Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / ComputeLibrary / 7f09cf7dfba61aad6d6d588b1b096f56bafaad5e / . / documentation / index.xhtml
blob: 83b46b3b132c9d1f447f0c05eaba353c96d61cba [file] [log] [blame]
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1<!-- HTML header for doxygen 1.8.15-->
2<!-- Remember to use version doxygen 1.8.15 +-->
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]3<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
4<html xmlns="http://www.w3.org/1999/xhtml">
5<head>
6<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
7<meta http-equiv="X-UA-Compatible" content="IE=9"/>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]8<meta name="generator" content="Doxygen 1.8.15"/>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]9<meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]10<title>Compute Library: Introduction</title>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]11<link href="tabs.css" rel="stylesheet" type="text/css"/>
12<script type="text/javascript" src="jquery.js"></script>
13<script type="text/javascript" src="dynsections.js"></script>
14<link href="navtree.css" rel="stylesheet" type="text/css"/>
15<script type="text/javascript" src="resize.js"></script>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]16<script type="text/javascript" src="navtreedata.js"></script>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]17<script type="text/javascript" src="navtree.js"></script>
18<script type="text/javascript">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]19/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]20 $(document).ready(initResizable);
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]21/* @license-end */</script>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]22<link href="search/search.css" rel="stylesheet" type="text/css"/>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]23<script type="text/javascript" src="search/searchdata.js"></script>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]24<script type="text/javascript" src="search/search.js"></script>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]25<script type="text/x-mathjax-config">
26 MathJax.Hub.Config({
27 extensions: ["tex2jax.js"],
28 jax: ["input/TeX","output/HTML-CSS"],
29});
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]30</script><script type="text/javascript" async="async" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]31<link href="doxygen.css" rel="stylesheet" type="text/css" />
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]32<link href="stylesheet.css" rel="stylesheet" type="text/css"/>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]33</head>
34<body>
35<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
36<div id="titlearea">
37<table cellspacing="0" cellpadding="0">
38 <tbody>
39 <tr style="height: 56px;">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]40 <img alt="Compute Library" src="https://raw.githubusercontent.com/ARM-software/ComputeLibrary/gh-pages/ACL_logo.png" style="max-width: 100%;margin-top: 15px;margin-left: 10px"/>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]41 <td style="padding-left: 0.5em;">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]42 <div id="projectname">
Jenkins7f09cf72020-01-22 18:08:16 +0000[diff] [blame^]43 &#160;<span id="projectnumber">19.11.1</span>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]44 </div>
45 </td>
46 </tr>
47 </tbody>
48</table>
49</div>
50<!-- end header part -->
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]51<!-- Generated by Doxygen 1.8.15 -->
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]52<script type="text/javascript">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]53/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]54var searchBox = new SearchBox("searchBox", "search",false,'Search');
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]55/* @license-end */
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]56</script>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]57<script type="text/javascript" src="menudata.js"></script>
58<script type="text/javascript" src="menu.js"></script>
59<script type="text/javascript">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]60/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]61$(function() {
62 initMenu('',true,false,'search.php','Search');
63 $(document).ready(function() { init_search(); });
64});
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]65/* @license-end */</script>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]66<div id="main-nav"></div>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]67</div><!-- top -->
68<div id="side-nav" class="ui-resizable side-nav-resizable">
69 <div id="nav-tree">
70 <div id="nav-tree-contents">
71 <div id="nav-sync" class="sync"></div>
72 </div>
73 </div>
74 <div id="splitbar" style="-moz-user-select:none;"
75 class="ui-resizable-handle">
76 </div>
77</div>
78<script type="text/javascript">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]79/* @license magnet:?xt=urn:btih:cf05388f2679ee054f2beb29a391d25f4e673ac3&amp;dn=gpl-2.0.txt GPL-v2 */
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]80$(document).ready(function(){initNavTree('index.xhtml','');});
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]81/* @license-end */
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]82</script>
83<div id="doc-content">
84<!-- window showing the filter options -->
85<div id="MSearchSelectWindow"
86 onmouseover="return searchBox.OnSearchSelectShow()"
87 onmouseout="return searchBox.OnSearchSelectHide()"
88 onkeydown="return searchBox.OnSearchSelectKey(event)">
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]89</div>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]90
91<!-- iframe showing the search results (closed by default) -->
92<div id="MSearchResultsWindow">
93<iframe src="javascript:void(0)" frameborder="0"
94 name="MSearchResults" id="MSearchResults">
95</iframe>
96</div>
97
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]98<div class="PageDoc"><div class="header">
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]99 <div class="headertitle">
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]100<div class="title">Introduction </div> </div>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]101</div><!--header-->
102<div class="contents">
103<div class="toc"><h3>Table of Contents</h3>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]104<ul><li class="level1"><a href="#S0_1_contact">Contact / Support</a></li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]105<li class="level1"><a href="#S0_2_prebuilt_binaries">Pre-built binaries</a></li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]106<li class="level1"><a href="#S1_file_organisation">File organisation</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]107<li class="level1"><a href="#S2_versions_changelog">Release versions and changelog</a><ul><li class="level2"><a href="#S2_1_versions">Release versions</a></li>
108<li class="level2"><a href="#S2_2_changelog">Changelog</a></li>
109</ul>
110</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]111<li class="level1"><a href="#S3_how_to_build">How to build the library and the examples</a><ul><li class="level2"><a href="#S3_1_build_options">Build options</a></li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]112<li class="level2"><a href="#S3_2_linux">Building for Linux</a><ul><li class="level3"><a href="#S3_2_1_library">How to build the library ?</a></li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]113<li class="level3"><a href="#S3_2_2_examples">How to manually build the examples ?</a></li>
114</ul>
115</li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]116<li class="level2"><a href="#S3_3_android">Building for Android</a><ul><li class="level3"><a href="#S3_3_1_library">How to build the library ?</a></li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]117<li class="level3"><a href="#S3_3_2_examples">How to manually build the examples ?</a></li>
118</ul>
119</li>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]120<li class="level2"><a href="#S3_4_bare_metal">Building for bare metal</a><ul><li class="level3"><a href="#S3_4_1_library">How to build the library ?</a></li>
121<li class="level3"><a href="#S3_4_2_examples">How to manually build the examples ?</a></li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]122</ul>
123</li>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]124<li class="level2"><a href="#S3_5_windows_host">Building on a Windows host system</a><ul><li class="level3"><a href="#S3_5_1_ubuntu_on_windows">Bash on Ubuntu on Windows</a></li>
125<li class="level3"><a href="#S3_5_2_cygwin">Cygwin</a></li>
126</ul>
127</li>
128<li class="level2"><a href="#S3_6_cl_stub_library">The OpenCL stub library</a></li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]129<li class="level2"><a href="#S3_7_gles_stub_library">The Linux OpenGLES and EGL stub libraries</a></li>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]130<li class="level2"><a href="#S3_8_cl_requirements">OpenCL DDK Requirements</a><ul><li class="level3"><a href="#S3_8_1_cl_hard_requirements">Hard Requirements</a></li>
131<li class="level3"><a href="#S3_8_2_cl_performance_requirements">Performance improvements</a></li>
132</ul>
133</li>
134<li class="level2"><a href="#S3_9_cl_tuner">OpenCL Tuner</a><ul><li class="level3"><a href="#S3_9_1_cl_tuner_how_to">How to use it</a></li>
135</ul>
136</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]137</ul>
138</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]139</ul>
140</div>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]141<div class="textblock"><p>The Computer Vision and Machine Learning library is a set of functions optimised for both ARM CPUs and GPUs using SIMD technologies.Several builds of the library are available using various configurations:</p><ul>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]142<li>OS: Linux, Android or bare metal.</li>
143<li>Architecture: armv7a (32bit) or arm64-v8a (64bit)</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]144<li>Technology: NEON / OpenCL / GLES_COMPUTE / NEON and OpenCL and GLES_COMPUTE</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]145<li>Debug / Asserts / Release: Use a build with asserts enabled to debug your application and enable extra validation. Once you are sure your application works as expected you can switch to a release build of the library for maximum performance.</li>
146</ul>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]147<h1><a class="anchor" id="S0_1_contact"></a>
148Contact / Support</h1>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]149<p>Please email <a href="#" onclick="location.href='mai'+'lto:'+'dev'+'el'+'ope'+'r@'+'arm'+'.c'+'om'; return false;">devel<span style="display: none;">.nosp@m.</span>oper<span style="display: none;">.nosp@m.</span>@arm.<span style="display: none;">.nosp@m.</span>com</a></p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]150<p>In order to facilitate the work of the support team please provide the build information of the library you are using. To get the version of the library you are using simply run: </p><pre class="fragment">$ strings android-armv7a-cl-asserts/libarm_compute.so | grep arm_compute_version
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]151arm_compute_version=v16.12 Build options: {'embed_kernels': '1', 'opencl': '1', 'arch': 'armv7a', 'neon': '0', 'asserts': '1', 'debug': '0', 'os': 'android', 'Werror': '1'} Git hash=f51a545d4ea12a9059fe4e598a092f1fd06dc858
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]152</pre><h1><a class="anchor" id="S0_2_prebuilt_binaries"></a>
153Pre-built binaries</h1>
154<p>For each release we provide some pre-built binaries of the library <a href="https://github.com/ARM-software/ComputeLibrary/releases">here</a></p>
155<p>These binaries have been built using the following toolchains:</p><ul>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]156<li>Linux armv7a: gcc-linaro-4.9-2016.02-x86_64_arm-linux-gnueabihf</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]157<li>Linux arm64-v8a: gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu</li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]158<li>Android armv7a: clang++ / libc++ NDK r17c</li>
159<li>Android am64-v8a: clang++ / libc++ NDK r17c</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]160</ul>
161<dl class="section warning"><dt>Warning</dt><dd>Make sure to use a compatible toolchain to build your application or you will get some std::bad_alloc errors at runtime.</dd></dl>
162<h1><a class="anchor" id="S1_file_organisation"></a>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]163File organisation</h1>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]164<p>This archive contains:</p><ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]165<li>The <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> header and source files</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]166<li>The latest Khronos OpenCL 1.2 C headers from the <a href="https://www.khronos.org/registry/cl/">Khronos OpenCL registry</a></li>
167<li>The latest Khronos cl2.hpp from the <a href="https://www.khronos.org/registry/cl/">Khronos OpenCL registry</a> (API version 2.1 when this document was written)</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]168<li>The latest Khronos OpenGL ES 3.1 C headers from the <a href="https://www.khronos.org/registry/gles/">Khronos OpenGL ES registry</a></li>
169<li>The latest Khronos EGL 1.5 C headers from the <a href="https://www.khronos.org/registry/gles/">Khronos EGL registry</a></li>
170<li>The sources for a stub version of libOpenCL.so, libGLESv1_CM.so, libGLESv2.so and libEGL.so to help you build your application.</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]171<li>An examples folder containing a few examples to compile and link against the library.</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]172<li>A <a class="el" href="dir_cbdb8362360e11eafe2fa3bc74cf0ffd.xhtml">utils</a> folder containing headers with some boiler plate code used by the examples.</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]173<li>This documentation.</li>
174</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]175<p>You should have the following file organisation: </p><pre class="fragment">.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]176├── arm_compute --&gt; All the arm_compute headers
Jenkins4ba87db2019-05-23 17:11:51 +0100[diff] [blame]177│ ├── graph.h --&gt; Includes all the Graph headers at once.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]178│   ├── core
179│   │   ├── CL
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]180│   │   │   ├── CLKernelLibrary.h --&gt; Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]181│   │   │   ├── CLKernels.h --&gt; Includes all the OpenCL kernels at once
182│   │   │   ├── CL specialisation of all the generic objects interfaces (ICLTensor, ICLImage, etc.)
183│   │   │   ├── kernels --&gt; Folder containing all the OpenCL kernels
184│   │   │   │   └── CL*Kernel.h
185│   │   │   └── OpenCL.h --&gt; Wrapper to configure the Khronos OpenCL C++ header
186│   │ ├── CPP
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]187│   │   │   ├── CPPKernels.h --&gt; Includes all the CPP kernels at once
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]188│   │ │   └── kernels --&gt; Folder containing all the CPP kernels
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]189│   │   │      └── CPP*Kernel.h
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]190│   │   ├── GLES_COMPUTE
191│   │   │   ├── GCKernelLibrary.h --&gt; Manages all the GLES kernels compilation and caching, provides accessors for the GLES Context.
192│   │   │   ├── GCKernels.h --&gt; Includes all the GLES kernels at once
193│   │   │   ├── GLES specialisation of all the generic objects interfaces (IGCTensor, IGCImage, etc.)
194│   │   │   ├── kernels --&gt; Folder containing all the GLES kernels
195│   │   │   │   └── GC*Kernel.h
196│   │   │   └── OpenGLES.h --&gt; Wrapper to configure the Khronos EGL and OpenGL ES C header
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]197│   │   ├── NEON
198│   │   │   ├── kernels --&gt; Folder containing all the NEON kernels
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]199│   │   │   │ ├── assembly --&gt; headers for assembly optimised NEON kernels.
200│   │   │   │ ├── convolution --&gt; headers for convolution assembly optimised NEON kernels.
201│   │   │   │   │   ├── common --&gt; headers for code which is common to several convolution implementations.
202│   │   │   │   │   ├── depthwise --&gt; headers for Depthwise convolultion assembly implementation
203│   │   │   │   │   └── winograd --&gt; headers for Winograd convolution assembly implementation
204│   │   │   │ ├── detail --&gt; Common code for several intrinsics implementations.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]205│   │   │   │   └── NE*Kernel.h
206│   │   │   └── NEKernels.h --&gt; Includes all the NEON kernels at once
207│   │   ├── All common basic types (Types.h, Window, Coordinates, Iterator, etc.)
208│   │   ├── All generic objects interfaces (ITensor, IImage, etc.)
209│   │   └── Objects metadata classes (ImageInfo, TensorInfo, MultiImageInfo)
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]210│   ├── graph
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]211│   │   ├── algorithms
212│   │   │   └── Generic algorithms used by the graph backend (e.g Order of traversal)
213│   │   ├── backends --&gt; The backend specific code
214│   │   │   ├── CL --&gt; OpenCL specific operations
215│   │   │   ├── GLES --&gt; OpenGLES Compute Shaders specific operations
216│   │   │   └── NEON --&gt; NEON specific operations
217│   │   ├── detail
218│   │   │   └── Collection of internal utilities.
219│   │   ├── frontend
220│   │   │   └── Code related to the stream frontend interface.
221│   │   ├── mutators
222│   │   │   └── Used to modify / optimise the Graph intermediate representation(Operator fusion, in place operations, etc.)
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]223│   │   ├── nodes
224│   │   │   └── The various nodes supported by the graph API
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]225│   │   ├── printers
226│   │   │   └── Debug printers
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]227│   │   └── Graph objects ( INode, ITensorAccessor, Graph, etc.)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]228│   └── runtime
229│   ├── CL
230│   │   ├── CL objects &amp; allocators (CLArray, CLImage, CLTensor, etc.)
231│   │   ├── functions --&gt; Folder containing all the OpenCL functions
232│   │   │   └── CL*.h
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]233│   │   ├── CLScheduler.h --&gt; Interface to enqueue OpenCL kernels and get/set the OpenCL CommandQueue and ICLTuner.
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]234│   │   ├── CLFunctions.h --&gt; Includes all the OpenCL functions at once
235│   │   └── tuners
236│   │      └── Local workgroup size tuners for specific architectures / GPUs
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]237│   ├── CPP
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]238│      │   ├── CPPKernels.h --&gt; Includes all the CPP functions at once.
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]239│   │   ├── CPPScheduler.h --&gt; Basic pool of threads to execute CPP/NEON code on several cores in parallel
240│   │   └── functions --&gt; Folder containing all the CPP functions
241│   │      └── CPP*.h
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]242│   ├── GLES_COMPUTE
243│   │   ├── GLES objects &amp; allocators (GCArray, GCImage, GCTensor, etc.)
244│   │   ├── functions --&gt; Folder containing all the GLES functions
245│   │   │   └── GC*.h
246│   │   ├── GCScheduler.h --&gt; Interface to enqueue GLES kernels and get/set the GLES CommandQueue.
247│   │   └── GCFunctions.h --&gt; Includes all the GLES functions at once
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]248│   ├── NEON
249│   │ ├── functions --&gt; Folder containing all the NEON functions
250│   │ │   └── NE*.h
251│   │ └── NEFunctions.h --&gt; Includes all the NEON functions at once
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]252│   ├── OMP
253│   │   └── OMPScheduler.h --&gt; OpenMP scheduler (Alternative to the CPPScheduler)
254│ ├── Memory manager files (LifetimeManager, PoolManager, etc.)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]255│   └── Basic implementations of the generic object interfaces (Array, Image, Tensor, etc.)
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]256├── data -&gt; Contains test images and reference data dumps used by validation tests
257├── docs -&gt; Contains Doxyfile and Doxygen sources used to generate the HTML pages in the documentation folder.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]258├── documentation
259│   ├── index.xhtml
260│   └── ...
261├── documentation.xhtml -&gt; documentation/index.xhtml
262├── examples
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]263│   ├── cl_*.cpp --&gt; OpenCL examples
264│   ├── gc_*.cpp --&gt; GLES compute shaders examples
265│   ├── graph_*.cpp --&gt; Graph examples
266│   ├── neoncl_*.cpp --&gt; NEON / OpenCL interoperability examples
267│   └── neon_*.cpp --&gt; NEON examples
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]268├── include
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]269│   ├── CL
270│   │ └── Khronos OpenCL C headers and C++ wrapper
271│   ├── half --&gt; FP16 library available from http://half.sourceforge.net
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]272│   ├── libnpy --&gt; Library to load / write npy buffers, available from https://github.com/llohse/libnpy
273│  └── linux --&gt; Headers only needed for Linux builds
274│   └── Khronos EGL and OpenGLES headers
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]275├── opencl-1.2-stubs
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]276│ └── opencl_stubs.c --&gt; OpenCL stubs implementation
277├── opengles-3.1-stubs
278│   ├── EGL.c --&gt; EGL stubs implementation
279│   └── GLESv2.c --&gt; GLESv2 stubs implementation
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]280├── scripts
281│   ├── caffe_data_extractor.py --&gt; Basic script to export weights from Caffe to npy files
282│   └── tensorflow_data_extractor.py --&gt; Basic script to export weights from Tensor Flow to npy files
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]283├── src
284│   ├── core
285│ │ └── ... (Same structure as headers)
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]286│   │ ├── CL
287│   │ │ └── cl_kernels --&gt; All the OpenCL kernels
288│   │ └── GLES_COMPUTE
289│   │ └── cs_shaders --&gt; All the OpenGL ES Compute Shaders
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]290│   ├── graph
291│ │ └── ... (Same structure as headers)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]292│ └── runtime
293│ └── ... (Same structure as headers)
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]294├── support
295│ └── Various headers to work around toolchains / platform issues.
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]296├── tests
297│   ├── All test related files shared between validation and benchmark
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]298│   ├── benchmark --&gt; Sources for benchmarking
299│ │ ├── Benchmark specific files
300│   │ ├── fixtures
301│ │ │ └── Backend agnostic fixtures to initialise and run the functions to test.
