Fix for ssr search dist

[e2002e]

This is a simple tweak to the ssr shader that allows to match the search distance.

In the rayCast loops on maxSteps which is an hardcoded, if the step size is too small we don't reflect far:

To match the searchDist we need to take the stepSize and search dist into account when setting the maxStep value:

#define maxSteps int(ceil(1.0 / ssrRayStep) * ssrSearchDist)

[MoritzBrueckner]

Looks good to me, but what's the reason you replaced const by #define? Defines are literally replaced in the code by the preprocessor, so instead of

for (int i = 0; i < maxSteps; i++) {

the compiler sees

for (int i = 0; i < int(ceil(1.0 / ssrRayStep) * ssrSearchDist); i++) {

Have you checked whether the compiler is able to optimize that so that the defined expression is only evaluated once? Otherwise using another solution might be faster. If const doesn't work because the value is not perceived by the compiler as being constant, you could just calculate it in Python and include the resulting value in compiled.inc like ssrRayStep and ssrSearchDist.

[e2002e]

You'r right I didn't think about glsl optimizations (if there are any is this case). So here i rolled back the value type.

[e2002e]

@MoritzBrueckner I realized I have missed the point about variable optimization.
Yes using a macro will recompute the values for each fragment at each frame instead of having a constant stored.
No reasons to do so.

[e2002e]

@luboslenco can you merge that ? It's reviewed by @MoritzBrueckner and fixes a bad bug !