arm_compute v17.10
[platform/upstream/armcl.git] / documentation / index.xhtml
1 <!-- HTML header for doxygen 1.8.9.1-->
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
3 <html xmlns="http://www.w3.org/1999/xhtml">
4 <head>
5 <meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
6 <meta http-equiv="X-UA-Compatible" content="IE=9"/>
7 <meta name="generator" content="Doxygen 1.8.6"/>
8 <meta name="robots" content="NOINDEX, NOFOLLOW" /> <!-- Prevent indexing by search engines -->
9 <title>Compute Library: Introduction</title>
10 <link href="tabs.css" rel="stylesheet" type="text/css"/>
11 <script type="text/javascript" src="jquery.js"></script>
12 <script type="text/javascript" src="dynsections.js"></script>
13 <link href="navtree.css" rel="stylesheet" type="text/css"/>
14 <script type="text/javascript" src="resize.js"></script>
15 <script type="text/javascript" src="navtree.js"></script>
16 <script type="text/javascript">
17   $(document).ready(initResizable);
18   $(window).load(resizeHeight);
19 </script>
20 <link href="search/search.css" rel="stylesheet" type="text/css"/>
21 <script type="text/javascript" src="search/search.js"></script>
22 <script type="text/javascript">
23   $(document).ready(function() { searchBox.OnSelectItem(0); });
24 </script>
25 <script type="text/x-mathjax-config">
26   MathJax.Hub.Config({
27     extensions: ["tex2jax.js"],
28     jax: ["input/TeX","output/HTML-CSS"],
29 });
30 </script><script src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
31 <link href="doxygen.css" rel="stylesheet" type="text/css" />
32 </head>
33 <body>
34 <div id="top"><!-- do not remove this div, it is closed by doxygen! -->
35 <div id="titlearea">
36 <table cellspacing="0" cellpadding="0">
37  <tbody>
38  <tr style="height: 56px;">
39   <td style="padding-left: 0.5em;">
40    <div id="projectname">Compute Library
41    &#160;<span id="projectnumber">17.10</span>
42    </div>
43   </td>
44  </tr>
45  </tbody>
46 </table>
47 </div>
48 <!-- end header part -->
49 <!-- Generated by Doxygen 1.8.6 -->
50 <script type="text/javascript">
51 var searchBox = new SearchBox("searchBox", "search",false,'Search');
52 </script>
53   <div id="navrow1" class="tabs">
54     <ul class="tablist">
55       <li class="current"><a href="index.xhtml"><span>Main&#160;Page</span></a></li>
56       <li><a href="pages.xhtml"><span>Related&#160;Pages</span></a></li>
57       <li><a href="namespaces.xhtml"><span>Namespaces</span></a></li>
58       <li><a href="annotated.xhtml"><span>Data&#160;Structures</span></a></li>
59       <li><a href="files.xhtml"><span>Files</span></a></li>
60       <li>
61         <div id="MSearchBox" class="MSearchBoxInactive">
62         <span class="left">
63           <img id="MSearchSelect" src="search/mag_sel.png"
64                onmouseover="return searchBox.OnSearchSelectShow()"
65                onmouseout="return searchBox.OnSearchSelectHide()"
66                alt=""/>
67           <input type="text" id="MSearchField" value="Search" accesskey="S"
68                onfocus="searchBox.OnSearchFieldFocus(true)" 
69                onblur="searchBox.OnSearchFieldFocus(false)" 
70                onkeyup="searchBox.OnSearchFieldChange(event)"/>
71           </span><span class="right">
72             <a id="MSearchClose" href="javascript:searchBox.CloseResultsWindow()"><img id="MSearchCloseImg" border="0" src="search/close.png" alt=""/></a>
73           </span>
74         </div>
75       </li>
76     </ul>
77   </div>
78 </div><!-- top -->
79 <div id="side-nav" class="ui-resizable side-nav-resizable">
80   <div id="nav-tree">
81     <div id="nav-tree-contents">
82       <div id="nav-sync" class="sync"></div>
83     </div>
84   </div>
85   <div id="splitbar" style="-moz-user-select:none;" 
86        class="ui-resizable-handle">
87   </div>
88 </div>
89 <script type="text/javascript">
90 $(document).ready(function(){initNavTree('index.xhtml','');});
91 </script>
92 <div id="doc-content">
93 <!-- window showing the filter options -->
94 <div id="MSearchSelectWindow"
95      onmouseover="return searchBox.OnSearchSelectShow()"
96      onmouseout="return searchBox.OnSearchSelectHide()"
97      onkeydown="return searchBox.OnSearchSelectKey(event)">
98 <a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(0)"><span class="SelectionMark">&#160;</span>All</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(1)"><span class="SelectionMark">&#160;</span>Data Structures</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(2)"><span class="SelectionMark">&#160;</span>Namespaces</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(3)"><span class="SelectionMark">&#160;</span>Files</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(4)"><span class="SelectionMark">&#160;</span>Functions</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(5)"><span class="SelectionMark">&#160;</span>Variables</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(6)"><span class="SelectionMark">&#160;</span>Typedefs</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(7)"><span class="SelectionMark">&#160;</span>Enumerations</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(8)"><span class="SelectionMark">&#160;</span>Enumerator</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(9)"><span class="SelectionMark">&#160;</span>Friends</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(10)"><span class="SelectionMark">&#160;</span>Macros</a><a class="SelectItem" href="javascript:void(0)" onclick="searchBox.OnSelectItem(11)"><span class="SelectionMark">&#160;</span>Pages</a></div>
99
100 <!-- iframe showing the search results (closed by default) -->
101 <div id="MSearchResultsWindow">
102 <iframe src="javascript:void(0)" frameborder="0" 
103         name="MSearchResults" id="MSearchResults">
104 </iframe>
105 </div>
106
107 <div class="header">
108   <div class="headertitle">
109 <div class="title">Introduction </div>  </div>
110 </div><!--header-->
111 <div class="contents">
112 <div class="toc"><h3>Table of Contents</h3>
113 <ul><li class="level1"><a href="#S0_1_contact">Contact / Support</a></li>
114 <li class="level1"><a href="#S1_file_organisation">File organisation</a></li>
115 <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>
116 <li class="level2"><a href="#S2_2_changelog">Changelog</a></li>
117 </ul>
118 </li>
119 <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>
120 <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>
121 <li class="level3"><a href="#S3_2_2_examples">How to manually build the examples ?</a></li>
122 </ul>
123 </li>
124 <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>
125 <li class="level3"><a href="#S3_3_2_examples">How to manually build the examples ?</a></li>
126 </ul>
127 </li>
128 <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>
129 <li class="level3"><a href="#S3_4_2_examples">How to manually build the examples ?</a></li>
130 </ul>
131 </li>
132 <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>
133 <li class="level3"><a href="#S3_5_2_cygwin">Cygwin</a></li>
134 </ul>
135 </li>
136 <li class="level2"><a href="#S3_6_cl_stub_library">The OpenCL stub library</a></li>
137 </ul>
138 </li>
139 </ul>
140 </div>
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>
142 <ul>
143 <li>OS: Linux, Android or bare metal.</li>
144 <li>Architecture: armv7a (32bit) or arm64-v8a (64bit)</li>
145 <li>Technology: NEON / OpenCL / NEON and OpenCL</li>
146 <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>
147 </ul>
148 <h1><a class="anchor" id="S0_1_contact"></a>
149 Contact / Support</h1>
150 <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>
151 <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>
152 <pre class="fragment">$ strings android-armv7a-cl-asserts/libarm_compute.so | grep arm_compute_version
153 arm_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
154 </pre><h1><a class="anchor" id="S1_file_organisation"></a>
155 File organisation</h1>
156 <p>This archive contains:</p>
157 <ul>
158 <li>The <a class="el" href="namespacearm__compute.xhtml">arm_compute</a> header and source files</li>
159 <li>The latest Khronos OpenCL 1.2 C headers from the <a href="https://www.khronos.org/registry/cl/">Khronos OpenCL registry</a></li>
160 <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>
161 <li>The sources for a stub version of libOpenCL.so to help you build your application.</li>
162 <li>An examples folder containing a few examples to compile and link against the library.</li>
163 <li>A <a class="el" href="dir_29327135a2a2ef098aeaaf2df19d2f1d.xhtml">utils</a> folder containing headers with some boiler plate code used by the examples.</li>
164 <li>This documentation.</li>
165 </ul>
166 <p>You should have the following file organisation: </p>
167 <pre class="fragment">.
