Thank you! I had no idea that this had such communities either, and I've now got some links to check out. I was just picking a small problem because I'm job hunting and had some time to show some of what I can do - for this problem I could give decent coverage of some of the optimization possibilities without writing a whole book.

I would fit far more naturally into the second community, as I will normally avoid using fixed point wherever possible. I only went with doubles because it made it so much easier to highlight what can be done with the pipelining.

Part of the attraction of floating point is that a lot more CPU silicon goes into floating point than integer - so for instance AVX2 supports multiplication of 64 bit doubles but not 64 bit integers. Before this post, the last time I looked at the Mandelbrot Set was on a PowerPC750 - the best I could do with 32 bit integers was 22 clock cycles, the best I could do with floating point was (I think) 11 clock cycles for doubles, 9 clock cycles for floats, simply because floating point pipelined and integers didn't.