The matrix construction was previously optimized for multi-threaded computation (by blocks of rows) using OpenMP in Ivan's version. Is this disabled, or perhaps, do we need to set the number of threads for OpenMP (as well as for MKL?)
I ran for a fairly large matrix (N=50,000) and CPU utilization stayed mostly at 100% on my Mac.
The matrix construction was previously optimized for multi-threaded computation (by blocks of rows) using OpenMP in Ivan's version. Is this disabled, or perhaps, do we need to set the number of threads for OpenMP (as well as for MKL?)
I ran for a fairly large matrix (N=50,000) and CPU utilization stayed mostly at 100% on my Mac.