Kernal-based PCF for Shadow Map Sampling

[bryanedds]

We experimented with the following -

const int SHADOW_SAMPLE_OFFSETS_COUNT = 20;
vec3 SHADOW_SAMPLE_OFFSETS[SHADOW_SAMPLE_OFFSETS_COUNT] = vec3[](
    vec3(1, 1, 1), vec3(1, -1, 1), vec3(-1, -1, 1), vec3(-1, 1, 1),
    vec3(1, 1, -1), vec3(1, -1, -1), vec3(-1, -1, -1), vec3(-1, 1, -1),
    vec3(1, 1, 0), vec3(1, -1, 0), vec3(-1, -1, 0), vec3(-1, 1, 0),
    vec3(1, 0, 1), vec3(-1, 0, 1), vec3(1, 0, -1), vec3(-1, 0, -1),
    vec3(0, 1, 1), vec3(0, -1, 1), vec3(0, -1, -1), vec3(0, 1, -1));

float computeShadowMapScalar(vec4 position, vec3 lightOrigin, samplerCube shadowMap)
{
    vec3 positionShadow = position.xyz - lightOrigin;
    float shadowZ = length(positionShadow);
    int shadowHits = 0;
    for (int i = 0; i < SHADOW_SAMPLE_OFFSETS_COUNT; ++i)
    {
        vec3 offset = SHADOW_SAMPLE_OFFSETS[i] * lightShadowSampleScalar;
        bool hit = shadowZ - lightShadowBias > texture(shadowMap, positionShadow + offset).x;
        if (hit) ++shadowHits;
    }
    return 1.0 - float(shadowHits) / SHADOW_SAMPLE_OFFSETS_COUNT;
}

...but the results looked worse than the naive approach we currently use and aren't user-adjustable.

Just figured I'd create a discussion for this as a place to park this code.

[bryanedds]

Maybe this approach will look better alongside dithering? - #935