/// /// Copyright (c) 2017-2020 Arm Limited. /// /// SPDX-License-Identifier: MIT /// /// Permission is hereby granted, free of charge, to any person obtaining a copy /// of this software and associated documentation files (the "Software"), to /// deal in the Software without restriction, including without limitation the /// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or /// sell copies of the Software, and to permit persons to whom the Software is /// furnished to do so, subject to the following conditions: /// /// The above copyright notice and this permission notice shall be included in all /// copies or substantial portions of the Software. /// /// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR /// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, /// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE /// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER /// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, /// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE /// SOFTWARE. /// namespace arm_compute { /** @mainpage Introduction @tableofcontents 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: - OS: Linux, Android or bare metal. - Architecture: armv7a (32bit) or arm64-v8a (64bit) - Technology: NEON / OpenCL / GLES_COMPUTE / NEON and OpenCL and GLES_COMPUTE - 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. @section S0_1_contact Contact / Support Please email developer@arm.com 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: $ strings android-armv7a-cl-asserts/libarm_compute.so | grep arm_compute_version 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 @section S0_2_prebuilt_binaries Pre-built binaries For each release we provide some pre-built binaries of the library [here](https://github.com/ARM-software/ComputeLibrary/releases) These binaries have been built using the following toolchains: - Linux armv7a: gcc-linaro-6.3.1-2017.05-x86_64_arm-linux-gnueabihf - Linux arm64-v8a: gcc-linaro-6.3.1-2017.05-x86_64_aarch64-linux-gnu - Android armv7a: clang++ / libc++ NDK r18b - Android am64-v8a: clang++ / libc++ NDK r18b @warning Make sure to use a compatible toolchain to build your application or you will get some std::bad_alloc errors at runtime. @section S1_file_organisation File organisation This archive contains: - The arm_compute header and source files - The latest Khronos OpenCL 1.2 C headers from the Khronos OpenCL registry - The latest Khronos cl2.hpp from the Khronos OpenCL registry (API version 2.1 when this document was written) - The latest Khronos OpenGL ES 3.1 C headers from the Khronos OpenGL ES registry - The latest Khronos EGL 1.5 C headers from the Khronos EGL registry - The sources for a stub version of libOpenCL.so, libGLESv1_CM.so, libGLESv2.so and libEGL.so to help you build your application. - An examples folder containing a few examples to compile and link against the library. - A @ref utils folder containing headers with some boiler plate code used by the examples. - This documentation. For detailed information about file organization, please refer to Files -> File List section of this documentation. @section S2_versions_changelog Release versions and changelog @subsection S2_1_versions Release versions 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: v17.03 (First release of March 2017) v17.03.1 (Second release of March 2017) v17.04 (First release of April 2017) @note We're aiming at releasing one major public release with new features per quarter. All releases in between will only contain bug fixes. @subsection S2_2_changelog Changelog v20.11 Public major release - Various bug fixes. - Various optimisations. - Performance regressions can be noted when executing Depthwise Convolution on Neon with a depth multiplier > 1 for quantized data type. This is planned to be resolved in 21.02 release. - Added new data type QASYMM8_SIGNED support for @ref NEROIAlignLayer. - Added new data type S32 support for: - @ref NEArithmeticSubtraction - @ref NEArithmeticSubtractionKernel - @ref NEPixelWiseMultiplication - @ref NEPixelWiseMultiplicationKernel - @ref NEElementwiseDivision - @ref NEDivisionOperationKernel - Interface change - Properly support softmax axis to have the same meaning as other major frameworks. That is, axis now defines the dimension on which Softmax/Logsoftmax is performed. E.g. for input of shape 4x5x6 and axis=1, softmax will be applied to 4x6=24 vectors of size 5. The supported value range of axis is [-rank, rank). This change applies to the following functions: - @ref NESoftmaxLayer - @ref NELogSoftmaxLayer - @ref CLSoftmaxLayer - @ref CLLogSoftmaxLayer - @ref GCSoftmaxLayer - New OpenCL kernels / functions: - @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel - @ref CLLogicalNot - @ref CLLogicalAnd - @ref CLLogicalOr - New NEON kernels / functions: - @ref NELogicalNot - @ref NELogicalAnd - @ref NELogicalOr - Removed padding from NEON kernels: - @ref NEComplexPixelWiseMultiplicationKernel - @ref NENonMaximaSuppression3x3Kernel - @ref NERemapKernel - @ref NEGEMMInterleave4x4Kernel - @ref NEDirectConvolutionLayerKernel - @ref NEScaleKernel - @ref NELocallyConnectedMatrixMultiplyKernel - @ref NEGEMMLowpOffsetContributionKernel - @ref NEGEMMTranspose1xWKernel - @ref NEPoolingLayerKernel - @ref NEConvolutionKernel - @ref NEDepthwiseConvolutionLayerNativeKernel - @ref NEGEMMLowpMatrixMultiplyKernel - @ref NEGEMMMatrixMultiplyKernel - @ref NEDirectConvolutionLayerOutputStageKernel - @ref NEReductionOperationKernel - @ref NEGEMMLowpMatrixAReductionKernel - @ref NEGEMMLowpMatrixBReductionKernel - Removed padding from OpenCL kernels: - @ref CLBatchConcatenateLayerKernel - @ref CLElementwiseOperationKernel - @ref CLBatchNormalizationLayerKernel - @ref CLPoolingLayerKernel - @ref CLWinogradInputTransformKernel - @ref CLGEMMLowpMatrixMultiplyNativeKernel - @ref CLGEMMLowpMatrixAReductionKernel - @ref CLGEMMLowpMatrixBReductionKernel - @ref CLGEMMLowpOffsetContributionOutputStageKernel - @ref CLGEMMLowpOffsetContributionKernel - @ref CLWinogradOutputTransformKernel - @ref CLGEMMLowpMatrixMultiplyReshapedKernel - @ref CLFuseBatchNormalizationKernel - @ref CLDepthwiseConvolutionLayerNativeKernel - @ref CLDepthConvertLayerKernel - @ref CLCopyKernel - @ref CLDepthwiseConvolutionLayer3x3NHWCKernel - @ref CLActivationLayerKernel - @ref CLWinogradFilterTransformKernel - @ref CLWidthConcatenateLayerKernel - @ref CLWidthConcatenate4TensorsKernel - @ref CLWidthConcatenate2TensorsKernel - @ref CLLogits1DMaxShiftExpSumKernel - @ref CLLogits1DNormKernel - @ref CLHeightConcatenateLayerKernel - @ref CLGEMMMatrixMultiplyKernel - @ref CLGEMMLowpQuantizeDownInt32ScaleKernel - @ref CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel - @ref CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel - @ref CLDepthConcatenateLayerKernel - @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel - Removed OpenCL kernels / functions: - CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel - CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel - CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel - Deprecated OpenCL kernels / functions (If a kernel is used only by the function that is being deprecated, the kernel is deprecated together): - CLLocallyConnectedLayer - CLLocallyConnectedMatrixMultiplyKernel - CLAbsoluteDifference - CLAbsoluteDifferenceKernel - CLAccumulate - CLAccumulateKernel - CLAccumulateSquared - CLAccumulateSquaredKernel - CLAccumulateWeighted - CLAccumulateWeightedKernel - CLAccumulateWeightedFP16Kernel - CLBox3x3 - CLBox3x3Kernel - CLBox3x3FP16Kernel - CLCannyEdge - CLChannelCombine - CLChannelCombineKernel - CLChannelExtract - CLChannelExtractKernel - CLColorConvert - CLColorConvertKernel - CLConvolution3x3 - CLConvolutionRectangle - CLConvolutionRectangleKernel - CLConvolutionSquare - CLConvolutionKernel - CLDerivative - CLDerivativeKernel - CLDilate - CLDilateKernel - CLEqualizeHistogram - CLErode - CLErodeKernel - CLFastCorners - CLFastCornersKernel - CLGaussian3x3 - CLGaussian3x3Kernel - CLGaussian5x5 - CLGaussian5x5HorKernel - CLGaussian5x5VertKernel - CLGaussianPyramid - CLGaussianPyramidHalf - CLGaussianPyramidOrb - CLHarrisCorners - CLHarrisScoreKernel - CLHarrisScoreFP16Kernel - CLHistogram - CLHistogramKernel - CLHOGOrientationBinningKernel - CLHOGBlockNormalizationKernel - CLHOGDetectorKernel - CLHOGNonMaximaSuppressionKernel - CLHOGDescriptor - CLHOGDetector - CLHOGGradient - CLHOGMultiDetection - CLHOGOrientationBinningKernel - CLHOGBlockNormalizationKernel - CLHOGDetectorKernel - CLIntegralImage - CLIntegralImageKernel - CLLaplacianReconstruct - CLLaplacianPyramid - CLMagnitude - CLMagnitudePhaseKernel - CLMedian3x3 - CLMedian3x3Kernel - CLMinMaxLocation - CLMinMaxLocationKernel - CLNonLinearFilter - CLNonLinearFilterKernel - CLNonMaximaSuppression3x3 - CLNonMaximaSuppression3x3FP16Kernel - CLNonMaximaSuppression3x3Kernel - CLOpticalFlow - CLPhase - CLRemap - CLRemapKernel - CLScharr3x3 - CLScharr3x3Kernel - CLSobel3x3 - CLSobel3x3Kernel - CLSobel5x5 - CLSobel5x5HorKernel - CLSobel5x5VertKernel - CLSobel7x7 - CLSobel7x7HorKernel - CLSobel7x7VertKernel - CLThreshold - CLThresholdKernel - CLWarpAffine - CLWarpAffineKernel - CLWarpPerspective - CLWarpPerspectiveKernel - Deprecated NEON kernels / functions (If a kernel is used only by the function that is being deprecated, the kernel is deprecated together): - NELocallyConnectedLayer - NELocallyConnectedMatrixMultiplyKernel - NEAbsoluteDifference - NEAbsoluteDifferenceKernel - NEAccumulate - NEAccumulateKernel - NEAccumulateSquared - NEAccumulateSquaredKernel - NEAccumulateWeighted - NEAccumulateWeightedKernel - NEAccumulateWeightedFP16Kernel - NEBox3x3 - NEBox3x3Kernel - NEBox3x3FP16Kernel - NECannyEdge - NEChannelCombine - NEChannelCombineKernel - NEChannelExtract - NEChannelExtractKernel - NEColorConvert - NEColorConvertKernel - NEConvolution3x3 - NEConvolutionRectangle - NEConvolutionRectangleKernel - NEConvolutionSquare - NEConvolutionKernel - NEDerivative - NEDerivativeKernel - NEDilate - NEDilateKernel - NEEqualizeHistogram - NEErode - NEErodeKernel - NEFastCorners - NEFastCornersKernel - NEGaussian3x3 - NEGaussian3x3Kernel - NEGaussian5x5 - NEGaussian5x5HorKernel - NEGaussian5x5VertKernel - NEGaussianPyramid - NEGaussianPyramidHalf - NEGaussianPyramidOrb - NEHarrisCorners - NEHarrisScoreKernel - NEHarrisScoreFP16Kernel - NEHistogram - NEHistogramKernel - NEHOGOrientationBinningKernel - NEHOGBlockNormalizationKernel - NEHOGDetectorKernel - NEHOGNonMaximaSuppressionKernel - NEHOGDescriptor - NEHOGDetector - NEHOGGradient - NEHOGMultiDetection - NEHOGOrientationBinningKernel - NEHOGBlockNormalizationKernel - NEHOGDetectorKernel - NEIntegralImage - NEIntegralImageKernel - NELaplacianReconstruct - NELaplacianPyramid - NEMagnitude - NEMagnitudePhaseKernel - NEMedian3x3 - NEMedian3x3Kernel - NEMinMaxLocation - NEMinMaxLocationKernel - NENonLinearFilter - NENonLinearFilterKernel - NENonMaximaSuppression3x3 - NENonMaximaSuppression3x3FP16Kernel - NENonMaximaSuppression3x3Kernel - NEOpticalFlow - NEPhase - NERemap - NERemapKernel - NEScharr3x3 - NEScharr3x3Kernel - NESobel3x3 - NESobel3x3Kernel - NESobel5x5 - NESobel5x5HorKernel - NESobel5x5VertKernel - NESobel7x7 - NESobel7x7HorKernel - NESobel7x7VertKernel - NEThreshold - NEThresholdKernel - NEWarpAffine - NEWarpAffineKernel - NEWarpPerspective - NEWarpPerspectiveKernel - Deprecated GLES kernels / functions (If a kernel is used only by the function that is being deprecated, the kernel is deprecated together): - GCAbsoluteDifference - GCActivationLayer - GCArithmeticAddition - GCBatchNormalizationLayer - GCConcatenateLayer - GCConvolutionLayer - GCDepthwiseConvolutionLayer - GCDirectConvolutionLayer - GCDropoutLayer - GCFillBorder - GCFullyConnectedLayer - GCGEMM - GCGEMMInterleave4x4 - GCGEMMTranspose1xW - GCNormalizationLayer - GCNormalizePlanarYUVLayer - GCPixelWiseMultiplication - GCPoolingLayer - GCScale - GCSoftmaxLayer - GCTensorShift - GCTranspose v20.08 Public major release - Various bug fixes. - Various optimisations. - Added new data type QASYMM8_SIGNED support for: - @ref CLArgMinMaxLayer - @ref CLArgMinMaxLayerKernel - Added new data type U8 support for: - @ref NECropKernel - @ref CLCropKernel - Added aligh_corner support for nearest neighbor interpolation in: - @ref NEScaleKernel - @ref CLScaleKernel - New OpenCL kernels / functions: - @ref CLMaxUnpoolingLayerKernel - New NEON kernels / functions: - @ref NEMaxUnpoolingLayerKernel - New graph example: - graph_yolov3_output_detector - GEMMTuner improvements: - Added fp16 support - Output json files for easier integration - Enabled tuning for export_to_cl_image_rhs option for RHS tensors - More robust script for running benchmarks - Removed padding from: - @ref NEPixelWiseMultiplicationKernel - @ref NEHeightConcatenateLayerKernel - @ref NEThresholdKernel - @ref NEBatchConcatenateLayerKernel - @ref NETransposeKernel - @ref NEBatchNormalizationLayerKernel - @ref NEArithmeticSubtractionKernel - @ref NEBoundingBoxTransformKernel - @ref NELogits1DMaxKernel - @ref NELogits1DSoftmaxKernel - @ref NEROIPoolingLayerKernel - @ref NEROIAlignLayerKernel - @ref NEYOLOLayerKernel - @ref NEUpsampleLayerKernel - @ref NEFloorKernel - @ref NEWidthConcatenateLayerKernel - @ref NEDepthConcatenateLayerKernel - @ref NENormalizationLayerKernel - @ref NEL2NormalizeLayerKernel - @ref NEFillArrayKernel - @ref NEDepthConvertLayerKernel - @ref NERangeKernel - @ref NEPriorBoxLayer - Removed OpenCL kernels / functions: - CLGEMMLowpQuantizeDownInt32ToUint8Scale - CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat - Removed NEON kernels / functions: - NEGEMMLowpQuantizeDownInt32ToUint8Scale - NEGEMMMatrixAccumulateBiasesKernel - Deprecated functions / interfaces: - Non-descriptor based interfaces for @ref NEThreshold, @ref CLThreshold - Non-descriptor based interfaces for @ref NEScale, @ref CLScale and @ref GCScale - In @ref NESoftmaxLayer, @ref NELogSoftmaxLayer, @ref CLSoftmaxLayer, @ref CLLogSoftmaxLayer and @ref GCSoftmaxLayer : The default "axis" value for @ref CLSoftmaxLayer, @ref CLLogSoftmaxLayer and @ref GCSoftmaxLayer is changed from 1 to 0. Only axis 0 is supported. The default "axis" value for @ref NESoftmaxLayer, @ref NELogSoftmaxLayer is changed from 1 to 0. Only axis 0 is supported. - The support for quantized data types has been removed from @ref CLLogSoftmaxLayer due to implementation complexity. - Removed padding requirement for the input (e.g. LHS of GEMM) and output in @ref CLGEMMMatrixMultiplyNativeKernel, @ref CLGEMMMatrixMultiplyReshapedKernel, @ref CLGEMMMatrixMultiplyReshapedOnlyRHSKernel and @ref CLIm2ColKernel (NHWC only) - This change allows to use @ref CLGEMMConvolutionLayer without extra padding for the input and output. - Only the weights/bias of @ref CLGEMMConvolutionLayer could require padding for the computation. - Only on Arm Mali Midgard GPUs, @ref CLGEMMConvolutionLayer could require padding since @ref CLGEMMMatrixMultiplyKernel is called and currently requires padding. - Added support for exporting the OpenCL buffer object to the OpenCL image object in @ref CLGEMMMatrixMultiplyReshapedKernel and @ref CLGEMMMatrixMultiplyReshapedOnlyRHSKernel. - This support allows to export the OpenCL buffer used for the reshaped RHS matrix to the OpenCL image object. - The padding requirement for the OpenCL image object is considered into the @ref CLGEMMReshapeRHSMatrixKernel. - The reshaped RHS matrix stores the weights when GEMM is used to accelerate @ref CLGEMMConvolutionLayer. v20.05 Public major release - Various bug fixes. - Various optimisations. - Updated recommended NDK version to r18b. - Updated recommended gcc version to Linaro 6.3.1. - Added Bfloat16 type support - Added Bfloat16 support in: - @ref NEWeightsReshapeKernel - @ref NEConvolutionLayerReshapeWeights - @ref NEIm2ColKernel - @ref NEIm2Col - @ref NEDepthConvertLayerKernel - @ref NEDepthConvertLayer - @ref NEGEMMConvolutionLayer - @ref NEGEMMAssemblyDispatch - Added new data type QASYMM8_SIGNED support for: - @ref CLDirectConvolutionLayer - @ref CLDeconvolutionLayer - @ref CLDirectDeconvolutionLayer - @ref CLGEMMDeconvolutionLayer - @ref CLGEMMLowpMatrixMultiplyReshapedKernel - @ref CLGEMMLowpQuantizeDownInt32ScaleKernel - @ref CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel - @ref CLReductionOperation - @ref CLReduceMean - @ref NEScale - @ref NEScaleKernel - @ref NEUpsampleLayer - @ref NECast - @ref NEReductionOperation - @ref NEReduceMean - @ref NEArgMinMaxLayer - @ref NEDeconvolutionLayer - @ref NEGEMMLowpQuantizeDownInt32ScaleKernel - @ref CPPBoxWithNonMaximaSuppressionLimit - @ref CPPDetectionPostProcessLayer - @ref CPPPermuteKernel - @ref CPPPermute - @ref CPPTopKVKernel - @ref CPPTopKV - @ref CPPUpsample - @ref CPPUpsampleKernel - New OpenCL kernels / functions: - @ref CLQLSTMLayer - @ref CLQLSTMLayerNormalizationKernel - New NEON kernels / functions: - @ref NEQLSTMLayer - @ref NEQLSTMLayerNormalizationKernel - Added HARD_SWISH support in: - @ref CLActivationLayerKernel - @ref NEActivationLayerKernel - Deprecated OpenCL kernels / functions: - CLGEMMLowpQuantizeDownInt32ToUint8Scale - CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat - Deprecated NEON kernels / functions: - NEGEMMLowpQuantizeDownInt32ToUint8Scale - Removed CPP kernels / functions: - CPPFlipWeightsKernel - Removed PoolingLayerInfo constructors without Data Layout. - Removed CLDepthwiseConvolutionLayer3x3 - Removed NEDepthwiseConvolutionLayerOptimized - Added support for Winograd 3x3,4x4 on NEON FP16: - @ref NEWinogradConvolutionLayer - @ref NEWinogradLayerTransformInputKernel - @ref NEWinogradLayerTransformOutputKernel - @ref NEWinogradLayerTransformWeightsKernel - Added CLCompileContext - Added NEON GEMM kernel with 2D window support v20.02.1 Maintenance release - Added Android-NN build script. v20.02 Public major release - Various bug fixes. - Various optimisations. - Added new data type QASYMM8_SIGNED support for: - @ref CLDepthwiseConvolutionLayer - CLDepthwiseConvolutionLayer3x3 - @ref CLGEMMConvolutionLayer - @ref CLGEMMLowpMatrixMultiplyCore - @ref CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel - @ref CLGEMMLowpMatrixMultiplyNativeKernel - @ref NEActivationLayer - @ref NEComparisonOperationKernel - @ref NEConvolutionLayer - @ref NEDepthwiseConvolutionLayer - NEDepthwiseConvolutionLayer3x3Kernel - @ref NEDirectConvolutionLayerOutputStageKernel - @ref NEElementwiseComparison - @ref NEElementwiseMax - @ref NEElementwiseMin - @ref NEElementwiseSquaredDiff - @ref NEFullyConnectedLayer - NEGEMMMatrixVectorMultiplyKernel - @ref NEPixelWiseMultiplication - @ref NEPoolingLayer - @ref NEPReluLayer - Added support for QSYMM8_PER_CHANNEL in: - NEDepthwiseConvolutionLayer3x3Kernel - Added support for split sizes in: - @ref CLSplit - @ref NESplit - New OpenCL kernels / functions: - @ref CLFill - CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel / @ref CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint - New NEON kernels / functions: - @ref NEFill - @ref NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel / @ref NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint - Deprecated NEON functions / interfaces: - CLDepthwiseConvolutionLayer3x3 - NEDepthwiseConvolutionLayerOptimized - PoolingLayerInfo constructors without Data Layout. - Added support for quantization with multiplier greater than 1 on NEON and CL. - Added support for quantized inputs of type QASYMM8_SIGNED and QASYMM8 to @ref CLQuantizationLayer. - Added the ability to build bootcode for bare metal. - Added support for generating synthetic QASYMM8 graphs. - Added support for F16 datatype in VGG16. - Removed pre-built binaries for GLES. v19.11.1 Public maintenance release - Fix offset calculation in NEReductionOperationKernel. - Fix data layout in NEScaleKernel for nhwc. - Retain configuration step data layout to avoid side-effects. - Perform sqrt in double domain for L2 pooling. - Fix output shape calculation for Reduce Mean - Restrict cases where optimized NEPadLayer runs. v19.11 Public major release - Various bug fixes. - Various optimisations. - Updated recommended NDK version to r17c. - Deprecated OpenCL kernels / functions: - CLDepthwiseConvolutionLayerReshapeWeightsGenericKernel - CLDepthwiseIm2ColKernel - CLDepthwiseSeparableConvolutionLayer - CLDepthwiseVectorToTensorKernel - CLDirectConvolutionLayerOutputStageKernel - Deprecated NEON kernels / functions: - NEDepthwiseWeightsReshapeKernel - NEDepthwiseIm2ColKernel - NEDepthwiseSeparableConvolutionLayer - NEDepthwiseVectorToTensorKernel - NEDepthwiseConvolutionLayer3x3 - New OpenCL kernels / functions: - @ref CLInstanceNormalizationLayerKernel / @ref CLInstanceNormalizationLayer - @ref CLDepthwiseConvolutionLayerNativeKernel to replace the old generic depthwise convolution (see Deprecated OpenCL kernels / functions) - @ref CLLogSoftmaxLayer - New NEON kernels / functions: - @ref NEBoundingBoxTransformKernel / @ref NEBoundingBoxTransform - @ref NEComputeAllAnchorsKernel / @ref NEComputeAllAnchors - @ref NEDetectionPostProcessLayer - @ref NEGenerateProposalsLayer - @ref NEInstanceNormalizationLayerKernel / @ref NEInstanceNormalizationLayer - @ref NELogSoftmaxLayer - @ref NEROIAlignLayerKernel / @ref NEROIAlignLayer - Added QASYMM8 support for: - @ref CLGenerateProposalsLayer - @ref CLROIAlignLayer - @ref CPPBoxWithNonMaximaSuppressionLimit - Added QASYMM16 support for: - @ref CLBoundingBoxTransform - Added FP16 support for: - @ref CLGEMMMatrixMultiplyReshapedKernel - Added new data type