# ZenDNN **Repository Path**: anjiang2020_admin/ZenDNN ## Basic Information - **Project Name**: ZenDNN - **Description**: AMD显卡的ZenDNN - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 1 - **Created**: 2022-02-01 - **Last Updated**: 2024-05-29 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README Zen Deep Neural Network Library (ZenDNN) ======================================== ZenDNN (Zen Deep Neural Network) Library accelerates deep learning inference applications on AMD CPUs. This library, which includes APIs for basic neural network building blocks optimized for AMD CPUs, targets deep learning application and framework developers with the goal of improving inference performance on AMD CPUs across a variety of workloads, including computer vision, natural language processing (NLP), and recommender systems. ZenDNN leverages oneDNN/DNNL v2.2's basic infrastructure and APIs. ZenDNN optimizes several APIs and adds new APIs, which are currently integrated into TensorFlow, ONNXRT, and PyTorch. ZenDNN uses AMD BLIS (BLAS-like Library Instantiation Software) library for its BLAS (Basic Linear Algebra Subprograms) API needs. # Table of Contents - [Zen Deep Neural Network Library (ZenDNN)](#zen-deep-neural-network-library-zendnn) - [Table of Contents](#table-of-contents) - [Scope](#scope) - [Release Highlights](#release-highlights) - [Supported OS and Compilers](#supported-os-and-compilers) - [OS](#os) - [Compilers](#compilers) - [Prerequisites](#prerequisites) - [AOCC and AMD-BLIS Library Installation](#aocc-and-amd-blis-library-installation) - [General Convention](#general-convention) - [AMD-BLIS Library Setup](#amd-blis-library-setup) - [AOCC Installation](#aocc-installation) - [Runtime Dependencies](#runtime-dependencies) - [Build from Source](#build-from-source) - [AOCC compiler](#aocc-compiler) - [GCC compiler](#gcc-compiler) - [Validate the build](#validate-the-build) - [Logs](#logs) - [License](#license) - [Technical Support](#technical-support) # Scope The scope of ZenDNN is to support AMD EPYC^TM CPUs on the Linux® platform. ZenDNN v3.2 offers optimized primitives, such as Convolution, MatMul, Elementwise, and Pool (Max and Average) that improve performance of many convolutional neural networks, recurrent neural networks, transformer-based models, and recommender system models. For the primitives not supported by ZenDNN, execution will fall back to the native path of the framework. # Release Highlights Following are the highlights of this release: * ZenDNN library is integrated with TensorFlow (v2.7), ONNXRT v1.8.0, and PyTorch (v1.9.0). * Python v3.6-v3.9 has been used to generate the TensorFlow v2.7 wheel files (*.whl). * Python v3.8 has been used to generate the ONNXRT v1.8.0 wheel file (*.whl). * Python v3.6-v3.9 has been used to generate the PyTorch v1.9.0 wheel files (*.whl). * NHWC (default format) and Blocked Format (NCHWc8) are supported. ZenDNN library is intended to be used in conjunction with the frameworks mentioned above and cannot be used independently. The latest information on the ZenDNN release and installers is available on AMD Developer Central (https://developer.amd.com/zendnn/). # Supported OS and Compilers This release of ZenDNN supports the following Operating Systems (OS) and compilers: ## OS * Ubuntu® 18.04 LTS and later * Red Hat® Enterprise Linux® (RHEL) 8.0 and later * CentOS 7.9 and later ## Compilers * GCC 7.5 and later * AOCC (AMD Optimizing C/C++ Compiler) 3.0 (https://developer.amd.com/amd-aocc/) Theoretically, any Linux based OS with GLIBC version later than 2.17 could be supported. # Prerequisites The following prerequisites must be met for this release of ZenDNN: * AOCC 3.0 and AOCL (AMD-BLIS) 3.0.6 must be installed for PyTorch and ONNXRT binary packages. * Note: While GCC 7.5 and later are also supported compilers, AOCC is recommended for optimal performance of the ZenDNN library. # AOCC