1 /** @mainpage Introduction
5 The Computer Vision and Machine Learning library is a set of functions optimised for both ARM CPUs and GPUs using SIMD technologies.
7 Several builds of the library are available using various configurations:
8 - OS: Linux, Android or bare metal.
9 - Architecture: armv7a (32bit) or arm64-v8a (64bit)
10 - Technology: NEON / OpenCL / GLES_COMPUTE / NEON and OpenCL and GLES_COMPUTE
11 - 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.
13 @section S0_1_contact Contact / Support
15 Please email developer@arm.com
17 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:
19 $ strings android-armv7a-cl-asserts/libarm_compute.so | grep arm_compute_version
20 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
22 @section S0_2_prebuilt_binaries Pre-built binaries
24 For each release we provide some pre-built binaries of the library [here](https://github.com/ARM-software/ComputeLibrary/releases)
26 These binaries have been built using the following toolchains:
27 - Linux armv7a: gcc-linaro-arm-linux-gnueabihf-4.9-2014.07_linux
28 - Linux arm64-v8a: gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu
29 - Android armv7a: clang++ / gnustl NDK r14
30 - Android am64-v8a: clang++ / gnustl NDK r14
32 @warning Make sure to use a compatible toolchain to build your application or you will get some std::bad_alloc errors at runtime.
34 @section S1_file_organisation File organisation
36 This archive contains:
37 - The arm_compute header and source files
38 - The latest Khronos OpenCL 1.2 C headers from the <a href="https://www.khronos.org/registry/cl/">Khronos OpenCL registry</a>
39 - 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)
40 - The latest Khronos OpenGL ES 3.1 C headers from the <a href="https://www.khronos.org/registry/gles/">Khronos OpenGL ES registry</a>
41 - The latest Khronos EGL 1.5 C headers from the <a href="https://www.khronos.org/registry/gles/">Khronos EGL registry</a>
42 - The sources for a stub version of libOpenCL.so, libGLESv1_CM.so, libGLESv2.so and libEGL.so to help you build your application.
43 - An examples folder containing a few examples to compile and link against the library.
44 - A @ref utils folder containing headers with some boiler plate code used by the examples.
47 You should have the following file organisation:
50 ├── arm_compute --> All the arm_compute headers
53 │ │ │ ├── CLKernelLibrary.h --> Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
54 │ │ │ ├── CLKernels.h --> Includes all the OpenCL kernels at once
55 │ │ │ ├── CL specialisation of all the generic objects interfaces (ICLTensor, ICLImage, etc.)
56 │ │ │ ├── kernels --> Folder containing all the OpenCL kernels
57 │ │ │ │ └── CL*Kernel.h
58 │ │ │ └── OpenCL.h --> Wrapper to configure the Khronos OpenCL C++ header
60 │ │ │ ├── CPPKernels.h --> Includes all the CPP kernels at once
61 │ │ │ └── kernels --> Folder containing all the CPP kernels
62 │ │ │ └── CPP*Kernel.h
64 │ │ │ ├── GCKernelLibrary.h --> Manages all the GLES kernels compilation and caching, provides accessors for the GLES Context.
65 │ │ │ ├── GCKernels.h --> Includes all the GLES kernels at once
66 │ │ │ ├── GLES specialisation of all the generic objects interfaces (IGCTensor, IGCImage, etc.)
67 │ │ │ ├── kernels --> Folder containing all the GLES kernels
68 │ │ │ │ └── GC*Kernel.h
69 │ │ │ └── OpenGLES.h --> Wrapper to configure the Khronos EGL and OpenGL ES C header
71 │ │ │ ├── kernels --> Folder containing all the NEON kernels
72 │ │ │ │ ├── arm64 --> Folder containing the interfaces for the assembly arm64 NEON kernels
73 │ │ │ │ ├── arm32 --> Folder containing the interfaces for the assembly arm32 NEON kernels
74 │ │ │ │ ├── assembly --> Folder containing the NEON assembly routines.
75 │ │ │ │ └── NE*Kernel.h
76 │ │ │ └── NEKernels.h --> Includes all the NEON kernels at once
77 │ │ ├── All common basic types (Types.h, Window, Coordinates, Iterator, etc.)
78 │ │ ├── All generic objects interfaces (ITensor, IImage, etc.)
79 │ │ └── Objects metadata classes (ImageInfo, TensorInfo, MultiImageInfo)
81 │ │ ├── CL --> OpenCL specific operations
82 │ │ │ └── CLMap.h / CLUnmap.h
84 │ │ │ └── The various nodes supported by the graph API
85 │ │ ├── Nodes.h --> Includes all the Graph nodes at once.
86 │ │ └── Graph objects ( INode, ITensorAccessor, Graph, etc.)
89 │ │ ├── CL objects & allocators (CLArray, CLImage, CLTensor, etc.)
90 │ │ ├── functions --> Folder containing all the OpenCL functions
92 │ │ ├── CLScheduler.h --> Interface to enqueue OpenCL kernels and get/set the OpenCL CommandQueue and ICLTuner.
93 │ │ └── CLFunctions.h --> Includes all the OpenCL functions at once
95 │ │ ├── CPPKernels.h --> Includes all the CPP functions at once.
96 │ │ └── CPPScheduler.h --> Basic pool of threads to execute CPP/NEON code on several cores in parallel
98 │ │ ├── GLES objects & allocators (GCArray, GCImage, GCTensor, etc.)
99 │ │ ├── functions --> Folder containing all the GLES functions
101 │ │ ├── GCScheduler.h --> Interface to enqueue GLES kernels and get/set the GLES CommandQueue.
102 │ │ └── GCFunctions.h --> Includes all the GLES functions at once
104 │ │ ├── functions --> Folder containing all the NEON functions
106 │ │ └── NEFunctions.h --> Includes all the NEON functions at once
108 │ │ └── OMPScheduler.h --> OpenMP scheduler (Alternative to the CPPScheduler)
109 │ ├── Memory manager files (LifetimeManager, PoolManager, etc.)
