=== a simple benchmark to evaluate drawing time of points with high overdraw ===

Upon writing a benchmark to evaluate the cost of using various data types
for vertex attributes I notices a strange behaviour in drawing huge amounts
of blended point with lots of overdraw: The naive expectation would be, that
when drawing a constant amount of point the more fragments are touched, the
higher to overall draw time.

As it turns out for at least a NVidia GeForce GTX980 things are not that simple.
The drawing times actually reduce with increasing viewport size... or rather
increasing vertical size; the horizontal size doesn't seem to have that much
of an effect.

To gather data from more GPUs I provide this source code repository and ask
people to run the program with their GPUs and submit the results for evaluation.

**IMPORTANT NOTICE:**
_The program depends on OpenGL-4.3, namely compute shaders. However since this
is more or less a quick hack I didn't add any version or feature checking. So
if trying to run this program on a system that lacks OpenGL-4.3 the results
are undefined and may range from a black window with nonsensical numbers to
immediate program crashes!_

== Build Dependencies ==
- OpenGL headers and linker stubs
  - Windows: compiler toolchains should have them without extra effort
  - X11/GLX (Linux / *BSD / Solaris): if OpenGL drivers are installed, they
      should be available. Otherwise install the Mesa development support
      package.

- GLUT (kind of smells, but everybody dabbling with GPUs has it around)
- GLEW (kind of sucks, but everybody dabbling with GPUs has it around)

== Build Instructions ==
- Using *make*: just run `make`
- Using *CMake*: `mkdir $BUILDDIR ; cd $BUILDDIR ; cmake $SOURCEDIR ; make`

== Run Instructions ==
Run the program and direct its stdout into a file, .e.g.

    ./pointoverdrawbench > statistics.log

OpenGL debug messages are emitted to stderr. If you see something noteworthy
there, add this to your report.

== How to report results ==
Submit them as a Github Issue