Difference between revisions of "インストールと実行"

From salmon
Jump to: navigation, search
(Created page with "== 必要条件 == このガイドでは、ユーザはUnixとそのコマンドライン操作について基礎的な知識を有していることを仮定します。 SAL...")
(No difference)

Revision as of 10:55, 6 January 2018

必要条件

このガイドでは、ユーザはUnixとそのコマンドライン操作について基礎的な知識を有していることを仮定します。 SALMONのインストールにあたり、以下のパッケージが必要です。

  • Fortran90/Cコンパイラ。このパッケージは、ユーザが以下のいずれかのコンパイラを利用できることを仮定します。
    • GCC (Gnu Compiler Collection)
    • Intel Fortran/C Compiler
    • Fujitsu Compiler (at FX100 / K-Computer)
  • 以下の線形代数パッケージのいずれか。
    • BLAS/LAPACK
    • Intel Math Kernel Library (MKL)
    • Fujitsu Scientific Subroutine Library 2 (SSL-II)
  • ビルドツール。
    • CMake

もし他のコンパイラを利用する場合は、ビルドスクリプト(Cmake)を変更することが必要です。#configure.pyスクリプトの付加的な選択肢をご覧ください。 もしあなたの計算機環境で数値ライブラリがインストールされていない場合は、あなた自身でBLAS/LAPACKをインストールすることが必要です。 インストールでのトラブルシューティングをご覧ください。

SALMONのインストールでは、CMakeの利用を第一の選択肢としています。 あなたの利用環境でCMakeツールを使うことに問題がある場合は、GNU makeツールの利用も可能です。 インストールでのトラブルシューティングをご覧ください。

ダウンロード

SALMONの最新バージョンは、ダウンロードからダウンロードすることができます。 ダウンロードしたファイルsalmon-<VERSION>.tar.gzからファイルを抽出するために、コマンドラインから以下のコマンドを入力します。

$ tar –zxvf ./salmon-<VERSION>.tar.gz

ファイルの抽出後、以下のディレクトリが生成されます。

SALMON
 |- src        Source codes
 |- example    Samples
 |- makefiles   GNU Makefiles for building
 |- cmakefiles   CMake related files
 |- makefiles   GNU Makefiles for building

ビルド

To compile SALMON to create executable the binary files, we adopt to use CMake tools as the first option. In case you fail to build SALMON using CMake in your environment, we may use Gnu Make. See #Build using GNU Makefile.

Checking CMake availability

First, examine whether CMake is usable in your environment or not. Type the following in Unix command-line:

$ cmake --version

If CMake is not installed in your system, an error message such as cmake: command not found will appear. If CMake is installed on your system, the version number will be shown. To build SALMON, CMake of version 3.0.2 or later is required. If you confirm that Came of version 3.0.2 or later is installed in your system, proceed to #Build using CMake. However, we realize that old versions of CMake are installed in many systems. If CMake is not installed or CMake of older versions is installed in your system, you need to install the new version by yourself. It is a simple procedure and explained below.

Installation of CMake

CMake is a cross-platform build tool. The simplest way to make CMake usable in your environment is to get the binary distribution of CMake from the download page. The file name of the binary distribution will be cmake-<VERSION>-<PLATFORM>.tar.gz). In standard Unix environment, a file for the platform of Linux x86_64 will be appropriate.

To download the file, proceed as follows: We assume that you are in the directory that you extracted files from the downloaded file of SALMON, and that you will use the version 3.8.2. First get the URL of the download link from your browser, and use wget command in your Unix command-line.

$ wget https://cmake.org/files/v3.8/cmake-3.8.2-Linux-x86_64.tar.gz

Next, unpack the archive by

$ tar -zxvf cmake-3.8.2-Linux-x86_64.tar.gz

and you will have the binary make-3.8.2-Linux-x86_64/bin/cmake in your directory.

To make the cmake command usable in your command-line, you need to modify the environment variable $PATH so that the executable of CMake are settled inside the directory specified in your $PATH. If you use the bash shell, you need to modify the file ~/.bashrc that specifies the $PATH variable. It can be done by typing the following command in your login directory,

$ export PATH=<SALMON_INSTALLATION_DIRECTORY>/cmake-3.8.2-Linux-x86_64/bin:$PATH

and then reload the configuration by typing:

$ source ~/.bashrc

Build using CMake

Confirming that CMake of version 3.0.2 or later can be usable in your environment, proceed the following steps. We assume that you are in the directory SALMON.

  • Create a new temporary directory build and move to the directory,
$ mkdir build
$ cd build
  • Execute the python script configure.py and then make,
$ python ../configure.py –arch=ARCHTECTURE --prefix=../
$ make
$ make install

In executing the python script, you need to specify ARCHITECTURE that indicates the architecture of the CPU in your computer system such as intel-avx. The options of the ARCHTECUTRE are as follows:

arch Detail Compiler Numerical Library
intel-knl Intel Knights Landing Intel Compiler Intel MKL
intel-knc Intel Knights Corner Intel Compiler Intel MKL
intel-avx Intel Processer (Ivy-, Sandy-Bridge) Intel Compiler Intel MKL
intel-avx2 Intel Processer (Haswell, Broadwell ..) Intel Compiler Intel MKL
fujitsu-fx100 FX100 Supercomputer Fujitsu Compiler SSL-II
fujitsu-k Fujitsu FX100 / K-computer Fujitsu Compiler SSL-II

If the build is successful, you will get a file salmon.cpu at the directory salmon/bin. If you specify many-core archtechtures, intel-knl or intel-knc, you find a file salmon.mic or both files salmon.cpu and salmon.mic.

