Avoiding branching within Cg shader

Questions & Answers
1

Within a fragment shader, I need some simple conditional evaluation - basically I'm checking the luminance of a pixel and want to perform an extra calculation if it is above some threshold value.

A simple example of the naive implementation would be:

if (lum >= threshold) {
    col *= some_value;
}

However, this will (as far as I'm aware) cause a branch and destroy the built-in parallelism.

I have instead come up with the following:

col *= lerp(1.0, some_value, step(threshold, lum));

My question is: will the trade-off between arithmetic complexity (more instructions used) and removal of branching be worthwhile? Is this a viable (common?) strategy?

2

That if-block is very unlikely to result in an actual dynamic branch. Shader compilers don't do dynamic branching unless they absolutely have to. Instead, it will likely use a fairly similar set of instructions (though it is likely to multiply before the lerp, not afterwards).