No announcement yet.

Revised Stack Space Patch For R600 Gallium3D Driver

  • Filter
  • Time
  • Show
Clear All
new posts

  • Revised Stack Space Patch For R600 Gallium3D Driver

    Phoronix: Revised Stack Space Patch For R600 Gallium3D Driver

    Vadim Girlin on Sunday posted a stack space patch for the AMD R600 Gallium3D driver that he reported to improve Unigine's performance by ~30% for this open-source Radeon graphics driver. He has since revised that patch, but testing has revealed it isn't too incredibly exciting at this point...

  • #2
    MIchael can you hit up a Unigen benchmark as well since thats what he was citing? Maybe its something specific to that engine that gets a big performance jump


    • #3
      Note: This patch had only been tested on Evergreen (as that is what Vadim has)... and the corruption report was from a Radeon HD 48xx, which is pre-Evergreen.

      And as stated in the previous post, this test probably affects Unigine more than other benchmarks. If I'm feeling ambitious, I'll do a before/after run when I get home.


      • #4
        Here ya go:

        Note: The mesa is git master as of commit 07cdfdb708ac28aa with the addition of my having reverted commit 1eedebc65b (enable DISCARD_RANGE for r600g)... The reverted commit triggers corruption in TF2 for me.

        The Before run is mesa as described above. The after run is with the addition of v3 of Vadim's stack space reservation patch (v3 was posted shortly before I got home from work). The difference is about 26% before you take standard deviation into account.