Get Started Guide

Get Started with Intel® oneAPI Math Kernel Library

ID 766875
Date 11/07/2023
Public

Get Started with Intel® oneAPI Math Kernel Library

The Intel® oneAPI Math Kernel Library (oneMKL) helps you achieve maximum performance with a math computing library of highly optimized, extensively parallelized routines for CPU and GPU. The library has C and Fortran interfaces for most routines on CPU, and SYCL interfaces for some routines on both CPU and GPU. You can find comprehensive support for several math operations in various interfaces including:

For C and Fortran on CPU

  • Linear algebra
  • Fast Fourier Transforms (FFT)
  • Vector math
  • Direct and iterative sparse solvers
  • Random number generators

For SYCL on CPU and GPU (Refer to the Intel® oneAPI Math Kernel Library—Data Parallel C++ Developer Reference for more details.)

  • Linear algebra
    • BLAS
    • Selected Sparse BLAS functionality
    • Selected LAPACK functionality
  • Fast Fourier Transforms (FFT)
    • 1D, 2D, and 3D
  • Random number generators
    • Selected functionality
  • Selected Vector Math functionality

Before You Begin

Visit the Release Notes page for the Known Issues and most up-to-date information.

Visit the Intel® oneAPI Math Kernel Library System Requirements page for system requirements.

Visit the Get Started with the Intel® oneAPI DPC++/C++ Compiler for DPC++ Compiler requirements.

Step 1: Install Intel® oneAPI Math Kernel Library

Download Intel® oneAPI Math Kernel Library from the Intel® oneAPI Base Toolkit.

For Python distributions, refer to Installing the Intel® Distribution for Python* and Intel® Performance Libraries with pip and PyPI.

For Python distributions, note the following limitation:

The oneMKL devel package (mkl-devel) for PIP distribution on Linux* and macOS* does not provide dynamic libraries symlinks (for more information see PIP GitHub issue #5919).

In the case of dynamic or single dynamic library linking with oneMKL devel package (for more information see oneMKL Link Line Advisor ) you must modify link line with oneMKL libraries full names and versions.

Refer to Intel® oneAPI Math Kernel Library and pkg-config tool for information about compiling and linking with the pkg-config tool.

oneMKL link line example with the oneAPI Base Toolkit via symlinks:

Linux:

icx app.obj -L${MKLROOT}/lib/intel64 -lmkl_intel_lp64-lmkl_intel_thread -lmkl_core -liomp5 -lpthread -lm -ldl

macOS:

icx app.obj -L${MKLROOT}/lib -Wl,-rpath,${MKLROOT}/lib-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread
-lm -ldl

The oneMKL link line example with PIP devel package via libraries full names and versions:

Linux:

icx app.obj ${MKLROOT}/lib/intel64/libmkl_intel_lp64.so.1 ${MKLROOT}/lib/intel64/libmkl_intel_thread.so.1 ${MKLROOT}/lib/intel64/libmkl_core.so.1 -liomp5 -lpthread -lm
-ldl

macOS:

icx app.obj -Wl,-rpath,${MKLROOT}/lib${MKLROOT}/lib/intel64/libmkl_intel_lp64.1.dylib ${MKLROOT}/lib/intel64/libmkl_intel_thread.1.dylib
${MKLROOT}/lib/intel64/libmkl_core.1.dylib -liomp5 -lpthread -lm-ldl

Step 2: Select a Function or Routine

Select a function or routine from oneMKL that is best suited for your problem. Use these resources:

Resource Link Contents

oneMKL Developer Guide for Linux*

oneMKL Developer Guide for Windows*

oneMKL Developer Guide for macOS*

The Developer Guide contains detailed information on several topics including:

  • Compiling and linking applications
  • Building custom DLLs
  • Threading
  • Memory Management

oneMKL Developer Reference - C

oneMKL Developer Reference - Fortran

oneMKL Developer Reference - DPC++

The Developer Reference (in C, Fortran, and DPC++ formats) contains detailed descriptions of the functions and interfaces for all library domains.

