Open-Source Radeon UVD Video Support On Fedora

[bridgman]

Yep. Remember that we make "big APUs" (A series with 2-4 Stars or Piledriver cores) and "small APUs" (E series with 1-2 Bobcat cores), and the big ones have 2-3x the CPU and GPU power of the small ones.

UVD is more helpful on the small APUs, since they have relatively lower CPU power available for software decoding.

[curaga]

How bad are these things? My laptop is 5 years old. Core2 Duo CPU. Software decoding of a 720p clip uses roughly 30% of the CPU. ... Living with 30% CPU usage for a few more months seemed OK but if AMD APUs are really this bad, I can understand why everybody is so eager to get this to work.

The e-350: think Atom + 30%. Not even close to your core 2 duo (but something like 5% of the price and 30% of the wattage).

[Ibidem]

How bad are these things? My laptop is 5 years old. Core2 Duo CPU. Software decoding of a 720p clip uses roughly 30% of the CPU.
My original intention with my comment was that getting VDPAU to work seems too complicated that it would be worth the effort until the patch lands in Linux distributions. Living with 30% CPU usage for a few more months seemed OK but if AMD APUs are really this bad, I can understand why everybody is so eager to get this to work.
Too bad compiling this stuff will take ages on those things?

Just out of curiosity-are you sure it's CPU decoding, or do you have an odd setup?
A Core2 Duo would usually be integrated Intel gfx or Nvidia, which would mean vaapi support or vdpau.
But I don't know your hardware as well as you, I realize.

On a 1.6 GHz dual-core Neo with make -j1 (thermal reasons), I get ~15 minute compiles for mesa:

 ./configure --prefix=/opt/mesa9x --enable-texture-float --with-dri-driverdir=/opt/mesa9x/lib --without-dri-drivers --with-gallium-drivers=r600 --with-clang-libdir=/usr/lib/llvm-3.3/lib

This is after installing the LLVM 3.3 snapshot from the appropriate repo.
Kernel compiles I can't do all in one run, but it's less than a half-hour including time to cool down.
And libdrm is at most five minutes.
Due to the GPU version I'm stuck with the vdpau shader implementation at best.

On Ubuntu, you can currently add two PPAs (Oibaf or xorg-edgers, and mainline), install the new kernel and mesa, reboot, and that should be enough to get it working.
On Fedora or Debian it takes more work for lack of packages (does someone have OBS builds setup?).

[Ibidem]

It's possible, it doesn't over heat with fglrx though so I'm assuming no.

I thought the same at first. My laptop runs warm, but it is useable with radeon and the following tweaks:
-linux-phc undervolting
-profile low
-"conservative" cpu frequency governor (ondemand is the oven governor, it seems)
And I need to blow the fan out more often than with fglrx.
Also, a fan control daemon is necessary if I have sustained high loads (Lenovo screwed up the default tables...)

But on the other hand, I use the VTs so much that the text-gfx modesetting switch speed is more important than OpenGL.

[smitty3268]

Once again, it's a technical risk review of releasing the IP. Is any of the information related to patent applications that may be in flight. Is any of the information licensed from 3rd parties where we may need some else's permission of release the information. Would the information put AMD at risk for with respect to contractual agreements with other companies. Would the information put our ability to sell into the windows market at risk, etc.

That sounds like the very definition of a legal review. It doesn't matter what kind of college degree the reviewer has, that's what it is.

[Ericg]

How bad are these things? My laptop is 5 years old. Core2 Duo CPU. Software decoding of a 720p clip uses roughly 30% of the CPU.
My original intention with my comment was that getting VDPAU to work seems too complicated that it would be worth the effort until the patch lands in Linux distributions. Living with 30% CPU usage for a few more months seemed OK but if AMD APUs are really this bad, I can understand why everybody is so eager to get this to work.
Too bad compiling this stuff will take ages on those things?

Try 1080p and see what the usage is. No matter the usage though your laptop is burning through more power to display that movie which is an issue when youre on battery power (plane ride, car ride, train ride, etc) Also if you are trying to do OTHER things while playing the movie (Say youre multitasking and the movie is more background noise) thats less CPU time for everything else. Which could be a big deal if you are say.... compiling software and watching a movie to kill time? With CPU + UVD, the CPU can compile while UVD decodes, without UVD the tasks are fighting over CPU time. OR If you are trying to rip a movie, the UVD could PROBABLY (read: im not 100% sure on this scenario) help in the decode process and possibly help in the re-encode process, thus improving rip times.

[Ericg]

Legal, for the most part, is not that familiar with the hardware details and what risks they may pose. Once again, it's a technical risk review of releasing the IP. Is any of the information related to patent applications that may be in flight. Is any of the information licensed from 3rd parties where we may need some else's permission of release the information. Would the information put AMD at risk for with respect to contractual agreements with other companies. Would the information put our ability to sell into the windows market at risk, etc. Legal is involved to a certain extent, but it mostly comes down to software and hardware architects who are more familiar with the low level details of the hardware and software stacks across OSes and what risks they may pose. Unfortunately, those software and hardware architects are very busy and a complex technical risk review takes time.

Okay so its a legal review but not by legal, so i was half right lol

[agd5f]

Why would an engineer care if info is released publicly? That makes no sense. Even if it's an engineer doing the work, it's obviously a legal review.

Legal, for the most part, is not that familiar with the hardware details and what risks they may pose. Once again, it's a technical risk review of releasing the IP. Is any of the information related to patent applications that may be in flight. Is any of the information licensed from 3rd parties where we may need some else's permission of release the information. Would the information put AMD at risk for with respect to contractual agreements with other companies. Would the information put our ability to sell into the windows market at risk, etc. Legal is involved to a certain extent, but it mostly comes down to software and hardware architects who are more familiar with the low level details of the hardware and software stacks across OSes and what risks they may pose. Unfortunately, those software and hardware architects are very busy and a complex technical risk review takes time.

[smitty3268]

For the lebenty-millionth time, it's primarily a technical review not a legal review.

Why would an engineer care if info is released publicly? That makes no sense. Even if it's an engineer doing the work, it's obviously a legal review.

[bridgman]

If I remember right... the code's been written, the documentation has been sanitized, its just sitting on legal's desk

For the lebenty-millionth time, it's primarily a technical review not a legal review.