168 ├── arm_compute --&gt; All the arm_compute headers
169 │   ├── core
170 │   │   ├── CL
171 │   │   │   ├── CLKernelLibrary.h --&gt; Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
172 │   │   │   ├── CLKernels.h --&gt; Includes all the OpenCL kernels at once
173 │   │   │   ├── CL specialisation of all the generic objects interfaces (ICLTensor, ICLImage, etc.)
174 │   │   │   ├── kernels --&gt; Folder containing all the OpenCL kernels
175 │   │   │   │   └── CL*Kernel.h
176 │   │   │   └── OpenCL.h --&gt; Wrapper to configure the Khronos OpenCL C++ header
177 │   │   ├── CPP
178 │   │   │   ├── CPPKernels.h --&gt; Includes all the CPP kernels at once
179 │   │   │   └── kernels --&gt; Folder containing all the CPP kernels
180 │   │   │       └── CPP*Kernel.h
181 │   │   ├── NEON
182 │   │   │   ├── kernels --&gt; Folder containing all the NEON kernels
183 │   │   │   │   ├── arm64 --&gt; Folder containing the interfaces for the assembly arm64 NEON kernels
184 │   │   │   │   ├── arm32 --&gt; Folder containing the interfaces for the assembly arm32 NEON kernels
185 │   │   │   │   ├── assembly --&gt; Folder containing the NEON assembly routines.
186 │   │   │   │   └── NE*Kernel.h
187 │   │   │   └── NEKernels.h --&gt; Includes all the NEON kernels at once
188 │   │   ├── All common basic types (Types.h, Window, Coordinates, Iterator, etc.)
189 │   │   ├── All generic objects interfaces (ITensor, IImage, etc.)
190 │   │   └── Objects metadata classes (ImageInfo, TensorInfo, MultiImageInfo)
191 │   ├── graph
192 │   │   ├── CL --&gt; OpenCL specific operations
193 │   │   │   └── CLMap.h / CLUnmap.h
194 │   │   ├── nodes
195 │   │   │   └── The various nodes supported by the graph API
196 │   │   ├── Nodes.h --&gt; Includes all the Graph nodes at once.
197 │   │   └── Graph objects ( INode, ITensorAccessor, Graph, etc.)
198 │   └── runtime
199 │       ├── CL
200 │       │   ├── CL objects &amp; allocators (CLArray, CLImage, CLTensor, etc.)
201 │       │   ├── functions --&gt; Folder containing all the OpenCL functions
202 │       │   │   └── CL*.h
203 │       │   ├── CLScheduler.h --&gt; Interface to enqueue OpenCL kernels and get/set the OpenCL CommandQueue and ICLTuner.
204 │       │   └── CLFunctions.h --&gt; Includes all the OpenCL functions at once
205 │       ├── CPP
206 │       │   ├── CPPKernels.h --&gt; Includes all the CPP functions at once.
207 │       │   └── CPPScheduler.h --&gt; Basic pool of threads to execute CPP/NEON code on several cores in parallel
208 │       ├── NEON
209 │       │   ├── functions --&gt; Folder containing all the NEON functions
210 │       │   │   └── NE*.h
211 │       │   └── NEFunctions.h --&gt; Includes all the NEON functions at once
212 │       ├── OMP
213 │       │   └── OMPScheduler.h --&gt; OpenMP scheduler (Alternative to the CPPScheduler)
214 │       ├── Memory manager files (LifetimeManager, PoolManager, etc.)
215 │       └── Basic implementations of the generic object interfaces (Array, Image, Tensor, etc.)
216 ├── documentation
217 │   ├── index.xhtml
218 │   └── ...
219 ├── documentation.xhtml -&gt; documentation/index.xhtml
220 ├── examples
221 │   ├── cl_convolution.cpp
222 │   ├── cl_events.cpp
223 │   ├── graph_lenet.cpp
224 │   ├── neoncl_scale_median_gaussian.cpp
225 │   ├── neon_cnn.cpp
226 │   ├── neon_copy_objects.cpp
227 │   ├── neon_convolution.cpp
228 │   └── neon_scale.cpp
229 ├── include
230 │   ├── CL
231 │   │   └── Khronos OpenCL C headers and C++ wrapper
232 │   ├── half --&gt; FP16 library available from http://half.sourceforge.net
233 │   └── libnpy --&gt; Library to load / write npy buffers, available from https://github.com/llohse/libnpy
234 ├── opencl-1.2-stubs
235 │   └── opencl_stubs.c
236 ├── scripts
237 │   ├── caffe_data_extractor.py --&gt; Basic script to export weights from Caffe to npy files
238 │   └── tensorflow_data_extractor.py --&gt; Basic script to export weights from Tensor Flow to npy files
239 ├── src
240 │   ├── core
241 │   │   └── ... (Same structure as headers)
242 │   │       └── CL
243 │   │           └── cl_kernels --&gt; All the OpenCL kernels
244 │   ├── graph
245 │   │   └── ... (Same structure as headers)
246 │   └── runtime
247 │       └── ... (Same structure as headers)
248 ├── support
249 │   └── Various headers to work around toolchains / platform issues.
250 ├── tests
251 │   ├── All test related files shared between validation and benchmark
252 │   ├── CL --&gt; OpenCL accessors
253 │   ├── NEON --&gt; NEON accessors
254 │   ├── benchmark --&gt; Sources for benchmarking
255 │   │   ├── Benchmark specific files
256 │   │   ├── CL --&gt; OpenCL benchmarking tests
257 │   │   └── NEON --&gt; NEON benchmarking tests
258 │   ├── datasets
259 │   │   └── Datasets for all the validation / benchmark tests, layer configurations for various networks, etc.
260 │   ├── framework
261 │   │   └── Boiler plate code for both validation and benchmark test suites (Command line parsers, instruments, output loggers, etc.)
262 │   ├── networks
263 │   │   └── Examples of how to instantiate networks.
264 │   ├── validation --&gt; Sources for validation
265 │   │   ├── Validation specific files
266 │   │   ├── CL --&gt; OpenCL validation tests
267 │   │   ├── CPP --&gt; C++ reference implementations
268 │   │   ├── fixtures
269 │   │   │   └── Fixtures to initialise and run the runtime Functions.