and AMD-BLIS Library Installation **AOCC** is a high performance, production quality code generation tool. AOCC can be downloaded from AMD Developer Central (https://developer.amd.com/amd-aocc/). ZenDNN compiled with AOCC may provide better performance as compared to the other open-source counterparts. **AMD-BLIS** is a portable open-source software framework for instantiating high-performance Basic Linear Algebra Subprograms (BLAS), such as, dense linear algebra libraries. AMD-BLIS is part of AOCL and can be downloaded from AMD Developer Central (https://developer.amd.com/amd-aocl/). Note: ZenDNN depends only on AMD-BLIS and has no dependency on any other AOCL library. ## General Convention The following points must be considered while installing AOCC and AMD-BLIS: * Change to the preferred directory where ZenDNN will be downloaded. * This parent folder is referred to as folder `` in the steps below. * It is good practice to keep AOCC 3.0 and AMD-BLIS 3.0.6 downloads in the same parent folder. * Assume that the parent folder for user setup follows this convention: `/home//my_work`. ## AMD-BLIS Library Setup Complete the following steps to setup the AOOC compiled BLIS library: 1. Execute the command `cd ` 2. Download aocl-linux-aocc-3.0-6.tar.gz. 3. Execute the following commands: ```bash tar -xvf aocl-linux-aocc-3.0-6.tar.gz cd aocl-linux-aocc-3.0-6 tar -xvf aocl-blis-linux-aocc-3.0-6.tar.gz cd amd-blis ``` This will set up the environment for BLIS AOCC path: ```bash export ZENDNN_BLIS_PATH=$(pwd) ``` For example: ```bash export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-aocc-3.0-6/amd-blis ``` Complete the following steps to setup the GCC compiled BLIS library: 1. Execute the command `cd `. 2. Download aocl-linux-gcc-3.0-6.tar.gz. 3. Execute the following commands: ```bash tar -xvf aocl-linux-gcc-3.0-6.tar.gz cd aocl-linux-gcc-3.0-6 tar -xvf aocl-blis-linux-gcc-3.0-6.tar.gz cd amd-blis ``` This will set up the environment for BLIS GCC path: ```bash export ZENDNN_BLIS_PATH=$(pwd) ``` For example: ```bash export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-gcc-3.0-6/amd-blis ``` ## AOCC Installation Complete the following steps to install AOCC: 1. Execute the command `cd `. 2. Download aocc-compiler-3.0.0.tar from the AMD Developer Central (https://developer.amd.com/amd-aocc/). 3. Execute the command ```bash tar -xvf aocc-compiler-3.0.0.tar ``` 4. Execute the command ```bash cd aocc-compiler-3.0.0 ``` This will install the compiler and display the AOCC set up instructions. 1. Execute the command ```bash bash install.sh ``` This will set up the environment for the AOCC path: ```bash export ZENDNN_AOCC_COMP_PATH=$(pwd) ``` For example: ```bash export ZENDNN_AOCC_COMP_PATH=/home//my_work/aocc-compiler-3.0.0 ``` The bashrc file can be edited to setup ZENDNN_AOCC_COMP_PATH environment path. For example, in the case of AOCC compiled AMD-BLIS: ```bash export ZENDNN_AOCC_COMP_PATH=/home//my_work/aocc-compiler-3.0.0 export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-aocc-3.0-6/amd-blis ``` For example, in the case of GCC compiled AMD-BLIS: ```bash export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-gcc-3.0-6/amd-blis ``` # Runtime Dependencies ZenDNN has the following runtime dependencies: * GNU C library (glibc.so) * GNU Standard C++ library (libstdc++.so) * Dynamic linking library (libdl.so) * POSIX Thread library (libpthread.so) * C Math Library (libm.so) * OpenMP (libomp.so) * Python v3.6-v3.9 for TensorFlow v2.7 * Python v3.6-v3.9 for PyTorch v1.9.0 * Python v3.8 for ONNXRT v1.8.0 Since ZenDNN is configured to use OpenMP, a C++ compiler with OpenMP 2.0 or later is required for runtime execution. # Build from Source Clone ZenDNN git: ```bash git clone https://github.com/amd/ZenDNN.git cd ZenDNN ``` ## AOCC compiler **ZENDNN_AOCC_COMP_PATH** and **ZENDNN_BLIS_PATH** should be defined. example: ```bash export