Installation¶
The code should be installed on Linux systems, on which it is developped and tested.
A few users have installed AMITEX_FFTP successfully on macOS: see macOS specificity.
For WINDOWS 10, the better way is to use the WSL (Windows Subsystem for Linux) that allows for the installation of Linux distributions: see Windows10-WSL specificity.
Installation on WINDOWS has also been tested once with CYGWIN (https://www.cygwin.com/)… try it if you have time ;-)
Prerequisites¶
If possible, install GCC, opemMPI and FFTW with package manager (such as yum, apt… depending on you linux distribution).
If really not possible, you can install them by your own in your home directory (see GCC, openMPI, fftw, mfront).
1. Fortran compiler : Intel 15.0.0 (minimum tested) or GCC > 5.0.0
The program (v8.17.8) has been tested successfully with Intel 19 and GCC 9.3.0.
2. MPI library : The program has been tested successfully with at least : Intel MPI 5.0.1 (version 2015) and version 2019 as well as Open MPI 2.0.4, 3.1.0, 4.0.3
Warning
- Observations made on intel-based clusters
Performances obtained with [Intel compiler + Intel MPI] are significantly better than with [GCC compiler + Open MPI].
While waiting for a clarification, use [Intel compiler + Intel MPI] if possible.
3. FFT libraries :
As mentioned, AMITEX_FFTP comes with the 2decomp&fft library which provides interfaces with various FFT libraries.
AMITEX_FFTP has been tested successfully with libraries fftw, versions 3.3.3 and 3.3.4. If 2decomp&fft offers the possibility to choose different FFT engines (acml, ffte, fftpack5, mkl), only fftw is used for the development of AMITEX_FFTP.
Note that fftw (http://www.fftw.org/) must be installed with both double (default) and simple precision.
4. The MFront library (optional)
The MFront library (http://tfel.sourceforge.net/documentations.html) is useful to easily develop behavior laws. These laws are available in a dynamic library and are based on the standard UMAT interface which is compatible with AMITEX_FFTP.
The MFront library should be installed and compiled with the same compiler than the one used to compile AMITEX_FFTP.
Installation guide¶
INSTALL!
The installation is straightforward : adust the install file and launch it.
Then the executable program amitex_fftp is located in (amitex_fftp)/libAmitex/bin folder
1. Adjust the following first lines of the install file located in the main folder (amitex_fftp):
# Compiler : ifort, gfortran
FC=ifort
# FC=gfortran
# MPI compiler
MPIFC=mpif90
#MPIFC=mpiifort
# FFTW Library
# fftw3 - FFTW version 3.x
# fftw3_f03 - FFTW 3.3-beta1 or later (with Fortran 2003 interface)
# generic - A general FFT algorithm (no 3rd-party library needed)
FFT=fftw3_f03
# FFTW library path
FFT_inc=/usr/include
FFT_lib=/usr/lib64
# OPTIMIZATION - TO BE ADJUSTED ONLY IN CASE OF TROUBLE AT EXECUTION
OPTIM_2decomp=-O3 #--- use OPTIM_2decomp=-O2 on ubuntu 20.04 (with native gcc9.3.0) or opensuse tumbleweed (with gcc10)
OPTIM_amitex=-O3
The user must choose :
- the compiler: ifort (Intel) or gfortran (GCC) (prefer the Intel compiler if available on your system),
- the wrapper compiler for the compilation with MPI : mpif90 or mpiifort
- the FFT library and inform the corresponding library path. Prefer fftw_f03 used during the development.
- ONLY IN CASE OF TROUBLE, the optimization level (-O3 by default). For example on ubuntu 20.04 or opensuse tumbleweed, the code fails at execution . Solution is to set OPTIM_2decomp=-O2 (instead of -O3).
