Announcement

Collapse
No announcement yet.

The RadeonSI Performance Tuning For AMD Zen CPUs Has Landed In Mesa

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

  • The RadeonSI Performance Tuning For AMD Zen CPUs Has Landed In Mesa

    Phoronix: The RadeonSI Performance Tuning For AMD Zen CPUs Has Landed In Mesa

    Earlier this week I reported on the RadeonSI Gallium3D code being tuned for AMD Zen CPUs in an attempt to deliver greater gaming performance for Ryzen processors. That work has now been merged into Mesa 18.3...

    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
    sweet! I hope you find the time soon!

    Comment


    • #3
      RadeonSI is an exception, because it has 1 internal thread.
      Nice one. In multithreaded world we are using 1 thread.

      Comment


      • #4
        Originally posted by atomsymbol



        RadeonSI, as well as other parts of Mesa, are written in C rather than C++. C++ nicely supports asynchronous computation (https://en.cppreference.com/w/cpp/thread/async). I suggested async to Mesa developers in the past as a solution to certain problems, but this has been rejected/ignored by the developers.
        Can you explain what exactly you are going to do in these additional threads? The most load in Mesa is generated when compiling shader code ( which is mitigated with shader cache, or SPIR-V ), and uploading stuff to VRAM. I think devs know much better than you about threading issues in Mesa.

        Comment


        • #5
          what about
          GL_ARB_gl_spirv* in Mesa?

          Comment


          • #6
            Originally posted by V1tol View Post
            Nice one. In multithreaded world we are using 1 thread.
            You are taking what was said in another thread out of context and wasting a perfectly good emoji

            Marek said that while most of the Mesa drivers don't use any extra threads besides the one the process is calling from the radeonsi driver uses 1 additional thread for dispatch. In addition, I believe (but am not sure) that the driver can also take advantage of having the application or game engine use multiple threads to call into the driver.
            Last edited by bridgman; 09 September 2018, 04:29 PM.
            Test signature

            Comment


            • #7
              Originally posted by V1tol View Post
              Nice one. In multithreaded world we are using 1 thread.
              What's the old saying? It used to be too slow, so I made it asynchronous. Now bigger I much probl haveems.

              Using more threads doesn't automatically provide any gains (in fact it costs extra overhead). Also, it seems like this particular change is about making sure it all runs on the same core so that the cache is hit. Spreading the processing across multiple cores would generally make that aspect worse. In general, you seem to have a good idea of what you're talking about, so you have to know that threading for threading's sake is a fool's errand.

              Comment


              • #8
                Originally posted by V1tol View Post
                Nice one. In multithreaded world we are using 1 thread.
                Indeed, let's look forward to when each application will use 1000 threads each!

                Comment


                • #9
                  Originally posted by geearf View Post
                  Indeed, let's look forward to when each application will use 1000 threads each!
                  In the future, 3D acceleration will be virtualized into the cloud, and will not require any hardware at all. All the greedy hardware companies will go out of business and everyone will have 3D for free. It is a human right after all; its time for free universal 3D for everyone.

                  Comment


                  • #10
                    Originally posted by torsionbar28 View Post
                    In the future, 3D acceleration will be virtualized into the cloud, and will not require any hardware at all. All the greedy hardware companies will go out of business and everyone will have 3D for free. It is a human right after all; its time for free universal 3D for everyone.
                    Man you're so behind the times (pun intended), it's "4D for all!".

                    Comment

                    Working...
                    X