270 │   │   └── NEON --&gt; NEON validation tests
271 │   └── dataset --&gt; Datasets defining common sets of input parameters
272 └── utils --&gt; Boiler plate code used by examples
273     └── Utils.h
274 </pre><h1><a class="anchor" id="S2_versions_changelog"></a>
275 Release versions and changelog</h1>
276 <h2><a class="anchor" id="S2_1_versions"></a>
277 Release versions</h2>
278 <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>
279 <pre class="fragment">v17.03 (First release of March 2017)
280 v17.03.1 (Second release of March 2017)
281 v17.04 (First release of April 2017)
282 </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>
283 <h2><a class="anchor" id="S2_2_changelog"></a>
284 Changelog</h2>
285 <p>v17.10 Public maintenance release</p>
286 <ul>
287 <li>Bug fixes:<ul>
288 <li>Check the maximum local workgroup size supported by OpenCL devices</li>
289 <li>Minor documentation updates (Fixed instructions to build the examples)</li>
290 <li>Introduced a <a class="el" href="classarm__compute_1_1graph_1_1_graph_context.xhtml" title="Graph context. ">arm_compute::graph::GraphContext</a></li>
291 <li>Added a few new Graph nodes and support for grouping.</li>
292 <li>Automatically enable cl_printf in debug builds</li>
293 <li>Fixed bare metal builds for armv7a</li>
294 <li>Added AlexNet and cartoon effect examples</li>
295 <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>
296 </ul>
297 </li>
298 </ul>
299 <p>v17.09 Public major release</p>
300 <ul>
301 <li>Experimental Graph support: initial implementation of a simple stream API to easily chain machine learning layers.</li>
302 <li>Memory Manager (<a class="el" href="classarm__compute_1_1_blob_lifetime_manager.xhtml">arm_compute::BlobLifetimeManager</a>, <a class="el" href="classarm__compute_1_1_blob_memory_pool.xhtml">arm_compute::BlobMemoryPool</a>, <a class="el" href="classarm__compute_1_1_i_lifetime_manager.xhtml">arm_compute::ILifetimeManager</a>, <a class="el" href="classarm__compute_1_1_i_memory_group.xhtml">arm_compute::IMemoryGroup</a>, <a class="el" href="classarm__compute_1_1_i_memory_manager.xhtml">arm_compute::IMemoryManager</a>, <a class="el" href="classarm__compute_1_1_i_memory_pool.xhtml">arm_compute::IMemoryPool</a>, <a class="el" href="classarm__compute_1_1_i_pool_manager.xhtml">arm_compute::IPoolManager</a>, <a class="el" href="classarm__compute_1_1_memory_manager_on_demand.xhtml">arm_compute::MemoryManagerOnDemand</a>, <a class="el" href="classarm__compute_1_1_pool_manager.xhtml">arm_compute::PoolManager</a>)</li>
303 <li>New validation and benchmark frameworks (Boost and Google frameworks replaced by homemade framework).</li>
304 <li>Most machine learning functions support both fixed point 8 and 16 bit (QS8, QS16) for both NEON and OpenCL.</li>
305 <li>New NEON kernels / functions:<ul>
306 <li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_assembly_base_kernel.xhtml">arm_compute::NEGEMMAssemblyBaseKernel</a> <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_a_arch64_kernel.xhtml">arm_compute::NEGEMMAArch64Kernel</a></li>
307 <li><a class="el" href="classarm__compute_1_1_n_e_dequantization_layer_kernel.xhtml">arm_compute::NEDequantizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_dequantization_layer.xhtml">arm_compute::NEDequantizationLayer</a></li>
308 <li><a class="el" href="classarm__compute_1_1_n_e_floor_kernel.xhtml">arm_compute::NEFloorKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_floor.xhtml">arm_compute::NEFloor</a></li>
309 <li><a class="el" href="classarm__compute_1_1_n_e_l2_normalize_kernel.xhtml">arm_compute::NEL2NormalizeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_l2_normalize.xhtml">arm_compute::NEL2Normalize</a></li>
310 <li><a class="el" href="classarm__compute_1_1_n_e_quantization_layer_kernel.xhtml">arm_compute::NEQuantizationLayerKernel</a> <a class="el" href="classarm__compute_1_1_n_e_min_max_layer_kernel.xhtml">arm_compute::NEMinMaxLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_quantization_layer.xhtml">arm_compute::NEQuantizationLayer</a></li>
311 <li><a class="el" href="classarm__compute_1_1_n_e_r_o_i_pooling_layer_kernel.xhtml">arm_compute::NEROIPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_r_o_i_pooling_layer.xhtml">arm_compute::NEROIPoolingLayer</a></li>
312 <li><a class="el" href="classarm__compute_1_1_n_e_reduction_operation_kernel.xhtml">arm_compute::NEReductionOperationKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_reduction_operation.xhtml">arm_compute::NEReductionOperation</a></li>
313 <li><a class="el" href="classarm__compute_1_1_n_e_reshape_layer_kernel.xhtml">arm_compute::NEReshapeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_reshape_layer.xhtml">arm_compute::NEReshapeLayer</a></li>
314 </ul>
315 </li>
316 <li>New OpenCL kernels / functions:<ul>
317 <li><a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution3x3_kernel.xhtml">arm_compute::CLDepthwiseConvolution3x3Kernel</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_im2_col_kernel.xhtml">arm_compute::CLDepthwiseIm2ColKernel</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_vector_to_tensor_kernel.xhtml">arm_compute::CLDepthwiseVectorToTensorKernel</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_weights_reshape_kernel.xhtml">arm_compute::CLDepthwiseWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution3x3.xhtml">arm_compute::CLDepthwiseConvolution3x3</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_convolution.xhtml">arm_compute::CLDepthwiseConvolution</a> <a class="el" href="classarm__compute_1_1_c_l_depthwise_separable_convolution_layer.xhtml">arm_compute::CLDepthwiseSeparableConvolutionLayer</a></li>
318 <li><a class="el" href="classarm__compute_1_1_c_l_dequantization_layer_kernel.xhtml">arm_compute::CLDequantizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_dequantization_layer.xhtml">arm_compute::CLDequantizationLayer</a></li>
319 <li><a class="el" href="classarm__compute_1_1_c_l_direct_convolution_layer_kernel.xhtml">arm_compute::CLDirectConvolutionLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_direct_convolution_layer.xhtml">arm_compute::CLDirectConvolutionLayer</a></li>
320 <li><a class="el" href="classarm__compute_1_1_c_l_flatten_layer.xhtml">arm_compute::CLFlattenLayer</a></li>
321 <li><a class="el" href="classarm__compute_1_1_c_l_floor_kernel.xhtml">arm_compute::CLFloorKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_floor.xhtml">arm_compute::CLFloor</a></li>
322 <li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_transpose1x_w.xhtml">arm_compute::CLGEMMTranspose1xW</a></li>
323 <li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_vector_multiply_kernel.xhtml">arm_compute::CLGEMMMatrixVectorMultiplyKernel</a></li>
324 <li><a class="el" href="classarm__compute_1_1_c_l_l2_normalize_kernel.xhtml">arm_compute::CLL2NormalizeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_l2_normalize.xhtml">arm_compute::CLL2Normalize</a></li>
325 <li><a class="el" href="classarm__compute_1_1_c_l_quantization_layer_kernel.xhtml">arm_compute::CLQuantizationLayerKernel</a> <a class="el" href="classarm__compute_1_1_c_l_min_max_layer_kernel.xhtml">arm_compute::CLMinMaxLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_quantization_layer.xhtml">arm_compute::CLQuantizationLayer</a></li>