110 │ └── Basic implementations of the generic object interfaces (Array, Image, Tensor, etc.)
114 ├── documentation.xhtml -> documentation/index.xhtml
116 │ ├── cl_*.cpp --> OpenCL examples
117 │ ├── gc_*.cpp --> GLES compute shaders examples
118 │ ├── graph_*.cpp --> Graph examples
119 │ ├── neoncl_*.cpp --> NEON / OpenCL interoperability examples
120 │ └── neon_*.cpp --> NEON examples
123 │ │ └── Khronos OpenCL C headers and C++ wrapper
124 │ ├── half --> FP16 library available from http://half.sourceforge.net
125 │ ├── libnpy --> Library to load / write npy buffers, available from https://github.com/llohse/libnpy
126 │ └── linux --> Headers only needed for Linux builds
127 │ └── Khronos EGL and OpenGLES headers
129 │ └── opencl_stubs.c --> OpenCL stubs implementation
130 ├── opengles-3.1-stubs
131 │ ├── EGL.c --> EGL stubs implementation
132 │ └── GLESv2.c --> GLESv2 stubs implementation
134 │ ├── caffe_data_extractor.py --> Basic script to export weights from Caffe to npy files
135 │ └── tensorflow_data_extractor.py --> Basic script to export weights from Tensor Flow to npy files
138 │ │ └── ... (Same structure as headers)
140 │ │ │ └── cl_kernels --> All the OpenCL kernels
142 │ │ └── cs_shaders --> All the OpenGL ES Compute Shaders
144 │ │ └── ... (Same structure as headers)
146 │ └── ... (Same structure as headers)
148 │ └── Various headers to work around toolchains / platform issues.
150 │ ├── All test related files shared between validation and benchmark
151 │ ├── CL --> OpenCL accessors
152 │ ├── GLES_COMPUTE --> GLES accessors
153 │ ├── NEON --> NEON accessors
154 │ ├── benchmark --> Sources for benchmarking
155 │ │ ├── Benchmark specific files
156 │ │ ├── CL --> OpenCL benchmarking tests
157 │ │ ├── GLES_COMPUTE --> GLES benchmarking tests
158 │ │ └── NEON --> NEON benchmarking tests
160 │ │ └── Datasets for all the validation / benchmark tests, layer configurations for various networks, etc.
162 │ │ └── Boiler plate code for both validation and benchmark test suites (Command line parsers, instruments, output loggers, etc.)
164 │ │ └── Examples of how to instantiate networks.
165 │ ├── validation --> Sources for validation
166 │ │ ├── Validation specific files
167 │ │ ├── CL --> OpenCL validation tests
168 │ │ ├── GLES_COMPUTE --> GLES validation tests
169 │ │ ├── CPP --> C++ reference implementations
171 │ │ │ └── Fixtures to initialise and run the runtime Functions.
172 │ │ └── NEON --> NEON validation tests
173 │ └── dataset --> Datasets defining common sets of input parameters
174 └── utils --> Boiler plate code used by examples
175 └── Various utilities to print types, load / store assets, etc.
177 @section S2_versions_changelog Release versions and changelog
179 @subsection S2_1_versions Release versions
181 All releases are numbered vYY.MM Where YY are the last two digits of the year, and MM the month number.
182 If there is more than one release in a month then an extra sequential number is appended at the end:
184 v17.03 (First release of March 2017)
185 v17.03.1 (Second release of March 2017)
186 v17.04 (First release of April 2017)
188 @note We're aiming at releasing one major public release with new features per quarter. All releases in between will only contain bug fixes.
190 @subsection S2_2_changelog Changelog
192 v18.02 Public major release
193 - Various NEON / OpenCL / GLES optimisations.
195 - Changed default number of threads on big LITTLE systems.
196 - Refactored examples and added:
197 - graph_mobilenet_qassym8
199 - graph_squeezenet_v1_1
200 - Renamed @ref arm_compute::CLConvolutionLayer into @ref arm_compute::CLGEMMConvolutionLayer and created a new @ref arm_compute::CLConvolutionLayer to select the fastest convolution method.
201 - Renamed @ref arm_compute::NEConvolutionLayer into @ref arm_compute::NEGEMMConvolutionLayer and created a new @ref arm_compute::NEConvolutionLayer to select the fastest convolution method.
202 - Added in place support to:
203 - @ref arm_compute::CLActivationLayer
204 - @ref arm_compute::CLBatchNormalizationLayer
205 - Added QASYMM8 support to:
206 - @ref arm_compute::CLActivationLayer
207 - @ref arm_compute::CLDepthwiseConvolutionLayer
208 - @ref arm_compute::NEDepthwiseConvolutionLayer
209 - @ref arm_compute::NESoftmaxLayer
210 - Added FP16 support to:
211 - @ref arm_compute::CLDepthwiseConvolutionLayer3x3
212 - @ref arm_compute::CLDepthwiseConvolutionLayer
213 - Added broadcasting support to @ref arm_compute::NEArithmeticAddition / @ref arm_compute::CLArithmeticAddition / @ref arm_compute::CLPixelWiseMultiplication
214 - Added fused batched normalization and activation to @ref arm_compute::CLBatchNormalizationLayer and @ref arm_compute::NEBatchNormalizationLayer
215 - Added support for non-square pooling to @ref arm_compute::NEPoolingLayer and @ref arm_compute::CLPoolingLayer
216 - New OpenCL kernels / functions:
217 - @ref arm_compute::CLDirectConvolutionLayerOutputStageKernel
218 - New NEON kernels / functions
219 - Added name() method to all kernels.
220 - Added support for Winograd 5x5.
