend shaders to do the work then it would not have been implemented on gallium3d. Thats what gallium3d does. It's just a matter of the driver being able to compile TGSI into native hardware instructions. My understanding of that situation is that they are currently working on a new LLVM based compiler that should resolve alot of the problems that the old shader compiler had. Both radeon and nouvaou are working towards this objective now... Its exactly the same situation with OpenCL. As soon as an adequate compile infrastructure is implemented then it should work
Like I said it is just a matter of time... They are in desperate need of coders that know how the graphics stack works and knows how compiler infrastructures work.
EDIT: I think this whole situation should improve tremendously once GCN based architectures get good support. The architecture was designed more for computing than for throughput. Where as the old VLIW architectures were designed more for throughput. It should be a whole lot easier to code an adequate compile infrastructure for the newer architecture. |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Fri Jan 04, 2013 9:35 pm Post subject: |
|
|
| duby2291 wrote: | | My understanding is that AMD is not going to release documents. It won't happen. It has something to do with DRM, as in restrictions management. |
They're trying. I wouldn't get my hopes up, but who knows, it might happen.
| duby2291 wrote: | | As far as shaders doing the work, that is kinda the whole point of gallium3d. If they didnt intend shaders to do the work then it would not have been implemented on gallium3d. Thats what gallium3d does. It's just a matter of the driver being able to compile TGSI into native hardware instructions. |
Doesn't matter what gallium does or doesn't do. GPUs aren't suitable for video decoding, that's what matters.
| duby2291 wrote: | | My understanding of that situation is that they are currently working on a new LLVM based compiler that should resolve alot of the problems that the old shader compiler had. Both radeon and nouvaou are working towards this objective now... Its exactly the same situation with OpenCL. As soon as an adequate compile infrastructure is implemented then it should work |
No, it will not work. GPUs excel at massive parallelism, but decoding is pretty much a serial process. No amount of compiler optimizations will change that. GPUs simply aren't built for the kind of task that is video decoding. That's why graphic cards have dedicated decoder units. Intel/Nvidia/AMD wouldn't bother with those if the GPU could do it. But they did bother with them, because the GPU cannot do it. |
|
| Back to top |
|
|
ivan2k
n00b
Joined: 10 Sep 2010
Posts: 50
Location: Codognè (TV), Italy
|
| Posted: Fri Jan 04, 2013 9:37 pm Post subject: |
|
|
@duby2291: thank you, but activating xvmc on mplayer, and using it with -vo xvmc results in:
| Code: | The output video driver you choose is incompatible with this codec.
Try adding scale filter in filter sequence, for example: -vf spp,scale |
*It's translated from Italian, so I hope you understand.
@Gusar:
Yes, probably my google-fu are very weak, or probably I was using wrong keywords on searching, however I'm amazed like you to find out 2 recent threads ask similar question and have not seen them.
The main problem for me, is that there are a lot of terms (xv, xvmc, xvba, vaapi, vdpau, mesa, gallium, ecc...) that I don't know what they do exactly, and don't know how the cooperate. Several times, someone refers to "vdpau" as the API and someone as the hardware support, ending with confusion.
So any attempt to search anything ends up with too much and generic results or too little and advanced (for me) discussions.
Yes, I'm a noob, no one born expert.
Anyway thank you, now I know a little more about AMD/ATI and their (un)support.
So, for HD3200, USE="-vdpau -vaapi" and probably -xvmc too. |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Sat Jan 05, 2013 12:16 am Post subject: |
|
|
| Gusar wrote: | | duby2291 wrote: | | My understanding is that AMD is not going to release documents. It won't happen. It has something to do with DRM, as in restrictions management. |
They're trying. I wouldn't get my hopes up, but who knows, it might happen.
| duby2291 wrote: | | As far as shaders doing the work, that is kinda the whole point of gallium3d. If they didnt intend shaders to do the work then it would not have been implemented on gallium3d. Thats what gallium3d does. It's just a matter of the driver being able to compile TGSI into native hardware instructions. |
Doesn't matter what gallium does or doesn't do. GPUs aren't suitable for video decoding, that's what matters.