QASYMM8_PER_CHANNEL support for: - @ref CLDequantizationLayer - @ref NEDequantizationLayer - Added new data type QSYMM8_PER_CHANNEL support for: - @ref CLConvolutionLayer - @ref NEConvolutionLayer - @ref CLDepthwiseConvolutionLayer - @ref NEDepthwiseConvolutionLayer - Added FP16 mixed-precision support for: - @ref CLGEMMMatrixMultiplyReshapedKernel - @ref CLPoolingLayerKernel - Added FP32 and FP16 ELU activation for: - @ref CLActivationLayer - @ref NEActivationLayer - Added asymmetric padding support for: - @ref CLDirectDeconvolutionLayer - @ref CLGEMMDeconvolutionLayer - @ref NEDeconvolutionLayer - Added SYMMETRIC and REFLECT modes for @ref CLPadLayerKernel / @ref CLPadLayer. - Replaced the calls to @ref NECopyKernel and @ref NEMemsetKernel with @ref NEPadLayer in @ref NEGenerateProposalsLayer. - Replaced the calls to @ref CLCopyKernel and @ref CLMemsetKernel with @ref CLPadLayer in @ref CLGenerateProposalsLayer. - Improved performance for CL Inception V3 - FP16. - Improved accuracy for CL Inception V3 - FP16 by enabling FP32 accumulator (mixed-precision). - Improved NEON performance by enabling fusing batch normalization with convolution and depth-wise convolution layer. - Improved NEON performance for MobileNet-SSD by improving the output detection performance. - Optimized @ref CLPadLayer. - Optimized CL generic depthwise convolution layer by introducing @ref CLDepthwiseConvolutionLayerNativeKernel. - Reduced memory consumption by implementing weights sharing. v19.08.1 Public maintenance release - Fix offset calculation in NEReductionOperationKernel. - Fix data layout in NEScaleKernel for nhwc. - Retain configuration step data layout to avoid side-effects. - Perform sqrt in double domain for L2 pooling. - Fix output shape calculation for Reduce Mean - Fix broadcast CLPixelwiseMultiplication with 5D tensors v19.08 Public major release - Various bug fixes. - Various optimisations. - Deprecated NEON functions - NEDepthConcatenateLayer - NEWidthConcatenateLayer - Deprecated OpenCL kernels / functions - CLDepthConcatenateLayer - CLGEMMInterleave4x4Kernel / CLGEMMInterleave4x4 - CLGEMMTranspose1xWKernel / CLGEMMTranspose1xW - CLWidthConcatenateLayer - New NEON kernels / functions: - @ref NEAbsLayer - @ref NECast - @ref NEElementwisePower - @ref NELogLayer - @ref NELSTMLayerQuantized - @ref NENegLayer - @ref NEPReluLayer - @ref NESinLayer - @ref NEBatchConcatenateLayerKernel - @ref NEDepthToSpaceLayerKernel / @ref NEDepthToSpaceLayer - @ref NEDepthwiseConvolutionLayerNativeKernel - @ref NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel - @ref NEMeanStdDevNormalizationKernel / @ref NEMeanStdDevNormalizationLayer - @ref NESpaceToDepthLayerKernel / @ref NESpaceToDepthLayer - New OpenCL kernels / functions: - @ref CLAbsLayer - @ref CLElementwisePower - @ref CLLogLayer - @ref CLLSTMLayerQuantized - @ref CLNegLayer - @ref CLPReluLayer - @ref CLSinLayer - @ref CLBatchConcatenateLayerKernel - @ref CLDepthToSpaceLayerKernel / @ref CLDepthToSpaceLayer - @ref CLGEMMLowpMatrixMultiplyNativeKernel - CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel - @ref CLGEMMMatrixMultiplyNativeKernel - @ref CLMeanStdDevNormalizationKernel / @ref CLMeanStdDevNormalizationLayer - @ref CLSpaceToDepthLayerKernel / @ref CLSpaceToDepthLayer - New examples: - neon_opticalflow - cl_cache - neon_permute - Added support for FP16 in @ref NEDeconvolutionLayer - Added support for FP16 in @ref CLDeconvolutionLayer - Added support for REDUCE_MIN and REDUCE_MAX in @ref ReductionOperation - Enable the fusion of batch normalization with convolution and depthwise convolution layer for FP32 in the graph API (OpenCL only) - Added support for fusing activation function and broadcast addition with the matrix multiplication for FP32 (OpenCL only) - Re-factored the depthwise convolution layer kernel on NEON for generic cases - Added an optimized depthwise convolution layer kernel for 5x5 filters (NEON only) - Added support to enable OpenCL kernel cache. Added example showing how to load the prebuilt OpenCL kernels from a binary cache file - Altered @ref QuantizationInfo interface to support per-channel quantization. - The CLDepthwiseConvolutionLayer3x3 will be included by @ref CLDepthwiseConvolutionLayer to accommodate for future optimizations. - The NEDepthwiseConvolutionLayerOptimized will be included by @ref NEDepthwiseConvolutionLayer to accommodate for future optimizations. - Removed inner_border_right and inner_border_top parameters from @ref CLDeconvolutionLayer interface - Removed inner_border_right and inner_border_top parameters from @ref NEDeconvolutionLayer interface - Optimized the NEON assembly kernel for GEMMLowp. The new implementation fuses the output stage and quantization with the matrix multiplication kernel v19.05 Public major release - Various bug fixes. - Various optimisations. - New Neon kernels / functions: - @ref NEBatchToSpaceLayerKernel / @ref NEBatchToSpaceLayer - @ref NEComplexPixelWiseMultiplicationKernel / @ref NEComplexPixelWiseMultiplication - @ref NECropKernel / @ref NECropResize - @ref NEDepthwiseConvolutionAssemblyDispatch - @ref NEFFTDigitReverseKernel - @ref NEFFTRadixStageKernel - @ref NEFFTScaleKernel - @ref NEGEMMLowpOffsetContributionOutputStageKernel - @ref NEHeightConcatenateLayerKernel - @ref NESpaceToBatchLayerKernel / @ref NESpaceToBatchLayer - @ref NEFFT1D - @ref NEFFT2D - @ref NEFFTConvolutionLayer - New OpenCL kernels / functions: - @ref CLComplexPixelWiseMultiplicationKernel / @ref CLComplexPixelWiseMultiplication - @ref CLCropKernel / @ref CLCropResize - @ref CLDeconvolutionReshapeOutputKernel - @ref CLFFTDigitReverseKernel - @ref CLFFTRadixStageKernel - @ref CLFFTScaleKernel - @ref CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel - @ref CLGEMMMatrixMultiplyReshapedOnlyRHSKernel - @ref CLHeightConcatenateLayerKernel - @ref CLDirectDeconvolutionLayer - @ref CLFFT1D - @ref CLFFT2D - @ref CLFFTConvolutionLayer - @ref CLGEMMDeconvolutionLayer - New OpenGLES kernels / functions: - @ref GCConcatenateLayer - Deprecated functions/interfaces - GCDepthConcatenateLayer - NEWidthConcatenateLayer - NEDepthConcatenateLayer - CLWidthConcatenateLayer - CLDepthConcatenateLayer - CLGEMMInterleave4x4 - CLGEMMTranspose1xW - Support different quantization info in CLConcatLayer. - Add checks on different input/output quantization info were not supported. - Tensors have different quantization information. - Add FP16 support checks. - Fix output quantization CLDeptwiseConv3x3 when activation is fused. - New graph examples: - graph_convolution - graph_fully_connected - graph_depthwise_convolution - Deepspeech v0.4.1 - Add support for QASYMM8 in NEArithmeticSubtractionKernel. - Add support for QASYMM8 in NEPixelWiseMultiplicationKernel. - Add support for QASYMM8 NEDeconvolution. - Add support for DequantizationLayer for NEON/CL. - Add support for dilation in CLDepthwiseConvolution. - Fuse offset contribution with the output stage when we use NEGEMMLowpMatrixMultiplyCore. - Optimize CLDeconvolution. - Add StackLayer to the graph API. - Add support for "reflect" padding mode in NEPad. - Winograd 7x7 NHWC on OpenCL. - Rework CL ML layers to run exclusively on CL. - Support different quantization info in PoolingLayer. - Implement and test import memory interfaces. - Added new tests and removed old ones. - Various clang-tidy fixes. v19.02 Public major release - Various bug fixes. - Various optimisations. - New Neon kernels / functions: - @ref NETileKernel / @ref NETile - @ref NEFuseBatchNormalizationKernel / @ref NEFuseBatchNormalization - @ref NEElementwiseOperationKernel - @ref NEElementwiseMax - @ref NEElementwiseMin - @ref NEElementwiseSquaredDiff - @ref NESelectKernel / @ref NESelect - @ref NESplit - @ref NESlice - @ref NEUnstack - @ref NEStridedSliceKernel / @ref NEStridedSlice - @ref NEElementwiseUnaryKernel - @ref NERsqrtLayer - @ref NEExpLayer - @ref NEReverseKernel / @ref NEReverse - @ref NEArgMinMaxLayer - @ref NEStackLayerKernel / @ref NEStackLayer - @ref NERangeKernel / @ref NERange - @ref NEPadLayer - @ref NEMemsetKernel - @ref NEGatherKernel / @ref NEGather - @ref NEElementwiseComparison - @ref NEElementwiseComparisonStatic - @ref NEComparisonOperationKernel - @ref NEElementwiseDivision - New OpenCL kernels / functions: - @ref CLSelectKernel / @ref CLSelect - @ref CLTileKernel / @ref CLTile - @ref CLComparisonKernel / @ref CLComparison - @ref CLArgMinMaxLayer - @ref CLElementwiseMax - @ref CLElementwiseMin - @ref CLElementwiseSquaredDiff - @ref CLStackLayerKernel / @ref CLStackLayer - @ref CLReverse / @ref CLReverseKernel - @ref CLRsqrtLayer - @ref CLExpLayer - @ref CLElementWiseUnaryLayerKernel - @ref CLGEMMReshapeLHSMatrixKernel - @ref CLGEMMReshapeRHSMatrixKernel - @ref CLGEMMMatrixMultiplyReshapedKernel - @ref CLRangeKernel / @ref CLRange - @ref CLUnstack - @ref CLGatherKernel / @ref CLGather - @ref CLGEMMLowpMatrixMultiplyReshapedKernel - New CPP kernels / functions: - @ref CPPDetectionOutputLayer - @ref CPPTopKV / @ref CPPTopKVKernel - Added new examples: - graph_ssd_mobilenet.cpp - graph_mobilenet_v2.cpp - graph_resnet12.cpp - graph_srcnn955.cpp - graph_vgg_vdsr.cpp - graph_inception_resnet_v1.cpp - Add 4D tensors support to - @ref NESoftmaxLayer - Fused activation in @ref CLWinogradConvolutionLayer - Extented @ref NEPermute to support more cases - Added NEON/SVE GEMM Hybrid kernels - Added u8 and s8 hybrid assembly kernels - Introduced GEMM strategy name in NEGEMMAssemblyWrapper - Improved @ref CLTuner - Fused the bias addition within @ref CLGEMM - Added support for QASYMM8 LOGISTIC activation in @ref NEActivationLayer - Added NHWC data layout support to: - @ref NEScale for F16 - @ref CLNormalizationLayer IN_MAP_2D for FP32/FP16 - @ref NEL2NormalizeLayer for FP32/FP16 - @ref NENormalizationLayer IN_MAP_2D for FP32/FP16 - @ref CLROIAlignLayer - @ref CLGenerateProposalsLayer - Added QASYMM8 support to the following kernels: - @ref NEArithmeticAdditionKernel - @ref NEScale - Added new tests and improved validation and benchmarking suites. - Deprecated functions/interfaces - Usage of inner_border_right and inner_border_top has been deprecated in @ref CLDeconvolutionLayer and @ref NEDeconvolutionLayer v18.11 Public major release - Various bug fixes. - Various optimisations. - New Neon kernels / functions: - @ref NEChannelShuffleLayer / @ref NEChannelShuffleLayerKernel - @ref NEReduceMean - @ref NEReorgLayer / @ref NEReorgLayerKernel - @ref NEPriorBoxLayer / @ref NEPriorBoxLayerKernel - @ref NEUpsampleLayer / @ref NEUpsampleLayerKernel - @ref NEYOLOLayer / @ref NEYOLOLayerKernel - New OpenCL kernels / functions: - @ref CLBatchToSpaceLayer / @ref CLBatchToSpaceLayerKernel - @ref CLBoundingBoxTransform / @ref CLBoundingBoxTransformKernel - @ref CLComputeAllAnchorsKernel - @ref