######################
Installation of Scilab
######################

************
Installation
************
Download linux binary from `https://www.scilab.org/download/scilab-2024.1.0`
and unpack (on lilo)::

  # 01-Aug-2024
  cd /vol/thchem
  tar xf scilab-2024.1.0.bin.x86_64-linux-gnu.tar.xz

Using the Intel Math Kernel Library (MKL)
=========================================
The `Scilab` binary package includes `openblas` for blas and lapack routines.
The MKL is faster and allows parallel computation, but I found few instructions
on the internet on how to link. The procedure described here makes
it possible to switch between different blas/lapack libraries by
setting an environment variable.

Creating MKL shared object with blas and lapack routines
--------------------------------------------------------
Instructions are given in the `Developer Guide for Intel oneAPI Math
Kernel Library Linux <https://www.intel.com/content/www/us/en/docs/onemkl/developer-guide-linux/2024-2/building-custom-dynamic-linked-shared-libraries.html>`_ on building custom shared objects::

  # 01-Aug-2024, see lilo8:/vol/thchem/lib/oneapi/sequential/job1.sh
  # Copy the lists of blas and lapack routines to include in shared object
  rsync -v /vol/opt/intel/oneapi/mkl/2023.2.0/tools/builder/{makefile,blas_example_list,lapack_example_list} .

  # Merge the lists
  cat blas_example_list lapack_example_list > blas_lapack_list

  # Select intel compilers environment

  module add mkl/latest
  # Output:
  # Loading mkl version 2023.2.0
  # Loading tbb version 2021.10.0
  # Loading compiler-rt version 2023.2.1
  # Loading mkl/latest
  #   Loading requirement: tbb/latest compiler-rt/latest

  # Create shared object

  make libintel64 export=blas_lapack_list interface=lp64 threading=sequential name=libblas_lapack_mkl_sequential

  # This creates shared object:

  -rwxr-x--- 1 gerritg thchem 66627032 Aug  2 13:08 libblas_lapack_mkl_.so

  # To make sure this shared object is use for blas and lapack under several
  # possible names set a number of soft links:

  for lib in libblas.so.3 libopenblas.so.0 libopenblas_pthreads.so.0 liblapack.so.3 liblapack.so.3gf
  do
    ln -sf libblas_lapack_mkl_$SP.so "$lib"
  done

  # For the parallel version of MKL use "threading=parallel" (and directory ../oneapi/parallel etc)

Linking MKL to Scilab

The script to start Scilab is in this directory::

  cd /vol/thchem/scilab-2024.1.0/bin

  # Make a copy of the bash script that starts scilab:

  cp scilab scilab-mkl

  # Edit scilab-mkl. Search for "export LD_LIBRARY_PATH"
  # This occur on line 675 and on line 839. Just before line 839 add:

  export LD_LIBRARY_PATH=${SCI_LD_PATH:+$SCI_LD_PATH:}$LD_LIBRARY_PATH

  # Look up "bash shell parameter expansion" for an explanation: if
  # $SCI_LD_PATH is not set, $LD_LIBRARY_PATH is not modified

To run Scilab with the MKL library::

  # SCI_LD_PATH is the directory
  export SCI_LD_PATH=/vol/thchem/lib/oneapi/sequential
  /vol/thchem/scilab-2024.1.0/bin/scilab-mkl

It is also possible to switch between versions using the `runscilab6` bash script,
which is part of the `gsci_lib6` library.

Technical details on the `scilab` command in where it links blas/lapack
-----------------------------------------------------------------------
Understanding these details should not be necessary to use `Scilab`,
but it may be helpful in case of trouble.