> Any notion of why the change from bool to Word8Array is so incredibly costly? > I realize that sieve is a pretty idio-syncratic benchmark, but I would have > thought that it wouldn't make that much of a difference. > ... > Ah, I get it: it must be that the factor of 8 in space caused us to be too > big for some L1 cache, right? Nope. I have no idea. Bool arrays in mlton use 32 bits per entry.