@hackage jalla0.2

Higher level functions for linear algebra. Wraps BLAS and LAPACKE.

Jalla aims at providing high level functions for linear algebra computations, which should be fast and easy enough to use. It also contains a Haskell wrapper for cblas and lapacke, a C interface for LAPACK written by Intel. Jalla is written and maintained by Christian Gosch (github at goschs dot de) and was started as a little project to learn Haskell.

Jalla (probably) needs ghc, it has not been tested with other Haskell implementations. It is developed with ghc > 7.0.

You need to have CBLAS and LAPACK installed, e.g. by installing ATLAS generated libraries on you computer. In Ubuntu, there are packages for CBLAS and LAPACK which you can install with apt-get, synaptic, or any other such tool. Lapacke must be compiled "by hand".

  1. Get the jalla sources; say they are located in the directory "jalla/"
  2. If you do not have LAPACKE installed, follow the instructions below under "Getting and building LAPACKE".
  3. In jalla/, call cabal configure cabal build cabal haddock
  4. Look at the generated documentation, which should be quite informative now.

Please contact the author at github at goschs dot de if anything does not work!

Using Arch Linux

On a current Arch Linux, you just need to install the lapack package, and you should be good to go. LAPACKE is included in that package.

Getting and building LAPACKE

LAPACKE is part of LAPACK since version 3.4. If you have that version, use that. Otherwise:

The following steps of getting and building LAPACKE can be done with the script get_lapacke.sh. You must have wget and scons installed on your system. From jalla's top level directory, call ./get_lapacke.sh This should download lapacke and build it with our SConstruct file, in the sub-directory ./lapacke. You can source the file set_env.bash if you are using Bash, in order to set the LD_LIBRARY_PATH and LIBRARY_PATH variables.

Note that you might as well get the newest sources of LAPACK from netlib, and copy the lapacke subdirectory from there over to the jalla directory. Both ways worked for me.

If you want to get and build manually:

  1. Get Lapacke at http://www.netlib.org/lapack/lapacke.tgz or at http://www.netlib.org/lapack/#_standard_c_language_apis_for_lapack

  2. Unpack lapacke into "./lapacke" in the top jalla directory.

  3. (Follow lapacke's instructions to build) /OR/ use the shiny new SCons file provided to build lapacke for use with jalla, which can be found in the lapacke/ sub-directory in the jalla source code.

    Note that now, you can also let the Setup.hs script do the work right before configuring: runhaskell Setup.hs configure --user --flags="build_lapacke" should build lapacke right before configuring jalla.

    If you don't use the provided SCons file, note that Jalla expects the LAPACKE_... naming scheme, i.e. LAPACKE in upper case, the rest in lower case. That scheme is called LAPACK_NAME_PATTERN_MC in the LAPACKE make.inc file. Jalla also expects the define LAPACK_COMPLEX_STRUCTURE in the same make.inc. This means that complex numbers are represented as C struct.

Documentation

The haddock documentation is not complete, but not so bad anymore. I will add more. The most important things are documented.

Status

The code may change in some places, and will definitely be growing as soon as I need more functionality. Please let me know if you have suggestions.

Please also tell me if you have a greater interest in this project and want to participate!

License

See the file LICENSE.