221 - @ref arm_compute::NEPermuteKernel / @ref arm_compute::NEPermute
222 - @ref arm_compute::NEWinogradLayerTransformInputKernel / @ref arm_compute::NEWinogradLayer
223 - @ref arm_compute::NEWinogradLayerTransformOutputKernel / @ref arm_compute::NEWinogradLayer
224 - @ref arm_compute::NEWinogradLayerTransformWeightsKernel / @ref arm_compute::NEWinogradLayer
225 - Renamed arm_compute::NEWinogradLayerKernel into @ref arm_compute::NEWinogradLayerBatchedGEMMKernel
226 - New GLES kernels / functions:
227 - @ref arm_compute::GCTensorShiftKernel / @ref arm_compute::GCTensorShift
229 v18.01 Public maintenance release
231 - Added some of the missing validate() methods
232 - Added @ref arm_compute::CLDeconvolutionLayerUpsampleKernel / @ref arm_compute::CLDeconvolutionLayer @ref arm_compute::CLDeconvolutionLayerUpsample
233 - Added @ref arm_compute::CLPermuteKernel / @ref arm_compute::CLPermute
234 - Added method to clean the programs cache in the CL Kernel library.
235 - Added @ref arm_compute::GCArithmeticAdditionKernel / @ref arm_compute::GCArithmeticAddition
236 - Added @ref arm_compute::GCDepthwiseConvolutionLayer3x3Kernel / @ref arm_compute::GCDepthwiseConvolutionLayer3x3
237 - Added @ref arm_compute::GCNormalizePlanarYUVLayerKernel / @ref arm_compute::GCNormalizePlanarYUVLayer
238 - Added @ref arm_compute::GCScaleKernel / @ref arm_compute::GCScale
239 - Added @ref arm_compute::GCWeightsReshapeKernel / @ref arm_compute::GCConvolutionLayer
240 - Added FP16 support to the following GLES compute kernels:
241 - @ref arm_compute::GCCol2ImKernel
242 - @ref arm_compute::GCGEMMInterleave4x4Kernel
243 - @ref arm_compute::GCGEMMTranspose1xWKernel
244 - @ref arm_compute::GCIm2ColKernel
245 - Refactored NEON Winograd (arm_compute::NEWinogradLayerKernel)
246 - Added @ref arm_compute::NEDirectConvolutionLayerOutputStageKernel
247 - Added QASYMM8 support to the following NEON kernels:
248 - @ref arm_compute::NEDepthwiseConvolutionLayer3x3Kernel
249 - @ref arm_compute::NEFillBorderKernel
250 - @ref arm_compute::NEPoolingLayerKernel
251 - Added new examples:
252 - graph_cl_mobilenet_qasymm8.cpp
253 - graph_inception_v3.cpp
255 - More tests added to both validation and benchmarking suites.
257 v17.12 Public major release
258 - Most machine learning functions on OpenCL support the new data type QASYMM8
259 - Introduced logging interface
260 - Introduced opencl timer
261 - Reworked GEMMLowp interface
262 - Added new NEON assembly kernels for GEMMLowp, SGEMM and HGEMM
263 - Added validation method for most Machine Learning kernels / functions
264 - Added new graph examples such as googlenet, mobilenet, squeezenet, vgg16 and vgg19
265 - Added sgemm example for OpenCL
266 - Added absolute difference example for GLES compute
267 - Added new tests and benchmarks in validation and benchmark frameworks
268 - Added new kernels / functions for GLES compute
270 - New OpenGL ES kernels / functions
271 - @ref arm_compute::GCAbsoluteDifferenceKernel / @ref arm_compute::GCAbsoluteDifference
272 - @ref arm_compute::GCActivationLayerKernel / @ref arm_compute::GCActivationLayer
273 - @ref arm_compute::GCBatchNormalizationLayerKernel / @ref arm_compute::GCBatchNormalizationLayer
274 - @ref arm_compute::GCCol2ImKernel
275 - @ref arm_compute::GCDepthConcatenateLayerKernel / @ref arm_compute::GCDepthConcatenateLayer
276 - @ref arm_compute::GCDirectConvolutionLayerKernel / @ref arm_compute::GCDirectConvolutionLayer
277 - @ref arm_compute::GCDropoutLayerKernel / @ref arm_compute::GCDropoutLayer
278 - @ref arm_compute::GCFillBorderKernel / @ref arm_compute::GCFillBorder
279 - @ref arm_compute::GCGEMMInterleave4x4Kernel / @ref arm_compute::GCGEMMInterleave4x4
280 - @ref arm_compute::GCGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::GCGEMMMatrixAdditionKernel / @ref arm_compute::GCGEMMMatrixMultiplyKernel / @ref arm_compute::GCGEMM
281 - @ref arm_compute::GCGEMMTranspose1xWKernel / @ref arm_compute::GCGEMMTranspose1xW
282 - @ref arm_compute::GCIm2ColKernel
283 - @ref arm_compute::GCNormalizationLayerKernel / @ref arm_compute::GCNormalizationLayer
284 - @ref arm_compute::GCPixelWiseMultiplicationKernel / @ref arm_compute::GCPixelWiseMultiplication
285 - @ref arm_compute::GCPoolingLayerKernel / @ref arm_compute::GCPoolingLayer
286 - @ref arm_compute::GCLogits1DMaxKernel / @ref arm_compute::GCLogits1DShiftExpSumKernel / @ref arm_compute::GCLogits1DNormKernel / @ref arm_compute::GCSoftmaxLayer
287 - @ref arm_compute::GCTransposeKernel / @ref arm_compute::GCTranspose
289 - New NEON kernels / functions
290 - @ref arm_compute::NEGEMMLowpAArch64A53Kernel / @ref arm_compute::NEGEMMLowpAArch64Kernel / @ref arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / @ref arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore
291 - @ref arm_compute::NEHGEMMAArch64FP16Kernel
292 - @ref arm_compute::NEDepthwiseConvolutionLayer3x3Kernel / @ref arm_compute::NEDepthwiseIm2ColKernel / @ref arm_compute::NEGEMMMatrixVectorMultiplyKernel / @ref arm_compute::NEDepthwiseVectorToTensorKernel / @ref arm_compute::NEDepthwiseConvolutionLayer
293 - @ref arm_compute::NEGEMMLowpOffsetContributionKernel / @ref arm_compute::NEGEMMLowpMatrixAReductionKernel / @ref arm_compute::NEGEMMLowpMatrixBReductionKernel / @ref arm_compute::NEGEMMLowpMatrixMultiplyCore
294 - @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
295 - @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8Scale
296 - @ref arm_compute::NEWinogradLayer / arm_compute::NEWinogradLayerKernel
298 - New OpenCL kernels / functions
299 - @ref arm_compute::CLGEMMLowpOffsetContributionKernel / @ref arm_compute::CLGEMMLowpMatrixAReductionKernel / @ref arm_compute::CLGEMMLowpMatrixBReductionKernel / @ref arm_compute::CLGEMMLowpMatrixMultiplyCore
300 - @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
301 - @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8Scale
303 - New graph nodes for NEON and OpenCL
304 - @ref arm_compute::graph::BranchLayer
305 - @ref arm_compute::graph::DepthConvertLayer
306 - @ref arm_compute::graph::DepthwiseConvolutionLayer
307 - @ref arm_compute::graph::DequantizationLayer
308 - @ref arm_compute::graph::FlattenLayer
309 - @ref arm_compute::graph::QuantizationLayer
310 - @ref arm_compute::graph::ReshapeLayer
312 v17.10 Public maintenance release
314 - Check the maximum local workgroup size supported by OpenCL devices
315 - Minor documentation updates (Fixed instructions to build the examples)
316 - Introduced a arm_compute::graph::GraphContext
317 - Added a few new Graph nodes, support for branches and grouping.
