Efficient Parallel Nonnegative Least Squares on Multicore Architectures