Skip to content

oneapi-src/oneMKL

UXL Foundation Logo

oneAPI Math Kernel Library (oneMKL) Interfaces

oneMKL Interfaces is an open-source implementation of the oneMKL Data Parallel C++ (DPC++) interface according to the oneMKL specification. It works with multiple devices (backends) using device-specific libraries underneath.

oneMKL is part of the UXL Foundation.

User Application oneMKL Layer Third-Party Library Hardware Backend
oneMKL interface oneMKL selector Intel(R) oneAPI Math Kernel Library (oneMKL) x86 CPU, Intel GPU
NVIDIA cuBLAS NVIDIA GPU
NVIDIA cuSOLVER NVIDIA GPU
NVIDIA cuRAND NVIDIA GPU
NVIDIA cuFFT NVIDIA GPU
NETLIB LAPACK x86 CPU
AMD rocBLAS AMD GPU
AMD rocSOLVER AMD GPU
AMD rocRAND AMD GPU
AMD rocFFT AMD GPU
portBLAS x86 CPU, Intel GPU, NVIDIA GPU, AMD GPU, Other SYCL devices (unsupported)
portFFT x86 CPU, Intel GPU, NVIDIA GPU, AMD GPU, Other SYCL devices (unsupported)

Table of Contents


Support and Requirements

Supported Usage Models:

Host API

There are two oneMKL selector layer implementations:

  • Run-time dispatching: The application is linked with the oneMKL library and the required backend is loaded at run-time based on device vendor (all libraries should be dynamic).

    Example of app.cpp with run-time dispatching:

    #include "oneapi/mkl.hpp"
    
    ...
    cpu_dev = sycl::device(sycl::cpu_selector());
    gpu_dev = sycl::device(sycl::gpu_selector());
    
    sycl::queue cpu_queue(cpu_dev);
    sycl::queue gpu_queue(gpu_dev);
    
    oneapi::mkl::blas::column_major::gemm(cpu_queue, transA, transB, m, ...);
    oneapi::mkl::blas::column_major::gemm(gpu_queue, transA, transB, m, ...);

    How to build an application with run-time dispatching:

    if OS is Linux, use icpx compiler. If OS is Windows, use icx compiler. Linux example:

    $> icpx -fsycl –I$ONEMKL/include app.cpp
    $> icpx -fsycl app.o –L$ONEMKL/lib –lonemkl
  • Compile-time dispatching: The application uses a templated backend selector API where the template parameters specify the required backends and third-party libraries and the application is linked with the required oneMKL backend wrapper libraries (libraries can be static or dynamic).

    Example of app.cpp with compile-time dispatching:

    #include "oneapi/mkl.hpp"
    
    ...
    cpu_dev = sycl::device(sycl::cpu_selector());
    gpu_dev = sycl::device(sycl::gpu_selector());
    
    sycl::queue cpu_queue(cpu_dev);
    sycl::queue gpu_queue(gpu_dev);
    
    oneapi::mkl::backend_selector<oneapi::mkl::backend::mklcpu> cpu_selector(cpu_queue);
    
    oneapi::mkl::blas::column_major::gemm(cpu_selector, transA, transB, m, ...);
    oneapi::mkl::blas::column_major::gemm(oneapi::mkl::backend_selector<oneapi::mkl::backend::cublas> {gpu_queue}, transA, transB, m, ...);

    How to build an application with compile-time dispatching:

    $> clang++ -fsycl –I$ONEMKL/include app.cpp
    $> clang++ -fsycl app.o –L$ONEMKL/lib –lonemkl_blas_mklcpu –lonemkl_blas_cublas

Refer to Selecting a Compiler for the choice between icpx/icx and clang++ compilers.

Device API

Header-based and backend-independent Device API can be called within sycl kernel or work from Host code (device-rng-usage-model-example). Currently, the following domains support the Device API:

  • RNG. To use RNG Device API functionality it's required to include oneapi/mkl/rng/device.hpp header file.

Supported Configurations:

Supported domains include: BLAS, LAPACK, RNG, DFT, SPARSE_BLAS

Supported compilers include:

  • Intel(R) oneAPI DPC++ Compiler: Intel proprietary compiler that supports CPUs and Intel GPUs. Intel(R) oneAPI DPC++ Compiler will be referred to as "Intel DPC++" in the "Supported Compiler" column of the tables below.
  • oneAPI DPC++ Compiler: Open source compiler that supports CPUs and Intel, NVIDIA, and AMD GPUs. oneAPI DPC++ Compiler will be referred to as "Open DPC++" in the "Supported Compiler" column of the tables below.
  • AdaptiveCpp Compiler (formerly known as hipSYCL): Open source compiler that supports CPUs and Intel, NVIDIA, and AMD GPUs.
    Note: The source code and some documents in this project still use the previous name hipSYCL during this transition period.

Linux*

Domain Backend Library Supported Compiler Supported Link Type
BLAS x86 CPU Intel(R) oneMKL Intel DPC++
AdaptiveCpp
Dynamic, Static
NETLIB LAPACK Intel DPC++
Open DPC++
AdaptiveCpp
Dynamic, Static
portBLAS Intel DPC++
Open DPC++
Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
portBLAS Intel DPC++
Open DPC++
Dynamic, Static
NVIDIA GPU NVIDIA cuBLAS Open DPC++
AdaptiveCpp
Dynamic, Static
portBLAS Open DPC++ Dynamic, Static
AMD GPU AMD rocBLAS Open DPC++
AdaptiveCpp
Dynamic, Static
portBLAS Open DPC++ Dynamic, Static
Other SYCL devices (unsupported) portBLAS Intel DPC++
Open DPC++
Dynamic, Static
LAPACK x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
NVIDIA GPU NVIDIA cuSOLVER Open DPC++ Dynamic, Static
AMD GPU AMD rocSOLVER Open DPC++ Dynamic, Static
RNG x86 CPU Intel(R) oneMKL Intel DPC++
AdaptiveCpp
Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
NVIDIA GPU NVIDIA cuRAND Open DPC++
AdaptiveCpp
Dynamic, Static
AMD GPU AMD rocRAND Open DPC++
AdaptiveCpp
Dynamic, Static
DFT x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
portFFT (limited API support) Intel DPC++ Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
portFFT (limited API support) Intel DPC++ Dynamic, Static
NVIDIA GPU NVIDIA cuFFT Open DPC++ Dynamic, Static
portFFT (limited API support) Open DPC++ Dynamic, Static
AMD GPU AMD rocFFT Open DPC++ Dynamic, Static
portFFT (limited API support) Open DPC++ Dynamic, Static
Other SYCL devices (unsupported) portFFT Open DPC++
Open DPC++
Dynamic, Static
SPARSE_BLAS x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static

Windows*

Domain Backend Library Supported Compiler Supported Link Type
BLAS x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
NETLIB LAPACK Intel DPC++
Open DPC++
Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
LAPACK x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
RNG x86 CPU Intel(R) oneMKL Intel DPC++ Dynamic, Static
Intel GPU Intel(R) oneMKL Intel DPC++ Dynamic, Static

Hardware Platform Support

  • CPU
    • Intel Atom(R) Processors
    • Intel(R) Core(TM) Processor Family
    • Intel(R) Xeon(R) Processor Family
  • Accelerators
    • Intel(R) Arc(TM) A-Series Graphics
    • Intel(R) Data Center GPU Max Series
    • NVIDIA(R) A100 (Linux* only)
    • AMD(R) GPUs see here tested on AMD Vega 20 (gfx906)
    • Other SYCL devices can be used, but are not supported

Supported Operating Systems

Linux*

Backend Supported Operating System
x86 CPU Red Hat Enterprise Linux* 9 (RHEL* 9)
Intel GPU Ubuntu 22.04 LTS
NVIDIA GPU Ubuntu 22.04 LTS

Windows*

Backend Supported Operating System
x86 CPU Microsoft Windows* Server 2022
Intel GPU Microsoft Windows* 11

Software Requirements

What should I download?

General:

Functional Testing Build Only Documentation
CMake (version 3.13 or newer)
Linux* : GNU* GCC 5.1 or higher
Windows* : MSVS* 2017 or MSVS* 2019 (version 16.5 or newer)
Ninja (optional)
GNU* FORTRAN Compiler - Sphinx
NETLIB LAPACK - -

Hardware and OS Specific:

Operating System Device Package
Linux*/Windows* x86 CPU Intel(R) oneAPI DPC++ Compiler
or
oneAPI DPC++ Compiler
Intel(R) oneAPI Math Kernel Library
Intel GPU Intel(R) oneAPI DPC++ Compiler
Intel GPU driver
Intel(R) oneAPI Math Kernel Library
Linux* only NVIDIA GPU oneAPI DPC++ Compiler
or
AdaptiveCpp with CUDA backend and dependencies
AMD GPU oneAPI DPC++ Compiler
or
AdaptiveCpp with ROCm backend and dependencies

Product and Version Information:

Product Supported Version License
CMake 3.13 or higher The OSI-approved BSD 3-clause License
Ninja 1.10.0 Apache License v2.0
GNU* FORTRAN Compiler 7.4.0 or higher GNU General Public License, version 3
Intel(R) oneAPI DPC++ Compiler Latest End User License Agreement for the Intel(R) Software Development Products
AdaptiveCpp Later than 2cfa530 BSD-2-Clause License
oneAPI DPC++ Compiler binary for x86 CPU Daily builds Apache License v2
oneAPI DPC++ Compiler source for NVIDIA and AMD GPUs Daily source releases Apache License v2
Intel(R) oneAPI Math Kernel Library Latest Intel Simplified Software License
NVIDIA CUDA SDK 12.0 End User License Agreement
AMD rocBLAS 4.5 AMD License
AMD rocRAND 5.1.0 AMD License
AMD rocSOLVER 5.0.0 AMD License
AMD rocFFT rocm-5.4.3 AMD License
NETLIB LAPACK 5d4180c BSD like license
portBLAS 0.1 Apache License v2.0
portFFT 0.1 Apache License v2.0

Documentation


Governance

The oneMKL Interfaces project is governed by the UXL Foundation and you can get involved in this project in multiple ways. It is possible to join the Math Special Interest Group (SIG) meetings where the group discusses and demonstrates work using this project. Members can also join the Open Source and Specification Working Group meetings.

You can also join the mailing lists for the UXL Foundation to be informed of when meetings are happening and receive the latest information and discussions.


Contributing

You can contribute to this project and also contribute to the specification for this project. Please read the CONTRIBUTING page for more information. You can also contact oneMKL developers and maintainers via UXL Foundation Slack using #onemkl channel.


License

Distributed under the Apache license 2.0. See LICENSE for more information.


FAQs

oneMKL

Q: What is the difference between the following oneMKL items?

A:

  • The oneAPI Specification for oneMKL defines the DPC++ interfaces for performance math library functions. The oneMKL specification can evolve faster and more frequently than implementations of the specification.

  • The oneAPI Math Kernel Library (oneMKL) Interfaces Project is an open source implementation of the specification. The project goal is to demonstrate how the DPC++ interfaces documented in the oneMKL specification can be implemented for any math library and work for any target hardware. While the implementation provided here may not yet be the full implementation of the specification, the goal is to build it out over time. We encourage the community to contribute to this project and help to extend support to multiple hardware targets and other math libraries.

  • The Intel(R) oneAPI Math Kernel Library (oneMKL) product is the Intel product implementation of the specification (with DPC++ interfaces) as well as similar functionality with C and Fortran interfaces, and is provided as part of Intel® oneAPI Base Toolkit. It is highly optimized for Intel CPU and Intel GPU hardware.

Q: I'm trying to use oneMKL Interfaces in my project using FetchContent, but I keep running into ONEMKL::SYCL::SYCL target was not found problem when I try to build the project. What should I do?

A: Make sure you set the compiler when you configure your project. E.g. cmake -Bbuild . -DCMAKE_CXX_COMPILER=icpx.

Q: I'm trying to use oneMKL Interfaces in my project using find_package(oneMKL). I set oneMKL/oneTBB and Compiler environment first, then I built and installed oneMKL Interfaces, and finally I tried to build my project using installed oneMKL Interfaces (e.g. like this cmake -Bbuild -GNinja -DCMAKE_CXX_COMPILER=icpx -DoneMKL_ROOT=<path_to_installed_oneMKL_interfaces> .) and I noticed that cmake includes installed oneMKL Interfaces headers as a system include which ends up as a lower priority than the installed oneMKL package includes which I set before for building oneMKL Interfaces. As a result, I get conflicts between oneMKL and installed oneMKL Interfaces headers. What should I do?

A: Having installed oneMKL Interfaces headers as -I instead on system includes (as -isystem) helps to resolve this problem. We use INTERFACE_INCLUDE_DIRECTORIES to add paths to installed oneMKL Interfaces headers (check oneMKLTargets.cmake in lib/cmake to find it). It's a known limitation that INTERFACE_INCLUDE_DIRECTORIES puts headers paths as system headers. To avoid that:

  • Option 1: Use CMake >=3.25. In this case oneMKL Interfaces will be built with EXPORT_NO_SYSTEM property set to true and you won't see the issue.
  • Option 2: If you use CMake < 3.25, set PROPERTIES NO_SYSTEM_FROM_IMPORTED true for your target. E.g: set_target_properties(test PROPERTIES NO_SYSTEM_FROM_IMPORTED true).