318 - Automatically enable cl_printf in debug builds
319 - Fixed bare metal builds for armv7a
320 - Added AlexNet and cartoon effect examples
321 - 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)
323 v17.09 Public major release
324 - Experimental Graph support: initial implementation of a simple stream API to easily chain machine learning layers.
325 - Memory Manager (@ref arm_compute::BlobLifetimeManager, @ref arm_compute::BlobMemoryPool, @ref arm_compute::ILifetimeManager, @ref arm_compute::IMemoryGroup, @ref arm_compute::IMemoryManager, @ref arm_compute::IMemoryPool, @ref arm_compute::IPoolManager, @ref arm_compute::MemoryManagerOnDemand, @ref arm_compute::PoolManager)
326 - New validation and benchmark frameworks (Boost and Google frameworks replaced by homemade framework).
327 - Most machine learning functions support both fixed point 8 and 16 bit (QS8, QS16) for both NEON and OpenCL.
328 - New NEON kernels / functions:
329 - @ref arm_compute::NEGEMMAssemblyBaseKernel @ref arm_compute::NEGEMMAArch64Kernel
330 - @ref arm_compute::NEDequantizationLayerKernel / @ref arm_compute::NEDequantizationLayer
331 - @ref arm_compute::NEFloorKernel / @ref arm_compute::NEFloor
332 - @ref arm_compute::NEL2NormalizeLayerKernel / @ref arm_compute::NEL2NormalizeLayer
333 - @ref arm_compute::NEQuantizationLayerKernel @ref arm_compute::NEMinMaxLayerKernel / @ref arm_compute::NEQuantizationLayer
334 - @ref arm_compute::NEROIPoolingLayerKernel / @ref arm_compute::NEROIPoolingLayer
335 - @ref arm_compute::NEReductionOperationKernel / @ref arm_compute::NEReductionOperation
336 - @ref arm_compute::NEReshapeLayerKernel / @ref arm_compute::NEReshapeLayer
338 - New OpenCL kernels / functions:
339 - @ref arm_compute::CLDepthwiseConvolutionLayer3x3Kernel @ref arm_compute::CLDepthwiseIm2ColKernel @ref arm_compute::CLDepthwiseVectorToTensorKernel @ref arm_compute::CLDepthwiseWeightsReshapeKernel / @ref arm_compute::CLDepthwiseConvolutionLayer3x3 @ref arm_compute::CLDepthwiseConvolutionLayer @ref arm_compute::CLDepthwiseSeparableConvolutionLayer
340 - @ref arm_compute::CLDequantizationLayerKernel / @ref arm_compute::CLDequantizationLayer
341 - @ref arm_compute::CLDirectConvolutionLayerKernel / @ref arm_compute::CLDirectConvolutionLayer
342 - @ref arm_compute::CLFlattenLayer
343 - @ref arm_compute::CLFloorKernel / @ref arm_compute::CLFloor
344 - @ref arm_compute::CLGEMMTranspose1xW
345 - @ref arm_compute::CLGEMMMatrixVectorMultiplyKernel
346 - @ref arm_compute::CLL2NormalizeLayerKernel / @ref arm_compute::CLL2NormalizeLayer
347 - @ref arm_compute::CLQuantizationLayerKernel @ref arm_compute::CLMinMaxLayerKernel / @ref arm_compute::CLQuantizationLayer
348 - @ref arm_compute::CLROIPoolingLayerKernel / @ref arm_compute::CLROIPoolingLayer
349 - @ref arm_compute::CLReductionOperationKernel / @ref arm_compute::CLReductionOperation
350 - @ref arm_compute::CLReshapeLayerKernel / @ref arm_compute::CLReshapeLayer
352 v17.06 Public major release
354 - Added support for fixed point 8 bit (QS8) to the various NEON machine learning kernels.
355 - Added unit tests and benchmarks (AlexNet, LeNet)
356 - Added support for sub tensors.
357 - Added infrastructure to provide GPU specific optimisation for some OpenCL kernels.
358 - Added @ref arm_compute::OMPScheduler (OpenMP) scheduler for NEON
359 - Added @ref arm_compute::SingleThreadScheduler scheduler for NEON (For bare metal)
360 - User can specify his own scheduler by implementing the @ref arm_compute::IScheduler interface.