Build for single process calculations

In default, the python script assumes parallel execution. If you use a single processor machine, specify --disable-mpi in executing the python script:

 $ python ../configure.py --arch=<ARCHITECTURE> --disable-mpi

Files necessary to run SALMON

To run SALMON, at least two kinds of files are required for any calculations. One is an input file with the filename extension *.inp* that should be read from the standard input stdin. This file should be prepared in the Fortran90 namelist format. Pseudopotential files of relevant elements are also required. Depending on your purpose, some other files may also be necessary. For example, coordinates of atomic positions of the target material may be either written in the input file or prepared as a separate file.

Pseudopotentials

SALMON utilizes norm-conserving pseudpotentials. You may find pseudopotentials of some elements in the samples prepared in Exercises. In SALMON, several formats of pseudopotentials may be usable. Pseudopotentials with an extension .fhi that can be obtained from the website listed below. (This is a part of previous atomic data files for the abinit code.)

pseudopotential website
Pseudopotentials for the ABINIT code https://www.abinit.org/sites/default/files/PrevAtomicData/psp-links/psp-links/lda_fhi

Filenames of the pseudopotentials should be written in the input file.

input file

Input files are composed of several blocks of namelists,

&namelist1
  variable1 = int_value
  variable2 = 'char_value'
 /
&namelist2
  variable1 = real8_value
  variable2 = int_value1, int_value2, int_value3
/

A block of namelists starts with &namelist line and ends with / line. The blocks may appear in any order.

Between two lines of &namelist and /, descriptions of variables and their values appear. Note that many variables have their default values so that it is not necessary to give values for all variables. Descriptions of the variables may appear at any position if they are between &namelist and /.

SALMON describes electron dynamics in systems with both isolated and periodic boundary conditions. The boundary condition is specified by the variable iperiodic in the namelist &system.

Calculations are usually achieved in two steps; first, the ground state calculation is carried out and then electron dynamics calculations in real time is carried out. A choice of the calculation mode is specified by the variable calc_mode in the namelist &calculation. For isolated systems, the ground state and the electron dynamics calculations should be carried out as two separate executions. First the ground state calculation is carried out specifying calc_mode = 'GS' . Then the real-time electron dynamics calculation is carried out specifying calc_mode = 'RT' . For periodic systems, two calculations should be carried out as a single execution specifying calc_mode = 'GS_RT' .

In Exercises, we prepare six exercises that cover typical calculations feasible by SALMON. We prepare explanations of the input files of the tutorials that will help to prepare input files of your own interests.

There are more than 20 groups of namelists. A complete list of namelist variables is given in the file SALMON/manual/input_variables.md. Namelist variables that are used in our tutorials are explained at Input variables.

Run SALMON

Before running SALMON, the following preparations are required as described above: The executable file of salmon.cpu (and salmon.mic if your system is the many-core machine) should be built from the source file of SALMON. An input file inputfile.inp and pseudopotential files should also be prepared.

The execution of the calculation can be done as follows: In single process environment, type the following command:

$ salmon.cpu < inputfile.inp > fileout.out

In multiprocess environment in which the command to execute parallel calculations using MPI is mpiexec, type the following command:

$ mpiexec -n NPROC salmon.cpu < inputfile.inp > fileout.out

where NPROC is the number of MPI processes that you will use. In many-core processor (e.g. intel-knl) environment, the execution command is

$ mpiexec.hydra -n NPROC salmon.mic < inputfile.inp > fileout.out

The execution command and the job submission procedure depends much on local environment. We summarize general conditions to execute SALMON:

  • salmon runs in both single-process and multi-process environments using MPI.
  • executable files are prepared as /salmon/bin/salmon.cpu and/or /salmon/bin/salmon.mic in the standard build procedure.
  • to start calculations, inputfile.inp should be read through stdin.

Appendix

Additional options in configure.py script

Manual specifications of compiler and environment variables

In executing configure.py, you may manually specify compiler and environment variables instead of specifying the architecture, for example:

$ python ../configure.py FC=mpiifort CC=mpiicc FFLAGS="-xAVX" CFLAGS="-restrict -xAVX"

The major options of configure.py are as follows:

Commandline switch Detail
-a ARCH, --arch=ARCH Target architecture
--enable-mpi, --disable-mpi enable/disable MPI parallelization.
---enable-scalapack, --disable-scalapack enable/disable computations with ScaLAPACK library
FC, FFLAGS User-defined Fortran Compiler, and the compiler options
CC, CFLAGS User-defined C Compiler, and the compiler options


Build for single process calculations

If you use a single processor machine, specify --disable-mpi in executing the python script:

 $ python ../configure.py --arch=<ARCHITECTURE> --disable-mpi

Build in GCC/GFortran environemnt

If you use GCC/GFortran compiler, specify the following flags in executing the python script:

$ python ../configure.py FC=gfortran CC=gcc FFLAG=-O3 CFLAG=-O3

Build using GNU Makefile

If CMake build fails in your environment, we recommend you to try to use Gnu Make for the build process. First, enter the directory makefiles:

$ cd SALMON/makefiles

In the directory, Makefile files are prepared for several architectures:

  • fujitsu
  • gnu
  • gnu-without-mpi
  • intel
  • intel-avx
  • intel-avx2
  • intel-knc
  • intel-knl
  • intel-without-mpi

Makefile files with -without-mpi indicate that they are for single processor environment. Choose Makefile appropriate for your environment, and execute the make command:

$ make -f Makefile.PLATFORM

If the make proceeds successful, a binary file is created in the directory SALMON/bin/.