- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello everyone,
I was wondering whether
mkl_dbsrmm
has a parallel implementation. I have set both OMP_NUM_THREADS
and MKL_NUM_THREADS
to 8 and I checked that number with[fortran] nthr = mkl_domain_get_max_threads( MKL_BLAS )[/fortran]
However, when I check the CPU usage with top it never goes above 100%. I remember seeing higher CPU usage (400%) when using
mkl_dcsrmultd.
We really need the parallellism because we are operating on some 6000x6000 matrices and a single core is really slow.
Thanks a lot,
Ionut
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Ionut,
I do seethe User Guide says it should be threaded, so we can check if that is accurate. The scaling will depend on the sparsity pattern so it's possibleyour case is not optimal.Is performance pegged at 100% or just always below it?
Todd

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page