361 - New OpenCL kernels / functions:
362 - @ref arm_compute::CLBatchNormalizationLayerKernel / @ref arm_compute::CLBatchNormalizationLayer
363 - @ref arm_compute::CLDepthConcatenateLayerKernel / @ref arm_compute::CLDepthConcatenateLayer
364 - @ref arm_compute::CLHOGOrientationBinningKernel @ref arm_compute::CLHOGBlockNormalizationKernel, @ref arm_compute::CLHOGDetectorKernel / @ref arm_compute::CLHOGDescriptor @ref arm_compute::CLHOGDetector @ref arm_compute::CLHOGGradient @ref arm_compute::CLHOGMultiDetection
365 - @ref arm_compute::CLLocallyConnectedMatrixMultiplyKernel / @ref arm_compute::CLLocallyConnectedLayer
366 - @ref arm_compute::CLWeightsReshapeKernel / @ref arm_compute::CLConvolutionLayerReshapeWeights
368 - @ref arm_compute::CPPDetectionWindowNonMaximaSuppressionKernel
369 - New NEON kernels / functions:
370 - @ref arm_compute::NEBatchNormalizationLayerKernel / @ref arm_compute::NEBatchNormalizationLayer
371 - @ref arm_compute::NEDepthConcatenateLayerKernel / @ref arm_compute::NEDepthConcatenateLayer
372 - @ref arm_compute::NEDirectConvolutionLayerKernel / @ref arm_compute::NEDirectConvolutionLayer
373 - @ref arm_compute::NELocallyConnectedMatrixMultiplyKernel / @ref arm_compute::NELocallyConnectedLayer
374 - @ref arm_compute::NEWeightsReshapeKernel / @ref arm_compute::NEConvolutionLayerReshapeWeights
376 v17.05 Public bug fixes release
378 - Remaining of the functions ported to use accurate padding.
379 - Library does not link against OpenCL anymore (It uses dlopen / dlsym at runtime instead to determine whether or not OpenCL is available).
380 - Added "free" method to allocator.
381 - Minimum version of g++ required for armv7 Linux changed from 4.8 to 4.9
383 v17.04 Public bug fixes release
385 The following functions have been ported to use the new accurate padding:
386 - @ref arm_compute::CLColorConvertKernel
387 - @ref arm_compute::CLEdgeNonMaxSuppressionKernel
388 - @ref arm_compute::CLEdgeTraceKernel
389 - @ref arm_compute::CLGaussianPyramidHorKernel
390 - @ref arm_compute::CLGaussianPyramidVertKernel
391 - @ref arm_compute::CLGradientKernel
392 - @ref arm_compute::NEChannelCombineKernel
393 - @ref arm_compute::NEFillArrayKernel
394 - @ref arm_compute::NEGaussianPyramidHorKernel
395 - @ref arm_compute::NEGaussianPyramidVertKernel
396 - @ref arm_compute::NEHarrisScoreFP16Kernel
397 - @ref arm_compute::NEHarrisScoreKernel
398 - @ref arm_compute::NEHOGDetectorKernel
399 - @ref arm_compute::NELogits1DMaxKernel
400 - arm_compute::NELogits1DShiftExpSumKernel
401 - arm_compute::NELogits1DNormKernel
402 - @ref arm_compute::NENonMaximaSuppression3x3FP16Kernel
403 - @ref arm_compute::NENonMaximaSuppression3x3Kernel
405 v17.03.1 First Major public release of the sources
406 - Renamed the library to arm_compute
407 - New CPP target introduced for C++ kernels shared between NEON and CL functions.
408 - New padding calculation interface introduced and ported most kernels / functions to use it.
409 - New OpenCL kernels / functions:
410 - @ref arm_compute::CLGEMMLowpMatrixMultiplyKernel / arm_compute::CLGEMMLowp
411 - New NEON kernels / functions:
412 - @ref arm_compute::NENormalizationLayerKernel / @ref arm_compute::NENormalizationLayer
413 - @ref arm_compute::NETransposeKernel / @ref arm_compute::NETranspose
414 - @ref arm_compute::NELogits1DMaxKernel, arm_compute::NELogits1DShiftExpSumKernel, arm_compute::NELogits1DNormKernel / @ref arm_compute::NESoftmaxLayer
415 - @ref arm_compute::NEIm2ColKernel, @ref arm_compute::NECol2ImKernel, arm_compute::NEConvolutionLayerWeightsReshapeKernel / @ref arm_compute::NEConvolutionLayer
416 - @ref arm_compute::NEGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::NEFullyConnectedLayer
417 - @ref arm_compute::NEGEMMLowpMatrixMultiplyKernel / arm_compute::NEGEMMLowp
419 v17.03 Sources preview
420 - New OpenCL kernels / functions:
421 - @ref arm_compute::CLGradientKernel, @ref arm_compute::CLEdgeNonMaxSuppressionKernel, @ref arm_compute::CLEdgeTraceKernel / @ref arm_compute::CLCannyEdge
422 - GEMM refactoring + FP16 support: @ref arm_compute::CLGEMMInterleave4x4Kernel, @ref arm_compute::CLGEMMTranspose1xWKernel, @ref arm_compute::CLGEMMMatrixMultiplyKernel, @ref arm_compute::CLGEMMMatrixAdditionKernel / @ref arm_compute::CLGEMM
423 - @ref arm_compute::CLGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::CLFullyConnectedLayer
424 - @ref arm_compute::CLTransposeKernel / @ref arm_compute::CLTranspose
425 - @ref arm_compute::CLLKTrackerInitKernel, @ref arm_compute::CLLKTrackerStage0Kernel, @ref arm_compute::CLLKTrackerStage1Kernel, @ref arm_compute::CLLKTrackerFinalizeKernel / @ref arm_compute::CLOpticalFlow
426 - @ref arm_compute::CLNormalizationLayerKernel / @ref arm_compute::CLNormalizationLayer
427 - @ref arm_compute::CLLaplacianPyramid, @ref arm_compute::CLLaplacianReconstruct
428 - New NEON kernels / functions:
429 - @ref arm_compute::NEActivationLayerKernel / @ref arm_compute::NEActivationLayer
430 - GEMM refactoring + FP16 support (Requires armv8.2 CPU): @ref arm_compute::NEGEMMInterleave4x4Kernel, @ref arm_compute::NEGEMMTranspose1xWKernel, @ref arm_compute::NEGEMMMatrixMultiplyKernel, @ref arm_compute::NEGEMMMatrixAdditionKernel / @ref arm_compute::NEGEMM
