Intel offers a range of software tools from libraries for use on clusters. The University of Siegen has a license for the Intel Parallel Studio XE, which includes most of these tools and libraries. Some of the components of Intel Parallel Studio XE are installed on the OMNI cluster and can be used by all members of the university.

The documentation of Parallel Studio can be downloaded from here.

This page describes the components of Intel Parallel Studio installed on OMNI, namely Intel Compiler, Intel MPI and Intel MKL, and how to use them.

Intel Compiler

Intel offers compilers for C, C++ and Fortran. More information on the pages for C and C++, as well as the page for Fortran.

Intel MPI

The Intel MPI library is an implementation of the Message Passing Interface standard (MPI) supplied with Parallel Studio. MPI is a standard that describes the message exchange for parallel calculations on distributed computer systems. In contrast to OpenMP, MPI can also be used to work on a common problem on several separate computers.

Intel MPI is set up on the cluster as a module called impi. This module contains versions compiled with GCC and the Intel Compiler. You can replace the loaded compiler module ( gnu9 or intel) and the correct Intel MPI version will be used without having to reload the impi module.

MKL

The Intel MKL (Math Kernel Library) is a program library for mathematical calculations. It optimizes applications for Intel CPUs and GPUs, but also works on AMD CPUs such as those in the OMNI cluster.

The main components are BLAS (Basic Linear Algebra Subprograms), LAPACK (Linear Algebra Package), solvers for sparse matrices, fast Fourier transformations, statistics and vector mathematics. Vector math functions are implemented with single and double precision and are available for complex and floating point numbers. Note that MKL is not the only BLAS/LAPACK implementation. implementation.

In the area of statistics, for example, there are algorithms for random numbers and probability distributions and various splines for interpolation are included. Intel MKL also has functions for neural networks.

The Intel MKL supports C, C++, DPC++ and Fortran and is also compatible with Java, C#, Python and other languages.

The MKL is only available on the cluster in the version compiled with the Intel compiler, so the intel module must be loaded. The MKL installation folder is stored in the form of the MKLROOT environment variable. We recommend only specifying the following options when compiling and no extra library or include paths.
-liomp5-lpthread -ldl -lm -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core