halation.dctl

// Halation v22.09.7
//
// https://github.com/hotgluebanjo
//
// THIS DCTL REQUIRES OPEN DOMAIN SCENE-LINEAR INPUT VALUES.

// Size of the falloff filter. A higher value results in a larger pixel radius which will be slower.
DEFINE_UI_PARAMS(SIZE, Size, DCTLUI_SLIDER_FLOAT, 1.0, 0.0, 5.0, 0.001)

// How much light passes back through the green layer.
DEFINE_UI_PARAMS(COLOR, Color, DCTLUI_SLIDER_FLOAT, 0.15, 0.0, 0.5, 0.001)

// Method to use. Options are:
// - Proosa: https://liftgammagain.com/forum/index.php?threads/halation-and-gate-weave.13056/page-5#post-145519
// - Minuth: https://liftgammagain.com/forum/index.php?threads/halation-and-gate-weave.13056/page-6#post-149147
DEFINE_UI_PARAMS(METHOD, Method, DCTLUI_COMBO_BOX, 0, {PROOSA, MINUTH}, {Proosa, Minuth})

__DEVICE__ inline float3 _sample(__TEXTURE__ p_TexR, __TEXTURE__ p_TexG, __TEXTURE__ p_TexB, int p_X, int p_Y) {
    return make_float3(
        _tex2D(p_TexR, p_X, p_Y),
        _tex2D(p_TexG, p_X, p_Y),
        _tex2D(p_TexB, p_X, p_Y)
    );
}

__DEVICE__ inline float exp_k(float x, float y, float r0) {
    return _expf(-_sqrtf(x*x + y*y) / r0);
}

// There doesn't seem to be a way to set a pixel on a texture, so can't get faster.
// Pretty sure this kernel isn't separable anyway.
__DEVICE__ float3 exp_blur(__TEXTURE__ p_TexR, __TEXTURE__ p_TexG, __TEXTURE__ p_TexB, int p_X, int p_Y, float size) {
    // Radius is just large enough to capture most of the falloff.
    int radius = _ceil(4.5f * size);
    float3 sum = make_float3(0.0f, 0.0f, 0.0f);
    float weight_sum = 0.0f;

    for (int i = -radius; i <= radius; i += 1) {
        for (int j = -radius; j <= radius; j += 1) {
            float weight = exp_k(i, j, size);
            sum += weight * _sample(p_TexR, p_TexG, p_TexB, p_X + i, p_Y + j);
            weight_sum += weight;
        }
    }

    return sum / weight_sum;
}

__DEVICE__ float3 transform(int p_Width, int p_Height, int p_X, int p_Y, __TEXTURE__ p_TexR, __TEXTURE__ p_TexG, __TEXTURE__ p_TexB) {
    float3 rgb = _sample(p_TexR, p_TexG, p_TexB, p_X, p_Y);

    if (SIZE < 1e-10) return rgb;

    float3 blurred = exp_blur(p_TexR, p_TexG, p_TexB, p_X, p_Y, SIZE);
    float3 result, color, halated;

    switch (METHOD) {
    case PROOSA:
        color = make_float3(1.0f, COLOR, 0.0f);
        halated = rgb + (blurred * color);
        result = halated / (color + 1.0f);
        break;
    // Approximation of what the Texture Blend blend mode is doing.
    // Just frequency separation.
    case MINUTH:
        color = make_float3(0.7f, 1.0f - COLOR / 3.0f, 1.0f);
        halated = (rgb - blurred) * color;
        result = halated + blurred;
    }

    return result;
}