Librsb

Original author(s)	Michele Martone
Stable release	1.2.0 / September 2016 (2016-09)
Operating system	Cross-platform
Available in	C, C++, Fortran
Type	Software library
License	GPL License
Website	librsb.sourceforge.net

librsb is an open-source parallel library for sparse matrix computations using the Recursive Sparse Blocks (RSB) matrix format.

librsb provides cache efficient multi-threaded Sparse BLAS operations via OpenMP, and is best suited to large sparse matrices.

Features

librsb provides:

• Conversion from/to COO, CSR, CSC sparse matrix formats.
• Support for the four BLAS types.
• Support for general, symmetric, hermitian matrices.
• Parallel threaded, eventually strided:
  • Sparse matrix-vector multiplication.
  • Sparse matrix-dense matrix multiplication.
  • Sparse matrix-vector triangular solve.
  • Sparse matrix-dense matrix triangular solve.
• Sparse matrix-sparse matrix multiplication.
• Elemental sparse matrix operations (scaling, add, etc.).
• Row-wise or column-wise scaling.
• Rows / columns extraction.
• An online empirical autotuning function.
• File input/output in the Matrix Market format.
• Rendering of the RSB structure into EPS (Encapsulated Postscript) figures.
• A program for benchmarking / testing performance.
• Implements the Sparse BLAS standard, as specified in the BLAS Technical Forum. ^[1] documents.

System requirements

librsb can be used from:

• C and C++ (rsb.h interface)
• Fortran 90/95/2003 (module rsb)
• GNU Octave (sparsersb package for GNU Octave) GNU Octave package ^[2]

References

1. BLAS Technical Forum
2. sparsersb: "Interface to the librsb package implementing the RSB sparse matrix format for fast shared-memory sparse matrix computations."

• Martone, M. Efficient multithreaded untransposed, transposed or symmetric sparse matrix-vector multiplication with the Recursive Sparse Blocks format. Parallel Computing 40(7): 251-270 (2014)

External links

• librsb.sf.net
• librsb.sourceforge.net