Announcement

Collapse
No announcement yet.

Tips for Radeon OSS drivers

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

  • Tips for Radeon OSS drivers

    well guys if you want to squezee a bit more fps from your nicely done OSS driver you can use this flags at compilation time.

    export CFLAGS="-O3 -Wall -Wmissing-prototypes -std=c99 -ffast-math -fvisibility=hidden -fno-strict-aliasing -fPIC -march=native -mfpmath=sse -msse2 -msseregparm -maccumulate-outgoing-args -m64 "

    export CXXFLAGS="-O3 -Wall -fno-strict-aliasing -fPIC -march=native -mfpmath=sse -msse2 -msseregparm -maccumulate-outgoing-args -m64 -fprefetch-loop-arrays"

    it worked fine for me in the DDX, Mesa, DRM. ubuntu lucid user careful since latest gcc update seems to break the compiler in mesa/gallium/auxiliary when using -fprefetch-loop-arrays, if you encounter this bug just remove that flag or compile your own gcc 4.5

    have fun

  • #2
    Does this require compiling everything (mesa, drm, driver, etc) too?

    I tried it by compiling mesa with the r300g driver as described in the Radeon Gallium on Ubuntu thread.

    glxgears shows a blank window and Urban Terror shows a garbled screen.

    Comment


    • #3
      Originally posted by rob2687 View Post
      Does this require compiling everything (mesa, drm, driver, etc) too?

      I tried it by compiling mesa with the r300g driver as described in the Radeon Gallium on Ubuntu thread.

      glxgears shows a blank window and Urban Terror shows a garbled screen.
      well compiling only mesa works too, but remember radeon gallium dont support cards without vertex shader like the one in the laptops for example, at least for now. and i heard that only r500 series get gl 2.1 working properly for now. so stick for mesa regular driver for now

      Comment


      • #4
        This would probably be a lot more compelling with benchmarks proving an actual performance increase...

        Comment


        • #5
          Originally posted by whizse View Post
          This would probably be a lot more compelling with benchmarks proving an actual performance increase...
          Who needs benchmarks anyway. I just tried it and everything 'feels' a lot snappier. Nah, just kidding . You're right; if no benchmarks then not interested.

          Comment


          • #6
            I compiled xf86-video-ati git and here's some quick benchmarks.


            No extra flags

            840 frames 18.1 seconds 46.4 fps 6.0/21.6/86.0/8.6 ms
            840 frames 17.8 seconds 47.2 fps 5.0/21.2/119.0/9.6 ms
            840 frames 17.3 seconds 48.5 fps 7.0/20.6/88.0/8.3 ms
            840 frames 17.5 seconds 48.0 fps 6.0/20.8/144.0/8.9 ms
            840 frames 17.9 seconds 47.0 fps 6.0/21.3/66.0/8.1 ms

            avg fps 47.42


            Optional flags

            840 frames 18.2 seconds 46.2 fps 6.0/21.7/105.0/9.0 ms
            840 frames 18.0 seconds 46.6 fps 5.0/21.4/84.0/8.6 ms
            840 frames 18.4 seconds 45.7 fps 6.0/21.9/95.0/8.8 ms
            840 frames 18.5 seconds 45.3 fps 6.0/22.1/75.0/8.9 ms
            840 frames 18.6 seconds 45.1 fps 6.0/22.2/138.0/9.6 ms

            avg fps 45.78

            Comment


            • #7
              That's in OpenArena btw. The std=c99 flag gives an error during compile so I removed that flag.

              Comment


              • #8
                I'd recommend against -std=c99, use -std=gnu99 instead. Neither affects performance but c99 is strict C99 and missing extensions.

                Comment


                • #9
                  I loled :P It seems the -funroll-loops madness escaped Gentoo and is now in the wild

                  Comment


                  • #10
                    Originally posted by RealNC View Post
                    I loled :P It seems the -funroll-loops madness escaped Gentoo and is now in the wild
                    Yeah, scary, isn't it.

                    Comment

                    Working...
                    X