Intel Math Kernel Library

Intel Math Kernel Library
Name	Intel Math Kernel Library
Developer	Intel Corporation
Released	1991
Programming language	C, Fortran, Assembly language
Operating system	Microsoft Windows, Linux, macOS
Platform	x86-64, Intel Xeon Phi, Intel Core
License	Proprietary software, Open-source software (components)

Intel Math Kernel Library

Intel Math Kernel Library is a software library of highly optimized math routines for science, engineering, and financial applications developed by Intel Corporation. It provides routines for linear algebra, fast Fourier transforms, vector math, and random number generation commonly used in applications from High Performance Computing centers to workstation environments. The library is often paired with compilers, debuggers, and performance tools from companies and projects such as GNU Compiler Collection, Microsoft Visual Studio, and Intel Parallel Studio.

Overview

Intel Math Kernel Library implements a suite of numerical algorithms derived from established projects and standards including implementations compatible with BLAS, LAPACK, and FFTPACK interfaces. The library concentrates on CPU microarchitecture-specific optimizations for families like Intel Xeon and Intel Core, and historically extended support to accelerators such as Intel Xeon Phi. It is distributed in binary and source component forms and is integrated into ecosystems used by vendors such as Hewlett Packard Enterprise, Dell Technologies, and cloud providers like Amazon Web Services and Microsoft Azure.

Features and Components

The library includes implementations of level 1–3 BLAS operations, LAPACK routines for linear algebra, multi-threaded FFTs compatible with standards like DFT libraries, and vector math (VML) functions for elementary transcendental operations. Additional components cover random number generation (RNG) with methods referenced in classical texts from authors such as Donald Knuth, and sparse solvers comparable to algorithms in PETSc and Trilinos. The package bundles threading runtimes and interoperability hooks for tools like OpenMP, MPI, and works with compilers including Intel C++ Compiler and GCC.

Performance and Optimization

Performance tuning in the library targets instruction sets such as SSE, AVX, AVX2, and AVX-512, leveraging microarchitecture features from Intel processors including Skylake and Cascade Lake. Optimizations exploit cache hierarchies and vector pipelines referenced in publications from processor architects at Intel Corporation and research labs like Lawrence Livermore National Laboratory and Argonne National Laboratory. Benchmarks often compare against implementations from projects such as OpenBLAS and vendor libraries from AMD to demonstrate throughput on workloads like dense matrix multiplication, FFT workloads used in applications from MATLAB and NumPy.

Programming Interfaces and Language Bindings

Interfaces are exposed in C and Fortran with headers and modules that follow calling conventions compatible with legacy scientific codebases originating from institutions like CERN and Los Alamos National Laboratory. Language bindings and wrappers enable use from environments and projects such as Python via SciPy, Julia, and R through packages maintained by community organizations and vendors including Anaconda, Inc. and The Apache Software Foundation projects. Integration guides reference build systems and package managers such as CMake and RPM Package Manager.

Supported Platforms and Hardware Acceleration

Primary support targets 64-bit x86-64 processors from Intel Corporation and related server platforms like Intel Xeon and legacy Intel Xeon Phi coprocessors. Hardware acceleration takes advantage of SIMD extensions and multi-core scaling on platforms provided by OEMs such as Supermicro and cloud infrastructures like Google Cloud Platform. Cross-platform ports and adaptations are used in research collaborations involving supercomputing centers such as Oak Ridge National Laboratory and national laboratories funded by agencies like U.S. Department of Energy.

Licensing and Packaging

The library is distributed under Intel's licensing terms with binary redistributions for commercial and academic users, and select components released with permissive licenses maintained by Intel Corporation in source form. Packaging formats include native installers for Microsoft Windows, packages for RPM-based and Debian-based Linux distributions, and containers used by orchestration platforms like Kubernetes. Licensing models and redistribution policies are often considered alongside alternatives from projects such as OpenBLAS and commercial offerings from NVIDIA for GPU-accelerated math.

Adoption and Use Cases

Adoption spans academic research groups at universities such as Massachusetts Institute of Technology, Stanford University, and University of California, Berkeley; national laboratories like Los Alamos National Laboratory and Sandia National Laboratories; and commercial users in finance, media, and simulation from firms such as Goldman Sachs, Siemens, and Lockheed Martin. Common use cases include large-scale simulations in computational fluid dynamics referenced by practitioners at American Institute of Aeronautics and Astronautics, signal processing in telecommunications companies like Ericsson, machine learning pipelines in organizations such as Facebook and Google LLC, and computational chemistry tools developed at research institutes such as Max Planck Society.

Category:Numerical libraries