I have a hunch that the actually fastest version is to change all those "||" to "|" in the original version. Short-circuiting and adding branches probably just

I have a hunch that the actually fastest version is to change all those "||" to "|" in the original version. Short-circuiting and adding branches probably just slows the whole thing down compared to just running the comparisons, which will probably pipeline really well. Interestingly the compiler doesn't do this optimization, even though it could. So maybe I'm wrong. No real way to know without benchmarking, though.