302│ │ ├── CL --&gt; OpenCL benchmarking tests
303│ │ ├── GLES_COMPUTE --&gt; GLES benchmarking tests
304│ │ └── NEON --&gt; NEON benchmarking tests
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]305│   ├── CL --&gt; OpenCL accessors
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]306│   ├── GLES_COMPUTE --&gt; GLES accessors
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]307│   ├── NEON --&gt; NEON accessors
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]308│   ├── datasets
309│ │ └── Datasets for all the validation / benchmark tests, layer configurations for various networks, etc.
310│   ├── framework
311│ │ └── Boiler plate code for both validation and benchmark test suites (Command line parsers, instruments, output loggers, etc.)
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]312│   └── validation --&gt; Sources for validation
313│ ├── Validation specific files
314│   ├── fixtures
315│ │ └── Backend agnostic fixtures to initialise and run the functions to test.
316│   ├── reference
317│ │ └── Reference implementation used to validate the results of the various backends.
318│ ├── CL --&gt; OpenCL validation tests
319│ ├── GLES_COMPUTE --&gt; GLES validation tests
320│ ├── CPP --&gt; C++ reference implementations
321│ └── NEON --&gt; NEON validation tests
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]322└── utils --&gt; Boiler plate code used by examples
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]323 └── Various utilities to print types, load / store assets, etc.
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]324</pre><h1><a class="anchor" id="S2_versions_changelog"></a>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]325Release versions and changelog</h1>
326<h2><a class="anchor" id="S2_1_versions"></a>
327Release versions</h2>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]328<p>All releases are numbered vYY.MM Where YY are the last two digits of the year, and MM the month number. If there is more than one release in a month then an extra sequential number is appended at the end: </p><pre class="fragment">v17.03 (First release of March 2017)
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]329v17.03.1 (Second release of March 2017)
330v17.04 (First release of April 2017)
331</pre><dl class="section note"><dt>Note</dt><dd>We're aiming at releasing one major public release with new features per quarter. All releases in between will only contain bug fixes.</dd></dl>
332<h2><a class="anchor" id="S2_2_changelog"></a>
333Changelog</h2>
Jenkins7f09cf72020-01-22 18:08:16 +0000[diff] [blame^]334<p>v19.11.1 Public maintenance release</p><ul>
335<li>Fix offset calculation in <a class="el" href="classarm__compute_1_1_n_e_reduction_operation_kernel.xhtml" title="NEON kernel to perform a reduction operation.">NEReductionOperationKernel</a>.</li>
336<li>Fix data layout in <a class="el" href="classarm__compute_1_1_n_e_scale_kernel.xhtml" title="NEON kernel to perform scaling on a tensor.">NEScaleKernel</a> for nhwc.</li>
337<li>Retain configuration step data layout to avoid side-effects.</li>
338<li>Perform sqrt in double domain for L2 pooling.</li>
339<li>Fix output shape calculation for Reduce Mean</li>
340<li>Restrict cases where optimized <a class="el" href="classarm__compute_1_1_n_e_pad_layer.xhtml" title="Basic function to pad a tensor.">NEPadLayer</a> runs.</li>
341</ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]342<p>v19.11 Public major release</p><ul>
343<li>Various bug fixes.</li>
344<li>Various optimisations.</li>
345<li>Updated recommended NDK version to r17c.</li>
346<li>Deprecated OpenCL kernels / functions:<ul>
347<li>CLDepthwiseConvolutionLayerReshapeWeightsGenericKernel</li>
348<li>CLDepthwiseIm2ColKernel</li>
349<li>CLDepthwiseSeparableConvolutionLayer</li>
350<li>CLDepthwiseVectorToTensorKernel</li>
351<li>CLDirectConvolutionLayerOutputStageKernel</li>
352</ul>
353</li>
354<li>Deprecated NEON kernels / functions:<ul>
355<li>NEDepthwiseWeightsReshapeKernel</li>
356<li>NEDepthwiseIm2ColKernel</li>
357<li>NEDepthwiseSeparableConvolutionLayer</li>
358<li>NEDepthwiseVectorToTensorKernel</li>
359<li>NEDepthwiseConvolutionLayer3x3</li>
360</ul>
361</li>
362<li>New OpenCL kernels / functions:<ul>
363<li><a class="el" href="classarm__compute_1_1_c_l_instance_normalization_layer_kernel.xhtml">CLInstanceNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_instance_normalization_layer.xhtml">CLInstanceNormalizationLayer</a></li>
364<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer_native_kernel.xhtml">CLDepthwiseConvolutionLayerNativeKernel</a> to replace the old generic depthwise convolution (see Deprecated OpenCL kernels / functions)</li>
365<li><a class="el" href="namespacearm__compute.xhtml#aa02883dd85b75a6eb0d4878f266908dd">CLLogSoftmaxLayer</a></li>
366</ul>
367</li>
368<li>New NEON kernels / functions:<ul>
369<li><a class="el" href="classarm__compute_1_1_n_e_bounding_box_transform_kernel.xhtml">NEBoundingBoxTransformKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_bounding_box_transform.xhtml">NEBoundingBoxTransform</a></li>
370<li><a class="el" href="classarm__compute_1_1_n_e_compute_all_anchors_kernel.xhtml">NEComputeAllAnchorsKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_compute_all_anchors.xhtml">NEComputeAllAnchors</a></li>
371<li><a class="el" href="classarm__compute_1_1_n_e_detection_post_process_layer.xhtml">NEDetectionPostProcessLayer</a></li>
372<li><a class="el" href="classarm__compute_1_1_n_e_generate_proposals_layer.xhtml">NEGenerateProposalsLayer</a></li>
373<li><a class="el" href="classarm__compute_1_1_n_e_instance_normalization_layer_kernel.xhtml">NEInstanceNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_instance_normalization_layer.xhtml">NEInstanceNormalizationLayer</a></li>
374<li><a class="el" href="namespacearm__compute.xhtml#a4478c830368ed024dc47a2bf48978616">NELogSoftmaxLayer</a></li>
375<li><a class="el" href="classarm__compute_1_1_n_e_r_o_i_align_layer_kernel.xhtml">NEROIAlignLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_r_o_i_align_layer.xhtml">NEROIAlignLayer</a></li>
376</ul>
377</li>
378<li>Added QASYMM8 support for:<ul>
379<li><a class="el" href="classarm__compute_1_1_c_l_generate_proposals_layer.xhtml">CLGenerateProposalsLayer</a></li>
380<li><a class="el" href="classarm__compute_1_1_c_l_r_o_i_align_layer.xhtml">CLROIAlignLayer</a></li>
381<li><a class="el" href="classarm__compute_1_1_c_p_p_box_with_non_maxima_suppression_limit.xhtml">CPPBoxWithNonMaximaSuppressionLimit</a></li>
382</ul>
383</li>
384<li>Added QASYMM16 support for:<ul>
385<li><a class="el" href="classarm__compute_1_1_c_l_bounding_box_transform.xhtml">CLBoundingBoxTransform</a></li>
386</ul>
387</li>
388<li>Added FP16 support for:<ul>
389<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_reshaped_kernel.xhtml">CLGEMMMatrixMultiplyReshapedKernel</a></li>
390</ul>
391</li>
392<li>Added new data type QASYMM8_PER_CHANNEL support for:<ul>
393<li><a class="el" href="classarm__compute_1_1_c_l_dequantization_layer.xhtml">CLDequantizationLayer</a></li>
394<li><a class="el" href="classarm__compute_1_1_n_e_dequantization_layer.xhtml">NEDequantizationLayer</a></li>
395</ul>
396</li>
397<li>Added new data type QSYMM8_PER_CHANNEL support for:<ul>
398<li><a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a></li>
399<li><a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">NEConvolutionLayer</a></li>
400<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a></li>
401<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a></li>
402</ul>
403</li>
404<li>Added FP16 mixed-precision support for:<ul>
405<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_reshaped_kernel.xhtml">CLGEMMMatrixMultiplyReshapedKernel</a></li>
406<li><a class="el" href="classarm__compute_1_1_c_l_pooling_layer_kernel.xhtml">CLPoolingLayerKernel</a></li>
407</ul>
408</li>
409<li>Added FP32 and FP16 ELU activation for:<ul>
410<li><a class="el" href="classarm__compute_1_1_c_l_activation_layer.xhtml">CLActivationLayer</a></li>
411<li><a class="el" href="classarm__compute_1_1_n_e_activation_layer.xhtml">NEActivationLayer</a></li>
412</ul>
413</li>
414<li>Added asymmetric padding support for:<ul>
415<li><a class="el" href="classarm__compute_1_1_c_l_direct_deconvolution_layer.xhtml">CLDirectDeconvolutionLayer</a></li>
416<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_deconvolution_layer.xhtml">CLGEMMDeconvolutionLayer</a></li>
417<li><a class="el" href="classarm__compute_1_1_n_e_deconvolution_layer.xhtml">NEDeconvolutionLayer</a></li>
418</ul>
419</li>
420<li>Added SYMMETRIC and REFLECT modes for <a class="el" href="classarm__compute_1_1_c_l_pad_layer_kernel.xhtml">CLPadLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_pad_layer.xhtml">CLPadLayer</a>.</li>
421<li>Replaced the calls to <a class="el" href="classarm__compute_1_1_n_e_copy_kernel.xhtml">NECopyKernel</a> and <a class="el" href="classarm__compute_1_1_n_e_memset_kernel.xhtml">NEMemsetKernel</a> with <a class="el" href="classarm__compute_1_1_n_e_pad_layer.xhtml">NEPadLayer</a> in <a class="el" href="classarm__compute_1_1_n_e_generate_proposals_layer.xhtml">NEGenerateProposalsLayer</a>.</li>
422<li>Replaced the calls to <a class="el" href="classarm__compute_1_1_c_l_copy_kernel.xhtml">CLCopyKernel</a> and <a class="el" href="classarm__compute_1_1_c_l_memset_kernel.xhtml">CLMemsetKernel</a> with <a class="el" href="classarm__compute_1_1_c_l_pad_layer.xhtml">CLPadLayer</a> in <a class="el" href="classarm__compute_1_1_c_l_generate_proposals_layer.xhtml">CLGenerateProposalsLayer</a>.</li>
423<li>Improved performance for CL Inception V3 - FP16.</li>
424<li>Improved accuracy for CL Inception V3 - FP16 by enabling FP32 accumulator (mixed-precision).</li>
425<li>Improved NEON performance by enabling fusing batch normalization with convolution and depth-wise convolution layer.</li>
426<li>Improved NEON performance for MobileNet-SSD by improving the output detection performance.</li>
427<li>Optimized <a class="el" href="classarm__compute_1_1_c_l_pad_layer.xhtml">CLPadLayer</a>.</li>
428<li>Optimized CL generic depthwise convolution layer by introducing <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer_native_kernel.xhtml">CLDepthwiseConvolutionLayerNativeKernel</a>.</li>
429<li>Reduced memory consumption by implementing weights sharing.</li>
430</ul>
Jenkins7f09cf72020-01-22 18:08:16 +0000[diff] [blame^]431<p>v19.08.1 Public maintenance release</p><ul>
432<li>Fix offset calculation in <a class="el" href="classarm__compute_1_1_n_e_reduction_operation_kernel.xhtml" title="NEON kernel to perform a reduction operation.">NEReductionOperationKernel</a>.</li>
433<li>Fix data layout in <a class="el" href="classarm__compute_1_1_n_e_scale_kernel.xhtml" title="NEON kernel to perform scaling on a tensor.">NEScaleKernel</a> for nhwc.</li>
434<li>Retain configuration step data layout to avoid side-effects.</li>
435<li>Perform sqrt in double domain for L2 pooling.</li>
436<li>Fix output shape calculation for Reduce Mean</li>
437<li>Fix broadcast CLPixelwiseMultiplication with 5D tensors</li>
438</ul>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]439<p>v19.08 Public major release</p><ul>
440<li>Various bug fixes.</li>
441<li>Various optimisations.</li>
442<li>Deprecated NEON functions<ul>
443<li>NEDepthConcatenateLayer</li>
444<li>NEWidthConcatenateLayer</li>
445</ul>
446</li>
447<li>Deprecated OpenCL kernels / functions<ul>
448<li>CLDepthConcatenateLayer</li>
449<li>CLGEMMInterleave4x4Kernel / CLGEMMInterleave4x4</li>
450<li>CLGEMMTranspose1xWKernel / CLGEMMTranspose1xW</li>
451<li>CLWidthConcatenateLayer</li>
452</ul>
453</li>
454<li>New NEON kernels / functions:<ul>
455<li><a class="el" href="classarm__compute_1_1_n_e_abs_layer.xhtml">NEAbsLayer</a></li>
456<li><a class="el" href="classarm__compute_1_1_n_e_cast.xhtml">NECast</a></li>
457<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_power.xhtml">NEElementwisePower</a></li>
458<li><a class="el" href="classarm__compute_1_1_n_e_log_layer.xhtml">NELogLayer</a></li>
459<li><a class="el" href="classarm__compute_1_1_n_e_l_s_t_m_layer_quantized.xhtml">NELSTMLayerQuantized</a></li>
460<li><a class="el" href="classarm__compute_1_1_n_e_neg_layer.xhtml">NENegLayer</a></li>
461<li><a class="el" href="classarm__compute_1_1_n_e_p_relu_layer.xhtml">NEPReluLayer</a></li>
462<li><a class="el" href="classarm__compute_1_1_n_e_sin_layer.xhtml">NESinLayer</a></li>
463<li><a class="el" href="classarm__compute_1_1_n_e_batch_concatenate_layer_kernel.xhtml">NEBatchConcatenateLayerKernel</a></li>
464<li><a class="el" href="classarm__compute_1_1_n_e_depth_to_space_layer_kernel.xhtml">NEDepthToSpaceLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_depth_to_space_layer.xhtml">NEDepthToSpaceLayer</a></li>
465<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer_native_kernel.xhtml">NEDepthwiseConvolutionLayerNativeKernel</a></li>
466<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_quantize_down_int32_to_int16_scale_by_fixed_point_kernel.xhtml">NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel</a></li>
467<li><a class="el" href="classarm__compute_1_1_n_e_mean_std_dev_normalization_kernel.xhtml">NEMeanStdDevNormalizationKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_mean_std_dev_normalization_layer.xhtml">NEMeanStdDevNormalizationLayer</a></li>
468<li><a class="el" href="classarm__compute_1_1_n_e_space_to_depth_layer_kernel.xhtml">NESpaceToDepthLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_space_to_depth_layer.xhtml">NESpaceToDepthLayer</a></li>
469</ul>
470</li>
471<li>New OpenCL kernels / functions:<ul>
472<li><a class="el" href="classarm__compute_1_1_c_l_abs_layer.xhtml">CLAbsLayer</a></li>
473<li><a class="el" href="classarm__compute_1_1_c_l_elementwise_power.xhtml">CLElementwisePower</a></li>
474<li><a class="el" href="classarm__compute_1_1_c_l_log_layer.xhtml">CLLogLayer</a></li>
475<li><a class="el" href="classarm__compute_1_1_c_l_l_s_t_m_layer_quantized.xhtml">CLLSTMLayerQuantized</a></li>
476<li><a class="el" href="classarm__compute_1_1_c_l_neg_layer.xhtml">CLNegLayer</a></li>
477<li><a class="el" href="classarm__compute_1_1_c_l_p_relu_layer.xhtml">CLPReluLayer</a></li>
478<li><a class="el" href="classarm__compute_1_1_c_l_sin_layer.xhtml">CLSinLayer</a></li>
479<li><a class="el" href="classarm__compute_1_1_c_l_batch_concatenate_layer_kernel.xhtml">CLBatchConcatenateLayerKernel</a></li>
480<li><a class="el" href="classarm__compute_1_1_c_l_depth_to_space_layer_kernel.xhtml">CLDepthToSpaceLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_depth_to_space_layer.xhtml">CLDepthToSpaceLayer</a></li>
481<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_native_kernel.xhtml">CLGEMMLowpMatrixMultiplyNativeKernel</a></li>
482<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_quantize_down_int32_to_int16_scale_by_fixed_point_kernel.xhtml">CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel</a></li>
483<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_native_kernel.xhtml">CLGEMMMatrixMultiplyNativeKernel</a></li>
484<li><a class="el" href="classarm__compute_1_1_c_l_mean_std_dev_normalization_kernel.xhtml">CLMeanStdDevNormalizationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_mean_std_dev_normalization_layer.xhtml">CLMeanStdDevNormalizationLayer</a></li>
485<li><a class="el" href="classarm__compute_1_1_c_l_space_to_depth_layer_kernel.xhtml">CLSpaceToDepthLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_space_to_depth_layer.xhtml">CLSpaceToDepthLayer</a></li>
486</ul>
487</li>
488<li>New examples:<ul>
489<li>neon_opticalflow</li>
490<li>cl_cache</li>
491<li>neon_permute</li>
492</ul>
493</li>
494<li>Added support for FP16 in <a class="el" href="classarm__compute_1_1_n_e_deconvolution_layer.xhtml">NEDeconvolutionLayer</a></li>
495<li>Added support for FP16 in <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer.xhtml">CLDeconvolutionLayer</a></li>
496<li>Added support for REDUCE_MIN and REDUCE_MAX in <a class="el" href="namespacearm__compute.xhtml#a5827eb9cb394e74af87f74bd354fb45b">ReductionOperation</a></li>
497<li>Enable the fusion of batch normalization with convolution and depthwise convolution layer for FP32 in the graph API (OpenCL only)</li>
498<li>Added support for fusing activation function and broadcast addition with the matrix multiplication for FP32 (OpenCL only)</li>
499<li>Re-factored the depthwise convolution layer kernel on NEON for generic cases</li>
500<li>Added an optimized depthwise convolution layer kernel for 5x5 filters (NEON only)</li>
501<li>Added support to enable OpenCL kernel cache. Added example showing how to load the prebuilt OpenCL kernels from a binary cache file</li>
502<li>Altered <a class="el" href="classarm__compute_1_1_quantization_info.xhtml">QuantizationInfo</a> interface to support per-channel quantization.</li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]503<li>The <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3.xhtml">CLDepthwiseConvolutionLayer3x3</a> will be included by <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a> to accommodate for future optimizations.</li>
504<li>The <a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer_optimized.xhtml">NEDepthwiseConvolutionLayerOptimized</a> will be included by <a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a> to accommodate for future optimizations.</li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]505<li>Removed inner_border_right and inner_border_top parameters from <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer.xhtml">CLDeconvolutionLayer</a> interface</li>
506<li>Removed inner_border_right and inner_border_top parameters from <a class="el" href="classarm__compute_1_1_n_e_deconvolution_layer.xhtml">NEDeconvolutionLayer</a> interface</li>
507<li>Optimized the NEON assembly kernel for GEMMLowp. The new implementation fuses the output stage and quantization with the matrix multiplication kernel</li>
508</ul>
Jenkins4ba87db2019-05-23 17:11:51 +0100[diff] [blame]509<p>v19.05 Public major release</p><ul>
510<li>Various bug fixes.</li>
511<li>Various optimisations.</li>
512<li>New Neon kernels / functions:<ul>
513<li><a class="el" href="classarm__compute_1_1_n_e_batch_to_space_layer_kernel.xhtml">NEBatchToSpaceLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_batch_to_space_layer.xhtml">NEBatchToSpaceLayer</a></li>
514<li><a class="el" href="classarm__compute_1_1_n_e_complex_pixel_wise_multiplication_kernel.xhtml">NEComplexPixelWiseMultiplicationKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_complex_pixel_wise_multiplication.xhtml">NEComplexPixelWiseMultiplication</a></li>
515<li><a class="el" href="classarm__compute_1_1_n_e_crop_kernel.xhtml">NECropKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_crop_resize.xhtml">NECropResize</a></li>
516<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_assembly_dispatch.xhtml">NEDepthwiseConvolutionAssemblyDispatch</a></li>
517<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t_digit_reverse_kernel.xhtml">NEFFTDigitReverseKernel</a></li>
518<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t_radix_stage_kernel.xhtml">NEFFTRadixStageKernel</a></li>
519<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t_scale_kernel.xhtml">NEFFTScaleKernel</a></li>
520<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_output_stage_kernel.xhtml">NEGEMMLowpOffsetContributionOutputStageKernel</a></li>
521<li><a class="el" href="classarm__compute_1_1_n_e_height_concatenate_layer_kernel.xhtml">NEHeightConcatenateLayerKernel</a></li>
522<li><a class="el" href="classarm__compute_1_1_n_e_space_to_batch_layer_kernel.xhtml">NESpaceToBatchLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_space_to_batch_layer.xhtml">NESpaceToBatchLayer</a></li>
523<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t1_d.xhtml">NEFFT1D</a></li>