Intel® oneAPI Math Kernel Library Function Finding Advisor

Use the LAPACK Function Finding Advisor to explore LAPACK routines that are useful for a particular problem. For example, if you specify an operation as:

  • Routine type: Computational
  • Computational problem: Orthogonal factorization
  • Matrix type: General
  • Operation: Perform QR factorization

Step 3: Link Your Code

Use the oneMKL Link Line Advisor to configure the link command according to your program features.

Some limitations and additional requirements:

Intel® oneAPI Math Kernel Library for SYCL supports only the use of the mkl_intel_ilp64 interface library and sequential or TBB threading.

For SYCL interfaces with static linking on Linux

icpx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 <typical user includes and linking flags and other libs> ${MKLROOT}/lib/intel64/libmkl_sycl.a -Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_ilp64.a ${MKLROOT}/lib/intel64/libmkl_<sequential|tbb_thread>.a ${MKLROOT}/lib/intel64/libmkl_core.a -Wl,--end-group -lsycl -lOpenCL -lpthread -ldl -lm

For example, building/statically linking main.cpp with ilp64 interfaces and TBB threading:

icpx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 -I${MKLROOT}/include main.cpp ${MKLROOT}/lib/intel64/libmkl_sycl.a -Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_ilp64.a ${MKLROOT}/lib/intel64/libmkl_tbb_thread.a ${MKLROOT}/lib/intel64/libmkl_core.a -Wl,--end-group -L${TBBROOT}/lib/intel64/gcc4.8 -ltbb -lsycl -lOpenCL -lpthread -lm -ldl

For SYCL interfaces with dynamic linking on Linux

icpx -fsycl -DMKL_ILP64 <typical user includes and linking flags and other libs> -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_<sequential|tbb_thread> -lmkl_core -lsycl -lOpenCL -lpthread -ldl -lm

For example, building/dynamically linking main.cpp with ilp64 interfaces and TBB threading including all SYCL domains:

icpx -fsycl -DMKL_ILP64 -I${MKLROOT}/include main.cpp -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_tbb_thread -lmkl_core -lsycl -lOpenCL -ltbb -lpthread -ldl -lm

Or the same configuration with the BLAS SYCL domain only (note that libraries specific to the SYCL domain are aligned with oneMKL domain namespaces):

icpx -fsycl -DMKL_ILP64 -I${MKLROOT}/include main.cpp -L${MKLROOT}/lib/intel64 -lmkl_sycl_blas -lmkl_intel_ilp64 -lmkl_tbb_thread -lmkl_core -lsycl -lOpenCL -ltbb -lpthread -ldl -lm

For SYCL interfaces with static linking on Windows

icpx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 <typical user includes and linking flags and other libs> "%MKLROOT%"\lib\intel64\mkl_sycl.lib mkl_intel_ilp64.lib mkl_<sequential|tbb_thread>.lib mkl_core_lib sycl.lib OpenCL.lib

For example, building/statically linking main.cpp with ilp64 interfaces and TBB threading:

icpx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 -I"%MKLROOT%\include" main.cpp"%MKLROOT%"\lib\intel64\mkl_sycl.lib  mkl_intel_ilp64.lib mkl_tbb_thread.lib mkl_core.lib sycl.lib OpenCL.lib tbb.lib

For SYCL interfaces with dynamic linking on Windows

icx -fsycl -DMKL_ILP64 <typical user includes and linking flags and other libs> "%MKLROOT%"\lib\intel64\mkl_sycl_dll.lib mkl_intel_ilp64_dll.lib mkl_<sequential|tbb_thread>_dll.lib mkl_core_dll.lib tbb.lib sycl.lib OpenCL.lib

For example, building/dynamically linking main.cpp with ilp64 interfaces and TBB threading including all SYCL domains:

icx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 -I"%MKLROOT%\include" main.cpp "%MKLROOT%"\lib\intel64\mkl_sycl_dll.lib mkl_intel_ilp64_dll.lib mkl_tbb_thread_dll.lib mkl_core_dll.lib tbb.lib sycl.lib OpenCL.lib