431 - @ref arm_compute::NEPoolingLayerKernel / @ref arm_compute::NEPoolingLayer
433 v17.02.1 Sources preview
434 - New OpenCL kernels / functions:
435 - @ref arm_compute::CLLogits1DMaxKernel, @ref arm_compute::CLLogits1DShiftExpSumKernel, @ref arm_compute::CLLogits1DNormKernel / @ref arm_compute::CLSoftmaxLayer
436 - @ref arm_compute::CLPoolingLayerKernel / @ref arm_compute::CLPoolingLayer
437 - @ref arm_compute::CLIm2ColKernel, @ref arm_compute::CLCol2ImKernel, arm_compute::CLConvolutionLayerWeightsReshapeKernel / @ref arm_compute::CLConvolutionLayer
438 - @ref arm_compute::CLRemapKernel / @ref arm_compute::CLRemap
439 - @ref arm_compute::CLGaussianPyramidHorKernel, @ref arm_compute::CLGaussianPyramidVertKernel / @ref arm_compute::CLGaussianPyramid, @ref arm_compute::CLGaussianPyramidHalf, @ref arm_compute::CLGaussianPyramidOrb
440 - @ref arm_compute::CLMinMaxKernel, @ref arm_compute::CLMinMaxLocationKernel / @ref arm_compute::CLMinMaxLocation
441 - @ref arm_compute::CLNonLinearFilterKernel / @ref arm_compute::CLNonLinearFilter
442 - New NEON FP16 kernels (Requires armv8.2 CPU)
443 - @ref arm_compute::NEAccumulateWeightedFP16Kernel
444 - @ref arm_compute::NEBox3x3FP16Kernel
445 - @ref arm_compute::NENonMaximaSuppression3x3FP16Kernel
447 v17.02 Sources preview
448 - New OpenCL kernels / functions:
449 - @ref arm_compute::CLActivationLayerKernel / @ref arm_compute::CLActivationLayer
450 - @ref arm_compute::CLChannelCombineKernel / @ref arm_compute::CLChannelCombine
451 - @ref arm_compute::CLDerivativeKernel / @ref arm_compute::CLChannelExtract
452 - @ref arm_compute::CLFastCornersKernel / @ref arm_compute::CLFastCorners
453 - @ref arm_compute::CLMeanStdDevKernel / @ref arm_compute::CLMeanStdDev
454 - New NEON kernels / functions:
455 - HOG / SVM: @ref arm_compute::NEHOGOrientationBinningKernel, @ref arm_compute::NEHOGBlockNormalizationKernel, @ref arm_compute::NEHOGDetectorKernel, arm_compute::NEHOGNonMaximaSuppressionKernel / @ref arm_compute::NEHOGDescriptor, @ref arm_compute::NEHOGDetector, @ref arm_compute::NEHOGGradient, @ref arm_compute::NEHOGMultiDetection
456 - @ref arm_compute::NENonLinearFilterKernel / @ref arm_compute::NENonLinearFilter
457 - Introduced a CLScheduler to manage the default context and command queue used by the runtime library and create synchronisation events.
458 - Switched all the kernels / functions to use tensors instead of images.
459 - Updated documentation to include instructions to build the library from sources.
461 v16.12 Binary preview release
464 @section S3_how_to_build How to build the library and the examples
466 @subsection S3_1_build_options Build options
468 scons 2.3 or above is required to build the library.
469 To see the build options available simply run ```scons -h```:
471 debug: Debug (yes|no)
475 asserts: Enable asserts (this flag is forced to 1 for debug=1) (yes|no)
479 arch: Target Architecture (armv7a|arm64-v8a|arm64-v8.2-a|x86_32|x86_64)
483 os: Target OS (linux|android|bare_metal)
487 build: Build type (native|cross_compile|embed_only)
488 default: cross_compile
489 actual: cross_compile
491 examples: Build example programs (yes|no)
495 Werror: Enable/disable the -Werror compilation flag (yes|no)
499 opencl: Enable OpenCL support (yes|no)
503 neon: Enable Neon support (yes|no)
507 gles_compute: Enable OpenGL ES Compute Shader support (yes|no)
511 embed_kernels: Embed OpenCL kernels and OpenGL ES compute shader in library binary (yes|no)
515 set_soname: Set the library's soname and shlibversion (requires SCons 2.4 or above) (yes|no)
519 openmp: Enable OpenMP backend (yes|no)
523 cppthreads: Enable C++11 threads backend (yes|no)
527 build_dir: Specify sub-folder for the build ( /path/to/build_dir )
531 extra_cxx_flags: Extra CXX flags to be appended to the build command
535 pmu: Enable PMU counters (yes|no)
539 mali: Enable Mali hardware counters (yes|no)
543 validation_tests: Build validation test programs (yes|no)
547 benchmark_tests: Build benchmark test programs (yes|no)
551 @b debug / @b asserts:
552 - With debug=1 asserts are enabled, and the library is built with symbols and no optimisations enabled.
553 - 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)
554 - 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).
556 @b arch: The x86_32 and x86_64 targets can only be used with neon=0 and opencl=1.
558 @b os: Choose the operating system you are targeting: Linux, Android or bare metal.
559 @note bare metal can only be used for NEON (not OpenCL), only static libraries get built and NEON's multi-threading support is disabled.
561 @b build: 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.
563 @note 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.
565 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.
567 @b Werror: 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 developer@arm.com so that the issue can be addressed).
569 @b opencl / @b neon / @b gles_compute: Choose which SIMD technology you want to target. (NEON for ARM Cortex-A CPUs or OpenCL / GLES_COMPUTE for ARM Mali GPUs)
571 @b embed_kernels: 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 CLKernelLibrary::init() / GCKernelLibrary::init(). By default the path is set to "./cl_kernels" / "./cs_shaders".
