Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo

Close

Tree [e3e264] master /
History



File Date Author Commit
buffer-performance 2010-03-17 bob bob [2644e1] Directories cleanup
experimental-cont-array 2010-03-17 bob bob [2644e1] Directories cleanup
experimental-glib 2010-03-06 bob bob [023b82] simplified src directory to just basics
experiments 2010-03-17 bob bob [2644e1] Directories cleanup
gtk 2010-04-25 bgibbs bgibbs [e3e264] Add LGPL license and INSTALL files
obsolete 2009-11-27 Robert Gibbs Robert Gibbs [aa44ce] gtk test program in progress
opengl 2010-04-25 bgibbs bgibbs [e3e264] Add LGPL license and INSTALL files
src 2010-04-25 bgibbs bgibbs [e3e264] Add LGPL license and INSTALL files
x11 2010-04-25 bgibbs bgibbs [e3e264] Add LGPL license and INSTALL files
.gitignore 2009-05-21 Robert Gibbs Robert Gibbs [15cc77] single complete poly test case working
Makefile 2010-04-25 bgibbs bgibbs [e3e264] Add LGPL license and INSTALL files
README 2010-03-06 bob bob [0142d4] Notes on dir structure

Read Me

----------------------------------------------------------------
Directory structure:

src: This is the main code for the cyl transform. It uses a custom array
     Implementation.

experimental-glib: This uses garray vice a custom array implementation.
                   It is much slower than the custom array.

experimental-cont-array: This is a slightly faster implementation on some
machines.  On other machines it doesn't makeany difference.


----------------------------------------------------------------
OpenGL 1600x1200 window Nvidia
> 40 FPS w/o  Tesselation
> 17.5 FPS w/ Tesselation
On slow T41 w/open source ATI
> 4.5 FPS

----------------------------------------------------------------
Major GTK performance improvement by doing custom double buffering.
> 50 FPS for 1600x1200 window when drawing area double buffering 
  flag is unset and a GdkPixmap is used instead
  GTK_WIDGET_UNSET_FLAGS (app.da, GTK_DOUBLE_BUFFERED);


-------------------------------------------
Performance using Xlib calls directly
50 FPS to (1600x1200 window with no double buffering
48 FPS when buffering to a pixmap
70 FPS with XCopyArea disabled

---------------------------------------------------------------------
Performance for large area (1600x1200) goes way down (10 FPS).

- Looked at reducing 14K pts down to 4K points but did not see a big
  difference.
- Looked at redrawing w/o transormation (no panning) and did not see
  a big difference.
- Bottleneck appears to be in gdk_draw_polygon().  Need to look at X11
  call that it wraps to see if there is a difference.

-------------------------------------------
Experiments with edge lines from polygon clipping.  This does not cause a
problem in X11 due to winding rule or polygon being convex not sure.

For not-filled shapes, then "open" flag can be set in xform to eliminate this.
Additionally the clip object could be set with bounds slightly larger than 
window bounds.  (This does not always work due to some artifact with X11.)