CLGenerateProposalsLayer - @ref CLNormalizePlanarYUVLayer / @ref CLNormalizePlanarYUVLayerKernel - @ref CLReorgLayer / @ref CLReorgLayerKernel - @ref CLSpaceToBatchLayer / @ref CLSpaceToBatchLayerKernel - @ref CLPadLayer - @ref CLReduceMean - @ref CLPriorBoxLayer / @ref CLPriorBoxLayerKernel - @ref CLROIAlignLayer / @ref CLROIAlignLayerKernel - @ref CLSlice - @ref CLSplit - @ref CLStridedSlice / @ref CLStridedSliceKernel - @ref CLUpsampleLayer / @ref CLUpsampleLayerKernel - @ref CLYOLOLayer / @ref CLYOLOLayerKernel - New CPP kernels / functions: - @ref CPPBoxWithNonMaximaSuppressionLimit / @ref CPPBoxWithNonMaximaSuppressionLimitKernel - Added the validate method in: - @ref NEDepthConvertLayer - @ref NEFloor / @ref CLFloor - @ref NEGEMMMatrixAdditionKernel - @ref NEReshapeLayer / @ref CLReshapeLayer - @ref CLScale - Added new examples: - graph_shufflenet.cpp - graph_yolov3.cpp - Added documentation for add a new function or kernel. - Improved doxygen documentation adding a list of the existing functions. - Add 4D tensors support to - CLWidthConcatenateLayer - @ref CLFlattenLayer - @ref CLSoftmaxLayer - Add dot product support for @ref CLDepthwiseConvolutionLayer3x3NHWCKernel non-unit stride - Add SVE support - Fused batch normalization into convolution layer weights in @ref CLFuseBatchNormalization - Fuses activation in @ref CLDepthwiseConvolutionLayer3x3NCHWKernel, @ref CLDepthwiseConvolutionLayer3x3NHWCKernel and @ref NEGEMMConvolutionLayer - Added NHWC data layout support to: - @ref CLChannelShuffleLayer - @ref CLDeconvolutionLayer - @ref CLL2NormalizeLayer - Added QASYMM8 support to the following kernels: - @ref CLScaleKernel - NEDepthwiseConvolutionLayer3x3Kernel - @ref CLPixelWiseMultiplicationKernel - Added FP16 support to the following kernels: - @ref CLDepthwiseConvolutionLayer3x3NHWCKernel - NEDepthwiseConvolutionLayer3x3Kernel - @ref CLNormalizePlanarYUVLayerKernel - @ref CLWinogradConvolutionLayer (5x5 kernel) - More tests added to both validation and benchmarking suites. v18.08 Public major release - Various bug fixes. - Various optimisations. - Updated recommended NDK version to r17b. - Removed support for QS8/QS16 data types. - Added support for grouped convolution in @ref CLConvolutionLayer. - Added NHWC data layout support to: - NEDepthConcatenateLayer / CLDepthConcatenateLayer - @ref NEWinogradConvolutionLayer / @ref CLWinogradConvolutionLayer - @ref CLDepthwiseConvolutionLayer - @ref CLDirectConvolutionLayer - @ref CLConvolutionLayer - @ref CLScale - @ref CLIm2ColKernel - New Neon kernels / functions: - @ref NERNNLayer - New OpenCL kernels / functions: - @ref CLArithmeticDivision - Introduced prepare() stage support in the graph API for GLES. - Added support for memory reusage when trying to allocate smaller CLTensors. - Enabled NHWC execution on graph examples. - Added JPEG accessor for validation purposes. - Added validate methods to some kernels / functions. v18.05 Public major release - Various bug fixes. - Various optimisations. - Major redesign in the interface for the neon kernels implemented in assembly. - Removed arm_compute::NEGEMMLowpAArch64A53Kernel / arm_compute::NEGEMMLowpAArch64Kernel / arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore / arm_compute::NEHGEMMAArch64FP16Kernel - Added NEGEMMAssemblyWrapper and AssemblyKernelGlue which are used to execute assembly kernels in neon functions. - Minor changes to the CPUInfo type to make it compatible with the new assembly gemm interface. - Moved neon assembly kernels to the folder src/core/NEON/kernels/arm_gemm. - Improved doxygen documentation. - Improved memory management for layer's transitions. - Added support for NHWC data layout in tensors. - Added NHWC data layout support to: - @ref NEGEMMConvolutionLayer - @ref NEDirectConvolutionLayer - @ref NEPoolingLayer / @ref CLPoolingLayer - @ref NEBatchNormalizationLayer / @ref CLBatchNormalizationLayer - @ref NEDepthwiseConvolutionLayer - @ref NEScale - @ref NEIm2Col - Added support for dilated convolutions in @ref NEConvolutionLayer and @ref CLConvolutionLayer. - New OpenCL kernels / functions: - @ref CLChannelShuffleLayer / @ref CLChannelShuffleLayerKernel - @ref CLConvertFullyConnectedWeightsKernel / @ref CLConvertFullyConnectedWeights - @ref CLCopy / @ref CLCopyKernel - @ref CLLSTMLayer - @ref CLRNNLayer - CLWidthConcatenateLayer / @ref CLWidthConcatenateLayerKernel - @ref CLWinogradFilterTransformKernel / @ref CLWinogradInputTransformKernel / @ref CLWinogradConvolutionLayer - @ref CLWinogradInputTransformKernel / @ref CLWinogradInputTransform - New Neon kernels / functions: - @ref NEConvertFullyConnectedWeightsKernel / @ref NEConvertFullyConnectedWeights. - Created the validate method in @ref CLDepthwiseConvolutionLayer. - Beta and gamma are no longer mandatory arguments in @ref NEBatchNormalizationLayer and @ref CLBatchNormalizationLayer. - Added depth multiplier support in @ref NEDepthwiseConvolutionLayer and @ref CLDepthwiseConvolutionLayer. - Added broadcast multiply support in @ref NEPixelWiseMultiplication / @ref NEPixelWiseMultiplicationKernel. - Port mobilenet example to NHWC data layout. - Enabled Winograd method in @ref CLConvolutionLayer. - Renamed NEWinogradLayer to @ref NEWinogradConvolutionLayer. - Updated @ref NEWinogradConvolutionLayer to use highly optimised assembly kernels in src/core/NEON/kernels/arm_gemm. - Added memory manager support