326 <li><a class="el" href="classarm__compute_1_1_c_l_r_o_i_pooling_layer_kernel.xhtml">arm_compute::CLROIPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_r_o_i_pooling_layer.xhtml">arm_compute::CLROIPoolingLayer</a></li>
327 <li><a class="el" href="classarm__compute_1_1_c_l_reduction_operation_kernel.xhtml">arm_compute::CLReductionOperationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_reduction_operation.xhtml">arm_compute::CLReductionOperation</a></li>
328 <li><a class="el" href="classarm__compute_1_1_c_l_reshape_layer_kernel.xhtml">arm_compute::CLReshapeLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_reshape_layer.xhtml">arm_compute::CLReshapeLayer</a></li>
329 </ul>
330 </li>
331 </ul>
332 <p>v17.06 Public major release</p>
333 <ul>
334 <li>Various bug fixes</li>
335 <li>Added support for fixed point 8 bit (QS8) to the various NEON machine learning kernels.</li>
336 <li>Added unit tests and benchmarks (AlexNet, LeNet)</li>
337 <li>Added support for sub tensors.</li>
338 <li>Added infrastructure to provide GPU specific optimisation for some OpenCL kernels.</li>
339 <li>Added <a class="el" href="classarm__compute_1_1_o_m_p_scheduler.xhtml">arm_compute::OMPScheduler</a> (OpenMP) scheduler for NEON</li>
340 <li>Added <a class="el" href="classarm__compute_1_1_single_thread_scheduler.xhtml">arm_compute::SingleThreadScheduler</a> scheduler for NEON (For bare metal)</li>
341 <li>User can specify his own scheduler by implementing the <a class="el" href="classarm__compute_1_1_i_scheduler.xhtml">arm_compute::IScheduler</a> interface.</li>
342 <li>New OpenCL kernels / functions:<ul>
343 <li><a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer_kernel.xhtml">arm_compute::CLBatchNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_batch_normalization_layer.xhtml">arm_compute::CLBatchNormalizationLayer</a></li>
344 <li><a class="el" href="classarm__compute_1_1_c_l_depth_concatenate_kernel.xhtml">arm_compute::CLDepthConcatenateKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_depth_concatenate.xhtml">arm_compute::CLDepthConcatenate</a></li>
345 <li><a class="el" href="classarm__compute_1_1_c_l_h_o_g_orientation_binning_kernel.xhtml">arm_compute::CLHOGOrientationBinningKernel</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_block_normalization_kernel.xhtml">arm_compute::CLHOGBlockNormalizationKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_h_o_g_detector_kernel.xhtml">arm_compute::CLHOGDetectorKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_h_o_g_descriptor.xhtml">arm_compute::CLHOGDescriptor</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_detector.xhtml">arm_compute::CLHOGDetector</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_gradient.xhtml">arm_compute::CLHOGGradient</a> <a class="el" href="classarm__compute_1_1_c_l_h_o_g_multi_detection.xhtml">arm_compute::CLHOGMultiDetection</a></li>
346 <li><a class="el" href="classarm__compute_1_1_c_l_locally_connected_matrix_multiply_kernel.xhtml">arm_compute::CLLocallyConnectedMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_locally_connected_layer.xhtml">arm_compute::CLLocallyConnectedLayer</a></li>
347 <li><a class="el" href="classarm__compute_1_1_c_l_weights_reshape_kernel.xhtml">arm_compute::CLWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_convolution_layer_reshape_weights.xhtml">arm_compute::CLConvolutionLayerReshapeWeights</a></li>
348 </ul>
349 </li>
350 <li>New C++ kernels:<ul>
351 <li><a class="el" href="classarm__compute_1_1_c_p_p_detection_window_non_maxima_suppression_kernel.xhtml">arm_compute::CPPDetectionWindowNonMaximaSuppressionKernel</a></li>
352 </ul>
353 </li>
354 <li>New NEON kernels / functions:<ul>
355 <li><a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer_kernel.xhtml">arm_compute::NEBatchNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_batch_normalization_layer.xhtml">arm_compute::NEBatchNormalizationLayer</a></li>
356 <li><a class="el" href="classarm__compute_1_1_n_e_depth_concatenate_kernel.xhtml">arm_compute::NEDepthConcatenateKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_depth_concatenate.xhtml">arm_compute::NEDepthConcatenate</a></li>
357 <li><a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer_kernel.xhtml">arm_compute::NEDirectConvolutionLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_direct_convolution_layer.xhtml">arm_compute::NEDirectConvolutionLayer</a></li>
358 <li><a class="el" href="classarm__compute_1_1_n_e_locally_connected_matrix_multiply_kernel.xhtml">arm_compute::NELocallyConnectedMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_locally_connected_layer.xhtml">arm_compute::NELocallyConnectedLayer</a></li>
359 <li><a class="el" href="classarm__compute_1_1_n_e_weights_reshape_kernel.xhtml">arm_compute::NEWeightsReshapeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_convolution_layer_reshape_weights.xhtml">arm_compute::NEConvolutionLayerReshapeWeights</a></li>
360 </ul>
361 </li>
362 </ul>
363 <p>v17.05 Public bug fixes release</p>
364 <ul>
365 <li>Various bug fixes</li>
366 <li>Remaining of the functions ported to use accurate padding.</li>
367 <li>Library does not link against OpenCL anymore (It uses dlopen / dlsym at runtime instead to determine whether or not OpenCL is available).</li>
368 <li>Added "free" method to allocator.</li>
369 <li>Minimum version of g++ required for armv7 Linux changed from 4.8 to 4.9</li>
370 </ul>
371 <p>v17.04 Public bug fixes release</p>
372 <p>The following functions have been ported to use the new accurate padding:</p>
373 <ul>
374 <li><a class="el" href="classarm__compute_1_1_c_l_color_convert_kernel.xhtml">arm_compute::CLColorConvertKernel</a></li>
375 <li><a class="el" href="classarm__compute_1_1_c_l_edge_non_max_suppression_kernel.xhtml">arm_compute::CLEdgeNonMaxSuppressionKernel</a></li>
376 <li><a class="el" href="classarm__compute_1_1_c_l_edge_trace_kernel.xhtml">arm_compute::CLEdgeTraceKernel</a></li>
377 <li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_hor_kernel.xhtml">arm_compute::CLGaussianPyramidHorKernel</a></li>
378 <li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_vert_kernel.xhtml">arm_compute::CLGaussianPyramidVertKernel</a></li>
379 <li><a class="el" href="classarm__compute_1_1_c_l_gradient_kernel.xhtml">arm_compute::CLGradientKernel</a></li>
380 <li><a class="el" href="classarm__compute_1_1_n_e_channel_combine_kernel.xhtml">arm_compute::NEChannelCombineKernel</a></li>
381 <li><a class="el" href="classarm__compute_1_1_n_e_fill_array_kernel.xhtml">arm_compute::NEFillArrayKernel</a></li>
382 <li><a class="el" href="classarm__compute_1_1_n_e_gaussian_pyramid_hor_kernel.xhtml">arm_compute::NEGaussianPyramidHorKernel</a></li>
383 <li><a class="el" href="classarm__compute_1_1_n_e_gaussian_pyramid_vert_kernel.xhtml">arm_compute::NEGaussianPyramidVertKernel</a></li>
384 <li><a class="el" href="classarm__compute_1_1_n_e_harris_score_f_p16_kernel.xhtml">arm_compute::NEHarrisScoreFP16Kernel</a></li>
385 <li><a class="el" href="classarm__compute_1_1_n_e_harris_score_kernel.xhtml">arm_compute::NEHarrisScoreKernel</a></li>
