Content Examples and Tutorial Scope of this Branch Compiler Requirements Git It Philosophy Mailing List

FLENS

FLENS is short for Flexible Library for Efficient Numerical Solutions. And that's exactly what it is:

A builing block for the implementation of other (higher-level) numerical libraries or numerical applications

Some key points

• C++ library (requires a C++11 conform compiler)

• Easy install as FLENS is headers only

• FLENS gives you

  • Matrix/vector types for dense linear algebra

  • Generic implementation of BLAS (see CXXBLAS)

  • Generic reimplementation of LAPACK (see FLENS-LAPACK)

  • Generic implementation means that you can use any element type for your matrices/vectors that “acts like a double”. So in particular types from the QD Library, mpfr, ...

• If high performance BLAS libraries like ATLAS/GotoBLAS/... are available you simply can link against them and boost performance.

• FLENS-BLAS is a high-level and easy-to-use BLAS interface

• FLENS-LAPACK is a high-level and easy-to-use C++ implementation of LAPACK.

• FLENS also allows using overloaded operators for accessing BLAS. Most important in this respect:

  • Overloaded operators come without performance penalty!

  • In a special debug mode you can exactly trace what is happening and how you linear algebra expressions gets evaluated through BLAS calls

  • More features are explained in the tutorial.

Examples and Tutorial

• Overview of all driver functions currently implemented in FLENS-LAPACK

• We just started with a tutorial:

  • Session 1: General matrices and dense vectors.

  • Session 2: BLAS, CXXBLAS, FLENS-BLAS and Overloaded Operators.

  • Session 3: More on matrices: Triangular and symmetric matrices, matrix views, ...

  • Session 4: Using FLENS-LAPACK.

  • Session 5: Using an external LAPACK implementation.

  • Session 6: Define your own matrix/vector types.

  • Session 7: How to define a kick-ass interface for FLENS-LAPACK.

Scope of this Branch

• Only real matrix types with full storage and dense vector types, i.e.

  • GeMatrix,

  • SyMatrix,

  • TrMatrix,

  • DenseVector.

• Our generic BLAS implementation CXXBLAS.

• Support for high and multipecission types from QD Library and mpfr.

• Support for native BLAS implementations like ATLAS or GotoBLAS.

• FLENS-BLAS and FLENS-LAPACK.

• Overloaded operators for BLAS working ontop of the FLENS-BLAS layer.

• We improved our doctool (look at the bottom of each page for document source). We now use restructured text.

Compiler Requirements

FLENS uses some of the C++11 features and therefore requires a recent C++ compiler:

• clang version 3.0 (or higher)

• gcc version 4.7 (or higher)

• icc version 12.1.2 (or higher)

Git It

You can clone a public branch from GitHub

Philosophy

This should answer the following questions:

• Why are the matrix/vector types in FLENS as they are?

• Why do we reimplement the LAPACK library with FLENS?

• What is the idea behind our CXXBLAS?

The answer is simple:

• LAPACK is the most advanced and best supported numerical library in the world. Improvements in numerical algorithms have a great chance to make it soon into LAPACK. If we keep FLENS-LAPACK up-to-date to current LAPACK releases we keep up with some of the most sophisticated numerical algorithms.

• LAPACK gains its performance from highly optimized BLAS kernel routines. These implementations achieve peak performance on certain platforms. When possible our CXXBLAS is just an interface to these implementations.

• For multi-precision data types CXXBLAS gives you a generic implementation that does not give high performance (yet) but allows using LAPACK in high precision.

Mailing List

Join the mailing list!

View document source

© 2012 Michael Lehn