Performance: strided sum of iterator (and rayon)

[scottmcm]

@grothesque

We can speed up the above program by performing strided iteration. For example, we could sum even and odd items into separate accumulators and only sum the two accumulators at the end.

The word you're looking for here is "vectorization".

See point 4 in the following post for why this usually doesn't happen automatically with f64:

Understanding Rusts Auto-Vectorization and Methods for speed increase

Don't use &Vec<T> parameters; use &[T] instead. It's just a pure win -- less indirection, more flexible for your callers, and doesn't actually keep you from doing anything. Use "reslicing" to make it as obvious as possible to LLVM that your array indexes are in-bounds, if you're going to use indexes into multiple slices. For example, this emits two bounds checks every iteration, because it needs to write the correct prefix of z and panic in the correct place if x or y is actually short…