386 <li><a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector_kernel.xhtml">arm_compute::NEHOGDetectorKernel</a></li>
387 <li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_max_kernel.xhtml">arm_compute::NELogits1DMaxKernel</a></li>
388 <li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_shift_exp_sum_kernel.xhtml">arm_compute::NELogits1DShiftExpSumKernel</a></li>
389 <li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_norm_kernel.xhtml">arm_compute::NELogits1DNormKernel</a></li>
390 <li><a class="el" href="classarm__compute_1_1_n_e_non_maxima_suppression3x3_f_p16_kernel.xhtml">arm_compute::NENonMaximaSuppression3x3FP16Kernel</a></li>
391 <li><a class="el" href="classarm__compute_1_1_n_e_non_maxima_suppression3x3_kernel.xhtml">arm_compute::NENonMaximaSuppression3x3Kernel</a></li>
392 </ul>
393 <p>v17.03.1 First Major public release of the sources</p>
394 <ul>
395 <li>Renamed the library to <a class="el" href="namespacearm__compute.xhtml">arm_compute</a></li>
396 <li>New CPP target introduced for C++ kernels shared between NEON and CL functions.</li>
397 <li>New padding calculation interface introduced and ported most kernels / functions to use it.</li>
398 <li>New OpenCL kernels / functions:<ul>
399 <li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp_matrix_multiply_kernel.xhtml">arm_compute::CLGEMMLowpMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_lowp.xhtml">arm_compute::CLGEMMLowp</a></li>
400 </ul>
401 </li>
402 <li>New NEON kernels / functions:<ul>
403 <li><a class="el" href="classarm__compute_1_1_n_e_normalization_layer_kernel.xhtml">arm_compute::NENormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_normalization_layer.xhtml">arm_compute::NENormalizationLayer</a></li>
404 <li><a class="el" href="classarm__compute_1_1_n_e_transpose_kernel.xhtml">arm_compute::NETransposeKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_transpose.xhtml">arm_compute::NETranspose</a></li>
405 <li><a class="el" href="classarm__compute_1_1_n_e_logits1_d_max_kernel.xhtml">arm_compute::NELogits1DMaxKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_logits1_d_shift_exp_sum_kernel.xhtml">arm_compute::NELogits1DShiftExpSumKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_logits1_d_norm_kernel.xhtml">arm_compute::NELogits1DNormKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_softmax_layer.xhtml">arm_compute::NESoftmaxLayer</a></li>
406 <li><a class="el" href="classarm__compute_1_1_n_e_im2_col_kernel.xhtml">arm_compute::NEIm2ColKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_col2_im_kernel.xhtml">arm_compute::NECol2ImKernel</a>, arm_compute::NEConvolutionLayerWeightsReshapeKernel / <a class="el" href="classarm__compute_1_1_n_e_convolution_layer.xhtml">arm_compute::NEConvolutionLayer</a></li>
407 <li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_accumulate_biases_kernel.xhtml">arm_compute::NEGEMMMatrixAccumulateBiasesKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_fully_connected_layer.xhtml">arm_compute::NEFullyConnectedLayer</a></li>
408 <li><a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp_matrix_multiply_kernel.xhtml">arm_compute::NEGEMMLowpMatrixMultiplyKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_lowp.xhtml">arm_compute::NEGEMMLowp</a></li>
409 </ul>
410 </li>
411 </ul>
412 <p>v17.03 Sources preview</p>
413 <ul>
414 <li>New OpenCL kernels / functions:<ul>
415 <li><a class="el" href="classarm__compute_1_1_c_l_gradient_kernel.xhtml">arm_compute::CLGradientKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_edge_non_max_suppression_kernel.xhtml">arm_compute::CLEdgeNonMaxSuppressionKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_edge_trace_kernel.xhtml">arm_compute::CLEdgeTraceKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_canny_edge.xhtml">arm_compute::CLCannyEdge</a></li>
416 <li>GEMM refactoring + FP16 support: <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_interleave4x4_kernel.xhtml">arm_compute::CLGEMMInterleave4x4Kernel</a>, <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_transpose1x_w_kernel.xhtml">arm_compute::CLGEMMTranspose1xWKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_multiply_kernel.xhtml">arm_compute::CLGEMMMatrixMultiplyKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_addition_kernel.xhtml">arm_compute::CLGEMMMatrixAdditionKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_g_e_m_m.xhtml">arm_compute::CLGEMM</a></li>
417 <li><a class="el" href="classarm__compute_1_1_c_l_g_e_m_m_matrix_accumulate_biases_kernel.xhtml">arm_compute::CLGEMMMatrixAccumulateBiasesKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_fully_connected_layer.xhtml">arm_compute::CLFullyConnectedLayer</a></li>
418 <li><a class="el" href="classarm__compute_1_1_c_l_transpose_kernel.xhtml">arm_compute::CLTransposeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_transpose.xhtml">arm_compute::CLTranspose</a></li>
419 <li><a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_init_kernel.xhtml">arm_compute::CLLKTrackerInitKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_stage0_kernel.xhtml">arm_compute::CLLKTrackerStage0Kernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_stage1_kernel.xhtml">arm_compute::CLLKTrackerStage1Kernel</a>, <a class="el" href="classarm__compute_1_1_c_l_l_k_tracker_finalize_kernel.xhtml">arm_compute::CLLKTrackerFinalizeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_optical_flow.xhtml">arm_compute::CLOpticalFlow</a></li>
420 <li><a class="el" href="classarm__compute_1_1_c_l_normalization_layer_kernel.xhtml">arm_compute::CLNormalizationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_normalization_layer.xhtml">arm_compute::CLNormalizationLayer</a></li>
421 <li><a class="el" href="classarm__compute_1_1_c_l_laplacian_pyramid.xhtml">arm_compute::CLLaplacianPyramid</a>, <a class="el" href="classarm__compute_1_1_c_l_laplacian_reconstruct.xhtml">arm_compute::CLLaplacianReconstruct</a></li>
422 </ul>
423 </li>
424 <li>New NEON kernels / functions:<ul>
425 <li><a class="el" href="classarm__compute_1_1_n_e_activation_layer_kernel.xhtml">arm_compute::NEActivationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_activation_layer.xhtml">arm_compute::NEActivationLayer</a></li>
426 <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">arm_compute::NEGEMMInterleave4x4Kernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_transpose1x_w_kernel.xhtml">arm_compute::NEGEMMTranspose1xWKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_multiply_kernel.xhtml">arm_compute::NEGEMMMatrixMultiplyKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m_matrix_addition_kernel.xhtml">arm_compute::NEGEMMMatrixAdditionKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_g_e_m_m.xhtml">arm_compute::NEGEMM</a></li>
427 <li><a class="el" href="classarm__compute_1_1_n_e_pooling_layer_kernel.xhtml">arm_compute::NEPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_pooling_layer.xhtml">arm_compute::NEPoolingLayer</a></li>
428 </ul>
429 </li>
430 </ul>
431 <p>v17.02.1 Sources preview</p>
432 <ul>
433 <li>New OpenCL kernels / functions:<ul>
