Haswell Video Acceleration Code Published

[phoronix]

Phoronix: Haswell Video Acceleration Code Published

Just minutes after writing about how Intel keeps releasing open-source Linux code for Haswell, their next-generation hardware for 2013, they ended up pushing out their initial video acceleration (VA-API) support code...

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

[89c51]

Will this include webm enc/dec. And BTW how come webm hasn't gained traction in the HW VA field.

[przemoli]

What is rationale behind VA API? Why can not we have VDPAU and be done with all other APIs? Any legal briers? Any functionality that would not be exposed? Conversion would be too hard with no benefits (apart from ONE(1) api that could be targeted as opposed to bazzillion now )

[blackiwid]

What is rationale behind VA API? Why can not we have VDPAU and be done with all other APIs? Any legal briers? Any functionality that would not be exposed? Conversion would be too hard with no benefits (apart from ONE(1) api that could be targeted as opposed to bazzillion now )

hmm va-api is the best api if you look it up... so why would not nvidia the smallest gpu-vendor from the 3 big switch to the same the biggest (intel) does use... so that we have not a bazzillion now.


I googled the two apis just a bit, it seems that vaapi is like a wrapper, it uses whatever your driver others and gives a api to the programs. so if your card can do VDPAU it can be accessed its features (or a big subset of it) through va-api. I think the api is better for player-developers, thats the reason why they use that. I guess, then I assume that vdpau is more lowlevel api, if not I dont know really why use this wrapper.


I think the biggest point against vdpau is that itself is maybe opensource yes, but nvidia forgot to implement a patchset for a free driver, as example the nouvou driver, or in earlier times in the nv driver, so because there was never a example opnesource implementation I guess doku is on such lowlevel stuff also not so good, so if you then have vaapi who has a implementation on intel gpus, which api wins is clear. And as a intel or amd developer I would not want to contact a nvidia developer to ask him anything. technicaly both reduces 95%? of the cpu load, so technicaly it doesnt matter much which one you use I guess, but on the one there is a example free implementation which is a docu and you dont need to ask a intel dev or somewhen else just look into the source. on the other site you have to ask a nvidia developer, which are people in a company that is a enemy of freedom stuff. So I think its clear why nobody uses VDPAU!

[Gusar]

Err, blackiwid... Gallium has VDPAU, but not VAAPI. Kinda negates what you're saying.

Also, what's the state of advanced deinterlacing in vaapi? From what I know it's only in a branch. No idea about the state of that branch, and when the main vaapi will get this stuff. Not that it'll help me, because VAAPI can't do deinterlacing of mpeg2 video on Ironlake. (before anyone asks why I want it - it's for watching live sports)

@przemoli: Well, vaapi began before there was vdpau. But for a long time there was nothing concrete, just some docs and the beginnings of a library, nothing end-users could actually use. Then, out of the blue came Nvidia with vdpau. And because they did it right - header files, full docs, and sample patches for ffmpeg/mplayer, it quickly got widespread support in various players. But eventually vaapi reached a point where it started doing something concrete. So now we have both.

[blackiwid]

Err, blackiwid... Gallium has VDPAU, but not VAAPI. Kinda negates what you're saying.

Also, what's the state of advanced deinterlacing in vaapi? From what I know it's only in a branch. No idea about the state of that branch, and when the main vaapi will get this stuff. Not that it'll help me, because VAAPI can't do deinterlacing of mpeg2 video on Ironlake. (before anyone asks why I want it - it's for watching live sports)

@przemoli: Well, vaapi began before there was vdpau. But for a long time there was nothing concrete, just some docs and the beginnings of a library, nothing end-users could actually use. Then, out of the blue came Nvidia with vdpau. And because they did it right - header files, full docs, and sample patches for ffmpeg/mplayer, it quickly got widespread support in various players. But eventually vaapi reached a point where it started doing something concrete. So now we have both.

so ok you say that I can watch with a videoplayer and my amd grafics card and the gallium driver with vdpau videos? or with nouvou? I dont think so, so there is praktical no support in gallium, a theoretical libery thats just there for theoretical aspects dont bring anyone anything.

[Gusar]

There is nothing "theoretical" about Gallium's VPDAU support. It implements the presentation part, and decoding of mpeg2. It works.

Your argument was "nobody uses VDPAU". This argument is wrong. That's all there is to it.

[blackiwid]

There is nothing "theoretical" about Gallium's VPDAU support. It implements the presentation part, and decoding of mpeg2. It works.

Your argument was "nobody uses VDPAU". This argument is wrong. That's all there is to it.

thats maybe for some very very special cases today still relevant for the majority of anything its just not, mpeg2 just istnt useful for 99.99999999% of the people so its basicly useless showcase... and it seams to make a new codec availbible is more work than defining on two or 100 such stupid apis...

[nachokb]

is there any application for encoding using vaapi?

nice, but it seems the only code using VAAPI's encoding features are just some Intel demos... there's no support in ffmpeg, gstreamer or anything (yet) it seems... anyone know how we can use it??