| duby2291 wrote: | | My understanding of that situation is that they are currently working on a new LLVM based compiler that should resolve alot of the problems that the old shader compiler had. Both radeon and nouvaou are working towards this objective now... Its exactly the same situation with OpenCL. As soon as an adequate compile infrastructure is implemented then it should work |
No, it will not work. GPUs excel at massive parallelism, but decoding is pretty much a serial process. No amount of compiler optimizations will change that. GPUs simply aren't built for the kind of task that is video decoding. That's why graphic cards have dedicated decoder units. Intel/Nvidia/AMD wouldn't bother with those if the GPU could do it. But they did bother with them, because the GPU cannot do it. |
It doesnt matter. AMD is not going to release UVD docs. It must be done with shaders or it isnt going to happen at all. I believe it will be difficult to do, but I don't believe it "cannot" be done. It is being worked on as we speak. It's going to happen. There is no other route to take. That's the whole reason for spending all this time developing this new LLVM compiler infrastructure is to make a new platform that is more suitable for compute tasks. It must be done in order to facilitate things like video decode and OpenCL and other things.
EDIT: I agree with you that current GPU architectures arent suitable for compute jobs. But they are capable of computing. This new round of architectures is even more capable of computing than the last. It only gets better from here. In the future (only god knows how far) the GPU will be capable of doing most co-processing. When that day come the compile infrastructure that is being built today is going to come in real handy. Thats why I say that even things that arent easy now, are in fact very worthwhile pursuits. Even though its hard it still needs done. |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Sat Jan 05, 2013 12:32 pm Post subject: |
|
|
| duby2291 wrote: | | It doesnt matter. AMD is not going to release UVD docs. |
And you know that how? Chances are slim, but flat out saying "they're not going to" without anything to back it up is silly.
| duby2291 wrote: | | It must be done with shaders or it isnt going to happen at all. |
Then it isn't going to happen.
| duby2291 wrote: | | I believe it will be difficult to do, but I don't believe it "cannot" be done. It is being worked on as we speak. |
No it isn't. There was a person at AMD working on it, but he, like everyone else who tried before, discovered it can't be done. So he instead switched to writing UVD code, but there's no guarantee he'll be able to release that code.
| duby2291 wrote: | | It's going to happen. There is no other route to take. |
Everyone who tried to do something with video on the GPU (be it decoding or encoding) gave up. Every single one. If you think you're smarter than all of them, go ahead, try yourself. Or you can trust them when they say GPUs aren't suitable for this kind of task.
| duby2291 wrote: | That's the whole reason for spending all this time developing this new LLVM compiler infrastructure is to make a new platform that is more suitable for compute tasks.
EDIT: I agree with you that current GPU architectures arent suitable for compute jobs. But they are capable of computing. This new round of architectures is even more capable of computing than the last. It only gets better from here. In the future (only god knows how far) the GPU will be capable of doing most co-processing. When that day come the compile infrastructure that is being built today is going to come in real handy. Thats why I say that even things that arent easy now, are in fact very worthwhile pursuits. Even though its hard it still needs done. |
Wow, are you stubborn. "Compute" isn't some sort of magic. You can't say "It's compute! It can do anything!"
I can only repeat: GPUs excel at massive parallelism. Keyword "massive". Video decoding is a serial process. The magic of "compute" can't change this. |
|
| Back to top |
|
|
_______0
Guru
Joined: 15 Oct 2012
Posts: 521
|
| Posted: Sat Jan 05, 2013 3:50 pm Post subject: |
|
|
| Gusar wrote: |
| duby2291 wrote: | | It must be done with shaders or it isnt going to happen at all. |
Then it isn't going to happen.
|
why not? some linux games already using shaders and so on. |
|
| Back to top |
|
|
Etal
Veteran
Joined: 15 Jul 2005
Posts: 1932
|
| Posted: Sat Jan 05, 2013 4:23 pm Post subject: |
|
|
| _______0 wrote: | | Gusar wrote: |
| duby2291 wrote: | | It must be done with shaders or it isnt going to happen at all. |
Then it isn't going to happen.
|
why not? some linux games already using shaders and so on. |
Did you just skip the whole thread?
