Page 3 of 3 FirstFirst 123
Results 21 to 29 of 29

Thread: RadeonSI GLAMOR 2D Performance vs. Catalyst

  1. #21
    Join Date
    Jun 2010
    Location
    ฿ 16LDJ6Hrd1oN3nCoFL7BypHSEYL84ca1JR
    Posts
    1,039

    Default

    Quote Originally Posted by Pontostroy View Post
    You use SNA, intel+glamor up to 100 times slower than intel+sna
    http://openbenchmarking.org/result/1...SO-1401199SO68
    I'm no expert but that looks more like one totally unoptimized operation.

    Anyway, sna has been extremly fine tuned for a long time now. Just look at only the amount of work they put into it: http://cgit.freedesktop.org/xorg/dri...?qt=grep&q=sna or some of the insane optimisations: http://cgit.freedesktop.org/xorg/dri...t=grep&q=optim

    Glamor being much younger probably hasn't seen that much optimization.

  2. #22
    Join Date
    Dec 2012
    Posts
    524

    Default

    Every time I read micro-optimize I want to hug my cpu more.

  3. #23
    Join Date
    Sep 2008
    Location
    Vilnius, Lithuania
    Posts
    2,537

    Default

    Quote Originally Posted by Spittie View Post
    He's using those repository (created by him!) to get the latest mesa/llvm from git, and the latest kernel rc.
    http://download.opensuse.org/reposit...openSUSE_13.1/
    http://download.opensuse.org/reposit...openSUSE_13.1/
    Yeap. I use it when I need to try the absolute latest, too. It's like the openSUSE equivalent for oibaf PPA.

  4. #24
    Join Date
    Jan 2010
    Location
    Portugal
    Posts
    945

    Default

    Quote Originally Posted by ChrisXY View Post
    Glamor being much younger probably hasn't seen that much optimization.
    Actually, Glamor's first commit is from November 2011, and SNA's is April 2011, so not much head-start:

    Glamor: http://cgit.freedesktop.org/xorg/dri...0bce3eea834620
    SNA: http://cgit.freedesktop.org/xorg/dri...fa2c5626adf4cb

    The difference is that Chris Wilson did a ton of work on improving SNA, while Glamor didn't see nearly as much work.

  5. #25
    Join Date
    Feb 2011
    Location
    Ukraine
    Posts
    132

    Default

    radeon exa vs glamor (r600, radeonsi)
    http://openbenchmarking.org/result/1...SO-1401219SO64

  6. #26
    Join Date
    Jan 2013
    Posts
    971

    Default

    Quote Originally Posted by Pontostroy View Post
    radeon exa vs glamor (r600, radeonsi)
    http://openbenchmarking.org/result/1...SO-1401219SO64
    Thanks a bunch for testing and posting!

    How did you build the environment for this?

    Summed it up a bit below. All are FPS - more better.

    Qgears2/OpenGL/text
    7790 gla - 305.27
    6770 gla - 307.52
    6770 exa - 316.25

    Qgears2/CPU Rast/text
    7790 gla - 149.41
    6770 gla - 178.45
    6770 exa - 184.58

    Qgears2/Xrender/text
    7790 gla - 156.05
    6770 gla - 186.97
    6770 exa - 193.60


    Qgears2/OpenGL/Gears
    7790 gla - 527.56
    6770 gla - 581.72
    6770 exa - 602.21

    Qgears2/CPU Rast/Gears
    7790 gla - 136.99
    6770 gla - 159.83
    6770 exa - 165.06

    Qgears2/Xrender/Gears
    7790 gla - 141.12
    6770 gla - 165.72
    6770 exa - 170.65


    Qgears2/OpenGL/Image Scaling
    7790 gla - 4370.40
    6770 gla - 4929.05
    6770 exa - 6496.89

    Qgears2/CPU Rast/Image Scaling
    7790 gla - 265.94
    6770 gla - 354.06
    6770 exa - 379.87

    Qgears2/Xrender/Image Scaling
    7790 gla - 281.58
    6770 gla - 373.39
    6770 exa - 400.17

    I skipped the x11perf tests.

  7. #27
    Join Date
    Feb 2011
    Location
    Ukraine
    Posts
    132

  8. #28
    Join Date
    Jan 2013
    Posts
    971

    Default

    So, judged from X11perf vs Qgears - we don't have bad performance due to driver problems,
    but we have bad performance due to Xorg stack unable to use 3D-based/OpenGL-based hardware efficiently, preferring simple 2D calls instead and OpenGL calls meet a lot of overhead.

    With Wayland the situation may change a lot to the good. Hm?

  9. #29
    Join Date
    Dec 2012
    Posts
    524

    Default

    Quote Originally Posted by brosis View Post
    So, judged from X11perf vs Qgears - we don't have bad performance due to driver problems,
    but we have bad performance due to Xorg stack unable to use 3D-based/OpenGL-based hardware efficiently, preferring simple 2D calls instead and OpenGL calls meet a lot of overhead.

    With Wayland the situation may change a lot to the good. Hm?
    Since Wayland uses EGL, we hope. It has a much reduced API and much less state mess and more effective pipelined calls than GLX, but the EGL implementations on Mesa haven't been beaten to death for years, so we shall see.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •