Repository files navigation

oneAPI Deep Neural Network Library (oneDNN) is an open-source cross-platformperformance library of basic building blocks for deep learning applications.oneDNN project is part of theUXL Foundation and is an implementationof theoneAPI specification for oneDNN component.

The library is optimized for Intel(R) Architecture Processors, Intel Graphics,and Arm(R) 64-bit Architecture (AArch64)-based processors. oneDNN hasexperimental support for the following architectures: NVIDIA* GPU,AMD* GPU, OpenPOWER* Power ISA (PPC64), IBMz* (s390x), and RISC-V.

oneDNN is intended for deep learning applications and frameworkdevelopers interested in improving application performance on CPUs and GPUs.

Deep learning practitioners should use one of the applications enabled with oneDNN:

• Apache SINGA
• DeepLearning4J*
• Flashlight*
• MATLAB* Deep Learning Toolbox
• ONNX Runtime
• OpenVINO(TM) toolkit
• PaddlePaddle*
• PyTorch*. Intel GPU support and additionaloptimizations are available withIntel® Extension for PyTorch*.
• Tensorflow*. Intel GPU support and additionaloptimizations are available withIntel® Extension for TensorFlow*.

• Documentation
• System Requirements
• Installation
• Validated Configurations
• Governance
• Support
• Contributing
• License
• Security
• Trademark Information

• oneDNN Developer Guide and Reference explains the programmingmodel, supported functionality, implementation details, and includesannotated examples.
• API Reference provides a comprehensive reference of the libraryAPI.
• Release Notes explains the new features, performanceoptimizations, and improvements implemented in each version ofoneDNN.

oneDNN supports platforms based on the following architectures:

• Intel 64 or AMD64,
• Arm 64-bit Architecture (AArch64).
• OpenPOWER /IBM Power ISA.
• IBMz z/Architecture (s390x).
• RISC-V 64-bit (RV64).

WARNING

Power ISA (PPC64), IBMz (s390x), and RISC-V (RV64) support isexperimental with limited testing validation.

The library is optimized for the following CPUs:

• Intel 64/AMD64 architecture
  • Intel Atom(R) processor (at least Intel SSE4.1 support is required)
  • Intel Core(TM) processor (at least Intel SSE4.1 support is required)
  • Intel Xeon(R) processor E3, E5, and E7 family (formerly Sandy Bridge,Ivy Bridge, Haswell, and Broadwell)
  • Intel Xeon Scalable processor (formerly Skylake, Cascade Lake, CooperLake, Ice Lake, Sapphire Rapids, and Emerald Rapids)
  • Intel Xeon CPU Max Series (formerly Sapphire Rapids HBM)
  • Intel Core Ultra processors (formerly Meteor Lake, Arrow Lake,and Lunar Lake)
  • Intel Xeon 6 processors (formerly Sierra Forest and Granite Rapids)
• AArch64 architecture
  • Arm Neoverse(TM) N1 and V1 processors

On a CPU based on Intel 64 or on AMD64 architecture, oneDNN detectsthe instruction set architecture (ISA) at runtime and uses just-in-time (JIT)code generation to deploy the code optimized for the latest supported ISA.Future ISAs may have initial support in the library disabled by default andrequire the use of run-time controls to enable them. SeeCPU dispatcher control for more details.

WARNING

On macOS, applications that use oneDNN may need to request specialentitlements if they use the hardened runtime. See theLinking Guide for more details.

The library is optimized for the following GPUs:

• Intel Graphics for 11th-14th Generation Intel Core Processors
• Intel Iris Xe MAX Graphics (formerly DG1)
• Intel Arc(TM) graphics (formerly Alchemist)
• Intel Data Center GPU Flex Series (formerly Arctic Sound)
• Intel Data Center GPU Max Series (formerly Ponte Vecchio)
• Intel Graphics and Intel Arc graphics for Intel Core Ultra processors(formerly Meteor Lake, Arrow Lake and Lunar Lake)
• future Intel Arc graphics (code name Battlemage)

oneDNN supports systems meeting the following requirements:

• Operating system with Intel 64 / Arm 64 / Power / IBMz architecture support
• C++ compiler with C++11 standard support
• CMake 3.13 or later

The following tools are required to build oneDNN documentation:

• Doxygen 1.8.5 or later
• Doxyrest 2.1.2 or later
• Sphinx 4.0.2 or later
• sphinx-book-theme 0.0.41 or later

Configurations of CPU and GPU engines may introduce additional build timedependencies.

oneDNN CPU engine is used to execute primitives on Intel ArchitectureProcessors, 64-bit Arm Architecture (AArch64) processors,64-bit Power ISA (PPC64) processors, IBMz (s390x), and compatible devices.

The CPU engine is built by default but can be disabled at build time by settingDNNL_CPU_RUNTIME toNONE. In this case, GPU engine must be enabled.The CPU engine can be configured to use the OpenMP, TBB or SYCL runtime.The following additional requirements apply:

• OpenMP runtime requires C++ compiler with OpenMP 2.0 or laterstandard support
• TBB runtime requiresThreading Building Blocks (TBB) 2017 or later.
• SYCL runtime requires
  • Intel oneAPI DPC++/C++ Compiler
  • Threading Building Blocks (TBB)

Some implementations rely on OpenMP 4.0 SIMD extensions. For the bestperformance results on Intel Architecture Processors we recommend using theIntel C++ Compiler.

On a CPU based on Arm AArch64 architecture, oneDNN CPU engine can be built withArm Compute Library (ACL) integration. ACL is an open-source library formachine learning applications and provides AArch64 optimized implementationsof core functions. This functionality currently requires that ACL is downloadedand built separately. SeeBuild from Source section of the Developer Guide fordetails. oneDNN only supports Compute Library versions 24.11.1 or later.

Intel Processor Graphics and Xe Architecture graphics are supported bythe oneDNN GPU engine. The GPU engine is disabled in the default buildconfiguration. The following additional requirements apply when GPU engineis enabled:

• OpenCL runtime requires
  • OpenCL* runtime library (OpenCL version 1.2 or later)
  • OpenCL driver (with kernel language support for OpenCL C 2.0 or later)with Intel subgroups and USM extensions support
• SYCL runtime requires
  • Intel oneAPI DPC++/C++ Compiler
  • OpenCL runtime library (OpenCL version 3.0 or later)
  • oneAPI Level Zero
• SYCL runtime with NVIDIA GPU support requires
  • oneAPI DPC++ Compiler with support for CUDA oroneAPI for NVIDIA GPUs
  • NVIDIA CUDA* driver
  • cuBLAS 10.1 or later
  • cuDNN 7.6 or later
• SYCL runtime with AMD GPU support requires
  • oneAPI DPC++ Compiler with support for HIP AMD oroneAPI for AMD GPUs
  • AMD ROCm version 5.3 or later
  • MIOpen version 2.18 or later (optional if AMD ROCm includes the requiredversion of MIOpen)
  • rocBLAS version 2.45.0 or later (optional if AMD ROCm includesthe required version of rocBLAS)
• SYCL runtime with a generic GPU support requires
  • oneAPI DPC++/C++ Compiler that supports the target GPU. Refer to thegeneric GPU vendor readme for more information.

WARNING

Linux will reset GPU when kernel runtime exceeds several seconds. The usercan prevent this behavior bydisabling hangcheck for Intel GPU driver.Windows has built-intimeout detection and recovery mechanism that resultsin similar behavior. The user can prevent this behavior by increasing theTdrDelay value.

WARNING

NVIDIA GPU support is experimental. General information, build instructions,and implementation limitations are available in theNVIDIA backend readme.

WARNING

AMD GPU support is experimental. General information, build instructions,and implementation limitations are available in theAMD backend readme.

When oneDNN is built from source, the library runtime dependencies and specificversions are defined by the build environment.

Common dependencies:

• GNU C Library (libc.so)
• GNU Standard C++ Library v3 (libstdc++.so)
• Dynamic Linking Library (libdl.so)
• C Math Library (libm.so)
• POSIX Threads Library (libpthread.so)

Runtime-specific dependencies:

Common dependencies:

• Microsoft Visual C++ Redistributable (msvcrt.dll)

Runtime-specific dependencies:

Common dependencies:

• System C/C++ runtime (libc++.dylib,libSystem.dylib)

Runtime-specific dependencies:

You can download and install the oneDNN library using one of the following options:

• Binary Distribution: You can download pre-built binary packages fromthe following sources:

