Next: Installation of Nero2d, Previous: Installation on GNU/Linux, Up: Installation on GNU/Linux
Nero2d is designed to be a performant simulator tool. Internally, it depends on a few high-performance libraries which are widely available. As most hardware vendors have their own optimized implementations of these libraries, we strongly recommend using them. It is up to you to find suitable versions which provide you with the maximum performance for your system. The required performance libraries are FFTW, BLAS and LAPACK. In the next sections, these are discussed in detail.
FFTWNero2d requires the FFTW library, or the Fastest Fourier Transform of the West, version 3.0 or higher. The library is likely to be present on your system already, if not, it can be installed using your package manager. If you use yum to install software (Fedora, Redhat, etc.), you will need to install the fftw, fftw-devel and optionally the fftw-static packages. If you use aptitude to install software (Debian, Ubuntu, etc.), you will need the fftw3 and fftw3-dev packages.
You can also opt to compile FFTW from source: it can be freely downloaded at www.fftw.org. Nero2d requires the installation of the double precision, non-threaded version of FFTW (the default installation). It is recommended to do a global install of FFTW i.e. install FFTW in the standard location (usually /usr/local) so that it is accessible to all users. This will however, require root access to your system. If you do a local installation (somewhere in your home directory) of the FFTW library, Nero2d provides a configuration flag to locate it (see Configure options for Nero2d). For a detailed refernce on the FFTW library, please consult its manual.
BLAS and LAPACKNero2d also requires the BLAS library, or the Basic Linear Algebra Subprograms, and the LAPACK, or the Linear Algebra PACKage. Most vendors provide optimized libraries for free and you should try to use these in the first place. The configure script can detect a great number of BLAS and LAPACK implementations. If everything fails however, and no BLAS or LAPACK are found on your system, Nero2d can compile the libraries ifself. This can (and most likely will) result in a poor performance.
Some packages install BLAS and/or LAPACK libraries in a non-standard location: Nero2d provides configuration flags to locate them (see Configure options for Nero2d). We specifically support two vendor specific implementations: the ACML from AMD and the MKL from Intel.
ACML
For people with AMD based systems, the ACML or AMD Core Math Library includes both the BLAS and the LAPACK routines. It can be freely downloaded (after registration) at http://developer.amd.com. Nero2d's configure script has built-in support for the ACML and will detect them if they are installed. As a default, ACML is installed in the /opt/acmlX.X.X/ directory and the libraries are inside /opt/acmlX.X.X/F77/lib (where F77 is the fortran compiler used to generate the ACML). This is a non-default path for libraries and you will need to point out this location to the configure script:
./configure --with-blas-dir=/opt/acmlX.X.X/F77/lib
Note that if you want to link against dynamic libraries (.so), you will first need to set the the LD_LIBRARY_PATH environment variable before running the configure script and before invoking the nero2d binary.
export LD_LIBRARY_PATH=/opt/acmlX.X.X/F77/lib:$LD_LIBRARY_PATH
or
setenv LD_LIBRARY_PATH "/opt/acmlX.X.X/F77/lib:$LD_LIBRARY_PATH"
You might want to automate this process by adding this line to the .profile file.
Optionally, you can copy the content from /opt/acmlX.X.X/F77/lib to /usr/lib or /usr/local/lib. In this case, there is no need to rename the libraries to libblas and/or liblapack and there is no need to set any environment variables.
Note that ACML fully supports the --enable-blasfunc if you use the same Fortran compiler to build Nero2d as was used to build ACML (see Configure options for Nero2d).
Please refer to the ACML manual for (/opt/acmlX.X.X/Doc/acml.pdf) for a full reference.
MKL
For people with Intel based systems, the MKL or Math Kernel Library includes both the BLAS and the LAPACK routines. It can be downloaded (free trial download) at http://www.intel.com. Nero2d's configure script has built-in support for the MKL and will detect them if they are installed. As a default, MKL is installed in the /opt/intel/mkl/X.X.X directory and the libraries are inside /opt/intel/mkl/X.X.X/lib/ARCH (where ARCH is 32, 64 or em64t depending on the processor your are using). This is a non-default path for libraries and you will need to point out this location to the configure script:
./configure --with-blas-dir=/opt/intel/mkl/X.X.X/lib/ARCH
Note that if you want to link against dynamic libraries (.so), you will first need to set the the LD_LIBRARY_PATH environment variable before running the configure script and before invoking the nero2d binary.
export LD_LIBRARY_PATH=/opt/intel/mkl/.../ARCH:$LD_LIBRARY_PATH
or
setenv LD_LIBRARY_PATH "/opt/intel/mkl/.../ARCH:$LD_LIBRARY_PATH"
Note that the shell scripts in /opt/intel/mkl/VERSION/tools/environment can take care of this for you. You might want to automate this process by adding this line to the .profile file.
Optionally, you can copy the content from /opt/intel/mkl/VERSION/lib/ARCH to /usr/lib or /usr/local/lib. In this case, there is no need to rename the libraries to libblas and/or liblapack and there is no need to set any environment variables.
Note that MKL fully supports the --enable-blasfunc if you use the same Fortran compiler to build Nero2d as was used to build MKL. In practice, we have noted (in version 10.0.1) that MKL supports the g77 Fortran calling conventions and not those of gfortran. You are advised to compile Nero2d using the g77 compiler if you can. However, it appears that the --enable-blasfunc is fully compatible with the combination of gfortran and MKL. The configure script will issue a warning, but on our test system, everything went smooth. Be carefull however that this might just be luck.
BLAS/LAPACK (not recommended)
If no BLAS and/or LAPACK are found on your system, Nero2d will compile these libraries from source. This can however, result in a poor performance. Please note that the source files inside the directory src/zofpack/machcons need to be compiled without any sort of optimization. This is forced by overriding the $FFLAGS variable and setting them to empty for that directory. If you should have a Fortran compiler with default optimization turned on, you will need to edit src/zofpack/machcons/Makefile.am manually and set the $FFLAGS to disable this optimization. You will need to rerun automake first to regenerate the Makefiles.