ZENDNN_AOCC_COMP_PATH=/home//my_work/aocc-compiler-3.0.0 export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-aocc-3.0-6/amd-blis make clean source scripts/zendnn_aocc_build.sh ``` When new terminal is opened, user need to set up environment variables: ```bash source scripts/zendnn_aocc_env_setup.sh ``` ## GCC compiler **ZENDNN_BLIS_PATH** should be defined. example: ```bash export ZENDNN_BLIS_PATH=/home//my_work/aocl-linux-gcc-3.0-6/amd-blis make clean source scripts/zendnn_gcc_build.sh ``` When new terminal is opened, user need to set up environment variables: ```bash source scripts/zendnn_gcc_env_setup.sh ``` Please note above scripts must be sourced only from ZenDNN Folder. ## Validate the build After the library is built on Linux host, user can run unit tests using: ```bash source scripts/runApiTest.sh ``` Corresponding tests are located in the **tests/api_tests** directory. These unit tests don't produce any information/logs in the terminal. Library logs can be enabled with: ```bash ZENDNN_LOG_OPTS=ALL:2 source scripts/runApiTest.sh ``` # Logs Logging is disabled in the ZenDNN library by default. It can be enabled using the environment variable **ZENDNN_LOG_OPTS** before running any tests. Logging behavior can be specified by setting the environment variable **ZENDNN_LOG_OPTS** to a comma-delimited list of ACTOR:DBGLVL pairs. The different ACTORS are as follows: | ACTORS | Usage | :------ | :----------- | ALGO | Logs all algorithms executed | CORE | Logs all the core ZenDNN library operations | API | Logs all the ZenDNN API calls | TEST | Logs used in API tests, functionality tests and regression tests | PROF | Logs the performance of operations in millisecond | FWK | Logs all the framework (TensorFlow, ONNXRT, and PyTorch) specific calls For example: * To turn on info logging, use **ZENDNN_LOG_OPTS=ALL:2** * To turn off all logging, use **ZENDNN_LOG_OPTS=ALL:-1** * To only log errors, use **ZENDNN_LOG_OPTS=ALL:0** * To only log info for ALGO, use **ZENDNN_LOG_OPTS=ALL:-1,ALGO:2** * To only log info for CORE, use **ZENDNN_LOG_OPTS=ALL:-1,CORE:2** * To only log info for API, use **ZENDNN_LOG_OPTS=ALL:-1,API:2** * To only log info for PROF (profile), use **ZENDNN_LOG_OPTS=ALL:-1,PROF:2** * To only log info for FWK, use **ZENDNN_LOG_OPTS=ALL:-1,FWK:2** The Different Debug Levels (DBGLVL) are as follows: ```bash enum LogLevel { LOG_LEVEL_DISABLED = -1, LOG_LEVEL_ERROR = 0, LOG_LEVEL_WARNING = 1, LOG_LEVEL_INFO = 2, }; ``` # License ZenDNN is licensed under [Apache License Version 2.0](LICENSE). Refer to the "[LICENSE](LICENSE)" file for the full license text and copyright notice. This distribution includes third party software governed by separate license terms. 3-clause BSD license: * [Xbyak](https://github.com/herumi/xbyak) * [Googletest](https://github.com/google/googletest) * [Instrumentation and Tracing Technology API (ITT API)](https://github.com/intel/IntelSEAPI/tree/master/ittnotify) Apache License Version 2.0: * [oneDNN](https://github.com/oneapi-src/oneDNN) * [Xbyak_aarch64](https://github.com/fujitsu/xbyak_aarch64) * [TensorFlow](https://github.com/tensorflow/tensorflow) Boost Software License, Version 1.0: * [Boost C++ Libraries](https://www.boost.org/) BSD 2-Clause license: * [Caffe](https://github.com/BVLC/caffe) This third party software, even if included with the distribution of the Advanced Micro Devices software, may be governed by separate license terms, including without limitation, third party license terms, and open source software license terms. These separate license terms govern your use of the third party programs as set forth in the **THIRD-PARTY-PROGRAMS** file. # Technical Support Please email zendnnsupport@amd.com for questions, issues, and feedback on ZenDNN. Please submit your questions, feature requests, and bug reports on the [GitHub issues](https://https://github.com/amd/ZenDNN/issues) page.