Mesa 9.0 Officially Released, Supports OpenGL 3.1

[aceman]

If you have a capable driver, MPlayer's OpenGL output is similarly efficient as Xv or VDPAU. That is, the overhead is negligible.

So what about mplayer -vo vdpau. Can it be used anywhere where -vo xv was used? If there is not the proper format support in the specific GPU driver for vdpau (e.g. radeon only has mpeg1/2) will the video be decoded in SW? But will the presentation features of VDPAU still be used?

[curaga]

If you have a capable driver, MPlayer's OpenGL output is similarly efficient as Xv or VDPAU. That is, the overhead is negligible.

Radeon FOSS. XV output is at 0.5% cpu, GL output is 35%. Hardly negligible, and the difference was much greater on older cards (this is a HD4k).

[Gusar]

So what about mplayer -vo vdpau. Can it be used anywhere where -vo xv was used? If there is not the proper format support in the specific GPU driver for vdpau (e.g. radeon only has mpeg1/2) will the video be decoded in SW? But will the presentation features of VDPAU still be used?

Yep, you can use -vo vdpau with any video, even software decoded ones.

Does Gallium have a high-quality scaler? With Nvidia, you can use -vo vdpau:hqscaling=1 for a much better picture than the default bilinear scaler.

Radeon FOSS. XV output is at 0.5% cpu, GL output is 35%. Hardly negligible, and the difference was much greater on older cards (this is a HD4k).

Then you're doing something wrong. Or radeon is crap
On Nvidia and Intel the difference is minimal. Run mplayer with -v -vo gl, see which options for gl were chosen. Play around with different ones, especially the ati-hack and the force-pbo ones.

[curaga]

force-pbo: no difference.
force-pbo + ati-hack: no difference.

It's simply impossible for a heavier path to be equal to a faster path, so I guess we're arguing about "how much slower is gl". The answer for radeon, a lot, far more than I'd accept for better-quality fonts.

[marek]

force-pbo: no difference.

AFAIK, PBOs are not accelerated. Internally, we might be able accelerate PBO copies with texture buffer objects.

[curaga]

Why does everyone assume I want to get GL output ? I want XV-level cpu usage, preferably with improved subtitle resolution.

Perhaps vdpau can provide that. If not, I'll continue to use XV.

[Gusar]

It's simply impossible for a heavier path to be equal to a faster path

How do you explain Nvidia and Intel then?

Besides, Xv _also_ uses the 3D engine (textured video) on modern graphic cards, so the only difference between xv and gl is how to access the graphic card. GL probably has some overhead here, but on Nvidia and Intel it's negligible.

so I guess we're arguing about "how much slower is gl". The answer for radeon, a lot,

Well, then, I guess the second thing I said is valid: Radeon sucks :P

[curaga]

Since I don't use a compositor, it's possible XV uses the video overlay. Which is absolutely the best option for video, all of the displaying steps are done by hardware.

[curaga]

*by dedicated hardware, instead of a shader converting from yuv to rgb.

[Gusar]

Since I don't use a compositor, it's possible XV uses the video overlay.

Your HD4K doesn't have a video overlay.