Can matlab coder generate code that uses a specified external uBLAS linear algebra library included with BOOST?

Can the matlab coder application be used to generate C/C++ code that works with the C/C++ uBLAS linear algebra library and matrix data types included with BOOST ?

If so, Can you please provide some examples of code generation integration with support for BOOST and uBLAS? If not, Can you please provide tips/recommendations for BOOST and uBLAS support?

I'm open to exploring other alternatives such as the following:

• Use the numeric binding adaptor for LPACK to UBLAS as indicated at: https://github.com/uBLAS/numeric_bindings How do I configure the matlab coder to support LAPACK interface? Is there any examples for this?
• Use the open source Octave https://www.gnu.org/software/octave/ application to do the Matlab code to C++ conversion which claims to have support for UBLAS?
• Use https://github.com/jonathf/matlab2cpp with Armadillo wrappers for UBLAS
• Use https://github.com/rboorgapally/boost-ublas-extensions with UBLAS to hand convert the matlab code

Answer 1

[Ryan Livingston]

Update Starting in R2018b you can generate code that calls a CBLAS as well as an LAPACK library:

https://www.mathworks.com/help/coder/ug/speed-up-matrix-operations-in-generated-standalone-code-by-using-blas-calls.html

Starting in R2016a, MATLAB Coder can generate code that calls an LAPACK library using the LAPACKE interface:

https://www.mathworks.com/help/coder/ug/generate-code-that-calls-lapack-functions.html?searchHighlight=lapacke&s_tid=doc_srchtitle

That will allow you to leverage an optimized LAPACK library for high-level linear algebra operations like solving systems, doing matrix decompositions, etc. There are various LAPACK implementations available for different hardware. You can also get an optimized BLAS library and then use the reference LAPACK from netlib.org and point it to your optimized BLAS.

OpenBLAS provides an optimized BLAS and partially optimized LAPACK library for several platforms:

https://www.openblas.net/

ATLAS provides automatically tuned BLAS and a few LAPACK routines for multiple platforms:

http://math-atlas.sourceforge.net/

And there are various other optimized libraries available which may or may not be free:

http://www.netlib.org/lapack/faq.html#_what_and_where_are_the_lapack_vendors_implementations

There are also many other optimized BLAS libraries available including:

https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Implementations

These could be linked with the Netlib reference LAPACK/LAPACKE:

http://www.netlib.org/lapack/

Also note that some of these libraries may be available in system package managers like apt, yum, homebrew, etc.

Question

MATLAB Coder does not support generating code which uses the uBLAS functions or types. Could you please expand on why you chose uBLAS? Is performance your main concern? Style of the generated code? Something else?

[Royi Avital]

What about linking to Intel MKL (As it is shipped with MATLAB)?

[Mats Larsson]

Does this work also for Sparse matrices ? E.g. MKL has BLAS functions for sparse matrices.