Announcement

Collapse
No announcement yet.

SDL3 Will Keep Wayland Default At Least For The Time Being

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

  • #81
    Originally posted by mSparks View Post
    WTF are you talking about
    I am talking about the approach used by Waypipe to tunnel Wayland and its similarities in how SSH tunnels X11.

    In both cases the neither application nor display server are aware of them not talking to each other directly but via proxy.
    In both cases the tunnel end points work with a local unix domain socket: on the application side the tunnel pretends to be the display server by creating the unix domain socket. On the display server side the tunnel pretends to be the application by connecting to the display server's socket.

    All messages that do not have associated file descriptors can be passed through either tunnel unchanged.
    The main difference of the two tunnels is how the deal with messages that do have those.

    The X11 tunnel filters all extensions that would require such messages, letting the application deal with falling back to data serialization.
    The Wayland tunnel accepts the file descriptor, serializes the data across the network, and recreates a compatible file descriptor on the other side.

    Originally posted by mSparks View Post
    the waypipe dev VERY CLEARLY states that wayland does most of its heavy lifting with DMABUF.
    It does. Just as it does on a Linux based X11 system with clients who do direct rendering.
    It is, after all, the system's interface for such render buffers.

    Originally posted by mSparks View Post
    DMABUF (few GB/s) is far slower than the local TCP stack (several TB/s)
    The point regarding speed is the avoidance of any need for data transfer.

    When the application and display server communicate via a unix domain socket, they can pass handles to buffers instead of the data inside the buffers.
    These handles are unix file descriptors, so just 32-bit values, i.e. 4 bytes.
    A 4k image with 4 color channels of 8 bit each has around 24 MBs.

    It is much faster to transmit 4 bytes handles than 6 million times as much image data.

    An optimization X11 has employed for decades via various extensions and which became a core concept of the Wayland specification.

    Comment


    • #82
      Originally posted by mrg666 View Post
      I did not say Wayland is faster than X11, who said that?
      Nobody on this thread did.

      However, it seems that mSparks might have misinterpreted our conversation on file descriptor passing and how it is used by both X11 and Wayland to avoid transmission of buffer data by communicating handles to that data instead.

      Already invaluable for buffers allocated in main memory but essential for buffers allocated in GPU memory.
      Transferring image data from one process to another via socket is already bad enough, but first loading the image data from the GPU to then transfer it to the other process which then has to re-upload it to the GPU would be much worse.

      I've written a clarification in a separate posting.

      Comment


      • #83
        Originally posted by anda_skoa View Post

        It does. Just as it does on a Linux based X11 system with clients who do direct rendering.

        DMABUF access is linux kernel syscalls, not only does no other OS support it, most linux distributions don't either



        there is nothing traditionally protocol about that at all, not remotely similar to anything X11 specifies, wayland defining DMABUFF like defining HTTP as only using 802.11b

        Comment


        • #84
          Originally posted by mSparks View Post
          DMABUF access is linux kernel syscalls, not only does no other OS support it, most linux distributions don't either



          there is nothing traditionally protocol about that at all, not remotely similar to anything X11 specifies, wayland defining DMABUFF like defining HTTP as only using 802.11b
          Yeah keep posting. When monkeys type Romeo and Juliet, you will make sense, I guess.

          Comment


          • #85
            Originally posted by mrg666 View Post

            Yeah keep posting. When monkeys type Romeo and Juliet, you will make sense, I guess.
            Im not the one confused by waylands epic failure over a decade and a half.

            the extra bit of time SDL has just given it wont last much longer.

            Comment


            • #86
              Originally posted by mSparks View Post
              DMABUF access is linux kernel syscalls
              I did write Linux based X11 system, didn't I?

              Originally posted by mSparks View Post
              not only does no other OS support it
              Other system have their own abstraction for buffers accessed over DMA such as GPU textures.

              Originally posted by mSparks View Post
              there is nothing traditionally protocol about that at all
              No, but the Wayland protocol is.

              Originally posted by mSparks View Post
              not remotely similar to anything X11 specifies
              Yes, it does.
              That's the DRI extension, more specifically DRI3

              From the Wikipedia article on DRI
              The main differences of DRI3 compared to DRI2 are:
              • DRI3 clients allocate themselves their render buffers instead of relying on the X Server for doing the allocation —that was the method supported by DRI2.
              • DRI3 gets rid of the old insecure GEM buffer sharing mechanism based on GEM names (global GEM handles) for passing buffer objects between a DRI client and the X Server in favor of the one more secure and versatile based on PRIME DMA-BUFs, which uses file descriptors instead

              You are trying to find difference between the two systems but looking at the facilities and mechanism they actually share.

              In both cases (X11 and Wayland) the buffer which resulted from a client's hardware accelerated rendering is transferred to the respective display server by passing a handle (file descriptor) of a dma_buf instead of the pixel data.

              In each case the respective protocol has defined a message that the file descriptor is sent along with:Both have also similar means to transfer handles of shared memory buffers for clients which render on the CPU.
              On X11 that is handled via the XShm protocol extension, on Wayland via the wl_shm core protocol.

              As I tried to explain before, the difference is not what or how the two system are doing this, but in X11's case it was added as an improvement while Wayland, being the "younger" system, could already deploy it as part of the core protocols.

              Comment


              • #87
                Originally posted by mSparks View Post

                Im not the one confused by waylands epic failure over a decade and a half.

                the extra bit of time SDL has just given it wont last much longer.
                Yeah keep throwing your seemingly endless supply of BS it will stick some day ... or not since Gnome, Plasma, Red Hat, Ubuntu ... are all default Wayland now. You need to work harder.

                Comment


                • #88
                  Originally posted by mrg666 View Post

                  Yeah keep throwing your seemingly endless supply of BS it will stick some day ... or not since Gnome, Plasma, Red Hat, Ubuntu ... are all default Wayland now. You need to work harder.
                  Im presume you are happy on a default salary to.
                  aka minimum wage..
                  rofl.

                  Comment


                  • #89
                    Originally posted by mSparks View Post

                    Im presume you are happy on a default salary to.
                    aka minimum wage..
                    rofl.
                    Try to focus, even that stupid glxgears benchmark made more sense.

                    Comment


                    • #90
                      Originally posted by mrg666 View Post

                      Try to keep frocused, even that stupid glxgears benchmark made more sense.
                      Ill take that as a yes.

                      Comment

                      Working...
                      X