524<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t2_d.xhtml">NEFFT2D</a></li>
525<li><a class="el" href="classarm__compute_1_1_n_e_f_f_t_convolution_layer.xhtml">NEFFTConvolutionLayer</a></li>
526</ul>
527</li>
528<li>New OpenCL kernels / functions:<ul>
529<li><a class="el" href="classarm__compute_1_1_c_l_complex_pixel_wise_multiplication_kernel.xhtml">CLComplexPixelWiseMultiplicationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_complex_pixel_wise_multiplication.xhtml">CLComplexPixelWiseMultiplication</a></li>
530<li><a class="el" href="classarm__compute_1_1_c_l_crop_kernel.xhtml">CLCropKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_crop_resize.xhtml">CLCropResize</a></li>
531<li><a class="el" href="classarm__compute_1_1_c_l_deconvolution_reshape_output_kernel.xhtml">CLDeconvolutionReshapeOutputKernel</a></li>
532<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t_digit_reverse_kernel.xhtml">CLFFTDigitReverseKernel</a></li>
533<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t_radix_stage_kernel.xhtml">CLFFTRadixStageKernel</a></li>
534<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t_scale_kernel.xhtml">CLFFTScaleKernel</a></li>
535<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_reshaped_only_r_h_s_kernel.xhtml">CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel</a></li>
536<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_reshaped_only_r_h_s_kernel.xhtml">CLGEMMMatrixMultiplyReshapedOnlyRHSKernel</a></li>
537<li><a class="el" href="classarm__compute_1_1_c_l_height_concatenate_layer_kernel.xhtml">CLHeightConcatenateLayerKernel</a></li>
538<li><a class="el" href="classarm__compute_1_1_c_l_direct_deconvolution_layer.xhtml">CLDirectDeconvolutionLayer</a></li>
539<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t1_d.xhtml">CLFFT1D</a></li>
540<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t2_d.xhtml">CLFFT2D</a></li>
541<li><a class="el" href="classarm__compute_1_1_c_l_f_f_t_convolution_layer.xhtml">CLFFTConvolutionLayer</a></li>
542<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_deconvolution_layer.xhtml">CLGEMMDeconvolutionLayer</a></li>
543</ul>
544</li>
545<li>New OpenGLES kernels / functions:<ul>
546<li><a class="el" href="classarm__compute_1_1_g_c_concatenate_layer.xhtml">GCConcatenateLayer</a></li>
547</ul>
548</li>
549<li>Deprecated functions/interfaces<ul>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]550<li>GCDepthConcatenateLayer</li>
551<li>NEWidthConcatenateLayer</li>
552<li>NEDepthConcatenateLayer</li>
553<li>CLWidthConcatenateLayer</li>
554<li>CLDepthConcatenateLayer</li>
555<li>CLGEMMInterleave4x4</li>
556<li>CLGEMMTranspose1xW</li>
Jenkins4ba87db2019-05-23 17:11:51 +0100[diff] [blame]557</ul>
558</li>
559<li>Support different quantization info in CLConcatLayer.</li>
560<li>Add checks on different input/output quantization info were not supported.</li>
561<li>Tensors have different quantization information.</li>
562<li>Add FP16 support checks.</li>
563<li>Fix output quantization CLDeptwiseConv3x3 when activation is fused.</li>
564<li>New graph examples:<ul>
565<li>graph_convolution</li>
566<li>graph_fully_connected</li>
567<li>graph_depthwise_convolution</li>
568<li>Deepspeech v0.4.1</li>
569</ul>
570</li>
571<li>Add support for QASYMM8 in <a class="el" href="classarm__compute_1_1_n_e_arithmetic_subtraction_kernel.xhtml" title="Interface for the kernel to perform subtraction between two tensors.">NEArithmeticSubtractionKernel</a>.</li>
572<li>Add support for QASYMM8 in <a class="el" href="classarm__compute_1_1_n_e_pixel_wise_multiplication_kernel.xhtml" title="Interface for the kernel to perform addition between two tensors.">NEPixelWiseMultiplicationKernel</a>.</li>
573<li>Add support for QASYMM8 NEDeconvolution.</li>
574<li>Add support for DequantizationLayer for NEON/CL.</li>
575<li>Add support for dilation in CLDepthwiseConvolution.</li>
576<li>Fuse offset contribution with the output stage when we use <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_multiply_core.xhtml" title="Basic function to execute GEMMLowpMatrixMultiplyCore on NEON.">NEGEMMLowpMatrixMultiplyCore</a>.</li>
577<li>Optimize CLDeconvolution.</li>
578<li>Add StackLayer to the graph API.</li>
579<li>Add support for "reflect" padding mode in NEPad.</li>
580<li>Winograd 7x7 NHWC on OpenCL.</li>
581<li>Rework CL ML layers to run exclusively on CL.</li>
582<li>Support different quantization info in PoolingLayer.</li>
583<li>Implement and test import memory interfaces.</li>
584<li>Added new tests and removed old ones.</li>
585<li>Various clang-tidy fixes.</li>
586</ul>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]587<p>v19.02 Public major release</p><ul>
588<li>Various bug fixes.</li>
589<li>Various optimisations.</li>
590<li>New Neon kernels / functions:<ul>
591<li><a class="el" href="classarm__compute_1_1_n_e_tile_kernel.xhtml">NETileKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_tile.xhtml">NETile</a></li>
592<li><a class="el" href="classarm__compute_1_1_n_e_fuse_batch_normalization_kernel.xhtml">NEFuseBatchNormalizationKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_fuse_batch_normalization.xhtml">NEFuseBatchNormalization</a></li>
593<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_operation_kernel.xhtml">NEElementwiseOperationKernel</a></li>
594<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_max.xhtml">NEElementwiseMax</a></li>
595<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_min.xhtml">NEElementwiseMin</a></li>
596<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_squared_diff.xhtml">NEElementwiseSquaredDiff</a></li>
597<li><a class="el" href="classarm__compute_1_1_n_e_select_kernel.xhtml">NESelectKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_select.xhtml">NESelect</a></li>
598<li><a class="el" href="classarm__compute_1_1_n_e_split.xhtml">NESplit</a></li>
599<li><a class="el" href="classarm__compute_1_1_n_e_slice.xhtml">NESlice</a></li>
600<li><a class="el" href="classarm__compute_1_1_n_e_unstack.xhtml">NEUnstack</a></li>
601<li><a class="el" href="classarm__compute_1_1_n_e_strided_slice_kernel.xhtml">NEStridedSliceKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_strided_slice.xhtml">NEStridedSlice</a></li>
602<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_unary_kernel.xhtml">NEElementwiseUnaryKernel</a></li>
603<li><a class="el" href="classarm__compute_1_1_n_e_rsqrt_layer.xhtml">NERsqrtLayer</a></li>
604<li><a class="el" href="classarm__compute_1_1_n_e_exp_layer.xhtml">NEExpLayer</a></li>
605<li><a class="el" href="classarm__compute_1_1_n_e_reverse_kernel.xhtml">NEReverseKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_reverse.xhtml">NEReverse</a></li>
606<li><a class="el" href="classarm__compute_1_1_n_e_arg_min_max_layer.xhtml">NEArgMinMaxLayer</a></li>
607<li><a class="el" href="classarm__compute_1_1_n_e_stack_layer_kernel.xhtml">NEStackLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_stack_layer.xhtml">NEStackLayer</a></li>
608<li><a class="el" href="classarm__compute_1_1_n_e_range_kernel.xhtml">NERangeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_range.xhtml">NERange</a></li>
609<li><a class="el" href="classarm__compute_1_1_n_e_pad_layer.xhtml">NEPadLayer</a></li>
610<li><a class="el" href="classarm__compute_1_1_n_e_memset_kernel.xhtml">NEMemsetKernel</a></li>
611<li><a class="el" href="classarm__compute_1_1_n_e_gather_kernel.xhtml">NEGatherKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_gather.xhtml">NEGather</a></li>
612<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_comparison.xhtml">NEElementwiseComparison</a></li>
613<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_comparison_static.xhtml">NEElementwiseComparisonStatic</a></li>
614<li><a class="el" href="classarm__compute_1_1_n_e_comparison_operation_kernel.xhtml">NEComparisonOperationKernel</a></li>
615<li><a class="el" href="classarm__compute_1_1_n_e_elementwise_division.xhtml">NEElementwiseDivision</a></li>
616</ul>
617</li>
618<li>New OpenCL kernels / functions:<ul>
619<li><a class="el" href="classarm__compute_1_1_c_l_select_kernel.xhtml">CLSelectKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_select.xhtml">CLSelect</a></li>
620<li><a class="el" href="classarm__compute_1_1_c_l_tile_kernel.xhtml">CLTileKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_tile.xhtml">CLTile</a></li>
621<li><a class="el" href="classarm__compute_1_1_c_l_comparison_kernel.xhtml">CLComparisonKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_comparison.xhtml">CLComparison</a></li>
622<li><a class="el" href="classarm__compute_1_1_c_l_arg_min_max_layer.xhtml">CLArgMinMaxLayer</a></li>
623<li><a class="el" href="classarm__compute_1_1_c_l_elementwise_max.xhtml">CLElementwiseMax</a></li>
624<li><a class="el" href="classarm__compute_1_1_c_l_elementwise_min.xhtml">CLElementwiseMin</a></li>
625<li><a class="el" href="classarm__compute_1_1_c_l_elementwise_squared_diff.xhtml">CLElementwiseSquaredDiff</a></li>
626<li><a class="el" href="classarm__compute_1_1_c_l_stack_layer_kernel.xhtml">CLStackLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_stack_layer.xhtml">CLStackLayer</a></li>
627<li><a class="el" href="classarm__compute_1_1_c_l_reverse.xhtml">CLReverse</a> / <a class="el" href="classarm__compute_1_1_c_l_reverse_kernel.xhtml">CLReverseKernel</a></li>
628<li><a class="el" href="classarm__compute_1_1_c_l_rsqrt_layer.xhtml">CLRsqrtLayer</a></li>
629<li><a class="el" href="classarm__compute_1_1_c_l_exp_layer.xhtml">CLExpLayer</a></li>
630<li><a class="el" href="classarm__compute_1_1_c_l_element_wise_unary_layer_kernel.xhtml">CLElementWiseUnaryLayerKernel</a></li>
631<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_reshape_l_h_s_matrix_kernel.xhtml">CLGEMMReshapeLHSMatrixKernel</a></li>
632<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_reshape_r_h_s_matrix_kernel.xhtml">CLGEMMReshapeRHSMatrixKernel</a></li>
633<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_reshaped_kernel.xhtml">CLGEMMMatrixMultiplyReshapedKernel</a></li>
634<li><a class="el" href="classarm__compute_1_1_c_l_range_kernel.xhtml">CLRangeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_range.xhtml">CLRange</a></li>
635<li><a class="el" href="classarm__compute_1_1_c_l_unstack.xhtml">CLUnstack</a></li>
636<li><a class="el" href="classarm__compute_1_1_c_l_gather_kernel.xhtml">CLGatherKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_gather.xhtml">CLGather</a></li>
637<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_reshaped_kernel.xhtml">CLGEMMLowpMatrixMultiplyReshapedKernel</a></li>
638</ul>
639</li>
640<li>New CPP kernels / functions:<ul>
641<li><a class="el" href="classarm__compute_1_1_c_p_p_detection_output_layer.xhtml">CPPDetectionOutputLayer</a></li>
642<li><a class="el" href="classarm__compute_1_1_c_p_p_top_k_v.xhtml">CPPTopKV</a> / <a class="el" href="classarm__compute_1_1_c_p_p_top_k_v_kernel.xhtml">CPPTopKVKernel</a></li>
643</ul>
644</li>
645<li>Added new examples:<ul>
646<li><a class="el" href="graph__ssd__mobilenet_8cpp.xhtml">graph_ssd_mobilenet.cpp</a></li>
647<li><a class="el" href="graph__mobilenet__v2_8cpp.xhtml">graph_mobilenet_v2.cpp</a></li>
648<li><a class="el" href="graph__resnet12_8cpp.xhtml">graph_resnet12.cpp</a></li>
649<li><a class="el" href="graph__srcnn955_8cpp.xhtml">graph_srcnn955.cpp</a></li>
650<li><a class="el" href="graph__vgg__vdsr_8cpp.xhtml">graph_vgg_vdsr.cpp</a></li>
651<li><a class="el" href="graph__inception__resnet__v1_8cpp.xhtml">graph_inception_resnet_v1.cpp</a></li>
652</ul>
653</li>
654<li>Add 4D tensors support to<ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]655<li><a class="el" href="namespacearm__compute.xhtml#a4df2143ca0a3bdbbbc54b440a52541cd">NESoftmaxLayer</a></li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]656</ul>
657</li>
658<li>Fused activation in <a class="el" href="classarm__compute_1_1_c_l_winograd_convolution_layer.xhtml">CLWinogradConvolutionLayer</a></li>
659<li>Extented <a class="el" href="classarm__compute_1_1_n_e_permute.xhtml">NEPermute</a> to support more cases</li>
660<li>Added NEON/SVE GEMM Hybrid kernels</li>
661<li>Added u8 and s8 hybrid assembly kernels</li>
662<li>Introduced GEMM strategy name in NEGEMMAssemblyWrapper</li>
663<li>Improved <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml">CLTuner</a></li>
664<li>Fused the bias addition within <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m.xhtml">CLGEMM</a></li>
665<li>Added support for QASYMM8 LOGISTIC activation in <a class="el" href="classarm__compute_1_1_n_e_activation_layer.xhtml">NEActivationLayer</a></li>
666<li>Added NHWC data layout support to:<ul>
667<li><a class="el" href="classarm__compute_1_1_n_e_scale.xhtml">NEScale</a> for F16</li>
668<li><a class="el" href="classarm__compute_1_1_c_l_normalization_layer.xhtml">CLNormalizationLayer</a> IN_MAP_2D for FP32/FP16</li>
669<li><a class="el" href="classarm__compute_1_1_n_e_l2_normalize_layer.xhtml">NEL2NormalizeLayer</a> for FP32/FP16</li>
670<li><a class="el" href="classarm__compute_1_1_n_e_normalization_layer.xhtml">NENormalizationLayer</a> IN_MAP_2D for FP32/FP16</li>
671<li><a class="el" href="classarm__compute_1_1_c_l_r_o_i_align_layer.xhtml">CLROIAlignLayer</a></li>
672<li><a class="el" href="classarm__compute_1_1_c_l_generate_proposals_layer.xhtml">CLGenerateProposalsLayer</a></li>
673</ul>
674</li>
675<li>Added QASYMM8 support to the following kernels:<ul>
676<li><a class="el" href="classarm__compute_1_1_n_e_arithmetic_addition_kernel.xhtml">NEArithmeticAdditionKernel</a></li>
677<li><a class="el" href="classarm__compute_1_1_n_e_scale.xhtml">NEScale</a></li>
678</ul>
679</li>
680<li>Added new tests and improved validation and benchmarking suites.</li>
681<li>Deprecated functions/interfaces<ul>
682<li>Usage of inner_border_right and inner_border_top has been deprecated in <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer.xhtml">CLDeconvolutionLayer</a> and <a class="el" href="classarm__compute_1_1_n_e_deconvolution_layer.xhtml">NEDeconvolutionLayer</a></li>
683</ul>
684</li>
685</ul>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]686<p>v18.11 Public major release</p><ul>
687<li>Various bug fixes.</li>
688<li>Various optimisations.</li>
689<li>New Neon kernels / functions:<ul>
690<li><a class="el" href="classarm__compute_1_1_n_e_channel_shuffle_layer.xhtml">NEChannelShuffleLayer</a> / <a class="el" href="classarm__compute_1_1_n_e_channel_shuffle_layer_kernel.xhtml">NEChannelShuffleLayerKernel</a></li>
691<li><a class="el" href="classarm__compute_1_1_n_e_reduce_mean.xhtml">NEReduceMean</a></li>
692<li><a class="el" href="classarm__compute_1_1_n_e_reorg_layer.xhtml">NEReorgLayer</a> / <a class="el" href="classarm__compute_1_1_n_e_reorg_layer_kernel.xhtml">NEReorgLayerKernel</a></li>
693<li><a class="el" href="classarm__compute_1_1_n_e_prior_box_layer.xhtml">NEPriorBoxLayer</a> / <a class="el" href="classarm__compute_1_1_n_e_prior_box_layer_kernel.xhtml">NEPriorBoxLayerKernel</a></li>
694<li><a class="el" href="classarm__compute_1_1_n_e_upsample_layer.xhtml">NEUpsampleLayer</a> / <a class="el" href="classarm__compute_1_1_n_e_upsample_layer_kernel.xhtml">NEUpsampleLayerKernel</a></li>
695<li><a class="el" href="classarm__compute_1_1_n_e_y_o_l_o_layer.xhtml">NEYOLOLayer</a> / <a class="el" href="classarm__compute_1_1_n_e_y_o_l_o_layer_kernel.xhtml">NEYOLOLayerKernel</a></li>
696</ul>
697</li>
698<li>New OpenCL kernels / functions:<ul>
699<li><a class="el" href="classarm__compute_1_1_c_l_batch_to_space_layer.xhtml">CLBatchToSpaceLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_batch_to_space_layer_kernel.xhtml">CLBatchToSpaceLayerKernel</a></li>
700<li><a class="el" href="classarm__compute_1_1_c_l_bounding_box_transform.xhtml">CLBoundingBoxTransform</a> / <a class="el" href="classarm__compute_1_1_c_l_bounding_box_transform_kernel.xhtml">CLBoundingBoxTransformKernel</a></li>
701<li><a class="el" href="classarm__compute_1_1_c_l_compute_all_anchors_kernel.xhtml">CLComputeAllAnchorsKernel</a></li>
702<li><a class="el" href="classarm__compute_1_1_c_l_generate_proposals_layer.xhtml">CLGenerateProposalsLayer</a></li>
703<li><a class="el" href="classarm__compute_1_1_c_l_normalize_planar_y_u_v_layer.xhtml">CLNormalizePlanarYUVLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_normalize_planar_y_u_v_layer_kernel.xhtml">CLNormalizePlanarYUVLayerKernel</a></li>
704<li><a class="el" href="classarm__compute_1_1_c_l_reorg_layer.xhtml">CLReorgLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_reorg_layer_kernel.xhtml">CLReorgLayerKernel</a></li>
705<li><a class="el" href="classarm__compute_1_1_c_l_space_to_batch_layer.xhtml">CLSpaceToBatchLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_space_to_batch_layer_kernel.xhtml">CLSpaceToBatchLayerKernel</a></li>
706<li><a class="el" href="classarm__compute_1_1_c_l_pad_layer.xhtml">CLPadLayer</a></li>
707<li><a class="el" href="classarm__compute_1_1_c_l_reduce_mean.xhtml">CLReduceMean</a></li>
708<li><a class="el" href="classarm__compute_1_1_c_l_prior_box_layer.xhtml">CLPriorBoxLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_prior_box_layer_kernel.xhtml">CLPriorBoxLayerKernel</a></li>
709<li><a class="el" href="classarm__compute_1_1_c_l_r_o_i_align_layer.xhtml">CLROIAlignLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_r_o_i_align_layer_kernel.xhtml">CLROIAlignLayerKernel</a></li>
710<li><a class="el" href="classarm__compute_1_1_c_l_slice.xhtml">CLSlice</a></li>
711<li><a class="el" href="classarm__compute_1_1_c_l_split.xhtml">CLSplit</a></li>
712<li><a class="el" href="classarm__compute_1_1_c_l_strided_slice.xhtml">CLStridedSlice</a> / <a class="el" href="classarm__compute_1_1_c_l_strided_slice_kernel.xhtml">CLStridedSliceKernel</a></li>
713<li><a class="el" href="classarm__compute_1_1_c_l_upsample_layer.xhtml">CLUpsampleLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_upsample_layer_kernel.xhtml">CLUpsampleLayerKernel</a></li>
714<li><a class="el" href="classarm__compute_1_1_c_l_y_o_l_o_layer.xhtml">CLYOLOLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_y_o_l_o_layer_kernel.xhtml">CLYOLOLayerKernel</a></li>
715</ul>
716</li>
717<li>New CPP kernels / functions:<ul>
718<li><a class="el" href="classarm__compute_1_1_c_p_p_box_with_non_maxima_suppression_limit.xhtml">CPPBoxWithNonMaximaSuppressionLimit</a> / <a class="el" href="classarm__compute_1_1_c_p_p_box_with_non_maxima_suppression_limit_kernel.xhtml">CPPBoxWithNonMaximaSuppressionLimitKernel</a></li>
719</ul>
720</li>
721<li>Added the validate method in:<ul>
722<li><a class="el" href="classarm__compute_1_1_n_e_depth_convert_layer.xhtml">NEDepthConvertLayer</a></li>
723<li><a class="el" href="classarm__compute_1_1_n_e_floor.xhtml">NEFloor</a> / <a class="el" href="classarm__compute_1_1_c_l_floor.xhtml">CLFloor</a></li>
724<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_addition_kernel.xhtml">NEGEMMMatrixAdditionKernel</a></li>
725<li><a class="el" href="classarm__compute_1_1_n_e_reshape_layer.xhtml">NEReshapeLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_reshape_layer.xhtml">CLReshapeLayer</a></li>
726<li><a class="el" href="classarm__compute_1_1_c_l_scale.xhtml">CLScale</a></li>
727</ul>
728</li>
729<li>Added new examples:<ul>
730<li><a class="el" href="graph__shufflenet_8cpp.xhtml">graph_shufflenet.cpp</a></li>
731<li><a class="el" href="graph__yolov3_8cpp.xhtml">graph_yolov3.cpp</a></li>
732</ul>
733</li>
734<li>Added documentation for add a new function or kernel.</li>
735<li>Improved doxygen documentation adding a list of the existing functions.</li>
736<li>Add 4D tensors support to<ul>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]737<li>CLWidthConcatenateLayer</li>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]738<li><a class="el" href="classarm__compute_1_1_c_l_flatten_layer.xhtml">CLFlattenLayer</a></li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]739<li><a class="el" href="namespacearm__compute.xhtml#a30ce3b40394b4f2d1e4cc31db7183425">CLSoftmaxLayer</a></li>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]740</ul>