434 <li><a class="el" href="classarm__compute_1_1_c_l_logits1_d_max_kernel.xhtml">arm_compute::CLLogits1DMaxKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_logits1_d_shift_exp_sum_kernel.xhtml">arm_compute::CLLogits1DShiftExpSumKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_logits1_d_norm_kernel.xhtml">arm_compute::CLLogits1DNormKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_softmax_layer.xhtml">arm_compute::CLSoftmaxLayer</a></li>
435 <li><a class="el" href="classarm__compute_1_1_c_l_pooling_layer_kernel.xhtml">arm_compute::CLPoolingLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_pooling_layer.xhtml">arm_compute::CLPoolingLayer</a></li>
436 <li><a class="el" href="classarm__compute_1_1_c_l_im2_col_kernel.xhtml">arm_compute::CLIm2ColKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_col2_im_kernel.xhtml">arm_compute::CLCol2ImKernel</a>, arm_compute::CLConvolutionLayerWeightsReshapeKernel / <a class="el" href="classarm__compute_1_1_c_l_convolution_layer.xhtml">arm_compute::CLConvolutionLayer</a></li>
437 <li><a class="el" href="classarm__compute_1_1_c_l_remap_kernel.xhtml">arm_compute::CLRemapKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_remap.xhtml">arm_compute::CLRemap</a></li>
438 <li><a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_hor_kernel.xhtml">arm_compute::CLGaussianPyramidHorKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_vert_kernel.xhtml">arm_compute::CLGaussianPyramidVertKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid.xhtml">arm_compute::CLGaussianPyramid</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_half.xhtml">arm_compute::CLGaussianPyramidHalf</a>, <a class="el" href="classarm__compute_1_1_c_l_gaussian_pyramid_orb.xhtml">arm_compute::CLGaussianPyramidOrb</a></li>
439 <li><a class="el" href="classarm__compute_1_1_c_l_min_max_kernel.xhtml">arm_compute::CLMinMaxKernel</a>, <a class="el" href="classarm__compute_1_1_c_l_min_max_location_kernel.xhtml">arm_compute::CLMinMaxLocationKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_min_max_location.xhtml">arm_compute::CLMinMaxLocation</a></li>
440 <li><a class="el" href="classarm__compute_1_1_c_l_non_linear_filter_kernel.xhtml">arm_compute::CLNonLinearFilterKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_non_linear_filter.xhtml">arm_compute::CLNonLinearFilter</a></li>
441 </ul>
442 </li>
443 <li>New NEON FP16 kernels (Requires armv8.2 CPU)<ul>
444 <li><a class="el" href="classarm__compute_1_1_n_e_accumulate_weighted_f_p16_kernel.xhtml">arm_compute::NEAccumulateWeightedFP16Kernel</a></li>
445 <li><a class="el" href="classarm__compute_1_1_n_e_box3x3_f_p16_kernel.xhtml">arm_compute::NEBox3x3FP16Kernel</a></li>
446 <li><a class="el" href="classarm__compute_1_1_n_e_non_maxima_suppression3x3_f_p16_kernel.xhtml">arm_compute::NENonMaximaSuppression3x3FP16Kernel</a></li>
447 </ul>
448 </li>
449 </ul>
450 <p>v17.02 Sources preview</p>
451 <ul>
452 <li>New OpenCL kernels / functions:<ul>
453 <li><a class="el" href="classarm__compute_1_1_c_l_activation_layer_kernel.xhtml">arm_compute::CLActivationLayerKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_activation_layer.xhtml">arm_compute::CLActivationLayer</a></li>
454 <li><a class="el" href="classarm__compute_1_1_c_l_channel_combine_kernel.xhtml">arm_compute::CLChannelCombineKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_channel_combine.xhtml">arm_compute::CLChannelCombine</a></li>
455 <li><a class="el" href="classarm__compute_1_1_c_l_derivative_kernel.xhtml">arm_compute::CLDerivativeKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_channel_extract.xhtml">arm_compute::CLChannelExtract</a></li>
456 <li><a class="el" href="classarm__compute_1_1_c_l_fast_corners_kernel.xhtml">arm_compute::CLFastCornersKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_fast_corners.xhtml">arm_compute::CLFastCorners</a></li>
457 <li><a class="el" href="classarm__compute_1_1_c_l_mean_std_dev_kernel.xhtml">arm_compute::CLMeanStdDevKernel</a> / <a class="el" href="classarm__compute_1_1_c_l_mean_std_dev.xhtml">arm_compute::CLMeanStdDev</a></li>
458 </ul>
459 </li>
460 <li>New NEON kernels / functions:<ul>
461 <li>HOG / SVM: <a class="el" href="classarm__compute_1_1_n_e_h_o_g_orientation_binning_kernel.xhtml">arm_compute::NEHOGOrientationBinningKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_block_normalization_kernel.xhtml">arm_compute::NEHOGBlockNormalizationKernel</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector_kernel.xhtml">arm_compute::NEHOGDetectorKernel</a>, arm_compute::NEHOGNonMaximaSuppressionKernel / <a class="el" href="classarm__compute_1_1_n_e_h_o_g_descriptor.xhtml">arm_compute::NEHOGDescriptor</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_detector.xhtml">arm_compute::NEHOGDetector</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_gradient.xhtml">arm_compute::NEHOGGradient</a>, <a class="el" href="classarm__compute_1_1_n_e_h_o_g_multi_detection.xhtml">arm_compute::NEHOGMultiDetection</a></li>
462 <li><a class="el" href="classarm__compute_1_1_n_e_non_linear_filter_kernel.xhtml">arm_compute::NENonLinearFilterKernel</a> / <a class="el" href="classarm__compute_1_1_n_e_non_linear_filter.xhtml">arm_compute::NENonLinearFilter</a></li>
463 </ul>
464 </li>
465 <li>Introduced a CLScheduler to manage the default context and command queue used by the runtime library and create synchronisation events.</li>
466 <li>Switched all the kernels / functions to use tensors instead of images.</li>
467 <li>Updated documentation to include instructions to build the library from sources.</li>
468 </ul>
469 <p>v16.12 Binary preview release</p>
470 <ul>
471 <li>Original release</li>
472 </ul>
473 <h1><a class="anchor" id="S3_how_to_build"></a>
474 How to build the library and the examples</h1>
475 <h2><a class="anchor" id="S3_1_build_options"></a>
476 Build options</h2>
477 <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>
478 <pre class="fragment">debug: Debug (yes|no)
479     default: False
480     actual: False
481
482 asserts: Enable asserts (this flag is forced to 1 for debug=1) (yes|no)
483     default: False
484     actual: False
485
486 arch: Target Architecture (armv7a|arm64-v8a|arm64-v8.2-a|x86_32|x86_64)
487     default: armv7a
488     actual: armv7a
489
490 os: Target OS (linux|android|bare_metal)
491     default: linux
492     actual: linux
493
494 build: Build type (native|cross_compile)
495     default: cross_compile
496     actual: cross_compile
497
498 examples: Build example programs (yes|no)
499     default: True
500     actual: True
501
502 Werror: Enable/disable the -Werror compilation flag (yes|no)
503     default: True
504     actual: True
505
506 opencl: Enable OpenCL support (yes|no)
507     default: True
508     actual: True
509
510 neon: Enable Neon support (yes|no)
511     default: False
512     actual: False
513
514 embed_kernels: Embed OpenCL kernels in library binary (yes|no)
515     default: False
516     actual: False
517
518 set_soname: Set the library's soname and shlibversion (requires SCons 2.4 or above) (yes|no)
519     default: False
520     actual: False
521
522 openmp: Enable OpenMP backend (yes|no)
523     default: False
524     actual: False
525
526 cppthreads: Enable C++11 threads backend (yes|no)
527     default: True
528     actual: True
529
530 build_dir: Specify sub-folder for the build ( /path/to/build_dir )
531     default: .
