Announcement

Collapse
No announcement yet.

Valve's Timothy Arceri Lands Gallium3D NIR Optimizations

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

  • Valve's Timothy Arceri Lands Gallium3D NIR Optimizations

    Phoronix: Valve's Timothy Arceri Lands Gallium3D NIR Optimizations

    Timothy Arceri who has been for the past year working on Linux GPU driver optimizations for Valve has just merged his latest patch series providing optimizations for the Gallium3D NIR linking phase...

    http://www.phoronix.com/scan.php?pag...-NIR-Link-Opts

  • #2
    Typo:

    Originally posted by phoronix View Post
    While Freedreno and VC4 currently make use if the NIR intermediate representation,
    (the second one is redundancy)

    Comment


    • #3
      How much work is needed for RadeonSI use NIR fully compared to bring RadeonSI up to OpenGL 4.2, 4.3 or other recent version?

      Comment


      • #4
        I'm glad regardless of why, but I wonder whether Valve has him on payroll for goodwill with the Linux community, or if Valve sees a future with AMD-based Steam Machines...

        Comment


        • #5
          I thought Intel was using Nir... ?

          Comment


          • #6
            Originally posted by ernstp View Post
            I thought Intel was using Nir... ?
            They are, NIR comes from them.

            Comment


            • #7
              Originally posted by andrei_me View Post
              How much work is needed for RadeonSI use NIR fully compared to bring RadeonSI up to OpenGL 4.2, 4.3 or other recent version?
              These new Mesa patches are in the Oibaf ppa Artful version so you can test. In the Tomb Raider 2013 benchmark at fullhd normal settings with RX560&Ryzen 5 1600, there is average 1 fps regression without R600_DEBUG=nir compared to Oibaf ppa from last week. R600_DEBUG=nir fixes the regression, so NIR does not look so good. Phoronix have tested NIR:
              http://openbenchmarking.org/result/1...TY-NIRRADEON28
              Last edited by debianxfce; 12-04-2017, 03:53 AM.

              Comment


              • #8
                Originally posted by tildearrow View Post
                Typo:



                (the second one is redundancy)
                I'm not sure it's an acronym although it looks like one

                Comment


                • #9
                  Originally posted by debianxfce View Post

                  These new Mesa patches are in the Oibaf ppa Artful version so you can test. In the Tomb Raider 2013 benchmark at fullhd normal settings with RX560&Ryzen 5 1600, there is average 1 fps regression without R600_DEBUG=nir compared to Oibaf ppa from last week. R600_DEBUG=nir fixes the regression, so NIR does not look so good. Phoronix have tested NIR:
                  http://openbenchmarking.org/result/1...TY-NIRRADEON28
                  I don't get it. How does nir not look so good? It's winning in most of the benchmarks and you said it improves a regression you noticed.

                  Comment


                  • #10
                    Originally posted by tarceri View Post

                    I don't get it. How does nir not look so good? It's winning in most of the benchmarks and you said it improves a regression you noticed.
                    https://lists.freedesktop.org/archiv...er/177658.html

                    "Tested without regression on radeonsi. ยจ
                    I tested with regression on radeonsi without R600_DEBUG=nir. Handy way to force people to use experimental code. I do not want to see rendering problems when my games runs fine now:

                    Radeonsi code sharing with intel is dangerous, sure Intel will do everything against Amd, even buys amd gpu for their new laptop APU to tamper down Raven Ridge.
                    https://www.phoronix.com/scan.php?pa...nSI-NIR-Future

                    " but going "all in" with using NIR could result in more code sharing with the Intel folks (among others"

                    Sure intel does not test with Amd hardware their code like they do no test the Linux kernel code and do not care of bug reports.
                    Last edited by debianxfce; 12-04-2017, 05:58 AM.

                    Comment

                    Working...
                    X