Announcement

Collapse
No announcement yet.

RADV Reworking Pipeline Emitting To Improve CPU Usage

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

  • RADV Reworking Pipeline Emitting To Improve CPU Usage

    Phoronix: RADV Reworking Pipeline Emitting To Improve CPU Usage

    While using Vulkan lowers the CPU utilization compared to OpenGL, in our testing of NVIDIA versus the open-source Radeon drivers we generally have found the red team's drivers to consume more CPU resources. Thus it's good to hear that RADV co-conspirator Bas Nieuwenhuizen is working on reworking how this Radeon Vulkan driver handles pipeline emitting...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    "with the healthy competition thanks to the newly-opened AMDVLK driver."
    open amdvlk actually means less competition - it allows just copying its code to radv

    Comment


    • #3
      Originally posted by pal666 View Post
      "with the healthy competition thanks to the newly-opened AMDVLK driver."
      open amdvlk actually means less competition - it allows just copying its code to radv
      Which is arguably the best kind of competitor :P

      Comment


      • #4
        Originally posted by pal666 View Post
        "with the healthy competition thanks to the newly-opened AMDVLK driver."
        open amdvlk actually means less competition - it allows just copying its code to radv
        Also the other way, copying from radv to amdvlk.

        Comment


        • #5
          Originally posted by pal666 View Post
          "with the healthy competition thanks to the newly-opened AMDVLK driver."
          open amdvlk actually means less competition - it allows just copying its code to radv
          Why would we want to bloat radv like that?

          Dave.

          Comment


          • #6
            RAVD is based on Intel's ANV. Will Intel GPUs be able to benefit from this optimization or is it Radeon specific?

            Comment


            • #7
              Originally posted by slacka View Post
              RAVD is based on Intel's ANV. Will Intel GPUs be able to benefit from this optimization or is it Radeon specific?
              It is just some code restructuring in radv in non-shared code, so anv will not benefit. The principle behind it might or might not benefit anv, would need to check if they do it already.

              Comment


              • #8
                Originally posted by airlied View Post

                Why would we want to bloat radv like that?

                Dave.
                Most of us have no idea what you mean. Is amdvlk bloated? And if not, if amdvlk has some good code that could potentially be used in radv would you use it? (Or is it more a coding style thing?)

                Comment


                • #9
                  Originally posted by duby229 View Post

                  Most of us have no idea what you mean. Is amdvlk bloated? And if not, if amdvlk has some good code that could potentially be used in radv would you use it? (Or is it more a coding style thing?)
                  You can't just copy the code between drivers, you can use the other driver for ideas but the codebases are in no way compatible, you still have to rewrite code, e.g. amdvlk has the xgl shim on top of pal, pal defines all its own API, radv has none of that separation.

                  They definitely have a lot larger code base doing mostly the same things, so bloat definitely exists, it may exist for reasons of course that only affect amdvlk.

                  Dave.

                  Comment


                  • #10
                    Originally posted by airlied View Post
                    They definitely have a lot larger code base doing mostly the same things, so bloat definitely exists, it may exist for reasons of course that only affect amdvlk. Dave.
                    It's not there so much for AMDVLK itself - it's there in order to let one set of core code support a number of different OS/API combinations.
                    Test signature

                    Comment

                    Working...
                    X