Or the same configuration with the BLAS SYCL domain only (note that libraries specific to the SYCL domain are aligned with oneMKL domain namespaces):

icx -fsycl -fsycl-device-code-split=per_kernel -DMKL_ILP64 -I"%MKLROOT%\include" main.cpp "%MKLROOT%"\lib\intel64\mkl_sycl_blas_dll.lib mkl_intel_ilp64_dll.lib mkl_tbb_thread_dll.lib mkl_core_dll.lib tbb.lib sycl.lib OpenCL.lib

For C/Fortran Interfaces with OpenMP Offload Support

Use the C/Fotran Intel® oneAPI Math Kernel Library interfaces with OpenMP offload feature to the GPU.

See the C OpenMP Offload Developer Guide for more details about this feature.

Add the following changes to the C/Fortran oneMKL compile/link lines to enable OpenMP offload feature to GPU:

  • Additional compile/link options: -fiopenmp -fopenmp-targets=spir64 -mllvm -vpo-paropt-use-raw-dev-ptr -fsycl
  • Additional oneMKL library: oneMKL SYCL library

For example, building/ dynamically linking main.cpp on Linux with ilp64 interfaces and OpenMP threading:

icx -fiopenmp -fopenmp-targets=spir64 -mllvm -vpo-paropt-use-raw-dev-ptr -fsycl  -DMKL_ILP64 -m64 -I$(MKLROOT)/include main.cpp L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lsycl -lOpenCL -lstdc++ -lpthread -lm -ldl

For all other supported configurations, see Intel® oneAPI Math Kernel Library Link Line Advisor.

Find More

Resource

Description

Tutorial: Using Intel® oneAPI Math Kernel Library for Matrix Multiplication:

This tutorial demonstrates how you can use oneMKL to multiply matrices, measure the performance of matrix multiplication, and control threading.

Intel® oneAPI Math Kernel Library (oneMKL) Release Notes

The release notes contain information specific to the latest release of oneMKL including new and changed features. The release notes include links to principal online information resources related to the release. You can also find information on:

  • What's new in the release
  • Product contents
  • Obtaining technical support
  • License definitions

Intel® oneAPI Math Kernel Library

The Intel® oneAPI Math Kernel Library (oneMKL) product page. See this page for support and online documentation.

Intel® oneAPI Math Kernel Library Cookbook The Intel® oneAPI Math Kernel Library contains many routines to help you solve various numerical problems, such as multiplying matrices, solving a system of equations, and performing a Fourier transform.
Notes for Intel® oneAPI Math Kernel Library Vector Statistics

This document includes an overview, a usage model and testing results of random number generators included in VS.

Intel® oneAPI Math Kernel Library Vector Statistics Random Number Generator Performance Data

Performance data obtained using vector statistics (VS) random number generator (RNG) including CPE (clocks per element) unit of measure, basic random number generators (BRNG), generated distribution generators, and length of generated vectors.

Intel® oneAPI Math Kernel Library Vector Mathematics Performance and Accuracy Data

Vector Mathematics (VM) computes elementary functions on vector arguments. VM includes a set of highly optimized implementations of computationally expensive core mathematical functions (power, trigonometric, exponential, hyperbolic, and others) that operate on vectors.

Application Notes for Intel® oneAPI Math Kernel Library Summary Statistics

Summary Statistics is a subcomponent of the Vector Statistics domain of Intel® oneAPI Math Kernel Library. Summary Statistics provides you with functions for initial statistical analysis, and offers solutions for parallel processing of multi-dimensional datasets.

LAPACK Examples

This document provides code examples for oneMKL LAPACK (Linear Algebra PACKage) routines.

oneAPI Samples Catalog

These samples were designed to help you develop, offload, and optimize multiarchitecture applications targeting CPUs, GPUs, and FPGAs.

Notices and Disclaimers

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit www.intel.com/benchmarks.

Intel technologies may require enabled hardware, software or service activation.

No product or component can be absolutely secure.

Your costs and results may vary.

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.

Product and Performance Information

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.

Notice revision #20201201

No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document.

The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request.

Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.