video playback quality


Does this means the Pyra wouldn't be a good candidate as a video player?

Sent from my SM-G900F using Tapatalk
The way it is now, the CPU alone can handle video playback for all the 1080p h264 encoded videos I've tested on the devboard. It loads the CPU heavily, but it's fully watchable as you can see in the prior video posted. I wouldn't recommend much multi-tasking when watching HD videos. This however can change for the better if we get a proper driver, but I believe if I recall that hinges on working 3D acceleration that still needs to get working with the latest linux kernels and possible may need a bunch of work on our end. Like ED said in the lilputing interview, there will be a ton of software that needs to be worked on once these units get to developers and eventually us end users.
 
The way it is now, the CPU alone can handle video playback for all the 1080p h264 encoded videos I've tested on the devboard. It loads the CPU heavily, but it's fully watchable as you can see in the prior video posted. I wouldn't recommend much multi-tasking when watching HD videos. This however can change for the better if we get a proper driver, but I believe if I recall that hinges on working 3D acceleration that still needs to get working with the latest linux kernels and possible may need a bunch of work on our end. Like ED said in the lilputing interview, there will be a ton of software that needs to be worked on once these units get to developers and eventually us end users.
Thanks for your reply and explanation.

Based on Android devices, even with same media player, i.e. MX Player, different CPU has different profile:
On Snapdragon 801 (Samsung G900F), hardware decoder or hardware assisted decoder uses fewer power, software decoder eat battery very fast.
But on Tiger K1 (Nvidia Shield Tablet), hardware decoder eat battery fast, pure software decoder uses very very few batter power.

Does these profile due to hardware or software? What is the Pyra's TI CPU profile on video decode?

Sent from my SM-G900F using Tapatalk
 
This however can change for the better if we get a proper driver
I wonder about that. Is the actually display such a substantial load? Can you do some simple benchmarks? Just run "time mplayer blah.avi" against "time mplayer blah.avi -dumpstream -dumpfile /dev/null"
edit: where blah.avi is some really short video, 10 seconds long or something, 1080p obviously. Just let it play out.
 
The decoding is the primary load, that's why you want that to happen in hardware (i.e. the GPU).
The decoding doesn't happen in the GPU though, that would still all be done in the CPU.
 
It doesn't support hardware accelerated decoding? That's odd.
Not really, it's generally not the GPUs duty to decode video, just display it. The OMAP has accelerators for doing the decoding separate from the GPU but those are as likely to get utilized on the Pyra as they were on the Pandora.
 
So GPU support to decode videos is very unlikely to happen atm. That's okay for me since even on desktop systems gpu-vide-decoding is not always enabled (I have very fond memories of trying to enable hardware decoding on my AMD E350 based netbook :D ). But at least there should be a hw-accelerated X-server. Would this ease the stress on the cpu?
 
So GPU support to decode videos is very unlikely to happen atm
Never. The GPU will never be able to decode videos. It doesn't have that capability at all. It displays 2D and renders 3D images, that's it's job. The hardware decoder is completely separate from the GPU, and if it's ever to be used someone will need to actually write a codec for it because TI certainly won't. In several years of having access to the Pandora's hardware it hasn't happened, I have very little faith that it'll happen on the Pyra either: writing a codec from scratch is HARD.
 
Never. The GPU will never be able to decode videos. It doesn't have that capability at all. It displays 2D and renders 3D images, that's it's job. The hardware decoder is completely separate from the GPU, and if it's ever to be used someone will need to actually write a codec for it because TI certainly won't. In several years of having access to the Pandora's hardware it hasn't happened, I have very little faith that it'll happen on the Pyra either: writing a codec from scratch is HARD.
Is it that kind of quest hundreds have started on and haven't been heard of since or has it not even been tried? Is there anything to show, yet?
 
Yes, exactly, but in the case of a niche like this it's probably more like 3 people, and one came back bloody and raving mad.
bsp did an amazing job of exposing the DSP interface, but he hasn't been heard from in years, and no one has really made any progress on actually utilizing the thing.
 
Notaz and M-HT have done further work on the DSP side.

edit: But I don't disagree with the assessment of the situation.
 
Video decoding can be done via shaders to some degree, Gallium already does that with MPEG2. Anything more complex than that is considered to be either way too complex or simply unfit for the common architecture of modern GPUs, though...
 
I just remember atc compk
I wonder about that. Is the actually display such a substantial load? Can you do some simple benchmarks? Just run "time mplayer blah.avi" against "time mplayer blah.avi -dumpstream -dumpfile /dev/null"
edit: where blah.avi is some really short video, 10 seconds long or something, 1080p obviously. Just let it play out.
Well current state of my devboard, I can't do much of anything.

The decoding doesn't happen in the GPU though, that would still all be done in the CPU.
I didn't mean to imply it did anything substantial on the GPU, I remember talking to aTc long ago and we we're made to believe for some reason the video decoding was dependent on the 3D accelerator driver to work properly. maybe to handle scaling and rotating? not sure.
 
You know, I've been thinking, how old was it when it broke? Was there no warranty on those things? It sucks that it broke. :(
It was just about 2 years old at that time and now it's 4 months later. Not entirely sure what killed it honestly, I need to do some serious hardware diagnostics. It could've been just a power surge/dip.
 
Video playback is partly dependent on the GPU. A 1080p video needs to be scaled down to 720p you could use OpenGL to at least do that (every normal player supports that).

Using SDL (which is not yet accelerated on the Pyra as notaz did with the Pandora) or X11, it would need to scale down additionally to decoding the video.

That's a lot of CPU power needed.
 
Back
Top