Community
cancel
Showing results for 
Search instead for 
Did you mean: 
zzzhhh
Beginner
194 Views

Does MKL support sparse matrices in LAPACK?

From the LAPACK's FAQ, we know it does not support sparse matrices:

Just to take a chance, does intel MKL break this restriction by introducing sparse matrices in LAPACK? If not, is there any other function domain of MKL that uses sparse matrices? Thanks.

0 Kudos
5 Replies
mecej4
Black Belt
194 Views

Not "in Lapack", but certainly in MKL: see https://software.intel.com/en-us/mkl-developer-reference-fortran-intel-mkl-pardiso-parallel-direct-s... . There is also a legacy DSS interface to Pardiso, but you probably should not use it in new projects.

zzzhhh
Beginner
194 Views

Can you be more specific? I don't understand your reply totally. I use C.

mecej4
Black Belt
194 Views

Lapack is defined, maintained and documented by organizations outside Intel, although Intel may have input to and representation in that group. It would not be appropriate for Intel to "break" Lapack. Sparse matrices (other than band matrices) have never been part of Lapack. It is important to keep the Lapack portions of MKL compatible with the public domain Lapack.

MKL has substantial capabilities that are not in Lapack, so let us put aside the issue of what Lapack does or does not have, and ask what MKL provides for your intended application. The answer is MKL-Pardiso.

The MKL-C documentation of Pardiso is available at https://software.intel.com/en-us/mkl-developer-reference-c-intel-mkl-pardiso-parallel-direct-sparse-....

Gennady_F_Intel
Moderator
194 Views

plus two our cents to what mecej4 said:

1. MKL also provides SpBLAS API ( 6 different Sparse Matrix Storage Formats - see here https://software.intel.com/en-us/mkl-developer-reference-c-sparse-blas-level-2-and-level-3-routines-...

2. MKL v.2019 is aligned with Netlib Lapack NETLIB LAPACK 3.8.0

zzzhhh
Beginner
194 Views

Thank you, mecej4 and gennady. I appreciate so much for your detailed description.