Announcement

Collapse
No announcement yet.

AMD ACO Begins Using Navi NGG For Tessellation + Vertex Shaders

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

  • AMD ACO Begins Using Navi NGG For Tessellation + Vertex Shaders

    Phoronix: AMD ACO Begins Using Navi NGG For Tessellation + Vertex Shaders

    The AMD "ACO" compiler backed by Valve for offering a faster shader compiler back-end than AMDGPU LLVM for the RADV open-source Radeon Vulkan driver has begun making use of Navi's NGG "Next-Gen Geometry" hardware...

    http://www.phoronix.com/scan.php?pag...-VS-TESS-Lands

  • #2
    this is great news. i hope by kernel 5.7 and mesa 2.1 we can finally have performance equal to that of the windows blob on linux.

    Comment


    • #3
      Thank you very much, Valve and all the people involved in ACO development!

      I am wondering if something like this: https://gitlab.freedesktop.org/mesa/...ts/580/commits and https://gitlab.freedesktop.org/mesa/..._requests/4269 can be moved to the backend so that both OpenGL and Vulkan would benefit (when ACO will also be wired up for OpenGL).

      Comment


      • #4
        Will this improve performance and in which cases?

        Comment


        • #5
          Originally posted by valici View Post
          Will this improve performance and in which cases?
          https://github.com/daniel-schuermann...ment-610381317

          Comment


          • #6
            While it's supposed to be faster, it seems for now at least it's just an alternative path with no impact on performance. So I'm not sure if the claim that it's faster is true. May be there are some other benefits like higher power efficiency?

            Comment


            • #7
              Originally posted by pieman View Post
              this is great news. i hope by kernel 5.7 and mesa 2.1 we can finally have performance equal to that of the windows blob on linux.
              Are you implying that NGG is used on Windowz to? First time hearing that. Can someone from AMD confirm?

              Comment


              • #8
                Originally posted by valici View Post
                Will this improve performance and in which cases?
                No, it probably won't. At least I didn't see a noticable perf difference when I made those patches.

                Comment


                • #9
                  Originally posted by xxmitsu View Post
                  I am wondering if something like this: https://gitlab.freedesktop.org/mesa/...ts/580/commits and https://gitlab.freedesktop.org/mesa/..._requests/4269 can be moved to the backend so that both OpenGL and Vulkan would benefit (when ACO will also be wired up for OpenGL).
                  it's not clear what are you referring to with "backend". also it uses llvm, i.e. it has to be implemented for aco before it could be used with aco

                  Comment


                  • #10
                    Originally posted by xxmitsu View Post
                    Thank you very much, Valve and all the people involved in ACO development!

                    I am wondering if something like this: https://gitlab.freedesktop.org/mesa/...ts/580/commits and https://gitlab.freedesktop.org/mesa/..._requests/4269 can be moved to the backend so that both OpenGL and Vulkan would benefit (when ACO will also be wired up for OpenGL).
                    Primitive culling isn't considered to be beneficial in games.

                    One reason is that many game engines already have their own way to implement primitive culling (eg. through tessellation or by other means), the other reason is that implementing primitive culling in the NGG GS is resource intensive and would compete with other shaders (eg. pixel shaders).

                    So it's not a priority for ACO currently. We have other things to do first.

                    Comment

                    Working...
                    X