What algorithm does MKL use in its implementation of LU factorization? Does it use the old right-looking algorithm in 1995 LAPACK? If so, does it block carefully? Does it use recursive partitioning?
MKL internallyuses severalimplementationsand switches between them (and block sizes used)according to problem type/size and environment(CPU, number of threads, etc) trying to select the best one.