  • conda-forge: If the configuration you need is not available onthe conda-forge channel, you can build the library using theSource Distribution.
  • Intel oneAPI:
    • Intel® oneAPI Base Toolkit
    • Intel® oneDNN standalone package

• Source Distribution: You can build the library from source byfollowing the instructions on theBuild from Source page.

x86-64 CPU engine was validated on RedHat* Enterprise Linux 8 with

• GNU Compiler Collection 8.5, 9.5, 11.1, 11.3
• Clang* 11.0, 14.0.6
• Intel oneAPI DPC++/C++ Compiler 2024.0

on Windows Server* 2019 with

• Microsoft Visual Studio 2022
• Intel oneAPI DPC++/C++ Compiler 2024.0

on macOS 11 (Big Sur) with

• Apple LLVM version 13.0

AArch64 CPU engine was validated on Ubuntu 22.04 with

• GNU Compiler Collection 10.0, 13.0
• Clang* 17.0
• Arm Compiler for Linux 24.04
• Arm Compute Library (ACL) built for armv8-a arch, latest stable versionavailable at the time of release

on macOS 14 (Sonoma) with

• Apple LLVM version 15.0

GPU engine was validated on Ubuntu* 22.04 with

• GNU Compiler Collection 8.5, and 9.5
• Clang 11.0
• Intel oneAPI DPC++/C++ Compiler 2024.0
• Intel Software for General Purpose GPU capabilities latest stable versionavailable at the time of release

on Windows Server 2019 with

• Microsoft Visual Studio 2022
• Intel oneAPI DPC++/C++ Compiler 2024.0
• Intel Arc & Iris Xe Graphics Driver latest stable version available at thetime of release

Submit questions, feature requests, and bug reports on theGitHub issues page.

You can also contact oneDNN developers viaUXL Foundation Slack using#onednn channel.

oneDNN project is governed by theUXL Foundation and you can get involved inthis project in multiple ways. It is possible to join theAI Special InterestGroup (SIG) meetings where the groups discuss and demonstrate work using thisproject. Members can also join the Open Source and Specification Working Groupmeetings.

You can also join themailing lists for the UXL Foundation to be informedof when meetings are happening and receive the latest information anddiscussions.

We welcome community contributions to oneDNN. You can find the oneDNN releaseschedule and work already in progress towards future milestones in Github'sMilestones section. If you are looking for a specific task to start,consider selecting from issues that are marked with thehelp wanted label.

Seecontribution guidelines to start contributingto oneDNN. You can also contact oneDNN developers and maintainers viaUXL Foundation Slack using#onednn channel.

This project is intended to be a safe, welcoming space forcollaboration, and contributors are expected to adhere to theContributor Covenant code of conduct.

oneDNN is licensed underApache License Version 2.0. Referto the "LICENSE" file for the full license text andcopyright notice.

This distribution includes third party software governed by separatelicense terms.

3-clause BSD license:

• Xbyak
• gtest
• Instrumentation and Tracing Technology API(ITT API)
• CMake

2-clause BSD license:

• Sphinx

Apache License Version 2.0:

• Xbyak_aarch64
• LLVM

Boost Software License, Version 1.0:

• Boost C++ Libraries

MIT License:

• Intel Graphics Compute Runtime for oneAPI Level Zeroand OpenCL Driver
• Intel Graphics Compiler
• oneAPI Level Zero
• Doxyrest
• Intel Metrics Discovery Application ProgrammingInterface
• spdlog

This third-party software, even if included with the distribution ofthe Intel software, may be governed by separate license terms,including without limitation,third party license terms, other Intelsoftware license terms, and open source software license terms. Theseseparate license terms govern your use of the third party programs asset forth in the "THIRD-PARTY-PROGRAMS" file.

Security Policy outlines our guidelines and proceduresfor ensuring the highest level of security and trust for our userswho consume oneDNN.

Intel, the Intel logo, Arc, Intel Atom, Intel Core, Iris,OpenVINO, the OpenVINO logo, Pentium, VTune, and Xeon are trademarksof Intel Corporation or its subsidiaries.

Arm and Neoverse are trademarks, or registered trademarks of Arm Ltd.

* Other names and brands may be claimed as the property of others.

Microsoft, Windows, and the Windows logo are trademarks, or registeredtrademarks of Microsoft Corporation in the United States and/or othercountries.

OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permissionby Khronos.

(C) Intel Corporation