741</li>
742<li>Add dot product support for <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_h_w_c_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NHWCKernel</a> non-unit stride</li>
743<li>Add SVE support</li>
744<li>Fused batch normalization into convolution layer weights in <a class="el" href="classarm__compute_1_1_c_l_fuse_batch_normalization.xhtml">CLFuseBatchNormalization</a></li>
745<li>Fuses activation in <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_c_h_w_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NCHWKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_h_w_c_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NHWCKernel</a> and <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_convolution_layer.xhtml">NEGEMMConvolutionLayer</a></li>
746<li>Added NHWC data layout support to:<ul>
747<li><a class="el" href="classarm__compute_1_1_c_l_channel_shuffle_layer.xhtml">CLChannelShuffleLayer</a></li>
748<li><a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer.xhtml">CLDeconvolutionLayer</a></li>
749<li><a class="el" href="classarm__compute_1_1_c_l_l2_normalize_layer.xhtml">CLL2NormalizeLayer</a></li>
750</ul>
751</li>
752<li>Added QASYMM8 support to the following kernels:<ul>
753<li><a class="el" href="classarm__compute_1_1_c_l_scale_kernel.xhtml">CLScaleKernel</a></li>
754<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer3x3_kernel.xhtml">NEDepthwiseConvolutionLayer3x3Kernel</a></li>
755<li><a class="el" href="classarm__compute_1_1_c_l_pixel_wise_multiplication_kernel.xhtml">CLPixelWiseMultiplicationKernel</a></li>
756</ul>
757</li>
758<li>Added FP16 support to the following kernels:<ul>
759<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_h_w_c_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NHWCKernel</a></li>
760<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer3x3_kernel.xhtml">NEDepthwiseConvolutionLayer3x3Kernel</a></li>
761<li><a class="el" href="classarm__compute_1_1_c_l_normalize_planar_y_u_v_layer_kernel.xhtml">CLNormalizePlanarYUVLayerKernel</a></li>
762<li><a class="el" href="classarm__compute_1_1_c_l_winograd_convolution_layer.xhtml">CLWinogradConvolutionLayer</a> (5x5 kernel)</li>
763</ul>
764</li>
765<li>More tests added to both validation and benchmarking suites.</li>
766</ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]767<p>v18.08 Public major release</p><ul>
768<li>Various bug fixes.</li>
769<li>Various optimisations.</li>
770<li>Updated recommended NDK version to r17b.</li>
771<li>Removed support for QS8/QS16 data types.</li>
772<li>Added support for grouped convolution in <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a>.</li>
773<li>Added NHWC data layout support to:<ul>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]774<li>NEDepthConcatenateLayer / CLDepthConcatenateLayer</li>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]775<li><a class="el" href="classarm__compute_1_1_n_e_winograd_convolution_layer.xhtml">NEWinogradConvolutionLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_winograd_convolution_layer.xhtml">CLWinogradConvolutionLayer</a></li>
776<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a></li>
777<li><a class="el" href="classarm__compute_1_1_c_l_direct_convolution_layer.xhtml">CLDirectConvolutionLayer</a></li>
778<li><a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a></li>
779<li><a class="el" href="classarm__compute_1_1_c_l_scale.xhtml">CLScale</a></li>
780<li><a class="el" href="classarm__compute_1_1_c_l_im2_col_kernel.xhtml">CLIm2ColKernel</a></li>
781</ul>
782</li>
783<li>New Neon kernels / functions:<ul>
784<li><a class="el" href="classarm__compute_1_1_n_e_r_n_n_layer.xhtml">NERNNLayer</a></li>
785</ul>
786</li>
787<li>New OpenCL kernels / functions:<ul>
788<li><a class="el" href="classarm__compute_1_1_c_l_arithmetic_division.xhtml">CLArithmeticDivision</a></li>
789</ul>
790</li>
791<li>Introduced prepare() stage support in the graph API for GLES.</li>
792<li>Added support for memory reusage when trying to allocate smaller CLTensors.</li>
793<li>Enabled NHWC execution on graph examples.</li>
794<li>Added JPEG accessor for validation purposes.</li>
795<li>Added validate methods to some kernels / functions.</li>
796</ul>
797<p>v18.05 Public major release</p><ul>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]798<li>Various bug fixes.</li>
799<li>Various optimisations.</li>
800<li>Major redesign in the interface for the neon kernels implemented in assembly.</li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]801<li>Removed arm_compute::NEGEMMLowpAArch64A53Kernel / arm_compute::NEGEMMLowpAArch64Kernel / arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_assembly_matrix_multiply_core.xhtml" title="Basic function to execute matrix multiply assembly kernels.">arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore</a> / arm_compute::NEHGEMMAArch64FP16Kernel</li>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]802<li>Added NEGEMMAssemblyWrapper and AssemblyKernelGlue which are used to execute assembly kernels in neon functions.</li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]803<li>Minor changes to the <a class="el" href="classarm__compute_1_1_c_p_u_info.xhtml">CPUInfo</a> type to make it compatible with the new assembly gemm interface.</li>
804<li>Moved neon assembly kernels to the folder src/core/NEON/kernels/arm_gemm.</li>
805<li>Improved doxygen documentation.</li>
806<li>Improved memory management for layer's transitions.</li>
807<li>Added support for NHWC data layout in tensors.</li>
808<li>Added NHWC data layout support to:<ul>
809<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_convolution_layer.xhtml">NEGEMMConvolutionLayer</a></li>
810<li><a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer.xhtml">NEDirectConvolutionLayer</a></li>
811<li><a class="el" href="classarm__compute_1_1_n_e_pooling_layer.xhtml">NEPoolingLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_pooling_layer.xhtml">CLPoolingLayer</a></li>
812<li><a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer.xhtml">NEBatchNormalizationLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">CLBatchNormalizationLayer</a></li>
813<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a></li>
814<li><a class="el" href="classarm__compute_1_1_n_e_scale.xhtml">NEScale</a></li>
815<li><a class="el" href="classarm__compute_1_1_n_e_im2_col.xhtml">NEIm2Col</a></li>
816</ul>
817</li>
818<li>Added support for dilated convolutions in <a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">NEConvolutionLayer</a> and <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a>.</li>
819<li>New OpenCL kernels / functions:<ul>
820<li><a class="el" href="classarm__compute_1_1_c_l_channel_shuffle_layer.xhtml">CLChannelShuffleLayer</a> / <a class="el" href="classarm__compute_1_1_c_l_channel_shuffle_layer_kernel.xhtml">CLChannelShuffleLayerKernel</a></li>
821<li><a class="el" href="classarm__compute_1_1_c_l_convert_fully_connected_weights_kernel.xhtml">CLConvertFullyConnectedWeightsKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_convert_fully_connected_weights.xhtml">CLConvertFullyConnectedWeights</a></li>
822<li><a class="el" href="classarm__compute_1_1_c_l_copy.xhtml">CLCopy</a> / <a class="el" href="classarm__compute_1_1_c_l_copy_kernel.xhtml">CLCopyKernel</a></li>
823<li><a class="el" href="classarm__compute_1_1_c_l_l_s_t_m_layer.xhtml">CLLSTMLayer</a></li>
824<li><a class="el" href="classarm__compute_1_1_c_l_r_n_n_layer.xhtml">CLRNNLayer</a></li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]825<li>CLWidthConcatenateLayer / <a class="el" href="classarm__compute_1_1_c_l_width_concatenate_layer_kernel.xhtml">CLWidthConcatenateLayerKernel</a></li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]826<li><a class="el" href="classarm__compute_1_1_c_l_winograd_filter_transform_kernel.xhtml">CLWinogradFilterTransformKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_winograd_input_transform_kernel.xhtml">CLWinogradInputTransformKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_winograd_convolution_layer.xhtml">CLWinogradConvolutionLayer</a></li>
827<li><a class="el" href="classarm__compute_1_1_c_l_winograd_input_transform_kernel.xhtml">CLWinogradInputTransformKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_winograd_input_transform.xhtml">CLWinogradInputTransform</a></li>
828</ul>
829</li>
830<li>New Neon kernels / functions:<ul>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]831<li><a class="el" href="classarm__compute_1_1_n_e_convert_fully_connected_weights_kernel.xhtml">NEConvertFullyConnectedWeightsKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_convert_fully_connected_weights.xhtml">NEConvertFullyConnectedWeights</a>.</li>
832</ul>
833</li>
834<li>Created the validate method in <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a>.</li>
835<li>Beta and gamma are no longer mandatory arguments in <a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer.xhtml">NEBatchNormalizationLayer</a> and <a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">CLBatchNormalizationLayer</a>.</li>
836<li>Added depth multiplier support in <a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a> and <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a>.</li>
837<li>Added broadcast multiply support in <a class="el" href="classarm__compute_1_1_n_e_pixel_wise_multiplication.xhtml">NEPixelWiseMultiplication</a> / <a class="el" href="classarm__compute_1_1_n_e_pixel_wise_multiplication_kernel.xhtml">NEPixelWiseMultiplicationKernel</a>.</li>
838<li>Port mobilenet example to NHWC data layout.</li>
839<li>Enabled Winograd method in <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a>.</li>
840<li>Renamed NEWinogradLayer to <a class="el" href="classarm__compute_1_1_n_e_winograd_convolution_layer.xhtml">NEWinogradConvolutionLayer</a>.</li>
841<li>Updated <a class="el" href="classarm__compute_1_1_n_e_winograd_convolution_layer.xhtml">NEWinogradConvolutionLayer</a> to use highly optimised assembly kernels in src/core/NEON/kernels/arm_gemm.</li>
842<li>Added memory manager support in GLES functions.</li>
843<li>Major refactoring of the graph API.</li>
844<li>Added GLES backend in the graph API.</li>
845<li>Added support for the memory manager in the graph API.</li>
846<li>Enabled Winograd Convolution method in the graph API.</li>
847<li>Added support for grouped convolutions in the graph API.</li>
848<li>Replaced NEDeconvolutionLayerUpsampleKernel with <a class="el" href="classarm__compute_1_1_n_e_scale_kernel.xhtml">NEScaleKernel</a> in <a class="el" href="classarm__compute_1_1_n_e_deconvolution_layer.xhtml">NEDeconvolutionLayer</a>.</li>
849<li>Added fast maths flag in <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a>.</li>
850<li>Added new tests and benchmarks in validation and benchmark frameworks</li>
851<li>Merge Activation layer with Convolution Layer (NEON. CL, GLES)</li>
852<li>Added support to OpenCL 2.0 SVM</li>
853<li>Added support to import memory in OpenCL tensors.</li>
854<li>Added the prepare() method to perform any one off pre-processing before running the function.</li>
855<li>Added new examples:<ul>
856<li><a class="el" href="graph__inception__v4_8cpp.xhtml">graph_inception_v4.cpp</a></li>
857<li><a class="el" href="graph__resnext50_8cpp.xhtml">graph_resnext50.cpp</a></li>
858</ul>
859</li>
860<li>Added memory measurement instrument for CL.</li>
861</ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]862<p>v18.03 Public maintenance release</p><ul>
863<li>Various bug fixes.</li>
864<li>Fixed bug in <a class="el" href="classarm__compute_1_1_n_e_activation_layer.xhtml">NEActivationLayer</a></li>
865<li>Fix in <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml">CLTuner</a> when using batches.</li>
866<li>Updated recommended NDK version to r16b (And fixed warnings).</li>
867<li>Fixed bug in validation code.</li>
868<li>Added Inception v4 graph example.</li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]869<li>Renamed NEWinogradLayer.cpp to <a class="el" href="classarm__compute_1_1_n_e_winograd_convolution_layer.xhtml">NEWinogradConvolutionLayer</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]870</ul>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]871<p>v18.02 Public major release</p><ul>
872<li>Various NEON / OpenCL / GLES optimisations.</li>
873<li>Various bug fixes.</li>
874<li>Changed default number of threads on big LITTLE systems.</li>
875<li>Refactored examples and added:<ul>
876<li>graph_mobilenet_qassym8</li>
877<li>graph_resnet</li>
878<li>graph_squeezenet_v1_1</li>
879</ul>
880</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]881<li>Renamed <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a> into <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_convolution_layer.xhtml">CLGEMMConvolutionLayer</a> and created a new <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a> to select the fastest convolution method.</li>
882<li>Renamed <a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">NEConvolutionLayer</a> into <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_convolution_layer.xhtml">NEGEMMConvolutionLayer</a> and created a new <a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">NEConvolutionLayer</a> to select the fastest convolution method.</li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]883<li>Added in place support to:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]884<li><a class="el" href="classarm__compute_1_1_c_l_activation_layer.xhtml">CLActivationLayer</a></li>
885<li><a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">CLBatchNormalizationLayer</a></li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]886</ul>
887</li>
888<li>Added QASYMM8 support to:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]889<li><a class="el" href="classarm__compute_1_1_c_l_activation_layer.xhtml">CLActivationLayer</a></li>
890<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a></li>
891<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a></li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]892<li><a class="el" href="namespacearm__compute.xhtml#a4df2143ca0a3bdbbbc54b440a52541cd">NESoftmaxLayer</a></li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]893</ul>
894</li>
895<li>Added FP16 support to:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]896<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3.xhtml">CLDepthwiseConvolutionLayer3x3</a></li>
897<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a></li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]898</ul>
899</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]900<li>Added broadcasting support to <a class="el" href="classarm__compute_1_1_n_e_arithmetic_addition.xhtml">NEArithmeticAddition</a> / <a class="el" href="classarm__compute_1_1_c_l_arithmetic_addition.xhtml">CLArithmeticAddition</a> / <a class="el" href="classarm__compute_1_1_c_l_pixel_wise_multiplication.xhtml">CLPixelWiseMultiplication</a></li>
901<li>Added fused batched normalization and activation to <a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">CLBatchNormalizationLayer</a> and <a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer.xhtml">NEBatchNormalizationLayer</a></li>
902<li>Added support for non-square pooling to <a class="el" href="classarm__compute_1_1_n_e_pooling_layer.xhtml">NEPoolingLayer</a> and <a class="el" href="classarm__compute_1_1_c_l_pooling_layer.xhtml">CLPoolingLayer</a></li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]903<li>New OpenCL kernels / functions:<ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]904<li>CLDirectConvolutionLayerOutputStageKernel</li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]905</ul>
906</li>
907<li>New NEON kernels / functions<ul>
908<li>Added name() method to all kernels.</li>
909<li>Added support for Winograd 5x5.</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]910<li><a class="el" href="classarm__compute_1_1_n_e_permute_kernel.xhtml">NEPermuteKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_permute.xhtml">NEPermute</a></li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]911<li><a class="el" href="classarm__compute_1_1_n_e_winograd_layer_transform_input_kernel.xhtml">NEWinogradLayerTransformInputKernel</a> / NEWinogradLayer</li>
912<li><a class="el" href="classarm__compute_1_1_n_e_winograd_layer_transform_output_kernel.xhtml">NEWinogradLayerTransformOutputKernel</a> / NEWinogradLayer</li>
913<li><a class="el" href="classarm__compute_1_1_n_e_winograd_layer_transform_weights_kernel.xhtml">NEWinogradLayerTransformWeightsKernel</a> / NEWinogradLayer</li>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]914<li>Renamed NEWinogradLayerKernel into NEWinogradLayerBatchedGEMMKernel</li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]915</ul>
916</li>
917<li>New GLES kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]918<li><a class="el" href="classarm__compute_1_1_g_c_tensor_shift_kernel.xhtml">GCTensorShiftKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_tensor_shift.xhtml">GCTensorShift</a></li>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]919</ul>
920</li>
921</ul>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]922<p>v18.01 Public maintenance release</p><ul>
923<li>Various bug fixes</li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]924<li>Added some of the missing <a class="el" href="namespacearm__compute.xhtml#a4feaaa70771629f4b5dcf3b219c8b647">validate()</a> methods</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]925<li>Added <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer_upsample_kernel.xhtml">CLDeconvolutionLayerUpsampleKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer.xhtml">CLDeconvolutionLayer</a> <a class="el" href="classarm__compute_1_1_c_l_deconvolution_layer_upsample.xhtml">CLDeconvolutionLayerUpsample</a></li>
926<li>Added <a class="el" href="classarm__compute_1_1_c_l_permute_kernel.xhtml">CLPermuteKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_permute.xhtml">CLPermute</a></li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]927<li>Added method to clean the programs cache in the CL <a class="el" href="classarm__compute_1_1_kernel.xhtml" title="Kernel class.">Kernel</a> library.</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]928<li>Added <a class="el" href="classarm__compute_1_1_g_c_arithmetic_addition_kernel.xhtml">GCArithmeticAdditionKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_arithmetic_addition.xhtml">GCArithmeticAddition</a></li>
929<li>Added <a class="el" href="classarm__compute_1_1_g_c_depthwise_convolution_layer3x3_kernel.xhtml">GCDepthwiseConvolutionLayer3x3Kernel</a> / <a class="el" href="classarm__compute_1_1_g_c_depthwise_convolution_layer3x3.xhtml">GCDepthwiseConvolutionLayer3x3</a></li>
930<li>Added <a class="el" href="classarm__compute_1_1_g_c_normalize_planar_y_u_v_layer_kernel.xhtml">GCNormalizePlanarYUVLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_normalize_planar_y_u_v_layer.xhtml">GCNormalizePlanarYUVLayer</a></li>
931<li>Added <a class="el" href="classarm__compute_1_1_g_c_scale_kernel.xhtml">GCScaleKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_scale.xhtml">GCScale</a></li>
932<li>Added <a class="el" href="classarm__compute_1_1_g_c_weights_reshape_kernel.xhtml">GCWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_convolution_layer.xhtml">GCConvolutionLayer</a></li>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]933<li>Added FP16 support to the following GLES compute kernels:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]934<li><a class="el" href="classarm__compute_1_1_g_c_col2_im_kernel.xhtml">GCCol2ImKernel</a></li>
