Announcement

Collapse
No announcement yet.

The ACO Radeon Compiler Alternative To AMDGPU LLVM Looks Good But Work Isn't Done Yet

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

  • The ACO Radeon Compiler Alternative To AMDGPU LLVM Looks Good But Work Isn't Done Yet

    Phoronix: The ACO Radeon Compiler Alternative To AMDGPU LLVM Looks Good But Work Isn't Done Yet

    In addition to Intel announcing their work on the new "IBC" compiler back-end for their OpenGL/Vulkan drivers, the developers working on the Radeon "ACO" in cooperation with Valve were presenting the latest work on their compiler back-end at this week's XDC 2019 event in Canada...

    http://www.phoronix.com/scan.php?pag...ACO-RADV-State

  • #2
    Outside of those odd hangs I was getting last week that seem to not be effecting me as of 0103d4747a3, I've had great results using ACO from Daniel and currently now with 19.3 using RADV_PERFTEST=aco.

    marek has also done quite a bit of NIR related commits which may or may not warrant some AMD_DEBUG=nir tests.

    Comment


    • #3
      I use ACO since it was announced and I must say it works really well :-)

      Comment


      • #4
        Can this be used for OpenCL and all GPGPU tasks too?

        Comment


        • #5
          Originally posted by timofonic View Post
          Can this be used for OpenCL and all GPGPU tasks too?
          Potentially, any driver that generates NIR could use it. That'd be quite a ways away, though, and it's doubtful Valve cares much about OpenCL.

          Comment


          • #6
            Originally posted by smitty3268 View Post

            Potentially, any driver that generates NIR could use it. That'd be quite a ways away, though, and it's doubtful Valve cares much about OpenCL.
            Too bad, OpenCL/Computing Vulkan (aka OpenCL merged into Vulkan, my wet dream) can be amazing for really ambitious and complex games.

            Comment


            • #7
              I just read the presentation, well the pages I understood...
              Page 34 we see that NIR takes more time than ACO in compile time. Is this something that will be also be worked on or is it about as it should be?

              Comment

              Working...
              X