573 @b set_soname: Do you want to build the versioned version of the library ?
575 If enabled the library will contain a SONAME and SHLIBVERSION and some symlinks will automatically be created between the objects.
577 libarm_compute_core.so -> libarm_compute_core.so.1.0.0
578 libarm_compute_core.so.1 -> libarm_compute_core.so.1.0.0
579 libarm_compute_core.so.1.0.0
581 @note This options is disabled by default as it requires SCons version 2.4 or above.
583 @b extra_cxx_flags: Custom CXX flags which will be appended to the end of the build command.
585 @b build_dir: Build the library in a subfolder of the "build" folder. (Allows to build several configurations in parallel).
587 @b examples: Build or not the examples
589 @b validation_tests: Enable the build of the validation suite.
591 @b benchmark_tests: Enable the build of the benchmark tests
593 @b pmu: Enable the PMU cycle counter to measure execution time in benchmark tests. (Your device needs to support it)
595 @b mali: 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)
597 @b openmp Build in the OpenMP scheduler for NEON.
599 @note Only works when building with g++ not clang++
601 @b cppthreads Build in the C++11 scheduler for NEON.
603 @sa arm_compute::Scheduler::set
605 @subsection S3_2_linux Building for Linux
607 @subsubsection S3_2_1_library How to build the library ?
609 For Linux, the library was successfully built and tested using the following Linaro GCC toolchain:
611 - gcc-linaro-arm-linux-gnueabihf-4.9-2014.07_linux
612 - gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu
613 - gcc-linaro-6.3.1-2017.02-i686_aarch64-linux-gnu
615 @note 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)
616 @note If you are building with gles_compute=1 then scons will expect to find libEGL.so / libGLESv1_CM.so / libGLESv2.so either in the current directory or in "build" (See the section below if you need a stub OpenCL library to link against)
618 To cross-compile the library in debug mode, with NEON only support, for Linux 32bit:
620 scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=linux arch=armv7a
622 To cross-compile the library in asserts mode, with OpenCL only support, for Linux 64bit:
624 scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=linux arch=arm64-v8a
626 To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Linux 64bit:
628 scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=0 gles_compute=1 embed_kernels=1 os=linux arch=arm64-v8a
630 You can also compile the library natively on an ARM device by using <b>build=native</b>:
632 scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=arm64-v8a build=native
633 scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=native
635 @note 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.
637 For example on a 64bit Debian based system you would have to install <b>g++-arm-linux-gnueabihf</b>
639 apt-get install g++-arm-linux-gnueabihf
643 scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=cross_compile
645 or simply remove the build parameter as build=cross_compile is the default value:
647 scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a
649 @attention To cross compile with opencl=1 you need to make sure to have a version of libOpenCL matching your target architecture.
651 @subsubsection S3_2_2_examples How to manually build the examples ?
653 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.
655 @note The following command lines assume the arm_compute 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.
657 To cross compile a NEON example for Linux 32bit:
659 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
661 To cross compile a NEON example for Linux 64bit:
663 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
665 (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)
667 To cross compile an OpenCL example for Linux 32bit:
669 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
671 To cross compile an OpenCL example for Linux 64bit:
673 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
675 To cross compile a GLES example for Linux 32bit:
677 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
679 To cross compile a GLES example for Linux 64bit:
681 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
683 (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)
685 To cross compile the examples with the Graph API, such as graph_lenet.cpp, you need to link the examples against arm_compute_graph.so too.
687 @note The compute library must currently be built with both neon and opencl enabled - neon=1 and opencl=1
689 i.e. to cross compile the "graph_lenet" example for Linux 32bit:
691 arm-linux-gnueabihf-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
693 i.e. to cross compile the "graph_lenet" example for Linux 64bit:
695 aarch64-linux-gnu-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp -I. -Iinclude -std=c++11 -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
697 (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)
699 @note If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, arm_compute, arm_compute_core
701 To compile natively (i.e directly on an ARM device) for NEON for Linux 32bit:
703 g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -larm_compute -larm_compute_core -o neon_convolution
705 To compile natively (i.e directly on an ARM device) for NEON for Linux 64bit:
707 g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o neon_convolution
709 (notice the only difference with the 32 bit command is that we don't need the -mfpu option)
711 To compile natively (i.e directly on an ARM device) for OpenCL for Linux 32bit or Linux 64bit:
713 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
715 To compile natively (i.e directly on an ARM device) for GLES for Linux 32bit or Linux 64bit:
717 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
719 To compile natively the examples with the Graph API, such as graph_lenet.cpp, you need to link the examples against arm_compute_graph.so too.
720 @note The compute library must currently be built with both neon and opencl enabled - neon=1 and opencl=1
722 i.e. to natively compile the "graph_lenet" example for Linux 32bit:
724 g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
726 i.e. to natively compile the "graph_lenet" example for Linux 64bit:
728 g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp -I. -Iinclude -std=c++11 L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet
730 (notice the only difference with the 32 bit command is that we don't need the -mfpu option)
732 @note If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, arm_compute, arm_compute_core
734 @note These two commands assume libarm_compute.so is available in your library path, if not add the path to it using -L
736 To run the built executable simply run:
738 LD_LIBRARY_PATH=build ./neon_convolution
742 LD_LIBRARY_PATH=build ./cl_convolution
744 @subsection S3_3_android Building for Android
746 For Android, the library was successfully built and tested using Google's standalone toolchains:
747 - NDK r14 arm-linux-androideabi-4.9 for armv7a (clang++)
748 - NDK r14 aarch64-linux-android-4.9 for arm64-v8a (clang++)
750 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>
752 - Download the NDK r14 from here: https://developer.android.com/ndk/downloads/index.html
753 - Make sure you have Python 2 installed on your machine.
754 - Generate the 32 and/or 64 toolchains by running the following commands:
757 $NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-4.9 --stl gnustl --api 21
758 $NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-androideabi-4.9 --stl gnustl --api 21
760 @attention Due to some NDK issues make sure you use clang++ & gnustl
762 @note Make sure to add the toolchains to your PATH: export PATH=$PATH:$MY_TOOLCHAINS/aarch64-linux-android-4.9/bin:$MY_TOOLCHAINS/arm-linux-androideabi-4.9/bin
764 @subsubsection S3_3_1_library How to build the library ?
