#ifdef ps_main17
PS_OUTPUT ps_main17(PS_INPUT input)
{
    PS_OUTPUT output;

    // Potential speed optimization. There is a high probability that
    // game only want to extract a single channel (blue). It will allow
    // to remove most of the conditional operation and yield a +2/3 fps
    // boost on MGS3
    //
    // Hypothesis wrong in Prince of Persia ... Seriously WTF !
    //#define ONLY_BLUE;

    // Convert a RGBA texture into a 8 bits packed texture
    // Input column: 8x2 RGBA pixels
    // 0: 8 RGBA
    // 1: 8 RGBA
    // Output column: 16x4 Index pixels
    // 0: 8 R | 8 B
    // 1: 8 R | 8 B
    // 2: 8 G | 8 A
    // 3: 8 G | 8 A
    float c;

    uint2 sel = uint2(input.p.xy) % uint2(16u, 16u);
    int2  tb  = ((int2(input.p.xy) & ~int2(15, 3)) >> 1);

    int ty   = tb.y | (int(input.p.y) & 1);
    int txN  = tb.x | (int(input.p.x) & 7);
    int txH  = tb.x | ((int(input.p.x) + 4) & 7);

    txN *= ScalingFactor.x;
    txH *= ScalingFactor.x;
    ty  *= ScalingFactor.y;

    // TODO investigate texture gather
    float4 cN = Texture.Sample(TextureSampler, int2(txN, ty), 0);
    float4 cH = Texture.Sample(TextureSampler, int2(txH, ty), 0);


    if ((sel.y & 4u) == 0u) {
        // Column 0 and 2
#ifdef ONLY_BLUE
        c = cN.b;
#else
        if ((sel.y & 3u) < 2u) {
            // first 2 lines of the col
            if (sel.x < 8u)
                c = cN.r;
            else
                c = cN.b;
        } else {
            if (sel.x < 8u)
                c = cH.g;
            else
                c = cH.a;
        }
#endif
    } else {
#ifdef ONLY_BLUE
        c = cH.b;
#else
        // Column 1 and 3
        if ((sel.y & 3u) < 2u) {
            // first 2 lines of the col
            if (sel.x < 8u)
                c = cH.r;
            else
                c = cH.b;
        } else {
            if (sel.x < 8u)
                c = cN.g;
            else
                c = cN.a;
        }
#endif
    }

    output.c = (float4)(c); // divide by something here?
}
#endif