2. Check your paths for FFTW3
Check your path to fftw3 includes, FFT_inc. You should find at least the following files : fftw3.f, fftw3.f03, fftw3.h`
Check your path to fftw3 libraries, FFT_lib. You should find at least the two dynamic libraries : libfftw3f.so, libfftw3.so
3. Launch install:
./install
In practice, ./install :
- launch the script clean (to clean the directories if a first installation was run before)
- fills correctly (amitex_fftp)/lib_extern/2decomp_fft-1.5.8472/src/Makefile.inc and various Makefiles and scripts,
- compiles the 2decomp&fft library,
- compiles the FoX library,
- compiles AMITEX_FFTP,
- builds the executable program: the file amitex_fftp in the (amitex_fftp)/libAmitex/bin folder.
Warning
If the optional MFront library is not installed, the compilation of (amitex_fftp)/cas_tests/comportements/mazars and (amitex_fftp)/cas_tests/comportements/mgurtin_hpp_glissement will fail, obviously.
If you do not wish to use MFront do not pay attention to this error.
Validation¶
The file script_tests.sh within the folder (amitex)/validation provides a list of simulations based on geometries, loadings, algorithms and material properties defined in xml files respectively in directories microstructures, chargements, algorithmes and materiaux under (amitex_fftp)/cas_tests.
The script script_tests.sh can be run either interactively ./script_tests.sh or in batch mode. In this cas, different batch scripts for different batch schedulers (Torque, LoadLeveler…) are provided. Of course, you will have to create your own batch script in agreement with the batch scheduler available on your platform.
Finally, script_test.sh compares simulations results to reference results stored within each (amitex)/resultats/simulation_name/reference folder and give the non-regression comparison in the file tests.log and total time in time.log.
GCC, openMPI, fftw, mfront¶
IF REALLY IMPOSSIBLE to install GCC, openMPI, fftw with a package manager, try the following installation procedures in you home directory.
For GCC
wget ftp://ftp.irisa.fr/pub/mirrors/gcc.gnu.org/gcc/releases/gcc-8.3.0/gcc-8.3.0.tar.gz
tar xvfz gcc-8.3.0.tar.gz
cd gcc-8.3.0
./contrib/download_prerequisites
mkdir build
cd build
../configure --enable-languages=c,c++,fortran --disable-multilib --prefix=<PATH TO NEW VERSION>
make -j 4 # distributed compilation on 4 process
make install
Then you must adjust the PATH and LD_LIBRARY_PATH, for example in a file env_gcc8.3.0.sh
#!/bin/bash
export PATH=<PATH TO NEW GCC VERSION>/bin:$PATH
export LD_LIBRARY_PATH=<PATH TO NEW GCC VERSION>/lib64:$LD_LIBRARY_PATH
To use te compiler
source env_gcc8.3.0.sh
gfortran -v #to check the version
For openMPI
See https://www.open-mpi.org/faq/.
Below an example of standard (no specific options) installation of openMPI
gunzip -c openmpi-4.0.2.tar.gz | tar xf -
cd openmpi-4.0.2
./configure --prefix=<PATH TO NEW VERSION>
make all install
Then you must adjust the PATH and LD_LIBRARY_PATH, for example in a file env_openmpi.sh
#!/bin/bash
export PATH=<PATH TO NEW OPENMPI VERSION>/bin:$PATH
export LD_LIBRARY_PATH=<PATH TO NEW OPENMPI VERSION>/lib:$LD_LIBRARY_PATH
To use the compiler
source env_openmpi.sh
mpif90 -v #to check the version
For FFTW
See http://www.fftw.org/fftw3_doc/Installation-on-Unix.html.