_________________
“And even in authoritarian countries, information networks are helping people discover new facts and making governments more accountable.”– Hillary Clinton, Jan. 21, 2010 |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Sat Jan 05, 2013 6:28 pm Post subject: |
|
|
| duby2291 wrote: | | I suppose if you refuse to do a bit of google searching I'll just have to post some links for you.... |
Now watch me debunk each and every one of them.
http://wiki.xbmc.org/index.php?title=Hardware_Accelerated_Video_Decoding_Development#Alternative_methods_of_hardware_accelerated_video_decoding <- Only theory there. But practice is what matters. And in practice, it was found that GPUs aren't suitable for video decoding.
http://forum.xbmc.org/showthread.php?tid=33802 <- Have you noticed the date on this one? Were there any results from this? Clearly not, otherwise we'd be using it.
http://emericdev.wordpress.com/about-the-project/ <- Didn't result in a working decoder either.
http://www.google.com/patents/US20070018979 <- Don't see how a patent application matters. Show me an implementation using it.
http://www.phoronix.com/scan.php?page=news_item&px=OTQ1MQ <- Again, have you noticed the date of the article? Christian König is the "AMD person" I mentioned above. He did what that article says, he wrote the vdpau state tracker and the mpeg2 decoder. But he is *not* working on an h264 decoder, he's writing UVD code.
| duby2291 wrote: | | It IS being done. It WILL happen.... Its just a matter of time. |
None of the projects you linked to above are active anymore. Corresponding precisely to what I said - everyone who tried, failed.
http://www.phoronix.com/scan.php?page=news_item&px=MTA3NzI
http://dri.freedesktop.org/wiki/R600ToDo#Compute
http://news.softpedia.com/news/AMD-Working-Hard-at-a-Compiler-That-Uses-Their-GPUs-for-Computing-261497.shtml
^^ General "compute" stuff. Relevant as such, but irrelevant for video decoding.
http://developer.amd.com/wordpress/media/2012/10/OpenVideo_Decode_API.PDF <- This is for accessing the dedicated decoder units, nothing to do with shader based decoders. It's even in the abstract itself: | Quote: | OpenVideo supports the GPU fixed-function hardware Unified Video Decoder (UVD), which
allows interoperability with OpenCL through a common API (OpenDecode API). OpenVideo
provides the way for all OpenCL-based video applications to access the fixed-function hardware
in GPUs. |
So it's about accessing UVD and using the video decoded by UVD in an OpenCL context, *not* for decoding the video itself in OpenCL. Nvidia has had something like this for a long time in CUDA, it's called nvcuvid.
| duby2291 wrote: | | And again the list goes on and on and on.... |
Yes, the list of unfinished/failed attempts.
| duby2291 wrote: | | As I said it IS being worked on. It WILL happen. |
Only if you live in denial. |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Sat Jan 05, 2013 10:02 pm Post subject: |
|
|
Clearly no amount of evidence is going to change your mind.... When it happens I'll be sure to post here though. If all else fails an OpenCL decoder will be written but first it needs to be well supported. Hell if MS can do it with HLSL, then it CAN obviously be done. It is being done right now. It will happen. The foundation work is simply hard and there isnt a large enough pool af skilled programmers that know how to do it. Once its done though it will work. It may not be as efficient as a special purpose hardware implementation, but it will get the job done and pull that load off the cpu.
Official stance on UVD docs.
http://www.phoronix.com/scan.php?page=news_item&px=MTA0MTY
http://www.phoronix.com/scan.php?page=news_item&px=MTAwNzg
Both are on Phoronix as that is bridgmans main place to post new information.
Last edited by duby2291 on Sat Jan 05, 2013 10:21 pm; edited 1 time in total |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Sat Jan 05, 2013 10:16 pm Post subject: |
|
|
| duby2291 wrote: | | Clearly no amount of evidence is going to change your mind |
Why are you attributing to me what is actually true for you?
| duby2291 wrote: | | If all else fails an OpenCL decoder will be written |
Because OpenCL is magic? There was an attempt at an OpenCL VP8 decoder. You can search the Phoronix forums for how that went. Hint: It didn't go well.
| duby2291 wrote: | | Hell if MS can do it with HLSL |
Wait, what? Where does MS have a HLSL decoder? They have a crappy software one and they have hooks into the dedicated hardware decoding units.
| duby2291 wrote: | | It is being done right now. |
The power of denial. It's amazing.
| duby2291 wrote: | | The foundation work is simply hard and there isnt a large enough pool af skilled programmers that know how to do it. |
There are skilled developers - Veerapan (worked on the OpenCL VP8 decoder); Christian König (worked on the gallium h264 decoder). They gave up.
