Intel Corp (INTC)

[mmoy]

This is just fun/hobby stuff for me in trying to speed up a commonly used algorithm. I could split things up and do things piecemeal but I was hoping that AVX-2 would be robust enough so that I didn't have to do that. I had to do that with SSE2 as well and you do pay a penalty for doing that. They seem to have unlimited transistors these days and it wouldn't have been hard to make some of this stuff more convenient.

AVX-512 would be very nice as it would allow me to replace an in-memory table with a vector which should allow for a slightly faster lookup.