Perform benchmark traverse of BLAS 3 routines. Use up to max_dim with each routine run at least min_time [s] on nthread.