Announcement

Collapse
No announcement yet.

Radeon DRM-Next Work: GPUVM Optimizations, Deep Color HDMI

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

  • #11
    Originally posted by zanny View Post
    https://en.wikipedia.org/wiki/Displayport

    It is speced for 24 bit 8 channel audio.

    I know at least the Intel DRM driver supports it. Not sure about the others. They show up as HDMI Audio under Pulseaudio, though.
    From your wikipedia link; "DisplayPort can be used to transmit audio and video simultaneously, but each one is optional and can be transmitted without the other."
    So it isn't a required spec for displayport that it carries audio. Its just an option.

    Comment


    • #12
      So it isn't a required spec for displayport that it carries audio. Its just an option.
      Of course? You can do HDMI without the audio part too. The point is that connectors (at least on the sender side) and cables will need to support the audio lanes, even if you don't use it.

      Comment


      • #13
        Originally posted by droidhacker View Post
        From your wikipedia link; "DisplayPort can be used to transmit audio and video simultaneously, but each one is optional and can be transmitted without the other."
        So it isn't a required spec for displayport that it carries audio. Its just an option.
        Transmitting is optional. Being able to transmit is not, it's part of the spec.

        Comment


        • #14
          Someone explain what this means?


          Comment


          • #15
            Originally posted by Anfdrea View Post
            Someone explain what this means?
            It means you're a spambot and that your posts need to be deleted...
            Test signature

            Comment


            • #16
              Originally posted by zanny View Post
              Of course? You can do HDMI without the audio part too. The point is that connectors (at least on the sender side) and cables will need to support the audio lanes, even if you don't use it.
              There are no audio lanes, audio is embedded in SDP (secondary data packets), transmitted during horizontal and vertical blank intervals using the main lanes.

              Comment


              • #17
                Originally posted by smitty3268 View Post
                Transmitting is optional. Being able to transmit is not, it's part of the spec.
                A source may be able to transmit audio, but it is not required. If it does send audio, it must at least support LPCM 2 channel 16bit and one of 32/44.1/48kHz. Before sending, it must query the sink for its capabilities.

                A sink may be able to receive audio, but it is not required. If it is able to, LPCM 2c 16b with all of 32/44.1/48kHz is required, others are optional. Supported formats must be queryable by e.g. EDID.

                Support for video is required, at least 640x480 (this may have changed with DP 1.2 and MST).

                Blank video, i.e. padding only, may be transmitted.

                DP spec 1.1a is still available, unfortunately, DP spec 1.2 is not.

                Comment


                • #18
                  Originally posted by agd5f View Post
                  It means we copy the PTEs from the gart table to the GPUVM tables if they are the same rather than populating them directly.
                  What is the usage different between GART and GPUVM?
                  As I knew that GART is used to translate AGP memory, how about GPUVM?

                  Comment


                  • #19
                    Originally posted by gic4107 View Post
                    What is the usage different between GART and GPUVM?
                    As I knew that GART is used to translate AGP memory, how about GPUVM?
                    Nothing really. GART = Graphics Address Remapping Table. GPUVM = GPU Virtual Memory. It's just a mechanism to make a set of discontiguous pages in system memory look like a contiguous linear aperture to the GPU. AGP implemented it on the northbridge chipset side and the driver just pointed an aperture at the AGP aperture. As AGP went away and in order to support the same functionality on PCI boards, the GART mechanism moved onto the GPU. In this case the GPU maintains the GART page table. GPUVM just takes the concept a step further and allows each graphics client to have their own GPU virtual address space. On radeons at least, the page tables have the same format either way. The kernel driver uses a special context that is sort of akin to physical GPU addresses. I.e., the physical GPU address space has several apertures, one that maps to vram, one that maps to AGP, and one that uses a page table to translate (provides access to system pages, but could be used for vram too). GPUVM GPU address spaces have just one aperture that is translated by a page table. Each PTE can point to a system page or a vram page.

                    Comment

                    Working...
                    X