Announcement

Collapse
No announcement yet.

Radeon Driver Gets Reworked Textured Video

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Radeon Driver Gets Reworked Textured Video

    Phoronix: Radeon Driver Gets Reworked Textured Video

    AMD's Alex Deucher has delivered a new Textured Video implementation in the xf86-video-ati driver. While this video acceleration is not ideal as say VDPAU, VA-API, or even just XvMC it will certainly be appreciated by ATI customers. The reworked TexturedVideo implementation is in the Git master code and now has native support for planar formats and X-Video attributes for brightness, contrast, hue, saturation, gamma, and color-space are implemented within the driver...

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

  • #2
    Thank you! Works like a charm on my mobility X2300.

    Comment


    • #3
      While this video acceleration is not ideal as say VDPAU, VA-API, or even just XvMC it will certainly be appreciated by ATI customers.
      I don't understand this part about XvMC. Isn't XvMC only for MPEG2?

      I tend to remember discussions about XvMC, where arguments were that there was very little motivation to implement it, as it only accelerated MPEG-2 and all modern CPU's can easily decode MPEG-2.

      ?

      Comment


      • #4
        Yeah, although I think there's a VIA driver out there with a full slice-level H.264 decode command added to the XvMC API.

        We're thinking about doing an XvMC implementation (which would be MPEG-2 as you say) to give developers a starting point for other codecs. The details are different from one codec to the next but the basic operations are all pretty much the same whether you're talking about MPEG2 or H.264.

        Comment


        • #5
          Originally posted by bridgman View Post
          Yeah, although I think there's a VIA driver out there with a full slice-level H.264 decode command added to the XvMC API.

          We're thinking about doing an XvMC implementation (which would be MPEG-2 as you say) to give developers a starting point for other codecs. The details are different from one codec to the next but the basic operations are all pretty much the same whether you're talking about MPEG2 or H.264.
          Okay, I thought XvMC was a "packet solution", and not as an acceleration of each of the functions. Fourier transformations and such.

          Could it even be extended to accelerate wavelets, which is the basics for Snow codec?

          Btw. I just want to say thank you SOOOOOO MUCH to all of AMD for making this happen!

          It is a tremendous testimony to Linux and the open source environment, that you have leaded the way that graphics drivers don't have to be a secret.

          Comment


          • #6
            It gets confusing. XvMC is normally a "bunch of blocks" API, originally designed primarily for motion compensation. The normal XvMC API deals with partially-decoded 8x8 blocks, either starting at IDCT or at MC. The Via extension for slice level decoding is the only VLD-level API.

            I don't know much about wavelet-based codecs but in general anything which involves running a filter kernel over an area of pixels is a pretty good match for a GPU.

            Comment


            • #7
              Originally posted by Louise View Post
              Btw. I just want to say thank you SOOOOOO MUCH to all of AMD for making this happen!

              It is a tremendous testimony to Linux and the open source environment, that you have leaded the way that graphics drivers don't have to be a secret.
              Yeah, I was kinda surprised it wasn't mentioned in the Phoronix update, but maybe it's just yet not an important enough step. :P

              Comment


              • #8
                We weren't able to give Michael much warning... the review and cleanup dragged on for a long time, then all of a sudden it was done

                Comment


                • #9
                  Originally posted by bridgman View Post
                  It gets confusing. XvMC is normally a "bunch of blocks" API, originally designed primarily for motion compensation. The normal XvMC API deals with partially-decoded 8x8 blocks, either starting at IDCT or at MC. The Via extension for slice level decoding is the only VLD-level API.

                  I don't know much about wavelet-based codecs but in general anything which involves running a filter kernel over an area of pixels is a pretty good match for a GPU.
                  Me either. But I know how to use wavelets to predict earth quakes Not a joke.

                  I guess time will tell, what people will end up using XvMC for

                  Comment


                  • #10
                    Originally posted by bridgman View Post
                    We weren't able to give Michael much warning... the review and cleanup dragged on for a long time, then all of a sudden it was done
                    Would it be too much bias towards the world to let you handle some of the AMD-related news updates on Phoronix? XD

                    Comment

                    Working...
                    X