Below an example of standard (no specific options) installation of FFTW with intel compiler (single and double precisions required for AMITEX_FFTP) in the main folder
#install double-precision
./configure CC=icc --prefix=<PATH TO NEW FFTW VERSION>
#remove CC=icc for install with gcc instead of intel
make
make install
#install single-precision
./configure CC=icc --prefix=<PATH TO NEW FFTW VERSION> --enable-float
#remove CC=icc for install with gcc instead of intel
make
make install
Then you must adjust the PATH and LD_LIBRARY_PATH, for example in a file env_fftw.sh
#!/bin/bash
export PATH=<PATH TO NEW FFTW VERSION>/bin:$PATH
export LD_LIBRARY_PATH=<PATH TO NEW FFTW VERSION>/lib:$LD_LIBRARY_PATH
For MFRONT
Download the code at https://sourceforge.net/projects/tfel/files/
For a complete and up-to-date description refer to the manual page http://tfel.sourceforge.net/install.html
Below is a summary that should work with you native gcc compiler or with intel compiler (choose the correct lines)
mkdir tfel-3.2.1-gcc # for installation with gcc (native gcc compiler)
mkdir tfel-3.2.1-intel # for installation with intel
tar -xf tfel-3.2.1.tar.bz2
cd tfel-3.2.1
mkdir build
cd build
#for native gcc
cmake ../ -DCMAKE_BUILD_TYPE=Release -Dlocal-castem-header=ON -Denable-fortran=ON -Denable-aster=ON -DCMAKE_INSTALL_PREFIX=path_to_tfel3.2.1-gcc & >cmake.log
#for intel
CXX=icpc CC=icc FC=ifort F77=ifort cmake ../ -DCMAKE_BUILD_TYPE=Release -Dlocal-castem-header=ON -Denable-fortran=ON -Denable-aster=ON -DCMAKE_ INSTALL_PREFIX=path_to_tfel-3.2.1-intel
make -j 8 # 8 is the number of core for a parallel compilation
make check -j 8 # launch tests
make install # finalize installation in tfel-3.2.1-gcc or tfel-3.2.1-intel
# For a non-native gcc compiler : add CXX=g++ CC=gcc FC=gfortran F77=gfortran
# to the cmake command line above
Then you must adjust the PATH and LD_LIBRARY_PATH, for example in a file env_mfront.sh
#!/bin/bash
export PATH=<PATH TO NEW TFEL VERSION>/bin:$PATH
export LD_LIBRARY_PATH=<PATH TO TFEL VERSION>/lib:$LD_LIBRARY_PATH
macOS specificity¶
macOS installation is not tested by developers but AMITEX_FFTP has been successfully installed by different users on different configurations.
An example of successfull configuration is, on a Catalina-OS (01/2020):
- Xcode + command line tools,
- gcc9,
- open-mpi for gcc9,
- FFTW 3.3.8.
A problem comes with macOS because the commande ‘sed’ used in the scripts install and clean, has not the same behavior on macOS and Linux.
To overcome this issue :
install ‘gsed’ on macOS (sudo port install gsed)
Adjust the variable SED in scripts install AND clean so that ‘gsed’ is used instead of ‘sed’
#------- For MAC USERS : uncomment #SED=gsed SED=sed SED=gsed
Once modified install AND clean, the installation procedure is the same as described in Installation guide.
Windows10-WSL specificity¶
Windows Subsystem for Linux allows to run Linux distributions directly within your Windows10 environement. It has been used successfully to install AMITEX_FFTP on a Windows10 PC through a Ubuntu 16.04 installed with WSL. On this Linux Windows10-WSL/Ubuntu16.04 (or Ubuntu20.04) the prerequesites are fullfiled by
#COMPILERS
sudo apt-get install gcc
sudo apt-get install gfortran
sudo apt-get install g++
#MAKE
sudo apt-get install make
#FFTW
sudo apt-get install fftw3
sudo apt-get install fftw3-dev
#OPENMPI
sudo apt-get install libopenmpi-dev
The corresponding paths for fftw to be adjusted in the install script are
# On WSL/UBUNTU
FFT_inc=/usr/include
FFT_lib=/usr/lib/x86_64-linux-gnu
Warning
On Ubuntu 20.04, the code is compiled without any error but fails at execution (when running validation tests for example).
To overcome this problem, the variable OPTIM_2decomp in ‘install’ must be set to -O2 instead of -O3 (see Installation guide).