This is an old revision of the document!
Table of Contents
Installation of Intel Libraries
Introduction
Our software is highly optimized using parallel computations and high-performance subroutines (called libraries). We are using Intel Math Kernel Library (MKL) for matrix computations in a combination with Intel Fortran Compiler (IFORT). We usually provide users a single executable (or “binary”) including MKL subroutines, and it should work at all environments. However, our program may stop due to interaction among MKL/IFORT, compilers, and operating systems. To solve this issue keeping the high performance, we would ask the users to install dependent libraries to their system. The binary will use the user-installed libraries which should be specialized in the computing environment.
Instruction
You need 2 packages to be installed: “Redistributable library package” and “Intel Math Kernel Library”.
1. Redistributable library package
Please download the “Update 3” of the redistributable library package from the Intel official website. Choose your system on the download page.
In Linux and macOS, please unpack the package and run install.sh
. In Windows, please execute the downloaded package. After the installation, please set up the post-installation configuration if needed.
For Linux users
There are two options: system-wise installation by the “root” user, or private installation to your home directory by yourself. The following is the brief instruction to install the library common to the both cases.
- Download the "Update 3" from the Intel website. The package name is like
l_fcompxe_2015.3.187_redist.tgz
. - Unpack the file and move to the package directory.
- Run the script
install.sh
. If you are the root, the script installs the package to the system for all users. If you are a regular user, it goes to your home directory. - Follow the message shown by the script. Finish the installation.
- Making the library accessible, run the following command on your shell (for 64bit Linux) where
/path/to/library
should be replaced with the actual directory to where you have installed the package.source /path/to/library/bin/compilervars.sh intel64
- This command can be in your
.bash_profile
so that you don't have to type it whenever you log-in to this system.
2. Intel Math Kernel Library (MKL)
Intel MKL is free to use in any commercial and academic purposes. Although it needs no charge, you have to register (free) to get the MKL package.
Please go to the Intel official website, and register your name, emails, and so on. You will find the download page immediately after the registration.
- Intel MKL: https://software.intel.com/en-us/mkl
Download the latest version of MKL package, install it, and follow the instruction for the post-installation process (if required). For Linux, the library name is “libmkl_intel_lp64.so” etc.
For Linux users
As explained above, there are two options: system-wise installation and private installation. The following is the brief instruction to install the library common to the both cases.
- Register your information at the Intel Website.
- Download the latest package. The name is like
l_mkl_2019.5.281.tgz
. - Unpack the file and move to the package directory.
- Run the script
install.sh
. If you are the root, the script installs the package to the system for all users. If you are a regular user, it goes to your home directory. - Follow the message shown by the script. Finish the installation.
- Making the library accessible, run the following command on your shell (for 64bit Linux) where
/path/to/library/intel
should be replaced with the actual directory to where you have installed the package.source /path/to/library/intel/bin/compilervars.sh intel64
- This command can be in your
.bash_profile
so that you don't have to type it whenever you log-in to this system.
Tests
After the installation of the above 2 libraries, we will provide am executable. Please run the program and see it works as usual. If the installation (and the post-installation configuration) is not successful, you will get a message like “error while loading shared libraries …”.