Announcement

Collapse
No announcement yet.

Radeon "sisched" Scheduler Is Made Obsolete By RADV's ACO Back-End

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

  • Radeon "sisched" Scheduler Is Made Obsolete By RADV's ACO Back-End

    Phoronix: Radeon "sisched" Scheduler Is Made Obsolete By RADV's ACO Back-End

    It's been years since last hearing anything about sisched as the SI machine instruction scheduler that started out for the RadeonSI OpenGL driver and was ultimately supported by the RADV Vulkan driver too...

    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
    Originally posted by phoronix View Post
    SISCHED was never enabled by default but was selectively white-listed or with the RADV_PERFTEST=sisched environment variable.
    Think that environment variable is a typo, I believe it's AMD_DEBUG=sisched for sisched scheduler, and then RADV_PERFTEST=aco to enable ACO or RADV_PERFTEST=llvm to enable llvm and disable ACO.

    Comment


    • #3
      Originally posted by perpetually high View Post

      Think that environment variable is a typo, I believe it's AMD_DEBUG=sisched for sisched scheduler
      Isn't that for radeonsi and not radv?

      Comment


      • #4
        So, will ACO ever be enabled by default in the future?

        Comment


        • #5
          Originally posted by geearf View Post
          Isn't that for radeonsi and not radv?
          https://cgit.freedesktop.org/mesa/me...c555d88e47708f
          Ah, my apologies, thanks for the correction.

          Comment


          • #6
            It would be nice to see ACO enabled by default now. It seems really stable besides offering better performance.

            Comment


            • #7
              As the author of sisched I have to agree with the statement it is unmaintained.

              My mistake was probably to decline requesting llvm push accesses soon after sisched was merged.
              Ever since the merge it has been an uphill battle to get the slightest patch merged. My patches took months, years or even infinite time to get reviewed and pushed.
              It's been two years now i gave up and just wish it was dropped.

              Comment


              • #8
                Originally posted by mannerov View Post
                As the author of sisched I have to agree with the statement it is unmaintained.

                My mistake was probably to decline requesting llvm push accesses soon after sisched was merged.
                Ever since the merge it has been an uphill battle to get the slightest patch merged. My patches took months, years or even infinite time to get reviewed and pushed.
                It's been two years now i gave up and just wish it was dropped.
                Thank you for all of your work. I find cool meeting here on Phoronix the devs of the stuff I use (or have used).

                Originally posted by TemplarGR
                So, will ACO ever be enabled by default in the future?
                Is there a reason why they don't merge all the open source drivers available for AMD?

                Comment


                • #9
                  Originally posted by TemplarGR View Post
                  So, will ACO ever be enabled by default in the future?
                  I'm guessing not before it supports all that the llvm backend supports (at least gaming-wise).
                  There are still some breaking issues with ACO so it would not make sense to default to that.

                  Originally posted by Buntolo View Post
                  Is there a reason why they don't merge all the open source drivers available for AMD?
                  How would you suppose they do that?
                  For instance radeonsi is gallium3d-based, radv isn't.

                  Comment


                  • #10
                    Originally posted by geearf View Post
                    There are still some breaking issues with ACO so it would not make sense to default to that.
                    What issues are you talking about? Not saying ACO is free of bugs but we fixed a bunch of issues recently. I would be glad to fix more.

                    Comment

                    Working...
                    X