935<li><a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_interleave4x4_kernel.xhtml">GCGEMMInterleave4x4Kernel</a></li>
936<li><a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_transpose1x_w_kernel.xhtml">GCGEMMTranspose1xWKernel</a></li>
937<li><a class="el" href="classarm__compute_1_1_g_c_im2_col_kernel.xhtml">GCIm2ColKernel</a></li>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]938</ul>
939</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]940<li>Refactored NEON Winograd (NEWinogradLayerKernel)</li>
941<li>Added <a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer_output_stage_kernel.xhtml">NEDirectConvolutionLayerOutputStageKernel</a></li>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]942<li>Added QASYMM8 support to the following NEON kernels:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]943<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer3x3_kernel.xhtml">NEDepthwiseConvolutionLayer3x3Kernel</a></li>
944<li><a class="el" href="classarm__compute_1_1_n_e_fill_border_kernel.xhtml">NEFillBorderKernel</a></li>
945<li><a class="el" href="classarm__compute_1_1_n_e_pooling_layer_kernel.xhtml">NEPoolingLayerKernel</a></li>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]946</ul>
947</li>
948<li>Added new examples:<ul>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]949<li>graph_cl_mobilenet_qasymm8.cpp</li>
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]950<li><a class="el" href="graph__inception__v3_8cpp.xhtml">graph_inception_v3.cpp</a></li>
951<li><a class="el" href="gc__dc_8cpp.xhtml">gc_dc.cpp</a></li>
952</ul>
953</li>
954<li>More tests added to both validation and benchmarking suites.</li>
955</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]956<p>v17.12 Public major release</p><ul>
957<li>Most machine learning functions on OpenCL support the new data type QASYMM8</li>
958<li>Introduced logging interface</li>
959<li>Introduced opencl timer</li>
960<li>Reworked GEMMLowp interface</li>
961<li>Added new NEON assembly kernels for GEMMLowp, SGEMM and HGEMM</li>
962<li>Added validation method for most Machine Learning kernels / functions</li>
963<li>Added new graph examples such as googlenet, mobilenet, squeezenet, vgg16 and vgg19</li>
964<li>Added sgemm example for OpenCL</li>
965<li>Added absolute difference example for GLES compute</li>
966<li>Added new tests and benchmarks in validation and benchmark frameworks</li>
967<li>Added new kernels / functions for GLES compute</li>
968<li>New OpenGL ES kernels / functions<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]969<li><a class="el" href="classarm__compute_1_1_g_c_absolute_difference_kernel.xhtml">GCAbsoluteDifferenceKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_absolute_difference.xhtml">GCAbsoluteDifference</a></li>
970<li><a class="el" href="classarm__compute_1_1_g_c_activation_layer_kernel.xhtml">GCActivationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_activation_layer.xhtml">GCActivationLayer</a></li>
971<li><a class="el" href="classarm__compute_1_1_g_c_batch_normalization_layer_kernel.xhtml">GCBatchNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_batch_normalization_layer.xhtml">GCBatchNormalizationLayer</a></li>
972<li><a class="el" href="classarm__compute_1_1_g_c_col2_im_kernel.xhtml">GCCol2ImKernel</a></li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]973<li><a class="el" href="classarm__compute_1_1_g_c_depth_concatenate_layer_kernel.xhtml">GCDepthConcatenateLayerKernel</a> / GCDepthConcatenateLayer</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]974<li><a class="el" href="classarm__compute_1_1_g_c_direct_convolution_layer_kernel.xhtml">GCDirectConvolutionLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_direct_convolution_layer.xhtml">GCDirectConvolutionLayer</a></li>
975<li><a class="el" href="classarm__compute_1_1_g_c_dropout_layer_kernel.xhtml">GCDropoutLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_dropout_layer.xhtml">GCDropoutLayer</a></li>
976<li><a class="el" href="classarm__compute_1_1_g_c_fill_border_kernel.xhtml">GCFillBorderKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_fill_border.xhtml">GCFillBorder</a></li>
977<li><a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_interleave4x4_kernel.xhtml">GCGEMMInterleave4x4Kernel</a> / <a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_interleave4x4.xhtml">GCGEMMInterleave4x4</a></li>
978<li><a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_matrix_accumulate_biases_kernel.xhtml">GCGEMMMatrixAccumulateBiasesKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_matrix_addition_kernel.xhtml">GCGEMMMatrixAdditionKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_matrix_multiply_kernel.xhtml">GCGEMMMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_g_e_m_m.xhtml">GCGEMM</a></li>
979<li><a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_transpose1x_w_kernel.xhtml">GCGEMMTranspose1xWKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_g_e_m_m_transpose1x_w.xhtml">GCGEMMTranspose1xW</a></li>
980<li><a class="el" href="classarm__compute_1_1_g_c_im2_col_kernel.xhtml">GCIm2ColKernel</a></li>
981<li><a class="el" href="classarm__compute_1_1_g_c_normalization_layer_kernel.xhtml">GCNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_normalization_layer.xhtml">GCNormalizationLayer</a></li>
982<li><a class="el" href="classarm__compute_1_1_g_c_pixel_wise_multiplication_kernel.xhtml">GCPixelWiseMultiplicationKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_pixel_wise_multiplication.xhtml">GCPixelWiseMultiplication</a></li>
983<li><a class="el" href="classarm__compute_1_1_g_c_pooling_layer_kernel.xhtml">GCPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_pooling_layer.xhtml">GCPoolingLayer</a></li>
984<li><a class="el" href="classarm__compute_1_1_g_c_logits1_d_max_kernel.xhtml">GCLogits1DMaxKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_logits1_d_shift_exp_sum_kernel.xhtml">GCLogits1DShiftExpSumKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_logits1_d_norm_kernel.xhtml">GCLogits1DNormKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_softmax_layer.xhtml">GCSoftmaxLayer</a></li>
985<li><a class="el" href="classarm__compute_1_1_g_c_transpose_kernel.xhtml">GCTransposeKernel</a> / <a class="el" href="classarm__compute_1_1_g_c_transpose.xhtml">GCTranspose</a></li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]986</ul>
987</li>
988<li>New NEON kernels / functions<ul>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]989<li>arm_compute::NEGEMMLowpAArch64A53Kernel / arm_compute::NEGEMMLowpAArch64Kernel / arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_assembly_matrix_multiply_core.xhtml" title="Basic function to execute matrix multiply assembly kernels.">arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore</a></li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]990<li>arm_compute::NEHGEMMAArch64FP16Kernel</li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]991<li><a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer3x3_kernel.xhtml">NEDepthwiseConvolutionLayer3x3Kernel</a> / NEDepthwiseIm2ColKernel / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_vector_multiply_kernel.xhtml">NEGEMMMatrixVectorMultiplyKernel</a> / NEDepthwiseVectorToTensorKernel / <a class="el" href="classarm__compute_1_1_n_e_depthwise_convolution_layer.xhtml">NEDepthwiseConvolutionLayer</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]992<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_offset_contribution_kernel.xhtml">NEGEMMLowpOffsetContributionKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_a_reduction_kernel.xhtml">NEGEMMLowpMatrixAReductionKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_b_reduction_kernel.xhtml">NEGEMMLowpMatrixBReductionKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_multiply_core.xhtml">NEGEMMLowpMatrixMultiplyCore</a></li>
993<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_by_fixed_point_kernel.xhtml">NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_by_fixed_point.xhtml">NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint</a></li>
994<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_kernel.xhtml">NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale.xhtml">NEGEMMLowpQuantizeDownInt32ToUint8Scale</a></li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]995<li>NEWinogradLayer / NEWinogradLayerKernel</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]996</ul>
997</li>
998<li>New OpenCL kernels / functions<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]999<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_offset_contribution_kernel.xhtml">CLGEMMLowpOffsetContributionKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_a_reduction_kernel.xhtml">CLGEMMLowpMatrixAReductionKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_b_reduction_kernel.xhtml">CLGEMMLowpMatrixBReductionKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_core.xhtml">CLGEMMLowpMatrixMultiplyCore</a></li>
1000<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_by_fixed_point_kernel.xhtml">CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_by_fixed_point.xhtml">CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint</a></li>
1001<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale_kernel.xhtml">CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_quantize_down_int32_to_uint8_scale.xhtml">CLGEMMLowpQuantizeDownInt32ToUint8Scale</a></li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1002</ul>
1003</li>
1004<li>New graph nodes for NEON and OpenCL<ul>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1005<li>graph::BranchLayer</li>
1006<li>graph::DepthConvertLayer</li>
1007<li>graph::DepthwiseConvolutionLayer</li>
1008<li>graph::DequantizationLayer</li>
1009<li>graph::FlattenLayer</li>
1010<li>graph::QuantizationLayer</li>
1011<li>graph::ReshapeLayer</li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1012</ul>
1013</li>
1014</ul>
1015<p>v17.10 Public maintenance release</p><ul>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1016<li>Bug fixes:<ul>
1017<li>Check the maximum local workgroup size supported by OpenCL devices</li>
1018<li>Minor documentation updates (Fixed instructions to build the examples)</li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1019<li>Introduced a <a class="el" href="classarm__compute_1_1graph_1_1_graph_context.xhtml" title="Graph context.">graph::GraphContext</a></li>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1020<li>Added a few new Graph nodes, support for branches and grouping.</li>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1021<li>Automatically enable cl_printf in debug builds</li>
1022<li>Fixed bare metal builds for armv7a</li>
1023<li>Added AlexNet and cartoon effect examples</li>
1024<li>Fixed library builds: libraries are no longer built as supersets of each other.(It means application using the Runtime part of the library now need to link against both libarm_compute_core and libarm_compute)</li>
1025</ul>
1026</li>
1027</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1028<p>v17.09 Public major release</p><ul>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1029<li>Experimental Graph support: initial implementation of a simple stream API to easily chain machine learning layers.</li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1030<li><a class="el" href="classarm__compute_1_1_memory.xhtml" title="CPU implementation of memory object.">Memory</a> Manager (<a class="el" href="classarm__compute_1_1_blob_lifetime_manager.xhtml">BlobLifetimeManager</a>, <a class="el" href="classarm__compute_1_1_blob_memory_pool.xhtml">BlobMemoryPool</a>, <a class="el" href="classarm__compute_1_1_i_lifetime_manager.xhtml">ILifetimeManager</a>, <a class="el" href="classarm__compute_1_1_i_memory_group.xhtml">IMemoryGroup</a>, <a class="el" href="classarm__compute_1_1_i_memory_manager.xhtml">IMemoryManager</a>, <a class="el" href="classarm__compute_1_1_i_memory_pool.xhtml">IMemoryPool</a>, <a class="el" href="classarm__compute_1_1_i_pool_manager.xhtml">IPoolManager</a>, <a class="el" href="classarm__compute_1_1_memory_manager_on_demand.xhtml">MemoryManagerOnDemand</a>, <a class="el" href="classarm__compute_1_1_pool_manager.xhtml">PoolManager</a>)</li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1031<li>New validation and benchmark frameworks (Boost and Google frameworks replaced by homemade framework).</li>
1032<li>Most machine learning functions support both fixed point 8 and 16 bit (QS8, QS16) for both NEON and OpenCL.</li>
1033<li>New NEON kernels / functions:<ul>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1034<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_assembly_base_kernel.xhtml" title="Base class for GEMM NEON kernels implemented in Assembly.">arm_compute::NEGEMMAssemblyBaseKernel</a> arm_compute::NEGEMMAArch64Kernel</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1035<li><a class="el" href="classarm__compute_1_1_n_e_dequantization_layer_kernel.xhtml">NEDequantizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_dequantization_layer.xhtml">NEDequantizationLayer</a></li>
1036<li><a class="el" href="classarm__compute_1_1_n_e_floor_kernel.xhtml">NEFloorKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_floor.xhtml">NEFloor</a></li>
1037<li><a class="el" href="classarm__compute_1_1_n_e_l2_normalize_layer_kernel.xhtml">NEL2NormalizeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_l2_normalize_layer.xhtml">NEL2NormalizeLayer</a></li>
1038<li><a class="el" href="classarm__compute_1_1_n_e_quantization_layer_kernel.xhtml">NEQuantizationLayerKernel</a> <a class="el" href="classarm__compute_1_1_n_e_min_max_layer_kernel.xhtml">NEMinMaxLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_quantization_layer.xhtml">NEQuantizationLayer</a></li>
1039<li><a class="el" href="classarm__compute_1_1_n_e_r_o_i_pooling_layer_kernel.xhtml">NEROIPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_r_o_i_pooling_layer.xhtml">NEROIPoolingLayer</a></li>
1040<li><a class="el" href="classarm__compute_1_1_n_e_reduction_operation_kernel.xhtml">NEReductionOperationKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_reduction_operation.xhtml">NEReductionOperation</a></li>
1041<li><a class="el" href="classarm__compute_1_1_n_e_reshape_layer_kernel.xhtml">NEReshapeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_reshape_layer.xhtml">NEReshapeLayer</a></li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1042</ul>
1043</li>
1044<li>New OpenCL kernels / functions:<ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1045<li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_c_h_w_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NCHWKernel</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3_n_h_w_c_kernel.xhtml">CLDepthwiseConvolutionLayer3x3NHWCKernel</a> CLDepthwiseIm2ColKernel CLDepthwiseVectorToTensorKernel CLDepthwiseWeightsReshapeKernel / <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer3x3.xhtml">CLDepthwiseConvolutionLayer3x3</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution_layer.xhtml">CLDepthwiseConvolutionLayer</a> CLDepthwiseSeparableConvolutionLayer</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1046<li><a class="el" href="classarm__compute_1_1_c_l_dequantization_layer_kernel.xhtml">CLDequantizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_dequantization_layer.xhtml">CLDequantizationLayer</a></li>
1047<li><a class="el" href="classarm__compute_1_1_c_l_direct_convolution_layer_kernel.xhtml">CLDirectConvolutionLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_direct_convolution_layer.xhtml">CLDirectConvolutionLayer</a></li>
1048<li><a class="el" href="classarm__compute_1_1_c_l_flatten_layer.xhtml">CLFlattenLayer</a></li>
1049<li><a class="el" href="classarm__compute_1_1_c_l_floor_kernel.xhtml">CLFloorKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_floor.xhtml">CLFloor</a></li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]1050<li>CLGEMMTranspose1xW</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1051<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_vector_multiply_kernel.xhtml">CLGEMMMatrixVectorMultiplyKernel</a></li>
1052<li><a class="el" href="classarm__compute_1_1_c_l_l2_normalize_layer_kernel.xhtml">CLL2NormalizeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_l2_normalize_layer.xhtml">CLL2NormalizeLayer</a></li>
1053<li><a class="el" href="classarm__compute_1_1_c_l_quantization_layer_kernel.xhtml">CLQuantizationLayerKernel</a> <a class="el" href="classarm__compute_1_1_c_l_min_max_layer_kernel.xhtml">CLMinMaxLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_quantization_layer.xhtml">CLQuantizationLayer</a></li>
1054<li><a class="el" href="classarm__compute_1_1_c_l_r_o_i_pooling_layer_kernel.xhtml">CLROIPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_r_o_i_pooling_layer.xhtml">CLROIPoolingLayer</a></li>
1055<li><a class="el" href="classarm__compute_1_1_c_l_reduction_operation_kernel.xhtml">CLReductionOperationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_reduction_operation.xhtml">CLReductionOperation</a></li>
1056<li><a class="el" href="classarm__compute_1_1_c_l_reshape_layer_kernel.xhtml">CLReshapeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_reshape_layer.xhtml">CLReshapeLayer</a></li>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1057</ul>
1058</li>
1059</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1060<p>v17.06 Public major release</p><ul>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1061<li>Various bug fixes</li>
1062<li>Added support for fixed point 8 bit (QS8) to the various NEON machine learning kernels.</li>
1063<li>Added unit tests and benchmarks (AlexNet, LeNet)</li>
1064<li>Added support for sub tensors.</li>
1065<li>Added infrastructure to provide GPU specific optimisation for some OpenCL kernels.</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1066<li>Added <a class="el" href="classarm__compute_1_1_o_m_p_scheduler.xhtml">OMPScheduler</a> (OpenMP) scheduler for NEON</li>
1067<li>Added <a class="el" href="classarm__compute_1_1_single_thread_scheduler.xhtml">SingleThreadScheduler</a> scheduler for NEON (For bare metal)</li>
1068<li>User can specify his own scheduler by implementing the <a class="el" href="classarm__compute_1_1_i_scheduler.xhtml">IScheduler</a> interface.</li>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1069<li>New OpenCL kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1070<li><a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer_kernel.xhtml">CLBatchNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">CLBatchNormalizationLayer</a></li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]1071<li><a class="el" href="classarm__compute_1_1_c_l_depth_concatenate_layer_kernel.xhtml">CLDepthConcatenateLayerKernel</a> / CLDepthConcatenateLayer</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1072<li><a class="el" href="classarm__compute_1_1_c_l_h_o_g_orientation_binning_kernel.xhtml">CLHOGOrientationBinningKernel</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_block_normalization_kernel.xhtml">CLHOGBlockNormalizationKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_h_o_g_detector_kernel.xhtml">CLHOGDetectorKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_h_o_g_descriptor.xhtml">CLHOGDescriptor</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_detector.xhtml">CLHOGDetector</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_gradient.xhtml">CLHOGGradient</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_multi_detection.xhtml">CLHOGMultiDetection</a></li>
1073<li><a class="el" href="classarm__compute_1_1_c_l_locally_connected_matrix_multiply_kernel.xhtml">CLLocallyConnectedMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_locally_connected_layer.xhtml">CLLocallyConnectedLayer</a></li>
