Idempotent partition around median of 5?

[Ola Fosheim Grøstad via Digitalmars-d]

On Friday, 5 February 2016 at 15:41:11 UTC, Fool wrote:
> So we can do with seven moves instead of three swaps (nine 
> moves). ;-)

Yes, well, this all breaks down when you realize that this is all 
completely dominated by cache misses and that you can do median 
more effectively using SIMD instructions. :^P