ParentFull threaddavedx·> This is indeed twice as fast as the vectorized implementation, but, disappointingly, the naive implementation with loops is even faster.On CPU or GPU?View on HN