DRI2 Video Support Still Being Planned

[phoronix]

Phoronix: DRI2 Video Support Still Being Planned

Rob Clark of Texas Instruments talked more at XDC2012 about his plans for "DRI2 Video" support...

http://www.phoronix.com/vr.php?view=MTE5MTM

[liam]

Phoronix: DRI2 Video Support Still Being Planned

Rob Clark of Texas Instruments talked more at XDC2012 about his plans for "DRI2 Video" support...

http://www.phoronix.com/vr.php?view=MTE5MTM

So, the big win is tear-free playback? Don't we already have that when we use a GL surface as output?

[curaga]

I understand the main benefit is zero-copy. Tear-free was already possible a decade ago using the video overlay.

[brent]

Another problem with Linux is the lack of a modern standard API for video presentation. Xv is old, and it shows in many ways. We have VDPAU, but that is not supported on all chipsets.

[liam]

I understand the main benefit is zero-copy. Tear-free was already possible a decade ago using the video overlay.

Someone needs to tell my desktop that.
So, the use of swapbuffers implies zero copy? I thought it was simply about switchng out the front buffer. If you are running a conpositor it seems like you would still need to copy that swapped buffer.

[blackiwid]

does that reduce cpu usage in video playback... yesterday I tried because of that wayland in quantal but here (radeon) no pointer did appear

[daniels]

Yep, it reduces CPU usage and latency pretty greatly, by eliminating a couple of huge stalls. In the best case, you can have the overlay scan directly out of the buffer the DSP decoded into, which saves you two copies (in and out of the SHM area). This helps hugely with 1080p especially.

Using overlays rather than GL still helps you a lot because the overlays typically have much higher-quality scaling, filtering and colourspace conversion. They're pretty crucial in mobile because the you expend a great deal more power using the GPU for that.