// From http://http.developer.nvidia.com/GPUGems3/gpugems3_ch40.html

uniform sampler2D source;
uniform sampler2D depth;
uniform vec2 pixel;
layout(r16f) volatile restrict writeonly uniform image2D dest;
uniform float sigma = 5.;

layout (local_size_x = 8, local_size_y = 8) in;

shared float local_src[8][8 + 2 * 8];
shared float local_depth[8][8 + 2 * 8];

void main()
{
    int x = int(gl_LocalInvocationID.x), y = int(gl_LocalInvocationID.y);
    ivec2 iuv = ivec2(gl_GlobalInvocationID.x, gl_GlobalInvocationID.y);
    vec2 uv_m = (iuv - ivec2(0, 8)) * pixel;
    vec2 uv = iuv * pixel;
    vec2 uv_p = (iuv + ivec2(0, 8)) * pixel;

    local_src[x][y] = texture(source, uv_m).x;
    local_depth[x][y] = texture(depth, uv_m).x;
    local_src[x][y + 8] = texture(source, uv).x;
    local_depth[x][y + 8] = texture(depth, uv).x;
    local_src[x][y + 16] = texture(source, uv_p).x;
    local_depth[x][y + 16] = texture(depth, uv_p).x;

    barrier();

    float g0, g1, g2;
    g0 = 1.0 / (sqrt(2.0 * 3.14) * sigma);
    g1 = exp(-0.5 / (sigma * sigma));
    g2 = g1 * g1;
    float sum = local_src[x][y + 8] * g0;
    float pixel_depth = local_depth[x][y + 8];
    g0 *= g1;
    g1 *= g2;
    float tmp_weight, total_weight = g0;
    for (int j = 1; j < 8; j++) {
        tmp_weight = max(0.0, 1.0 - .001 * abs(local_depth[x][y + 8 + j] - pixel_depth));
        sum += local_src[x][y + 8 + j] * g0 * tmp_weight;
        total_weight += g0 * tmp_weight;
        tmp_weight = max(0.0, 1.0 - .001 * abs(local_depth[x][y + 8 - j] - pixel_depth));
        sum += local_src[x][y + 8 - j] * g0 * tmp_weight;
        total_weight += g0 * tmp_weight;
        g0 *= g1;
        g1 *= g2;

    }
    imageStore(dest, iuv, vec4(sum / total_weight));
}