#86 Limit visuals search to 24 planes depth

closed-fixed
nobody
None
5
2006-05-02
2005-10-22
No

On xorg with Composite extension, 32 plane depth
visuals would be slower than 24 plane depth due to some
overhead, possibly related to alpha-blending. Limiting
rdesktop to 24 plane depths, either by setting the
XLIB_SKIP_ARGB_VISUALS environment variable or by
applying this patch, fixes the performance problem.

Discussion

  • Patch to limit depth to 24 planes

     
  • Jeremy Y. Meng
    Jeremy Y. Meng
    2005-10-28

    Logged In: YES
    user_id=862010

    I had a similar issue before. On my SGI Octane2 V6
    graphics, the best visual found by rdesktop has 30 plane
    depth. This caused a totally black window. Does the depth
    has to be a multiple of 8?

     
  • Jeremy Y. Meng
    Jeremy Y. Meng
    2005-10-28

    Logged In: YES
    user_id=862010

    I had a similar issue before. On my SGI Octane2 V6
    graphics, the best visual found by rdesktop has 30 plane
    depth. This caused a totally black window. Does the depth
    has to be a multiple of 8?

     
  • Logged In: YES
    user_id=335423

    Judging from this output from an SGI workstation (xdpyinfo):
    visual:
    visual id: 0x45
    class: TrueColor
    depth: 30 planes
    available colormap entries: 1024 per subfield
    red, green, blue masks: 0x3ff, 0xffc00, 0x3ff00000
    significant bits in color specification: 10 bits

    ... those 30-plane visuals have 10 bits (1024 possible
    values) for each color, which is richer than the common PC
    TrueColor mode. If rdesktop didn't support this mode
    properly, then I'd expect the display to be mangled but not
    totally black...

    This is really quite curious but unfortunately I don't have
    such a cool workstation to test on :/ My NVIDIA card only
    offers 8-bit-per-color visuals. In any case, rdesktop should
    be fixed either to support this visual, or to refuse to take
    it. This patch is not enough to fix it since it only makes
    rdesktop *prefer* 24-bit visuals, but if a 24-bit visual is
    not available, it'll take the 30-bit visual.

     
  • Logged In: YES
    user_id=335423

    jymeng, can you fill out a new bug about this 30-bit visual?
    I think we'd like to track that one till it's solved.

     
  • Logged In: YES
    user_id=335423

    jymeng, I take that back -- there is a bug filled out
    already -- see #1055468. You're welcome to monitor that bug
    till it's fixed. (The cause was straightforward -- see my
    comment there.)

     
  • Logged In: YES
    user_id=335423

    Should be fixed by now.

     
    • status: open --> closed-fixed