Alternate bayer pattern generation

[torridgristle]

Years ago (so things might be different now) I made a PS1 bayer dither shader. At first I used an array of 16 ints for the 4x4 pattern, but then I tried recreating it from a set of booleans and modulos which looked more complicated, but ran faster (288 FPS with array, ~700 FPS with the boolean modulo mess). Here's the code if you're interested.

PixCoord = texCoord * sourceSize[0].xy - 0.5; // Turning -1,1 range to pixel range, not centered on the pixels though.

    float Modx4 = float(PixCoord.x - floor(PixCoord.x * 0.25) * 4.0 > 1.0);
    float Mody4 = float(PixCoord.y - floor(PixCoord.y * 0.25) * 4.0 > 1.0);
    float Modx2 = PixCoord.x - floor(PixCoord.x*0.5)*2.0;
    float Mody2 = PixCoord.y - floor(PixCoord.y*0.5)*2.0;

    float a = Mody2*6;
          a = round(a - a*1.333333333333333 * Modx2);

    float b = Mody4-4+Modx2*4;

    float c = Modx4 - Modx4 * Mody4 * 2;

    float Pattern = a+b+c;

AMD's GPU ShaderAnalyzer said it was somehow less instructions and cycles, I can't explain it.