766 @note 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)
768 To cross-compile the library in debug mode, with NEON only support, for Android 32bit:
770 CXX=clang++ CC=clang scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=android arch=armv7a
772 To cross-compile the library in asserts mode, with OpenCL only support, for Android 64bit:
774 CXX=clang++ CC=clang scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=android arch=arm64-v8a
776 To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Android 64bit:
778 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
780 @subsubsection S3_3_2_examples How to manually build the examples ?
782 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.
784 @note The following command lines assume the arm_compute 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.
786 Once you've got your Android standalone toolchain built and added to your path you can do the following:
788 To cross compile a NEON example:
791 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
793 aarch64-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
795 To cross compile an OpenCL example:
798 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
800 aarch64-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 -lOpenCL -DARM_COMPUTE_CL
802 To cross compile a GLES example:
805 arm-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
807 aarch64-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
809 To cross compile the examples with the Graph API, such as graph_lenet.cpp, you need to link the library arm_compute_graph also.
810 (notice the compute library has to be built with both neon and opencl enabled - neon=1 and opencl=1)
813 arm-linux-androideabi-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.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 -lOpenCL -DARM_COMPUTE_CL
815 aarch64-linux-android-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.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 -lOpenCL -DARM_COMPUTE_CL
817 @note Due to some issues in older versions of the Mali OpenCL DDK (<= r13p0), we recommend to link arm_compute statically on Android.
818 @note When linked statically the arm_compute_graph library currently needs the --whole-archive linker flag in order to work properly
820 Then you need to do is upload the executable and the shared library to the device using ADB:
822 adb push neon_convolution_arm /data/local/tmp/
823 adb push cl_convolution_arm /data/local/tmp/
824 adb push gc_absdiff_arm /data/local/tmp/
825 adb shell chmod 777 -R /data/local/tmp/
827 And finally to run the example:
829 adb shell /data/local/tmp/neon_convolution_arm
830 adb shell /data/local/tmp/cl_convolution_arm
831 adb shell /data/local/tmp/gc_absdiff_arm
835 adb push neon_convolution_aarch64 /data/local/tmp/
836 adb push cl_convolution_aarch64 /data/local/tmp/
837 adb push gc_absdiff_aarch64 /data/local/tmp/
838 adb shell chmod 777 -R /data/local/tmp/
840 And finally to run the example:
842 adb shell /data/local/tmp/neon_convolution_aarch64
843 adb shell /data/local/tmp/cl_convolution_aarch64
844 adb shell /data/local/tmp/gc_absdiff_aarch64
846 @subsection S3_4_bare_metal Building for bare metal
848 For bare metal, the library was successfully built using linaros's latest (gcc-linaro-6.3.1-2017.05) bare metal toolchains:
849 - arm-eabi for armv7a
850 - aarch64-elf for arm64-v8a
852 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>.
854 @note 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
856 @subsubsection S3_4_1_library How to build the library ?
858 To cross-compile the library with NEON support for baremetal arm64-v8a:
860 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
862 @subsubsection S3_4_2_examples How to manually build the examples ?
864 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>.
866 @subsection S3_5_windows_host Building on a Windows host system
868 Using `scons` directly from the Windows command line is known to cause
869 problems. The reason seems to be that if `scons` is setup for cross-compilation
870 it gets confused about Windows style paths (using backslashes). Thus it is
871 recommended to follow one of the options outlined below.
873 @subsubsection S3_5_1_ubuntu_on_windows Bash on Ubuntu on Windows
875 The best and easiest option is to use
876 <a href="https://msdn.microsoft.com/en-gb/commandline/wsl/about">Ubuntu on Windows</a>.
877 This feature is still marked as *beta* and thus might not be available.
878 However, if it is building the library is as simple as opening a *Bash on
879 Ubuntu on Windows* shell and following the general guidelines given above.
881 @subsubsection S3_5_2_cygwin Cygwin
883 If the Windows subsystem for Linux is not available <a href="https://www.cygwin.com/">Cygwin</a>
884 can be used to install and run `scons`. In addition to the default packages
885 installed by Cygwin `scons` has to be selected in the installer. (`git` might
886 also be useful but is not strictly required if you already have got the source
887 code of the library.) Linaro provides pre-built versions of
888 <a href="http://releases.linaro.org/components/toolchain/binaries/">GCC cross-compilers</a>
889 that can be used from the Cygwin terminal. When building for Android the
890 compiler is included in the Android standalone toolchain. After everything has
891 been set up in the Cygwin terminal the general guide on building the library
894 @subsection S3_6_cl_stub_library The OpenCL stub library
896 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 arm_compute against.
898 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.
900 @warning This OpenCL library provided is a stub and *not* a real implementation. You can use it to resolve OpenCL's symbols in arm_compute 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.
902 To cross-compile the stub OpenCL library simply run:
904 <target-prefix>-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
909 arm-linux-gnueabihf-gcc -o libOpenCL.so -Iinclude opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
911 aarch64-linux-gnu-gcc -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC
913 arm-linux-androideabi-clang -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
915 aarch64-linux-android-clang -o libOpenCL.so -Iinclude -shared opencl-1.2-stubs/opencl_stubs.c -fPIC -shared
917 @subsection S3_7_gles_stub_library The Linux OpenGLES and EGL stub libraries
919 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 arm_compute against.
921 @note The stub libraries are only needed on Linux. For Android, the NDK toolchains already provide the meta-EGL and meta-GLES libraries.
923 To cross-compile the stub OpenGLES and EGL libraries simply run:
925 <target-prefix>-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
926 <target-prefix>-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared
929 arm-linux-gnueabihf-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
930 arm-linux-gnueabihf-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared
933 aarch64-linux-gnu-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared
934 aarch64-linux-gnu-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared