Announcement

Collapse
No announcement yet.

RadeonSI Completes ACO Compiler Support With Mesa 24.0

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

  • #11
    Please add the latest Radv+Zink too in the latest RadeonSi benchmark.

    Comment


    • #12
      Originally posted by Teggs View Post

      Still it will be interesting to have the option. I have three games still using OpenGL, and one of them is easy to load the GPU at 100%. I intend to test these when 24.0 is available and see what happens in amdgpu_top.
      What games?

      Comment


      • #13
        Michael (Qiang Yu and karolherbst )

        '...I'll be running some RadeonSI ACO comparison benchmarks soon on Phoronix.'

        NOT needed, yet!

        It is NOT complete, 'only' 'part mode merged shader' as far as I know.

        vdpau (mplayer and mpv) do NOT work.

        mplayer -vo vdpau /data/Filme/test.mkv
        mpv --hwdec=vdpau -vo=vdpau /data/Filme/test.mkv
        ​Only blueish windows.

        OpenCL (RustiCL)
        is not complete, too.

        clinfo | less
        [-]
        Platform Name rusticl
        Number of devices 1
        Device Name AMD Radeon RX 580 Series (radeonsi, polaris10, ACO
        lines 1-20ACO ERROR:
        In file ../src/amd/compiler/aco_instruction_selection.cpp:9233
        Unimplemented intrinsic instr: div 32 %27 = @load_global (%26) (access=none, align_mul=4, align_offset=0)
        Abbruch (core dumped)

        Testing it from time to time...;-)

        Comment


        • #14
          Originally posted by nuetzel View Post
          Michael (Qiang Yu and karolherbst )

          '...I'll be running some RadeonSI ACO comparison benchmarks soon on Phoronix.'

          NOT needed, yet!

          It is NOT complete, 'only' 'part mode merged shader' as far as I know.

          vdpau (mplayer and mpv) do NOT work.

          mplayer -vo vdpau /data/Filme/test.mkv
          mpv --hwdec=vdpau -vo=vdpau /data/Filme/test.mkv
          ​Only blueish windows.

          OpenCL (RustiCL)
          is not complete, too.

          clinfo | less
          [-]
          Platform Name rusticl
          Number of devices 1
          Device Name AMD Radeon RX 580 Series (radeonsi, polaris10, ACO
          lines 1-20ACO ERROR:
          In file ../src/amd/compiler/aco_instruction_selection.cpp:9233
          Unimplemented intrinsic instr: div 32 %27 = @load_global (%26) (access=none, align_mul=4, align_offset=0)
          Abbruch (core dumped)

          Testing it from time to time...;-)
          You should report them here: https://gitlab.freedesktop.org/mesa/mesa/-/issues

          Comment


          • #15
            Originally posted by timofonic View Post

            What games?
            BATTLETECH (they chose all caps, not me), and Daggerfall Unity. Also one MMO written in Java that will most likely never update to Vulkan. I could probably get BATTLETECH as the second game to load GPU at 100% even on new equipment. They advertised as using Unity for that, but there's Flash in every single game screen and... let's just say it shows in more than visuals.

            Comment


            • #16
              Originally posted by kiffmet View Post
              ms178 It's worth questioning whether Unigine Heaven does even remotely represent a 2023 workload though. Optimizing drivers specifically for benchmarks usually only benefits their very specific corner cases.
              I found Unigine Heaven great for testing the overall functionality of the OpenGL stack and reported bugs to the developer in the ACO bring up. When it comes to performance, consistent lower results might point to shortcomings in the implementation like greater CPU overhead, worse code generation or missing optimizations. Only further analysis will tell, if it is worth fixing and might profit more relevant other OpenGL apps as well.

              Originally posted by kiffmet View Post
              Are there nowadays even any OpenGL applications, where a 6950XT doesn't yield enough performance? Funnily enough, the Windows version of Heaven run through Wine is faster with OpenGL (up to 40FPS more in 1080p), than with DXVK for my 6800XT btw.
              Performance in absolute terms is great, but that was not the point of my comment. My point was a performance evaluation between ACO vs. LLVM in that specific benchmark. While it is not the best benchmark, it is still a valid data point. Others shared results and found similar slight performance regressions in the MR on Gitlab. As more people than me hoped for a gain from the introduction from ACO for RadeonSI, the numbers were a bit of a let down. As of today, there is no real reason to switch to ACO for most people on RadeonSI.

              Comment


              • #17
                Originally posted by oibaf View Post

                [Full quote, really?]

                You should report them here: https://gitlab.freedesktop.org/mesa/mesa/-/issues
                It is 'upstream' ( ;-) ) and worked on for the last days.

                Comment


                • #18
                  Originally posted by nuetzel View Post
                  Michael (Qiang Yu and karolherbst )

                  '...I'll be running some RadeonSI ACO comparison benchmarks soon on Phoronix.'

                  NOT needed, yet!

                  It is NOT complete, 'only' 'part mode merged shader' as far as I know.

                  vdpau (mplayer and mpv) do NOT work.

                  mplayer -vo vdpau /data/Filme/test.mkv
                  mpv --hwdec=vdpau -vo=vdpau /data/Filme/test.mkv
                  ​Only blueish windows.

                  OpenCL (RustiCL)
                  is not complete, too.

                  clinfo | less
                  [-]
                  Platform Name rusticl
                  Number of devices 1
                  Device Name AMD Radeon RX 580 Series (radeonsi, polaris10, ACO
                  lines 1-20ACO ERROR:
                  In file ../src/amd/compiler/aco_instruction_selection.cpp:9233
                  Unimplemented intrinsic instr: div 32 %27 = @load_global (%26) (access=none, align_mul=4, align_offset=0)
                  Abbruch (core dumped)

                  Testing it from time to time...;-)
                  vdpau is basically dead now, use vaapi if you want acceleration in mpv

                  Comment


                  • #19
                    Originally posted by FireBurn View Post

                    vdpau is basically dead now, use vaapi if you want acceleration in mpv
                    Thanks, fixed my last hang!
                    mpv --hwdec=vaapi -vo=gpu /data/Filme/test.mkv

                    Comment

                    Working...
                    X