1. 21 Feb, 2011 2 commits
  2. 18 Feb, 2011 1 commit
    • Kenneth Moreland's avatar
      Split sizes based on final split size. · f5008166
      Kenneth Moreland authored
      Changed icetSparseImageSplit and associated functions to base the size
      of the splits based on the eventual number of splits done by the end
      of the algorithm.  Each of these functions takes a new
      eventual_num_partitions argument that is the number of times the
      image will be split by the end of the algorithm.  If you partition
      the image by factors of this number, you will get the same partition
      sizes and order as if you partitioned by that number to begin with.
  3. 15 Feb, 2011 3 commits
    • Kenneth Moreland's avatar
      Add sync render option to simple timing. · b08211cf
      Kenneth Moreland authored
      I am having trouble getting my crummy rendering routines to load balance
      even though they have about the same amount of work to do.  Instead,
      punt and add an option that will synchronize the renders so they all
      finish at the same time.  When you measure the composite performance,
      it will be roughly equivalent to a quick render.
    • Kenneth Moreland's avatar
      Reduce fov angle in simple timing. · c065d051
      Kenneth Moreland authored
      The field of view angle used in the simple timing test was a bit
      exaggerated.  Appart from not being well representative of a real
      application, it was causing load imbalances in the rendering
      (I think).
    • Kenneth Moreland's avatar
      Use an array for quick lookup of power function. · 0ac67f42
      Kenneth Moreland authored
      The pow function can take a while to complete.  The array lookup
      provides enough accuracy for our purposes and is generally faster.
  4. 04 Feb, 2011 2 commits
    • Kenneth Moreland's avatar
      Black backgrounds in SimpleTiming · 2746422d
      Kenneth Moreland authored
      Add a flag to SimpleTiming to turn black background on and off.  If
      in transparent mode and the background is anything other than black
      (with no opacity), then IceT has to "correct" the background by
      blending the composite image to the background.  Although important
      in practice, it is not worth timing when comparing algorithms.
      Now that I think about it, the correction is done completely on the
      display node.  Why not have the correction happen right before the
      collection so that this time is shared amongst processes?
    • Kenneth Moreland's avatar
      Don't return in a macro. · 47c237a6
      Kenneth Moreland authored
      In checkOpenGLError and checkIceTError, there was a macro that defined
      an if statement that returned from a function.  This messed up the code
      metric software tau, which missed this return from the function.  Instead
      use the macro to define a case for a switch statement.
  5. 01 Feb, 2011 1 commit
  6. 31 Jan, 2011 1 commit
  7. 26 Jan, 2011 1 commit
  8. 25 Jan, 2011 3 commits
  9. 21 Jan, 2011 1 commit
    • Kenneth Moreland's avatar
      Fix lock-up in DisplayNoDraw. · ac97073e
      Kenneth Moreland authored
      When an image failure occured, the root node broke out of some
      iterations, but the rest did not.  This often caused the rest of
      the processes to deadlock expecting the root to continue.
  10. 20 Jan, 2011 1 commit
  11. 18 Jan, 2011 1 commit
  12. 11 Jan, 2011 1 commit
    • Kenneth Moreland's avatar
      SimpleTiming now collects timing information in root process. · 7c8c62bf
      Kenneth Moreland authored
      Previously, each process wrote out its own log file, which were then
      to be collected by some other means.  However, the size is becoming
      problematic for large process counts.  Instead, the root process
      collects all the other timing information and computes the max for
      all processes, which is what I was doing in post processing anyway.
  13. 10 Jan, 2011 1 commit
  14. 20 Dec, 2010 2 commits
  15. 16 Dec, 2010 1 commit
  16. 15 Dec, 2010 3 commits
  17. 14 Dec, 2010 1 commit
  18. 10 Dec, 2010 1 commit
  19. 08 Dec, 2010 1 commit
  20. 07 Dec, 2010 4 commits
  21. 06 Dec, 2010 1 commit
  22. 02 Dec, 2010 3 commits
  23. 01 Dec, 2010 1 commit
  24. 30 Nov, 2010 2 commits
  25. 18 Nov, 2010 1 commit