#version 440

layout(local_size_x = 16, local_size_y = 16) in;
layout(rgba16f, binding = 1) readonly uniform image2D inputImage;
layout(rgba16f, binding = 2) writeonly uniform image2D outputImage;

// There is no equivalent of gl_NumWorkGroups in HLSL. So instead pass the
// values in in a uniform buffer.
layout(std140, binding = 0) uniform numWorkGroupsBuf {
    uvec3 numWorkGroups;
};

int wrapMod( in int a, in int base )
{
    return ( a >= 0 ) ? a % base : -(a % base) + base;
}

void getWrappedCoords( inout int sX, inout int sY, in int width, in int height )
{
    if (sY < 0) { sX -= width >> 1; sY = -sY; }
    if (sY >= height) { sX += width >> 1; sY = height - sY; }
    sX = wrapMod( sX, width );
}

void main()
{
    int prevWidth = int(numWorkGroups.x) << 1;
    int prevHeight = int(numWorkGroups.y) << 1;
    if (gl_GlobalInvocationID.x >= numWorkGroups.x || gl_GlobalInvocationID.y >= numWorkGroups.y)
        return;
    vec4 accumVal = vec4(0.0);
    for (int sy = -2; sy <= 2; ++sy) {
        for (int sx = -2; sx <= 2; ++sx) {
            int sampleX = sx + (int(gl_GlobalInvocationID.x) << 1);
            int sampleY = sy + (int(gl_GlobalInvocationID.y) << 1);
            getWrappedCoords(sampleX, sampleY, prevWidth, prevHeight);
            if ((sampleY * prevWidth + sampleX) < 0 )
                sampleY = prevHeight + sampleY;
            ivec2 pos = ivec2(sampleX, sampleY);
            vec4 value = imageLoad(inputImage, pos);
            float filterPdf = 1.0 / ( 1.0 + float(sx*sx + sy*sy)*2.0 );
            filterPdf /= 4.71238898;
            accumVal[0] += filterPdf * value.r;
            accumVal[1] += filterPdf * value.g;
            accumVal[2] += filterPdf * value.b;
            accumVal[3] += filterPdf * value.a;
        }
    }
    imageStore(outputImage, ivec2(gl_GlobalInvocationID.xy), accumVal);
}