- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
On my platform, the MKL_INT macro must yield an integral type that is binary-compatible with a 32-bit signed int, or 64-bit signed int if I link with the ILP64 libraries [MKL doc]
1) Is there a way to call into both the LP64 and ILP64 versions of MKL Sparse BLAS routines like mkl_dcsrmv(), from the same C++ program?
2) In future releases, would the MKL developers consider allowing users to redefine MKL_INT to different integral types? In my use-case, I have many small sparse matrices whose indices can be represented by 8- and 16-bit integers, so there's potential 2-to-8-fold memory footprint savings (w.r.t. indexing data) versus using the MKL defaults.Thanks! --Nick
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi, these are interesting questions. Please give me a little time to research it with the MKL engineering team. I'd like to get a definitive answer on the feasibility of #2. It may help me with a great answer to question #1.

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