On Thursday, 7 April 2016 at 13:09:22 UTC, Andrei Alexandrescu
wrote:
This is a good start but we'd need a more principled attack on
the problem. There are optimal sorting networks for a number of
small sizes; a good start is Knuth's TAoCP Volume 3 but there's
newer research as well, which needs to be investigated. A
sorting network in D can be nicely done with variadic template
parameters, e.g. this:
Could these fixed-length overloads be reused in the deepest
recursions of existing Phobos algorithms, such as
std.algorithm.sorting.sort() provided that the
comparison-predicate matches.