1074<li><a class="el" href="classarm__compute_1_1_c_l_weights_reshape_kernel.xhtml">CLWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_convolution_layer_reshape_weights.xhtml">CLConvolutionLayerReshapeWeights</a></li>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1075</ul>
1076</li>
1077<li>New C++ kernels:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1078<li><a class="el" href="classarm__compute_1_1_c_p_p_detection_window_non_maxima_suppression_kernel.xhtml">CPPDetectionWindowNonMaximaSuppressionKernel</a></li>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1079</ul>
1080</li>
1081<li>New NEON kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1082<li><a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer_kernel.xhtml">NEBatchNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer.xhtml">NEBatchNormalizationLayer</a></li>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]1083<li><a class="el" href="classarm__compute_1_1_n_e_depth_concatenate_layer_kernel.xhtml">NEDepthConcatenateLayerKernel</a> / NEDepthConcatenateLayer</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1084<li><a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer_kernel.xhtml">NEDirectConvolutionLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer.xhtml">NEDirectConvolutionLayer</a></li>
1085<li><a class="el" href="classarm__compute_1_1_n_e_locally_connected_matrix_multiply_kernel.xhtml">NELocallyConnectedMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_locally_connected_layer.xhtml">NELocallyConnectedLayer</a></li>
1086<li><a class="el" href="classarm__compute_1_1_n_e_weights_reshape_kernel.xhtml">NEWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_convolution_layer_reshape_weights.xhtml">NEConvolutionLayerReshapeWeights</a></li>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1087</ul>
1088</li>
1089</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1090<p>v17.05 Public bug fixes release</p><ul>
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1091<li>Various bug fixes</li>
1092<li>Remaining of the functions ported to use accurate padding.</li>
1093<li>Library does not link against OpenCL anymore (It uses dlopen / dlsym at runtime instead to determine whether or not OpenCL is available).</li>
1094<li>Added "free" method to allocator.</li>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1095<li>Minimum version of g++ required for armv7 Linux changed from 4.8 to 4.9</li>
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1096</ul>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1097<p>v17.04 Public bug fixes release</p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1098<p>The following functions have been ported to use the new accurate padding:</p><ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1099<li><a class="el" href="classarm__compute_1_1_c_l_color_convert_kernel.xhtml">CLColorConvertKernel</a></li>
1100<li><a class="el" href="classarm__compute_1_1_c_l_edge_non_max_suppression_kernel.xhtml">CLEdgeNonMaxSuppressionKernel</a></li>
1101<li><a class="el" href="classarm__compute_1_1_c_l_edge_trace_kernel.xhtml">CLEdgeTraceKernel</a></li>
1102<li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_hor_kernel.xhtml">CLGaussianPyramidHorKernel</a></li>
1103<li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_vert_kernel.xhtml">CLGaussianPyramidVertKernel</a></li>
1104<li><a class="el" href="classarm__compute_1_1_c_l_gradient_kernel.xhtml">CLGradientKernel</a></li>
1105<li><a class="el" href="classarm__compute_1_1_n_e_channel_combine_kernel.xhtml">NEChannelCombineKernel</a></li>
1106<li><a class="el" href="classarm__compute_1_1_n_e_fill_array_kernel.xhtml">NEFillArrayKernel</a></li>
1107<li><a class="el" href="classarm__compute_1_1_n_e_gaussian_pyramid_hor_kernel.xhtml">NEGaussianPyramidHorKernel</a></li>
1108<li><a class="el" href="classarm__compute_1_1_n_e_gaussian_pyramid_vert_kernel.xhtml">NEGaussianPyramidVertKernel</a></li>
Jenkinsb9abeae2018-11-22 11:58:08 +0000[diff] [blame]1109<li>NEHarrisScoreFP16Kernel</li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1110<li><a class="el" href="classarm__compute_1_1_n_e_harris_score_kernel.xhtml">NEHarrisScoreKernel</a></li>
1111<li><a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector_kernel.xhtml">NEHOGDetectorKernel</a></li>
1112<li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_max_kernel.xhtml">NELogits1DMaxKernel</a></li>
1113<li>NELogits1DShiftExpSumKernel</li>
1114<li>NELogits1DNormKernel</li>
1115<li><a class="el" href="namespacearm__compute.xhtml#a38cad49e6beaef76bc1ec5064c9e9dba">NENonMaximaSuppression3x3FP16Kernel</a></li>
1116<li><a class="el" href="classarm__compute_1_1_n_e_non_maxima_suppression3x3_kernel.xhtml">NENonMaximaSuppression3x3Kernel</a></li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1117</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1118<p>v17.03.1 First Major public release of the sources</p><ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1119<li>Renamed the library to <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a></li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1120<li>New CPP target introduced for C++ kernels shared between NEON and CL functions.</li>
1121<li>New padding calculation interface introduced and ported most kernels / functions to use it.</li>
1122<li>New OpenCL kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1123<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_kernel.xhtml">CLGEMMLowpMatrixMultiplyKernel</a> / CLGEMMLowp</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1124</ul>
1125</li>
1126<li>New NEON kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1127<li><a class="el" href="classarm__compute_1_1_n_e_normalization_layer_kernel.xhtml">NENormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_normalization_layer.xhtml">NENormalizationLayer</a></li>
1128<li><a class="el" href="classarm__compute_1_1_n_e_transpose_kernel.xhtml">NETransposeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_transpose.xhtml">NETranspose</a></li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1129<li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_max_kernel.xhtml">NELogits1DMaxKernel</a>, NELogits1DShiftExpSumKernel, NELogits1DNormKernel / <a class="el" href="namespacearm__compute.xhtml#a4df2143ca0a3bdbbbc54b440a52541cd">NESoftmaxLayer</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1130<li><a class="el" href="classarm__compute_1_1_n_e_im2_col_kernel.xhtml">NEIm2ColKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_col2_im_kernel.xhtml">NECol2ImKernel</a>, NEConvolutionLayerWeightsReshapeKernel / <a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">NEConvolutionLayer</a></li>
1131<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_accumulate_biases_kernel.xhtml">NEGEMMMatrixAccumulateBiasesKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_fully_connected_layer.xhtml">NEFullyConnectedLayer</a></li>
1132<li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_multiply_kernel.xhtml">NEGEMMLowpMatrixMultiplyKernel</a> / NEGEMMLowp</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1133</ul>
1134</li>
1135</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1136<p>v17.03 Sources preview</p><ul>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1137<li>New OpenCL kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1138<li><a class="el" href="classarm__compute_1_1_c_l_gradient_kernel.xhtml">CLGradientKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_edge_non_max_suppression_kernel.xhtml">CLEdgeNonMaxSuppressionKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_edge_trace_kernel.xhtml">CLEdgeTraceKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_canny_edge.xhtml">CLCannyEdge</a></li>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1139<li>GEMM refactoring + FP16 support: CLGEMMInterleave4x4Kernel, CLGEMMTranspose1xWKernel, <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_kernel.xhtml">CLGEMMMatrixMultiplyKernel</a>, CLGEMMMatrixAdditionKernel / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m.xhtml">CLGEMM</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1140<li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_accumulate_biases_kernel.xhtml">CLGEMMMatrixAccumulateBiasesKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_fully_connected_layer.xhtml">CLFullyConnectedLayer</a></li>
1141<li><a class="el" href="classarm__compute_1_1_c_l_transpose_kernel.xhtml">CLTransposeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_transpose.xhtml">CLTranspose</a></li>
1142<li><a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_init_kernel.xhtml">CLLKTrackerInitKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_stage0_kernel.xhtml">CLLKTrackerStage0Kernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_stage1_kernel.xhtml">CLLKTrackerStage1Kernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_finalize_kernel.xhtml">CLLKTrackerFinalizeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_optical_flow.xhtml">CLOpticalFlow</a></li>
1143<li><a class="el" href="classarm__compute_1_1_c_l_normalization_layer_kernel.xhtml">CLNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_normalization_layer.xhtml">CLNormalizationLayer</a></li>
1144<li><a class="el" href="classarm__compute_1_1_c_l_laplacian_pyramid.xhtml">CLLaplacianPyramid</a>, <a class="el" href="classarm__compute_1_1_c_l_laplacian_reconstruct.xhtml">CLLaplacianReconstruct</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1145</ul>
1146</li>
1147<li>New NEON kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1148<li><a class="el" href="classarm__compute_1_1_n_e_activation_layer_kernel.xhtml">NEActivationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_activation_layer.xhtml">NEActivationLayer</a></li>
1149<li>GEMM refactoring + FP16 support (Requires armv8.2 CPU): <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_interleave4x4_kernel.xhtml">NEGEMMInterleave4x4Kernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_transpose1x_w_kernel.xhtml">NEGEMMTranspose1xWKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_multiply_kernel.xhtml">NEGEMMMatrixMultiplyKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_addition_kernel.xhtml">NEGEMMMatrixAdditionKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m.xhtml">NEGEMM</a></li>
1150<li><a class="el" href="classarm__compute_1_1_n_e_pooling_layer_kernel.xhtml">NEPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_pooling_layer.xhtml">NEPoolingLayer</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1151</ul>
1152</li>
1153</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1154<p>v17.02.1 Sources preview</p><ul>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1155<li>New OpenCL kernels / functions:<ul>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1156<li><a class="el" href="classarm__compute_1_1_c_l_logits1_d_max_kernel.xhtml">CLLogits1DMaxKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_logits1_d_shift_exp_sum_kernel.xhtml">CLLogits1DShiftExpSumKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_logits1_d_norm_kernel.xhtml">CLLogits1DNormKernel</a> / <a class="el" href="namespacearm__compute.xhtml#a30ce3b40394b4f2d1e4cc31db7183425">CLSoftmaxLayer</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1157<li><a class="el" href="classarm__compute_1_1_c_l_pooling_layer_kernel.xhtml">CLPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_pooling_layer.xhtml">CLPoolingLayer</a></li>
1158<li><a class="el" href="classarm__compute_1_1_c_l_im2_col_kernel.xhtml">CLIm2ColKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_col2_im_kernel.xhtml">CLCol2ImKernel</a>, CLConvolutionLayerWeightsReshapeKernel / <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">CLConvolutionLayer</a></li>
1159<li><a class="el" href="classarm__compute_1_1_c_l_remap_kernel.xhtml">CLRemapKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_remap.xhtml">CLRemap</a></li>
1160<li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_hor_kernel.xhtml">CLGaussianPyramidHorKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_vert_kernel.xhtml">CLGaussianPyramidVertKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid.xhtml">CLGaussianPyramid</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_half.xhtml">CLGaussianPyramidHalf</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_orb.xhtml">CLGaussianPyramidOrb</a></li>
1161<li><a class="el" href="classarm__compute_1_1_c_l_min_max_kernel.xhtml">CLMinMaxKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_min_max_location_kernel.xhtml">CLMinMaxLocationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_min_max_location.xhtml">CLMinMaxLocation</a></li>
1162<li><a class="el" href="classarm__compute_1_1_c_l_non_linear_filter_kernel.xhtml">CLNonLinearFilterKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_non_linear_filter.xhtml">CLNonLinearFilter</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1163</ul>
1164</li>
1165<li>New NEON FP16 kernels (Requires armv8.2 CPU)<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1166<li><a class="el" href="namespacearm__compute.xhtml#aff99c045b07329b332b1cb97a2dd1518">NEAccumulateWeightedFP16Kernel</a></li>
1167<li><a class="el" href="namespacearm__compute.xhtml#a96f7f6f98dc47e0dc3b928bf87397ebf">NEBox3x3FP16Kernel</a></li>
1168<li><a class="el" href="namespacearm__compute.xhtml#a38cad49e6beaef76bc1ec5064c9e9dba">NENonMaximaSuppression3x3FP16Kernel</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1169</ul>
1170</li>
1171</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1172<p>v17.02 Sources preview</p><ul>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1173<li>New OpenCL kernels / functions:<ul>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1174<li><a class="el" href="classarm__compute_1_1_c_l_activation_layer_kernel.xhtml">CLActivationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_activation_layer.xhtml">CLActivationLayer</a></li>
1175<li><a class="el" href="classarm__compute_1_1_c_l_channel_combine_kernel.xhtml">CLChannelCombineKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_channel_combine.xhtml">CLChannelCombine</a></li>
1176<li><a class="el" href="classarm__compute_1_1_c_l_derivative_kernel.xhtml">CLDerivativeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_channel_extract.xhtml">CLChannelExtract</a></li>
1177<li><a class="el" href="classarm__compute_1_1_c_l_fast_corners_kernel.xhtml">CLFastCornersKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_fast_corners.xhtml">CLFastCorners</a></li>
1178<li><a class="el" href="classarm__compute_1_1_c_l_mean_std_dev_kernel.xhtml">CLMeanStdDevKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_mean_std_dev.xhtml">CLMeanStdDev</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1179</ul>
1180</li>
1181<li>New NEON kernels / functions:<ul>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1182<li><a class="el" href="classarm__compute_1_1_h_o_g.xhtml" title="CPU implementation of HOG data-object.">HOG</a> / SVM: <a class="el" href="classarm__compute_1_1_n_e_h_o_g_orientation_binning_kernel.xhtml">NEHOGOrientationBinningKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_block_normalization_kernel.xhtml">NEHOGBlockNormalizationKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector_kernel.xhtml">NEHOGDetectorKernel</a>, NEHOGNonMaximaSuppressionKernel / <a class="el" href="classarm__compute_1_1_n_e_h_o_g_descriptor.xhtml">NEHOGDescriptor</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector.xhtml">NEHOGDetector</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_gradient.xhtml">NEHOGGradient</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_multi_detection.xhtml">NEHOGMultiDetection</a></li>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1183<li><a class="el" href="classarm__compute_1_1_n_e_non_linear_filter_kernel.xhtml">NENonLinearFilterKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_non_linear_filter.xhtml">NENonLinearFilter</a></li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1184</ul>
1185</li>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1186<li>Introduced a <a class="el" href="classarm__compute_1_1_c_l_scheduler.xhtml" title="Provides global access to a CL context and command queue.">CLScheduler</a> to manage the default context and command queue used by the runtime library and create synchronisation events.</li>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1187<li>Switched all the kernels / functions to use tensors instead of images.</li>
1188<li>Updated documentation to include instructions to build the library from sources.</li>
1189</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1190<p>v16.12 Binary preview release</p><ul>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1191<li>Original release</li>
1192</ul>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1193<h1><a class="anchor" id="S3_how_to_build"></a>
1194How to build the library and the examples</h1>
1195<h2><a class="anchor" id="S3_1_build_options"></a>
1196Build options</h2>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1197<p>scons 2.3 or above is required to build the library. To see the build options available simply run <code>scons -h</code>: </p><pre class="fragment">debug: Debug (yes|no)
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1198 default: False
1199 actual: False
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1200
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1201asserts: Enable asserts (this flag is forced to 1 for debug=1) (yes|no)
1202 default: False
1203 actual: False
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1204
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1205arch: Target Architecture (armv7a|arm64-v8a|arm64-v8.2-a|x86_32|x86_64)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1206 default: armv7a
1207 actual: armv7a
1208
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1209os: Target OS (linux|android|bare_metal)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1210 default: linux
1211 actual: linux
1212
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]1213build: Build type (native|cross_compile|embed_only)
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1214 default: cross_compile
1215 actual: cross_compile
1216
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1217examples: Build example programs (yes|no)
1218 default: True
1219 actual: True
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1220
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1221Werror: Enable/disable the -Werror compilation flag (yes|no)
1222 default: True
1223 actual: True
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1224
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1225opencl: Enable OpenCL support (yes|no)
1226 default: True
1227 actual: True
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1228
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1229neon: Enable Neon support (yes|no)
1230 default: False
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1231 actual: False
1232
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1233gles_compute: Enable OpenGL ES Compute Shader support (yes|no)
1234 default: False
1235 actual: False
1236
1237embed_kernels: Embed OpenCL kernels and OpenGL ES compute shader in library binary (yes|no)
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]1238 default: True
1239 actual: True
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1240
1241set_soname: Set the library's soname and shlibversion (requires SCons 2.4 or above) (yes|no)
1242 default: False
1243 actual: False
1244
1245openmp: Enable OpenMP backend (yes|no)
1246 default: False
1247 actual: False
1248
1249cppthreads: Enable C++11 threads backend (yes|no)
1250 default: True
1251 actual: True
1252
1253build_dir: Specify sub-folder for the build ( /path/to/build_dir )
1254 default: .
1255 actual: .