532     actual: .
533
534 extra_cxx_flags: Extra CXX flags to be appended to the build command
535     default:
536     actual:
537
538 pmu: Enable PMU counters (yes|no)
539     default: False
540     actual: False
541
542 mali: Enable Mali hardware counters (yes|no)
543     default: False
544     actual: False
545
546 validation_tests: Build validation test programs (yes|no)
547     default: False
548     actual: False
549
550 benchmark_tests: Build benchmark test programs (yes|no)
551     default: False
552     actual: False
553 </pre><p><b>debug</b> / <b>asserts:</b> </p>
554 <ul>
555 <li>With debug=1 asserts are enabled, and the library is built with symbols and no optimisations enabled.</li>
556 <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>
557 <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>
558 </ul>
559 <p><b>arch:</b> The x86_32 and x86_64 targets can only be used with neon=0 and opencl=1.</p>
560 <p><b>os:</b> Choose the operating system you are targeting: Linux, Android or bare metal. </p>
561 <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>
562 <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>
563 <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>
564 <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>
565 <p><b>opencl</b> / <b>neon:</b> Choose which SIMD technology you want to target. (NEON for ARM Cortex-A CPUs or OpenCL for ARM Mali GPUs)</p>
566 <p><b>embed_kernels:</b> For OpenCL only: set embed_kernels=1 if you want the OpenCL kernels to be built in the library's binaries instead of being read from separate ".cl" files. If embed_kernels is set to 0 then the application can set the path to the folder containing the OpenCL kernel files by calling CLKernelLibrary::init(). By default the path is set to "./cl_kernels".</p>
567 <p><b>set_soname:</b> Do you want to build the versioned version of the library ?</p>
568 <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>
569 <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>
570 <p><b>extra_cxx_flags:</b> Custom CXX flags which will be appended to the end of the build command.</p>
571 <p><b>build_dir:</b> Build the library in a subfolder of the "build" folder. (Allows to build several configurations in parallel).</p>
572 <p><b>examples:</b> Build or not the examples</p>
573 <p><b>validation_tests:</b> Enable the build of the validation suite.</p>
574 <p><b>benchmark_tests:</b> Enable the build of the benchmark tests</p>
575 <p><b>pmu:</b> Enable the PMU cycle counter to measure execution time in benchmark tests. (Your device needs to support it)</p>
576 <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>
577 <p><b>openmp</b> Build in the OpenMP scheduler for NEON.</p>
578 <dl class="section note"><dt>Note</dt><dd>Only works when building with g++ not clang++</dd></dl>
579 <p><b>cppthreads</b> Build in the C++11 scheduler for NEON.</p>
580 <dl class="section see"><dt>See Also</dt><dd><a class="el" href="classarm__compute_1_1_scheduler.xhtml#aa35fa7aa123444c798c28d4ac8fe7546" title="Sets the user defined scheduler and makes it the active scheduler. ">arm_compute::Scheduler::set</a></dd></dl>
581 <h2><a class="anchor" id="S3_2_linux"></a>
582 Building for Linux</h2>
583 <h3><a class="anchor" id="S3_2_1_library"></a>
584 How to build the library ?</h3>
585 <p>For Linux, the library was successfully built and tested using the following Linaro GCC toolchain:</p>
586 <ul>
587 <li>gcc-linaro-arm-linux-gnueabihf-4.9-2014.07_linux</li>
588 <li>gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu</li>
589 <li>gcc-linaro-6.3.1-2017.02-i686_aarch64-linux-gnu</li>
590 </ul>
591 <dl class="section note"><dt>Note</dt><dd>If you are building with opencl=1 then scons will expect to find libOpenCL.so either in the current directory or in "build" (See the section below if you need a stub OpenCL library to link against)</dd></dl>
592 <p>To cross-compile the library in debug mode, with NEON only support, for Linux 32bit: </p>
593 <pre class="fragment">scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=linux arch=armv7a
594 </pre><p>To cross-compile the library in asserts mode, with OpenCL only support, for Linux 64bit: </p>
595 <pre class="fragment">scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=linux arch=arm64-v8a
596 </pre><p>You can also compile the library natively on an ARM device by using <b>build=native</b>: </p>
597 <pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=arm64-v8a build=native
598 scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=native
599 </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>
600 <p>For example on a 64bit Debian based system you would have to install <b>g++-arm-linux-gnueabihf</b> </p>
601 <pre class="fragment">apt-get install g++-arm-linux-gnueabihf
602 </pre><p>Then run </p>
603 <pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=cross_compile
604 </pre><p>or simply remove the build parameter as build=cross_compile is the default value: </p>
605 <pre class="fragment">scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a
606 </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>
607 <h3><a class="anchor" id="S3_2_2_examples"></a>
608 How to manually build the examples ?</h3>
609 <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>
610 <dl class="section note"><dt>Note</dt><dd>The following command lines assume the <a class="el" href="namespacearm__compute.xhtml">arm_compute</a> and libOpenCL 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>
611 <p>To cross compile a NEON example for Linux 32bit: </p>
612 <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
613 </pre><p>To cross compile a NEON example for Linux 64bit: </p>
614 <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
615 </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>
616 <p>To cross compile an OpenCL example for Linux 32bit: </p>
617 <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 -lOpenCL -o cl_convolution -DARM_COMPUTE_CL
618 </pre><p>To cross compile an OpenCL example for Linux 64bit: </p>
619 <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 -lOpenCL -o cl_convolution -DARM_COMPUTE_CL
620 </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>
621 <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.so also. (notice the compute library has to be built with both neon and opencl enabled - neon=1 and opencl=1)</p>
622 <p>i.e. to cross compile the "graph_lenet" example for Linux 32bit: </p>
623 <pre class="fragment">arm-linux-gnueabihf-g++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -lOpenCL -o graph_lenet -DARM_COMPUTE_CL
624 </pre><p>i.e. to cross compile the "graph_lenet" example for Linux 64bit: </p>
625 <pre class="fragment">aarch64-linux-gnu-g++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -L. -larm_compute_graph -larm_compute -larm_compute_core -lOpenCL -o graph_lenet -DARM_COMPUTE_CL
626 </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>
627 <p>To compile natively (i.e directly on an ARM device) for NEON for Linux 32bit: </p>
628 <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
629 </pre><p>To compile natively (i.e directly on an ARM device) for NEON for Linux 64bit: </p>
630 <pre class="fragment">g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o neon_convolution
631 </pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option)</p>
632 <p>To compile natively (i.e directly on an ARM device) for OpenCL for Linux 32bit or Linux 64bit: </p>
633 <pre class="fragment">g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -lOpenCL -o cl_convolution -DARM_COMPUTE_CL
634 </pre><p>To compile natively (i.e directly on an ARM device) 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.so also. (notice the compute library has to be built with both neon and opencl enabled - neon=1 and opencl=1)</p>
635 <p>i.e. to cross compile the "graph_lenet" example for Linux 32bit: </p>
636 <pre class="fragment">g++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -lOpenCL -o graph_lenet -DARM_COMPUTE_CL
637 </pre><p>i.e. to cross compile the "graph_lenet" example for Linux 64bit: </p>
