Re: [Openexr-devel] luts

[Lutz Latta]

Ken McGaugh wrote:

> I'm now need to use luts in my fragment shader, and I
> think this pack_2half/unpack_4ubyte trick is very cool.  But
> my fragment shaders are arbfp1 instead of fp30 (for portability).
> Is there anyway to do the same thing using arbfp1?

The trick is just a speed up of the way you would do it normally with integer 
arithmetic on the CPU, using a modulus and integer division. Since current GPUs 
don't have such integer arithmetic natively (although Cg can emulate it), the 
fastest way is to use the floating point functions fmod and floor. I've used 
the following in a different context:

        float2 p = float2(fmod(i, 256), floor(i / 256) + 0.5);

Note that the "+ 0.5" was to get away from the nearest neighbor sampling 
boundaries. Check whether it is needed in your context, or whether you need similar 
offsets somewhere else in the computation.

Cheers,
  Lutz