Anatomy of High-Performance Many-Threaded Matrix Multiplication