638 <pre class="fragment">g++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 L. -larm_compute_graph -larm_compute -larm_compute_core -lOpenCL -o graph_lenet -DARM_COMPUTE_CL
639 </pre><p>(notice the only difference with the 32 bit command is that we don't need the -mfpu option)</p>
640 <dl class="section note"><dt>Note</dt><dd>These two commands assume libarm_compute.so is available in your library path, if not add the path to it using -L</dd></dl>
641 <p>To run the built executable simply run: </p>
642 <pre class="fragment">LD_LIBRARY_PATH=build ./neon_convolution
643 </pre><p>or </p>
644 <pre class="fragment">LD_LIBRARY_PATH=build ./cl_convolution
645 </pre><dl class="section note"><dt>Note</dt><dd>If you built the library with support for both OpenCL and NEON you will need to link against OpenCL even if your application only uses NEON.</dd></dl>
646 <h2><a class="anchor" id="S3_3_android"></a>
647 Building for Android</h2>
648 <p>For Android, the library was successfully built and tested using Google's standalone toolchains:</p>
649 <ul>
650 <li>arm-linux-androideabi-4.9 for armv7a (clang++)</li>
651 <li>aarch64-linux-android-4.9 for arm64-v8a (g++)</li>
652 </ul>
653 <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>
654 <ul>
655 <li>Download the NDK r14 from here: <a href="https://developer.android.com/ndk/downloads/index.html">https://developer.android.com/ndk/downloads/index.html</a></li>
656 <li>Make sure you have Python 2 installed on your machine.</li>
657 <li>Generate the 32 and/or 64 toolchains by running the following commands:</li>
658 </ul>
659 <pre class="fragment">$NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-4.9 --stl gnustl
660 $NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-androideabi-4.9 --stl gnustl
661 </pre><dl class="section attention"><dt>Attention</dt><dd>Due to some NDK issues make sure you use g++ &amp; gnustl for aarch64 and clang++ &amp; gnustl for armv7</dd></dl>
662 <dl class="section note"><dt>Note</dt><dd>Make sure to add the toolchains to your PATH: export PATH=$PATH:$MY_TOOLCHAINS/aarch64-linux-android-4.9/bin:$MY_TOOLCHAINS/arm-linux-androideabi-4.9/bin</dd></dl>
663 <h3><a class="anchor" id="S3_3_1_library"></a>
664 How to build the library ?</h3>
665 <dl class="section note"><dt>Note</dt><dd>If you are building with opencl=1 then scons will expect to find libOpenCL.so either in the current directory or in "build" (See the section below if you need a stub OpenCL library to link against)</dd></dl>
666 <p>To cross-compile the library in debug mode, with NEON only support, for Android 32bit: </p>
667 <pre class="fragment">CXX=clang++ CC=clang scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=android arch=armv7a
668 </pre><p>To cross-compile the library in asserts mode, with OpenCL only support, for Android 64bit: </p>
669 <pre class="fragment">scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=android arch=arm64-v8a
670 </pre><h3><a class="anchor" id="S3_3_2_examples"></a>
671 How to manually build the examples ?</h3>
672 <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>
673 <dl class="section note"><dt>Note</dt><dd>The following command lines assume the <a class="el" href="namespacearm__compute.xhtml">arm_compute</a> and libOpenCL 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>
674 <p>Once you've got your Android standalone toolchain built and added to your path you can do the following:</p>
675 <p>To cross compile a NEON example: </p>
676 <pre class="fragment">#32 bit:
677 arm-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
678 #64 bit:
679 aarch64-linux-android-g++ 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
680 </pre><p>To cross compile an OpenCL example: </p>
681 <pre class="fragment">#32 bit:
682 arm-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 -lOpenCL -DARM_COMPUTE_CL
683 #64 bit:
684 aarch64-linux-android-g++ 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 -lOpenCL -DARM_COMPUTE_CL
685 </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>
686 <pre class="fragment">#32 bit:
687 arm-linux-androideabi-clang++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute_graph-static -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_arm -static-libstdc++ -pie -lOpenCL -DARM_COMPUTE_CL
688 #64 bit:
689 aarch64-linux-android-g++ examples/graph_lenet.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute_graph-static -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_aarch64 -static-libstdc++ -pie -lOpenCL -DARM_COMPUTE_CL
690 </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">arm_compute</a> statically on Android.</dd></dl>
691 <p>Then you need to do is upload the executable and the shared library to the device using ADB: </p>
692 <pre class="fragment">adb push neon_convolution_arm /data/local/tmp/
693 adb push cl_convolution_arm /data/local/tmp/
694 adb shell chmod 777 -R /data/local/tmp/
695 </pre><p>And finally to run the example: </p>
696 <pre class="fragment">adb shell /data/local/tmp/neon_convolution_arm
697 adb shell /data/local/tmp/cl_convolution_arm
698 </pre><p>For 64bit: </p>
699 <pre class="fragment">adb push neon_convolution_aarch64 /data/local/tmp/
700 adb push cl_convolution_aarch64 /data/local/tmp/
701 adb shell chmod 777 -R /data/local/tmp/
702 </pre><p>And finally to run the example: </p>
703 <pre class="fragment">adb shell /data/local/tmp/neon_convolution_aarch64
704 adb shell /data/local/tmp/cl_convolution_aarch64
705 </pre><h2><a class="anchor" id="S3_4_bare_metal"></a>
706 Building for bare metal</h2>
707 <p>For bare metal, the library was successfully built using linaros's latest (gcc-linaro-6.3.1-2017.05) bare metal toolchains:</p>
708 <ul>
709 <li>arm-eabi for armv7a</li>
710 <li>aarch64-elf for arm64-v8a</li>
711 </ul>
712 <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>
713 <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>
714 <h3><a class="anchor" id="S3_4_1_library"></a>
715 How to build the library ?</h3>
716 <p>To cross-compile the library with NEON support for baremetal arm64-v8a: </p>
717 <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
718 </pre><h3><a class="anchor" id="S3_4_2_examples"></a>
719 How to manually build the examples ?</h3>
720 <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>
721 <h2><a class="anchor" id="S3_5_windows_host"></a>
722 Building on a Windows host system</h2>
723 <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>
724 <h3><a class="anchor" id="S3_5_1_ubuntu_on_windows"></a>
725 Bash on Ubuntu on Windows</h3>
726 <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>
727 <h3><a class="anchor" id="S3_5_2_cygwin"></a>
728 Cygwin</h3>
729 <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>. 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>
730 <h2><a class="anchor" id="S3_6_cl_stub_library"></a>
731 The OpenCL stub library</h2>
732 <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">arm_compute</a> against.</p>
733 <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>
734 <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">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>
735 <p>To cross-compile the stub OpenCL library simply run: </p>
736 <pre class="fragment">&lt;target-prefix&gt;-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
737 </pre><p>For example: </p>
738 <pre class="fragment">&lt;target-prefix&gt;-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
739 #Linux 32bit
740 arm-linux-gnueabihf-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
741 #Linux 64bit
742 aarch64-linux-gnu-gcc -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC
743 #Android 32bit
744 arm-linux-androideabi-clang -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
745 #Android 64bit
746 aarch64-linux-android-gcc -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared</pre> </div></div><!-- contents -->
747 </div><!-- doc-content -->
748 <!-- start footer part -->
749 <div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
750   <ul>
751     <li class="footer">Generated on Thu Oct 12 2017 14:26:39 for Compute Library by
752     <a href="http://www.doxygen.org/index.html">
753     <img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.6 </li>
754   </ul>
755 </div>
756 </body>
757 </html>