I've optimized the glAccum function for the following common scenario:
glAccum(GL_LOAD, w); // or glAccum(GL_ACCUM, w) on a zeroed buffer
glAccum(GL_ACCUM, w);
...
glAccum(GL_ACCUM, w);
glAccum(GL_RETURN, 1.0);
Where w = 1/n and n = number of images to blend.
This is the usual sequence of commands for full-scene antialiasing.
Basically, I store unscaled integer color components in the accum
buffer instead of scaled pseudo-floats. This avoids a _lot_ of
int->float->int conversions. A lookup table avoids integer divides
in GL_RETURN.
Benchmarking on my IRIX system I've found a 33% speed improvement.
That is, if the old implmentation took 3 seconds to execute glAccum,
the new code should take about 2 seconds. Speed-up on other kinds
of CPUs should be at least as good.
I encourage anyone with a program that uses glAccum to test this
optimization.
-Brian
----------------------------------------------------------------------
Brian Paul Avid Technology / Softimage [EMAIL PROTECTED]
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