Last edited by Gusar on Sat Jan 05, 2013 11:03 pm; edited 2 times in total |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Sat Jan 05, 2013 10:23 pm Post subject: |
|
|
| Gusar wrote: | | duby2291 wrote: | | Clearly no amount of evidence is going to change your mind |
Why are you attributing to me what is actually true for you?
| duby2291 wrote: | | If all else fails an OpenCL decoder will be written |
Because OpenCL is magic? There was an attempt at an OpenCL VP8 decoder. You can search the Phoronix forums for how that went. Hint: It didn't go well.
| duby2291 wrote: | | Hell if MS can do it with HLSL |
Wait, what? Where does MS have a HLSL decoder? They have a crappy software one and they have hooks into the dedicated hardware decoding units.
| duby2291 wrote: | | It is being done right now. |
The power of denial. It's amazing. |
Clearly you didnt read anything I posted... If you didnt read anything, how could you debunk anything? You didnt post any evidence.. All you did was post a link to a forum thread where -you- stated that it couldnt be done... That isnt evidence....
I on the other hand have posted multiple links to multiple places that state explicitly that it is being worked on. |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Sat Jan 05, 2013 10:29 pm Post subject: |
|
|
| duby2291 wrote: | | Clearly you didnt read anything I posted... If you didnt read anything, how could you debunk anything? |
I clicked on all the links. I commented on each of them. Please do the same, go through each of my comments and point out what is wrong with them.
| duby2291 wrote: | | You didn't post any evidence. |
http://phoronix.com/forums/showthread.php?73824-Bridgman-Is-No-Longer-quot-The-AMD-Open-Source-Guy-quot/page5&p=287982#post287982
| duby2291 wrote: | | I on the other hand have posted multiple links to multiple places that state explicitly that it is being worked on. |
No, you posted links to attempts that went nowhere. |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Sat Jan 05, 2013 10:46 pm Post subject: |
|
|
eh I give up.... That was work he was doing for an embedded version of windows...
Like I said when we get a functional decoder I'll be sure to post it here. |
|
| Back to top |
|
|
cord
Guru
Joined: 28 Apr 2007
Posts: 346
|
| Posted: Sat Jan 12, 2013 8:01 pm Post subject: |
|
|
Some cut from my /var/log/Xorg.0.log
| Code: | [ 11.479] (II) RADEON(0): [DRI2] Setup complete
[ 11.479] (II) RADEON(0): [DRI2] DRI driver: r600
[ 11.479] (II) RADEON(0): [DRI2] VDPAU driver: r600
|
what does it mean? |
|
| Back to top |
|
|
duby2291
Guru
Joined: 17 Oct 2004
Posts: 583
|
| Posted: Sat Jan 12, 2013 8:06 pm Post subject: |
|
|
| VDPAU is working right now. It's just that the only thing it supports is XvMC at the moment. say you use mplayer for your media player backend and you select the vdpau video output and then try to play an x264 video it won't work. But it will work on mpeg and mpeg2 video, things like VCD and DVD. |
|
| Back to top |
|
|
Gusar
Advocate
Joined: 09 Apr 2005
Posts: 2665
Location: Slovenia
|
| Posted: Sun Jan 13, 2013 12:56 pm Post subject: |
|
|
VDPAU supporting XvMC makes no sense. XvMC is a framework for hardware video decoding, just like VDPAU. You either use one or the other.
@cord: It means you have VDPAU, plain and simple. There's only a shader-based mpeg2 decoder, so it's practical use is limited. Flash benefits from it though, it'll be able to use VDPAU for hardware presentation (the "P" in VDPAU) instead of doing everything in software. |
|
| Back to top |
|
|
cord
Guru
Joined: 28 Apr 2007
Posts: 346
|
| Posted: Sun Jan 13, 2013 2:38 pm Post subject: |
|
|
| Yes, it's very sorry. Decoding DVD by VDPAU is needless, because it doesn't load CPU very hard any way. |
|
| Back to top |
|
|
|
Desktop Environments