1256
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1257extra_cxx_flags: Extra CXX flags to be appended to the build command
1258 default:
1259 actual:
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1260
1261pmu: Enable PMU counters (yes|no)
1262 default: False
1263 actual: False
1264
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1265mali: Enable Mali hardware counters (yes|no)
1266 default: False
1267 actual: False
1268
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1269validation_tests: Build validation test programs (yes|no)
1270 default: False
1271 actual: False
1272
1273benchmark_tests: Build benchmark test programs (yes|no)
1274 default: False
1275 actual: False
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1276</pre><p><b>debug</b> / <b>asserts:</b> </p><ul>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1277<li>With debug=1 asserts are enabled, and the library is built with symbols and no optimisations enabled.</li>
1278<li>With debug=0 and asserts=1: Optimisations are enabled and symbols are removed, however all the asserts are still present (This is about 20% slower than the release build)</li>
1279<li>With debug=0 and asserts=0: All optimisations are enable and no validation is performed, if the application misuses the library it is likely to result in a crash. (Only use this mode once you are sure your application is working as expected).</li>
1280</ul>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1281<p><b>arch:</b> The x86_32 and x86_64 targets can only be used with neon=0 and opencl=1.</p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1282<p><b>os:</b> Choose the operating system you are targeting: Linux, Android or bare metal. </p><dl class="section note"><dt>Note</dt><dd>bare metal can only be used for NEON (not OpenCL), only static libraries get built and NEON's multi-threading support is disabled.</dd></dl>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1283<p><b>build:</b> you can either build directly on your device (native) or cross compile from your desktop machine (cross-compile). In both cases make sure the compiler is available in your path.</p>
1284<dl class="section note"><dt>Note</dt><dd>If you want to natively compile for 32bit on a 64bit ARM device running a 64bit OS then you will have to use cross-compile too.</dd></dl>
Anthony Barbier06ea0482018-02-22 15:45:35 +0000[diff] [blame]1285<p>There is also an 'embed_only' option which will generate all the .embed files for the OpenCL kernels and / or OpenGLES compute shaders. This might be useful if using a different build system to compile the library.</p>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1286<p><b>Werror:</b> If you are compiling using the same toolchains as the ones used in this guide then there shouldn't be any warning and therefore you should be able to keep Werror=1. If with a different compiler version the library fails to build because of warnings interpreted as errors then, if you are sure the warnings are not important, you might want to try to build with Werror=0 (But please do report the issue either on Github or by an email to <a href="#" onclick="location.href='mai'+'lto:'+'dev'+'el'+'ope'+'r@'+'arm'+'.c'+'om'; return false;">devel<span style="display: none;">.nosp@m.</span>oper<span style="display: none;">.nosp@m.</span>@arm.<span style="display: none;">.nosp@m.</span>com</a> so that the issue can be addressed).</p>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1287<p><b>opencl</b> / <b>neon</b> / <b>gles_compute:</b> Choose which SIMD technology you want to target. (NEON for ARM Cortex-A CPUs or OpenCL / GLES_COMPUTE for ARM Mali GPUs)</p>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1288<p><b>embed_kernels:</b> For OpenCL / GLES_COMPUTE only: set embed_kernels=1 if you want the OpenCL / GLES_COMPUTE kernels to be built in the library's binaries instead of being read from separate ".cl" / ".cs" files. If embed_kernels is set to 0 then the application can set the path to the folder containing the OpenCL / GLES_COMPUTE kernel files by calling <a class="el" href="classarm__compute_1_1_c_l_kernel_library.xhtml#a9f976367edcd9ab787375373e050b94b" title="Initialises the kernel library.">CLKernelLibrary::init()</a> / <a class="el" href="classarm__compute_1_1_g_c_kernel_library.xhtml#abe24625d55f2fb35da7e293e5e28d483" title="Initialises the kernel library.">GCKernelLibrary::init()</a>. By default the path is set to "./cl_kernels" / "./cs_shaders".</p>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1289<p><b>set_soname:</b> Do you want to build the versioned version of the library ?</p>
1290<p>If enabled the library will contain a SONAME and SHLIBVERSION and some symlinks will automatically be created between the objects. Example: libarm_compute_core.so -&gt; libarm_compute_core.so.1.0.0 libarm_compute_core.so.1 -&gt; libarm_compute_core.so.1.0.0 libarm_compute_core.so.1.0.0</p>
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1291<dl class="section note"><dt>Note</dt><dd>This options is disabled by default as it requires SCons version 2.4 or above.</dd></dl>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1292<p><b>extra_cxx_flags:</b> Custom CXX flags which will be appended to the end of the build command.</p>
1293<p><b>build_dir:</b> Build the library in a subfolder of the "build" folder. (Allows to build several configurations in parallel).</p>
1294<p><b>examples:</b> Build or not the examples</p>
1295<p><b>validation_tests:</b> Enable the build of the validation suite.</p>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1296<p><b>benchmark_tests:</b> Enable the build of the benchmark tests</p>
1297<p><b>pmu:</b> Enable the PMU cycle counter to measure execution time in benchmark tests. (Your device needs to support it)</p>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1298<p><b>mali:</b> Enable the collection of Mali hardware counters to measure execution time in benchmark tests. (Your device needs to have a Mali driver that supports it)</p>
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1299<p><b>openmp</b> Build in the OpenMP scheduler for NEON.</p>
1300<dl class="section note"><dt>Note</dt><dd>Only works when building with g++ not clang++</dd></dl>
1301<p><b>cppthreads</b> Build in the C++11 scheduler for NEON.</p>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1302<dl class="section see"><dt>See also</dt><dd><a class="el" href="classarm__compute_1_1_scheduler.xhtml#ad2fc671b2772dd9e28b81cf0e2514e85" title="Sets the user defined scheduler and makes it the active scheduler.">Scheduler::set</a></dd></dl>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1303<h2><a class="anchor" id="S3_2_linux"></a>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1304Building for Linux</h2>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1305<h3><a class="anchor" id="S3_2_1_library"></a>
1306How to build the library ?</h3>
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1307<p>For Linux, the library was successfully built and tested using the following Linaro GCC toolchain:</p>
1308<ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1309<li>gcc-linaro-4.9-2016.02-x86_64_arm-linux-gnueabihf</li>
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1310<li>gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu</li>
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1311</ul>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1312<p>To cross-compile the library in debug mode, with NEON only support, for Linux 32bit: </p><pre class="fragment">scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=linux arch=armv7a
1313</pre><p>To cross-compile the library in asserts mode, with OpenCL only support, for Linux 64bit: </p><pre class="fragment">scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=linux arch=arm64-v8a
1314</pre><p>To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Linux 64bit: </p><pre class="fragment">scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=0 gles_compute=1 embed_kernels=1 os=linux arch=arm64-v8a
1315</pre><p>You can also compile the library natively on an ARM device by using <b>build=native</b>: </p><pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=arm64-v8a build=native
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1316scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=native
Anthony Barbierdbdab852017-06-23 15:42:00 +0100[diff] [blame]1317</pre><dl class="section note"><dt>Note</dt><dd>g++ for ARM is mono-arch, therefore if you want to compile for Linux 32bit on a Linux 64bit platform you will have to use a cross compiler.</dd></dl>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1318<p>For example on a 64bit Debian based system you would have to install <b>g++-arm-linux-gnueabihf</b> </p><pre class="fragment">apt-get install g++-arm-linux-gnueabihf
1319</pre><p>Then run </p><pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=cross_compile
1320</pre><p>or simply remove the build parameter as build=cross_compile is the default value: </p><pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a
Anthony Barbiera4376382017-04-12 15:12:46 +0100[diff] [blame]1321</pre><dl class="section attention"><dt>Attention</dt><dd>To cross compile with opencl=1 you need to make sure to have a version of libOpenCL matching your target architecture.</dd></dl>
1322<h3><a class="anchor" id="S3_2_2_examples"></a>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1323How to manually build the examples ?</h3>
1324<p>The examples get automatically built by scons as part of the build process of the library described above. This section just describes how you can build and link your own application against our library.</p>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1325<dl class="section note"><dt>Note</dt><dd>The following command lines assume the <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> binaries are present in the current directory or in the system library path. If this is not the case you can specify the location of the pre-built library with the compiler option -L. When building the OpenCL example the commands below assume that the CL headers are located in the include folder where the command is executed.</dd></dl>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1326<p>To cross compile a NEON example for Linux 32bit: </p><pre class="fragment">arm-linux-gnueabihf-g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute -larm_compute_core -o neon_convolution
1327</pre><p>To cross compile a NEON example for Linux 64bit: </p><pre class="fragment">aarch64-linux-gnu-g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -L. -larm_compute -larm_compute_core -o neon_convolution
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1328</pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option and the compiler's name is different)</p>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1329<p>To cross compile an OpenCL example for Linux 32bit: </p><pre class="fragment">arm-linux-gnueabihf-g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL
1330</pre><p>To cross compile an OpenCL example for Linux 64bit: </p><pre class="fragment">aarch64-linux-gnu-g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -L. -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1331</pre><p>To cross compile a GLES example for Linux 32bit: </p><pre class="fragment">arm-linux-gnueabihf-g++ examples/gc_absdiff.cpp utils/Utils.cpp -I. -Iinclude/ -L. -larm_compute -larm_compute_core -std=c++11 -mfpu=neon -DARM_COMPUTE_GC -Iinclude/linux/ -o gc_absdiff
1332</pre><p>To cross compile a GLES example for Linux 64bit: </p><pre class="fragment">aarch64-linux-gnu-g++ examples/gc_absdiff.cpp utils/Utils.cpp -I. -Iinclude/ -L. -larm_compute -larm_compute_core -std=c++11 -DARM_COMPUTE_GC -Iinclude/linux/ -o gc_absdiff
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1333</pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option and the compiler's name is different)</p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1334<p>To cross compile the examples with the Graph API, such as <a class="el" href="graph__lenet_8cpp.xhtml">graph_lenet.cpp</a>, you need to link the examples against arm_compute_graph.so too.</p>
1335<dl class="section note"><dt>Note</dt><dd>The compute library must currently be built with both neon and opencl enabled - neon=1 and opencl=1</dd></dl>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1336<p>i.e. to cross compile the "graph_lenet" example for Linux 32bit: </p><pre class="fragment">arm-linux-gnueabihf-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
1337</pre><p>i.e. to cross compile the "graph_lenet" example for Linux 64bit: </p><pre class="fragment">aarch64-linux-gnu-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1338</pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option and the compiler's name is different)</p>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1339<dl class="section note"><dt>Note</dt><dd>If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a>, arm_compute_core</dd></dl>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1340<p>To compile natively (i.e directly on an ARM device) for NEON for Linux 32bit: </p><pre class="fragment">g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -larm_compute -larm_compute_core -o neon_convolution
1341</pre><p>To compile natively (i.e directly on an ARM device) for NEON for Linux 64bit: </p><pre class="fragment">g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o neon_convolution
Anthony Barbier46d59272017-05-04 09:15:15 +0100[diff] [blame]1342</pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option)</p>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1343<p>To compile natively (i.e directly on an ARM device) for OpenCL for Linux 32bit or Linux 64bit: </p><pre class="fragment">g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1344</pre><p>To compile natively (i.e directly on an ARM device) for GLES for Linux 32bit or Linux 64bit: </p><pre class="fragment">g++ examples/gc_absdiff.cpp utils/Utils.cpp -I. -Iinclude/ -L. -larm_compute -larm_compute_core -std=c++11 -DARM_COMPUTE_GC -Iinclude/linux/ -o gc_absdiff
1345</pre><p>To compile natively the examples with the Graph API, such as <a class="el" href="graph__lenet_8cpp.xhtml">graph_lenet.cpp</a>, you need to link the examples against arm_compute_graph.so too. </p><dl class="section note"><dt>Note</dt><dd>The compute library must currently be built with both neon and opencl enabled - neon=1 and opencl=1</dd></dl>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1346<p>i.e. to natively compile the "graph_lenet" example for Linux 32bit: </p><pre class="fragment">g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
1347</pre><p>i.e. to natively compile the "graph_lenet" example for Linux 64bit: </p><pre class="fragment">g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1348</pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option)</p>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1349<dl class="section note"><dt>Note</dt><dd>If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a>, arm_compute_core</dd>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1350<dd>
1351These two commands assume libarm_compute.so is available in your library path, if not add the path to it using -L</dd></dl>
1352<p>To run the built executable simply run: </p><pre class="fragment">LD_LIBRARY_PATH=build ./neon_convolution
1353</pre><p>or </p><pre class="fragment">LD_LIBRARY_PATH=build ./cl_convolution
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1354</pre><dl class="section note"><dt>Note</dt><dd>Examples accept different types of arguments, to find out what they are run the example with <em>&ndash;help</em> as an argument. If no arguments are specified then random values will be used to execute the graph.</dd></dl>
1355<p>For example: </p><pre class="fragment">LD_LIBRARY_PATH=. ./graph_lenet --help
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]1356</pre><p>Below is a list of the common parameters among the graph examples : </p><div class="fragment"><div class="line"><span class="comment">/* Common graph parameters</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * --help : Print the example&#39;s help message.</span></div><div class="line"><span class="comment"> * --threads : The number of threads to be used by the example during execution.</span></div><div class="line"><span class="comment"> * --target : Execution target to be used by the examples. Supported target options: NEON, CL, GC.</span></div><div class="line"><span class="comment"> * --type : Data type to be used by the examples. Supported data type options: QASYMM8, F16, F32.</span></div><div class="line"><span class="comment"> * --layout : Data layout to be used by the examples. Supported data layout options : NCHW, NHWC.</span></div><div class="line"><span class="comment"> * --enable-tuner : Toggle option to enable the OpenCL dynamic tuner.</span></div><div class="line"><span class="comment"> * --enable-cl-cache : Toggle option to load the prebuilt opencl kernels from a cache file.</span></div><div class="line"><span class="comment"> * --fast-math : Toggle option to enable the fast math option.</span></div><div class="line"><span class="comment"> * --data : Path that contains the trainable parameter files of graph layers.</span></div><div class="line"><span class="comment"> * --image : Image to load and operate on. Image types supported: PPM, JPEG, NPY.</span></div><div class="line"><span class="comment"> * --labels : File that contains the labels that classify upon.</span></div><div class="line"><span class="comment"> * --validation-file : File that contains a list of image names with their corresponding label id (e.g. image0.jpg 5).</span></div><div class="line"><span class="comment"> * This is used to run the graph over a number of images and report top-1 and top-5 metrics.</span></div><div class="line"><span class="comment"> * --validation-path : The path where the validation images specified in the validation file reside.</span></div><div class="line"><span class="comment"> * --validation-range : The range of the images to validate from the validation file (e.g 0,9).</span></div><div class="line"><span class="comment"> * If not specified all the images will be validated.</span></div><div class="line"><span class="comment"> * --tuner-file : The file to store the OpenCL dynamic tuner tuned parameters.</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Note that data, image and labels options should be provided to perform an inference run on an image.</span></div><div class="line"><span class="comment"> * Note that validation-file and validation-path should be provided to perform a graph accuracy estimation.</span></div><div class="line"><span class="comment"> * Note GLES target is not supported for most of the networks.</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Example execution commands:</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Execute a single inference given an image and a file containing the correspondence between label ids and human readable labels:</span></div><div class="line"><span class="comment"> * ./graph_vgg16 --data=data/ --target=cl --layout=nhwc --image=kart.jpeg --labels=imagenet1000_clsid_to_human.txt</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Perform a graph validation on a list of images:</span></div><div class="line"><span class="comment"> * ./graph_vgg16 --data=data/ --target=neon --threads=4 --layout=nchw --validation-file=val.txt --validation-path=ilsvrc_test_images/</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * File formats:</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Validation file should be a plain file containing the names of the images followed by the correct label id.</span></div><div class="line"><span class="comment"> * For example:</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * image0.jpeg 882</span></div><div class="line"><span class="comment"> * image1.jpeg 34</span></div><div class="line"><span class="comment"> * image2.jpeg 354</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * Labels file should be a plain file where each line is the respective human readable label (counting starts from 0).</span></div><div class="line"><span class="comment"> * For example:</span></div><div class="line"><span class="comment"> *</span></div><div class="line"><span class="comment"> * 0: label0_name label0_name</span></div><div class="line"><span class="comment"> * 1: label1_name or label1_name</span></div><div class="line"><span class="comment"> * 2: label2_name label2_name</span></div><div class="line"><span class="comment"> */</span></div></div><!-- fragment --> <h2><a class="anchor" id="S3_3_android"></a>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1357Building for Android</h2>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1358<p>For Android, the library was successfully built and tested using Google's standalone toolchains:</p><ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1359<li>clang++ from NDK r17b for armv7a</li>
1360<li>clang++ from NDK r17b for arm64-v8a</li>
1361<li>clang++ from NDK r18-beta1 for arm64-v8.2-a with FP16 support</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1362</ul>
1363<p>Here is a guide to <a href="https://developer.android.com/ndk/guides/standalone_toolchain.html">create your Android standalone toolchains from the NDK</a></p>
1364<ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1365<li>Download the NDK r17b from here: <a href="https://developer.android.com/ndk/downloads/index.html">https://developer.android.com/ndk/downloads/index.html</a></li>
Jenkins4ba87db2019-05-23 17:11:51 +0100[diff] [blame]1366<li>Make sure you have Python 2.7 installed on your machine.</li>
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1367<li>Generate the 32 and/or 64 toolchains by running the following commands:</li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1368</ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1369<pre class="fragment">$NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-ndk-r17b --stl libc++ --api 21
1370$NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-android-ndk-r17b --stl libc++ --api 21
1371</pre><dl class="section attention"><dt>Attention</dt><dd>We used to use gnustl but as of NDK r17 it is deprecated so we switched to libc++</dd></dl>
1372<dl class="section note"><dt>Note</dt><dd>Make sure to add the toolchains to your PATH: <pre class="fragment">export PATH=$PATH:$MY_TOOLCHAINS/aarch64-linux-android-ndk-r17b/bin:$MY_TOOLCHAINS/arm-linux-android-ndk-r17b/bin
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1373</pre></dd></dl>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1374<h3><a class="anchor" id="S3_3_1_library"></a>
1375How to build the library ?</h3>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1376<p>To cross-compile the library in debug mode, with NEON only support, for Android 32bit: </p><pre class="fragment">CXX=clang++ CC=clang scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=android arch=armv7a
1377</pre><p>To cross-compile the library in asserts mode, with OpenCL only support, for Android 64bit: </p><pre class="fragment">CXX=clang++ CC=clang scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=android arch=arm64-v8a
1378</pre><p>To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Android 64bit: </p><pre class="fragment">CXX=clang++ CC=clang scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=0 gles_compute=1 embed_kernels=1 os=android arch=arm64-v8a
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1379</pre><h3><a class="anchor" id="S3_3_2_examples"></a>
