Why is processing a sorted array faster than an unsorted array? (stackoverflow.com)

This is one of the classic posts here:

https://hn.algolia.com/?query=Why%20is%20processing%20a%20so...

Very good comments in some of the submissions

  int t = (data[c] - 128) >> 31;
  sum += ~t & data[c];

> They've gone to plaid

For completeness, a) make the array very big (much much bigger than the LLC) b) split the work amongst cores (make it parallel with your favorite methodology)