1380How to manually build the examples ?</h3>
1381<p>The examples get automatically built by scons as part of the build process of the library described above. This section just describes how you can build and link your own application against our library.</p>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1382<dl class="section note"><dt>Note</dt><dd>The following command lines assume the <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> binaries are present in the current directory or in the system library path. If this is not the case you can specify the location of the pre-built library with the compiler option -L. When building the OpenCL example the commands below assume that the CL headers are located in the include folder where the command is executed.</dd></dl>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1383<p>Once you've got your Android standalone toolchain built and added to your path you can do the following:</p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1384<p>To cross compile a NEON example: </p><pre class="fragment">#32 bit:
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1385arm-linux-androideabi-clang++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o neon_convolution_arm -static-libstdc++ -pie
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1386#64 bit:
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1387aarch64-linux-android-clang++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o neon_convolution_aarch64 -static-libstdc++ -pie
1388</pre><p>To cross compile an OpenCL example: </p><pre class="fragment">#32 bit:
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1389arm-linux-androideabi-clang++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o cl_convolution_arm -static-libstdc++ -pie -DARM_COMPUTE_CL
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1390#64 bit:
Jenkinsb3a371b2018-05-23 11:36:53 +0100[diff] [blame]1391aarch64-linux-android-clang++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o cl_convolution_aarch64 -static-libstdc++ -pie -DARM_COMPUTE_CL
Anthony Barbierf45d5a92018-01-24 16:23:15 +0000[diff] [blame]1392</pre><p>To cross compile a GLES example: </p><pre class="fragment">#32 bit:
1393arm-linux-androideabi-clang++ examples/gc_absdiff.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o gc_absdiff_arm -static-libstdc++ -pie -DARM_COMPUTE_GC
1394#64 bit:
1395aarch64-linux-android-clang++ examples/gc_absdiff.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute-static -larm_compute_core-static -L. -o gc_absdiff_aarch64 -static-libstdc++ -pie -DARM_COMPUTE_GC
1396</pre><p>To cross compile the examples with the Graph API, such as <a class="el" href="graph__lenet_8cpp.xhtml">graph_lenet.cpp</a>, you need to link the library arm_compute_graph also. (notice the compute library has to be built with both neon and opencl enabled - neon=1 and opencl=1) </p><pre class="fragment">#32 bit:
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1397arm-linux-androideabi-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -Wl,--whole-archive -larm_compute_graph-static -Wl,--no-whole-archive -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_arm -static-libstdc++ -pie -DARM_COMPUTE_CL
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1398#64 bit:
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1399aarch64-linux-android-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -Wl,--whole-archive -larm_compute_graph-static -Wl,--no-whole-archive -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_aarch64 -static-libstdc++ -pie -DARM_COMPUTE_CL
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1400</pre><dl class="section note"><dt>Note</dt><dd>Due to some issues in older versions of the Mali OpenCL DDK (&lt;= r13p0), we recommend to link <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> statically on Android. </dd>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1401<dd>
1402When linked statically the arm_compute_graph library currently needs the &ndash;whole-archive linker flag in order to work properly</dd></dl>
1403<p>Then you need to do is upload the executable and the shared library to the device using ADB: </p><pre class="fragment">adb push neon_convolution_arm /data/local/tmp/
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1404adb push cl_convolution_arm /data/local/tmp/
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1405adb push gc_absdiff_arm /data/local/tmp/
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1406adb shell chmod 777 -R /data/local/tmp/
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1407</pre><p>And finally to run the example: </p><pre class="fragment">adb shell /data/local/tmp/neon_convolution_arm
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1408adb shell /data/local/tmp/cl_convolution_arm
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1409adb shell /data/local/tmp/gc_absdiff_arm
1410</pre><p>For 64bit: </p><pre class="fragment">adb push neon_convolution_aarch64 /data/local/tmp/
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1411adb push cl_convolution_aarch64 /data/local/tmp/
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1412adb push gc_absdiff_aarch64 /data/local/tmp/
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1413adb shell chmod 777 -R /data/local/tmp/
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1414</pre><p>And finally to run the example: </p><pre class="fragment">adb shell /data/local/tmp/neon_convolution_aarch64
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1415adb shell /data/local/tmp/cl_convolution_aarch64
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1416adb shell /data/local/tmp/gc_absdiff_aarch64
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1417</pre><dl class="section note"><dt>Note</dt><dd>Examples accept different types of arguments, to find out what they are run the example with <em>&ndash;help</em> as an argument. If no arguments are specified then random values will be used to execute the graph.</dd></dl>
1418<p>For example: adb shell /data/local/tmp/graph_lenet &ndash;help</p>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1419<p>In this case the first argument of LeNet (like all the graph examples) is the target (i.e 0 to run on NEON, 1 to run on OpenCL if available, 2 to run on OpenCL using the <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a>), the second argument is the path to the folder containing the npy files for the weights and finally the third argument is the number of batches to run.</p>
Jenkinsc3f34a42018-03-02 12:38:09 +0000[diff] [blame]1420<h2><a class="anchor" id="S3_4_bare_metal"></a>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1421Building for bare metal</h2>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1422<p>For bare metal, the library was successfully built using linaros's latest (gcc-linaro-6.3.1-2017.05) bare metal toolchains:</p><ul>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1423<li>arm-eabi for armv7a</li>
1424<li>aarch64-elf for arm64-v8a</li>
1425</ul>
1426<p>Download linaro for <a href="https://releases.linaro.org/components/toolchain/binaries/6.3-2017.05/arm-eabi/">armv7a</a> and <a href="https://releases.linaro.org/components/toolchain/binaries/6.3-2017.05/aarch64-elf/">arm64-v8a</a>.</p>
1427<dl class="section note"><dt>Note</dt><dd>Make sure to add the toolchains to your PATH: export PATH=$PATH:$MY_TOOLCHAINS/gcc-linaro-6.3.1-2017.05-x86_64_aarch64-elf/bin:$MY_TOOLCHAINS/gcc-linaro-6.3.1-2017.05-x86_64_arm-eabi/bin</dd></dl>
1428<h3><a class="anchor" id="S3_4_1_library"></a>
1429How to build the library ?</h3>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1430<p>To cross-compile the library with NEON support for baremetal arm64-v8a: </p><pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=bare_metal arch=arm64-v8a build=cross_compile cppthreads=0 openmp=0 standalone=1
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1431</pre><h3><a class="anchor" id="S3_4_2_examples"></a>
1432How to manually build the examples ?</h3>
1433<p>Examples are disabled when building for bare metal. If you want to build the examples you need to provide a custom bootcode depending on the target architecture and link against the compute library. More information about bare metal bootcode can be found <a href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dai0527a/index.html">here</a>.</p>
1434<h2><a class="anchor" id="S3_5_windows_host"></a>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1435Building on a Windows host system</h2>
1436<p>Using <code>scons</code> directly from the Windows command line is known to cause problems. The reason seems to be that if <code>scons</code> is setup for cross-compilation it gets confused about Windows style paths (using backslashes). Thus it is recommended to follow one of the options outlined below.</p>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1437<h3><a class="anchor" id="S3_5_1_ubuntu_on_windows"></a>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1438Bash on Ubuntu on Windows</h3>
1439<p>The best and easiest option is to use <a href="https://msdn.microsoft.com/en-gb/commandline/wsl/about">Ubuntu on Windows</a>. This feature is still marked as <em>beta</em> and thus might not be available. However, if it is building the library is as simple as opening a <em>Bash on Ubuntu on Windows</em> shell and following the general guidelines given above.</p>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1440<h3><a class="anchor" id="S3_5_2_cygwin"></a>
Kaizen8938bd32017-09-28 14:38:23 +0100[diff] [blame]1441Cygwin</h3>
Jenkins975dfe12019-09-02 11:47:54 +0100[diff] [blame]1442<p>If the Windows subsystem for Linux is not available <a href="https://www.cygwin.com/">Cygwin</a> can be used to install and run <code>scons</code>, the minimum Cygwin version must be 3.0.7 or later. In addition to the default packages installed by Cygwin <code>scons</code> has to be selected in the installer. (<code>git</code> might also be useful but is not strictly required if you already have got the source code of the library.) Linaro provides pre-built versions of <a href="http://releases.linaro.org/components/toolchain/binaries/">GCC cross-compilers</a> that can be used from the Cygwin terminal. When building for Android the compiler is included in the Android standalone toolchain. After everything has been set up in the Cygwin terminal the general guide on building the library can be followed.</p>
Kaizenbf8b01d2017-10-12 14:26:51 +0100[diff] [blame]1443<h2><a class="anchor" id="S3_6_cl_stub_library"></a>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1444The OpenCL stub library</h2>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1445<p>In the opencl-1.2-stubs folder you will find the sources to build a stub OpenCL library which then can be used to link your application or <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> against.</p>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1446<p>If you preferred you could retrieve the OpenCL library from your device and link against this one but often this library will have dependencies on a range of system libraries forcing you to link your application against those too even though it is not using them.</p>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1447<dl class="section warning"><dt>Warning</dt><dd>This OpenCL library provided is a stub and <em>not</em> a real implementation. You can use it to resolve OpenCL's symbols in <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> while building the example but you must make sure the real libOpenCL.so is in your PATH when running the example or it will not work.</dd></dl>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1448<p>To cross-compile the stub OpenCL library simply run: </p><pre class="fragment">&lt;target-prefix&gt;-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
1449</pre><p>For example: </p><pre class="fragment">#Linux 32bit
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1450arm-linux-gnueabihf-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
1451#Linux 64bit
1452aarch64-linux-gnu-gcc -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC
1453#Android 32bit
1454arm-linux-androideabi-clang -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
1455#Android 64bit
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1456aarch64-linux-android-clang -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
1457</pre><h2><a class="anchor" id="S3_7_gles_stub_library"></a>
1458The Linux OpenGLES and EGL stub libraries</h2>
Jenkins0e205f72019-11-28 16:53:35 +0000[diff] [blame]1459<p>In the opengles-3.1-stubs folder you will find the sources to build stub EGL and OpenGLES libraries which then can be used to link your Linux application of <a class="el" href="namespacearm__compute.xhtml" title="Copyright (c) 2017-2019 ARM Limited.">arm_compute</a> against.</p>
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame]1460<dl class="section note"><dt>Note</dt><dd>The stub libraries are only needed on Linux. For Android, the NDK toolchains already provide the meta-EGL and meta-GLES libraries.</dd></dl>
1461<p>To cross-compile the stub OpenGLES and EGL libraries simply run: </p><pre class="fragment">&lt;target-prefix&gt;-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
1462&lt;target-prefix&gt;-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared
1463
1464#Linux 32bit
1465arm-linux-gnueabihf-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
1466arm-linux-gnueabihf-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared
1467
1468#Linux 64bit
1469aarch64-linux-gnu-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1470aarch64-linux-gnu-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared
1471</pre><h2><a class="anchor" id="S3_8_cl_requirements"></a>
1472OpenCL DDK Requirements</h2>
1473<h3><a class="anchor" id="S3_8_1_cl_hard_requirements"></a>
1474Hard Requirements</h3>
1475<p>Compute Library requires OpenCL 1.1 and above with support of non uniform workgroup sizes, which is officially supported in the Mali OpenCL DDK r8p0 and above as an extension (respective extension flag is <em>-cl-arm-non-uniform-work-group-size</em>).</p>
1476<p>Enabling 16-bit floating point calculations require <em>cl_khr_fp16</em> extension to be supported. All Mali GPUs with compute capabilities have native support for half precision floating points.</p>
1477<p>Use of <a class="el" href="classarm__compute_1_1_c_l_mean_std_dev.xhtml">CLMeanStdDev</a> function requires 64-bit atomics support, thus <em>cl_khr_int64_base_atomics</em> should be supported in order to use.</p>
1478<h3><a class="anchor" id="S3_8_2_cl_performance_requirements"></a>
1479Performance improvements</h3>
1480<p>Integer dot product built-in function extensions (and therefore optimized kernels) are available with Mali OpenCL DDK r22p0 and above for the following GPUs : G71, G76. The relevant extensions are <em>cl_arm_integer_dot_product_int8</em>, <em>cl_arm_integer_dot_product_accumulate_int8</em> and <em>cl_arm_integer_dot_product_accumulate_int16</em>.</p>
1481<p>OpenCL kernel level debugging can be simplified with the use of printf, this requires the <em>cl_arm_printf</em> extension to be supported.</p>
1482<p>SVM allocations are supported for all the underlying allocations in Compute Library. To enable this OpenCL 2.0 and above is a requirement.</p>
1483<h2><a class="anchor" id="S3_9_cl_tuner"></a>
1484OpenCL Tuner</h2>
Jenkins4ba87db2019-05-23 17:11:51 +0100[diff] [blame]1485<p>The OpenCL tuner, a.k.a. <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a>, is a module of Arm Compute Library that can improve the performance of the OpenCL kernels tuning the Local-Workgroup-Size (LWS). The optimal LWS for each unique OpenCL kernel configuration is stored in a table. This table can be either imported or exported from/to a file. The OpenCL tuner runs the same OpenCL kernel for a range of local workgroup sizes and keeps the local workgroup size of the fastest run to use in subsequent calls to the kernel. It supports three modes of tuning with different trade-offs between the time taken to tune and the kernel execution time achieved using the best LWS found. In the Exhaustive mode, it searches all the supported values of LWS. This mode takes the longest time to tune and is the most likely to find the optimal LWS. Normal mode searches a subset of LWS values to yield a good approximation of the optimal LWS. It takes less time to tune than Exhaustive mode. Rapid mode takes the shortest time to tune and finds an LWS value that is at least as good or better than the default LWS value. The mode affects only the search for the optimal LWS and has no effect when the LWS value is imported from a file. In order for the performance numbers to be meaningful you must disable the GPU power management and set it to a fixed frequency for the entire duration of the tuning phase.</p>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1486<p>If you wish to know more about LWS and the important role on improving the GPU cache utilization, we suggest having a look at the presentation "Even Faster CNNs: Exploring the New Class of Winograd Algorithms available at the following link:</p>
1487<p><a href="https://www.embedded-vision.com/platinum-members/arm/embedded-vision-training/videos/pages/may-2018-embedded-vision-summit-iodice">https://www.embedded-vision.com/platinum-members/arm/embedded-vision-training/videos/pages/may-2018-embedded-vision-summit-iodice</a></p>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1488<p>Tuning a network from scratch can be long and affect considerably the execution time for the first run of your network. It is recommended for this reason to store the <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a>'s result in a file to amortize this time when you either re-use the same network or the functions with the same configurations. The tuning is performed only once for each OpenCL kernel.</p>
1489<p><a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a> looks for the optimal LWS for each unique OpenCL kernel configuration. Since a function (i.e. Convolution Layer, Pooling Layer, Fully Connected Layer ...) can be called multiple times but with different parameters, we associate an "id" (called "config_id") to each kernel to distinguish the unique configurations. </p><pre class="fragment">#Example: 2 unique Matrix Multiply configurations
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1490</pre> <div class="fragment"><div class="line">TensorShape a0 = TensorShape(32,32);</div><div class="line">TensorShape b0 = TensorShape(32,32);</div><div class="line">TensorShape c0 = TensorShape(32,32);</div><div class="line">TensorShape a1 = TensorShape(64,64);</div><div class="line">TensorShape b1 = TensorShape(64,64);</div><div class="line">TensorShape c1 = TensorShape(64,64);</div><div class="line"></div><div class="line">Tensor a0_tensor;</div><div class="line">Tensor b0_tensor;</div><div class="line">Tensor c0_tensor;</div><div class="line">Tensor a1_tensor;</div><div class="line">Tensor b1_tensor;</div><div class="line">Tensor c1_tensor;</div><div class="line"></div><div class="line">a0_tensor.allocator()-&gt;init(TensorInfo(a0, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line">b0_tensor.allocator()-&gt;init(TensorInfo(b0, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line">c0_tensor.allocator()-&gt;init(TensorInfo(c0, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line">a1_tensor.allocator()-&gt;init(TensorInfo(a1, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line">b1_tensor.allocator()-&gt;init(TensorInfo(b1, 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line">c1_tensor.allocator()-&gt;init(TensorInfo(c1 1, <a class="code" href="namespacearm__compute.xhtml#ab4e88c89b3b7ea1735996cc4def22d58a44ad4ef5a76e6aa6fb3e3fa079a54fda">DataType::F32</a>));</div><div class="line"></div><div class="line">CLGEMM gemm0;</div><div class="line">CLGEMM gemm1;</div><div class="line"></div><div class="line"><span class="comment">// Configuration 0</span></div><div class="line">gemm0.configure(&amp;a0, &amp;b0, <span class="keyword">nullptr</span>, &amp;c0, 1.0f, 0.0f);</div><div class="line"></div><div class="line"><span class="comment">// Configuration 1</span></div><div class="line">gemm1.configure(&amp;a1, &amp;b1, <span class="keyword">nullptr</span>, &amp;c1, 1.0f, 0.0f);</div></div><!-- fragment --><h3><a class="anchor" id="S3_9_1_cl_tuner_how_to"></a>
1491How to use it</h3>
1492<p>All the graph examples in the ACL's folder "examples" and the arm_compute_benchmark accept an argument to enable the OpenCL tuner and an argument to export/import the LWS values to/from a file </p><pre class="fragment">#Enable CL tuner
1493./graph_mobilenet --enable-tuner –-target=CL
1494./arm_compute_benchmark --enable-tuner
1495
1496#Export/Import to/from a file
1497./graph_mobilenet --enable-tuner --target=CL --tuner-file=acl_tuner.csv
1498./arm_compute_benchmark --enable-tuner --tuner-file=acl_tuner.csv
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1499</pre><p>If you are importing the <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a>'results from a file, the new tuned LWS values will be appended to it.</p>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1500<p>Either you are benchmarking the graph examples or the test cases in the arm_compute_benchmark remember to: </p><pre class="fragment">-# Disable the power management
1501-# Keep the GPU frequency constant
1502-# Run multiple times the network (i.e. 10).
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1503</pre><p>If you are not using the graph API or the benchmark infrastructure you will need to manually pass a <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a> object to <a class="el" href="classarm__compute_1_1_c_l_scheduler.xhtml" title="Provides global access to a CL context and command queue.">CLScheduler</a> before configuring any function.</p>
1504<div class="fragment"><div class="line">CLTuner tuner;</div><div class="line"></div><div class="line"><span class="comment">// Setup Scheduler</span></div><div class="line"><a class="code" href="classarm__compute_1_1_c_l_scheduler.xhtml#a9b58d0eb9a2af8e6d7908695e1557d6c">CLScheduler::get</a>().<a class="code" href="classarm__compute_1_1_c_l_scheduler.xhtml#a46ecf9ef0fe80ba2ed35acfc29856b7d">default_init</a>(&amp;tuner);</div></div><!-- fragment --><p>After the first run, the <a class="el" href="classarm__compute_1_1_c_l_tuner.xhtml" title="Basic implementation of the OpenCL tuner interface.">CLTuner</a>'s results can be exported to a file using the method "save_to_file()".</p><ul>
Jenkins52ba29e2018-08-29 15:32:11 +0000[diff] [blame]1505<li>tuner.save_to_file("results.csv");</li>
1506</ul>
1507<p>This file can be also imported using the method "load_from_file("results.csv")".</p><ul>
1508<li>tuner.load_from_file("results.csv"); </li>
1509</ul>
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1510</div></div><!-- PageDoc -->
1511</div><!-- contents -->
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1512</div><!-- doc-content -->
1513<!-- start footer part -->
1514<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
1515 <ul>
Jenkins7f09cf72020-01-22 18:08:16 +0000[diff] [blame^]1516 <li class="footer">Generated on Wed Jan 22 2020 18:08:01 for Compute Library by
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1517 <a href="http://www.doxygen.org/index.html">
Jenkins514be652019-02-28 12:25:18 +0000[diff] [blame]1518 <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.15 </li>
Anthony Barbier871448e2017-03-24 14:54:29 +0000[diff] [blame]1519 </ul>
1520</div>
1521</body>
1522</html>