in GLES functions. - Major refactoring of the graph API. - Added GLES backend in the graph API. - Added support for the memory manager in the graph API. - Enabled Winograd Convolution method in the graph API. - Added support for grouped convolutions in the graph API. - Replaced NEDeconvolutionLayerUpsampleKernel with @ref NEScaleKernel in @ref NEDeconvolutionLayer. - Added fast maths flag in @ref CLConvolutionLayer. - Added new tests and benchmarks in validation and benchmark frameworks - Merge Activation layer with Convolution Layer (NEON. CL, GLES) - Added support to OpenCL 2.0 SVM - Added support to import memory in OpenCL tensors. - Added the prepare() method to perform any one off pre-processing before running the function. - Added new examples: - graph_inception_v4.cpp - graph_resnext50.cpp - Added memory measurement instrument for CL. v18.03 Public maintenance release - Various bug fixes. - Fixed bug in @ref NEActivationLayer - Fix in @ref CLTuner when using batches. - Updated recommended NDK version to r16b (And fixed warnings). - Fixed bug in validation code. - Added Inception v4 graph example. - Renamed NEWinogradLayer.cpp to @ref NEWinogradConvolutionLayer v18.02 Public major release - Various NEON / OpenCL / GLES optimisations. - Various bug fixes. - Changed default number of threads on big LITTLE systems. - Refactored examples and added: - graph_mobilenet_qassym8 - graph_resnet - graph_squeezenet_v1_1 - Renamed @ref CLConvolutionLayer into @ref CLGEMMConvolutionLayer and created a new @ref CLConvolutionLayer to select the fastest convolution method. - Renamed @ref NEConvolutionLayer into @ref NEGEMMConvolutionLayer and created a new @ref NEConvolutionLayer to select the fastest convolution method. - Added in place support to: - @ref CLActivationLayer - @ref CLBatchNormalizationLayer - Added QASYMM8 support to: - @ref CLActivationLayer - @ref CLDepthwiseConvolutionLayer - @ref NEDepthwiseConvolutionLayer - @ref NESoftmaxLayer - Added FP16 support to: - CLDepthwiseConvolutionLayer3x3 - @ref CLDepthwiseConvolutionLayer - Added broadcasting support to @ref NEArithmeticAddition / @ref CLArithmeticAddition / @ref CLPixelWiseMultiplication - Added fused batched normalization and activation to @ref CLBatchNormalizationLayer and @ref NEBatchNormalizationLayer - Added support for non-square pooling to @ref NEPoolingLayer and @ref CLPoolingLayer - New OpenCL kernels / functions: - CLDirectConvolutionLayerOutputStageKernel - New NEON kernels / functions - Added name() method to all kernels. - Added support for Winograd 5x5. - @ref NEPermuteKernel / @ref NEPermute - @ref NEWinogradLayerTransformInputKernel / NEWinogradLayer - @ref NEWinogradLayerTransformOutputKernel / NEWinogradLayer - @ref NEWinogradLayerTransformWeightsKernel / NEWinogradLayer - Renamed NEWinogradLayerKernel into NEWinogradLayerBatchedGEMMKernel - New GLES kernels / functions: - @ref GCTensorShiftKernel / @ref GCTensorShift v18.01 Public maintenance release - Various bug fixes - Added some of the missing validate() methods - Added @ref CLDeconvolutionLayerUpsampleKernel / @ref CLDeconvolutionLayer @ref CLDeconvolutionLayerUpsample - Added @ref CLPermuteKernel / @ref CLPermute - Added method to clean the programs cache in the CL Kernel library. - Added @ref GCArithmeticAdditionKernel / @ref GCArithmeticAddition - Added @ref GCDepthwiseConvolutionLayer3x3Kernel / @ref GCDepthwiseConvolutionLayer3x3 - Added @ref GCNormalizePlanarYUVLayerKernel / @ref GCNormalizePlanarYUVLayer - Added @ref GCScaleKernel / @ref GCScale - Added @ref GCWeightsReshapeKernel / @ref GCConvolutionLayer - Added FP16 support to the following GLES compute kernels: - @ref GCCol2ImKernel - @ref GCGEMMInterleave4x4Kernel - @ref GCGEMMTranspose1xWKernel - @ref GCIm2ColKernel - Refactored NEON Winograd (NEWinogradLayerKernel) - Added @ref NEDirectConvolutionLayerOutputStageKernel - Added QASYMM8 support to the following NEON kernels: - NEDepthwiseConvolutionLayer3x3Kernel - @ref NEFillBorderKernel - @ref NEPoolingLayerKernel - Added new examples: - graph_cl_mobilenet_qasymm8.cpp - graph_inception_v3.cpp - gc_dc.cpp - More tests added to both validation and benchmarking suites. v17.12 Public major release - Most machine learning functions on OpenCL support the new data type QASYMM8 - Introduced logging interface - Introduced opencl timer - Reworked GEMMLowp interface - Added new NEON assembly kernels for GEMMLowp, SGEMM and HGEMM - Added validation method for most Machine Learning kernels / functions - Added new graph examples such as googlenet, mobilenet, squeezenet, vgg16 and vgg19 - Added sgemm example for OpenCL - Added absolute difference example for GLES compute - Added new tests and benchmarks in validation and benchmark frameworks - Added new kernels / functions for GLES compute - New OpenGL ES kernels / functions - @ref GCAbsoluteDifferenceKernel / @ref GCAbsoluteDifference - @ref GCActivationLayerKernel / @ref GCActivationLayer - @ref GCBatchNormalizationLayerKernel / @ref GCBatchNormalizationLayer - @ref GCCol2ImKernel - @ref GCDepthConcatenateLayerKernel / GCDepthConcatenateLayer - @ref GCDirectConvolutionLayerKernel / @ref GCDirectConvolutionLayer - @ref GCDropoutLayerKernel / @ref GCDropoutLayer - @ref GCFillBorderKernel / @ref GCFillBorder - @ref GCGEMMInterleave4x4Kernel / @ref GCGEMMInterleave4x4 - @ref GCGEMMMatrixAccumulateBiasesKernel / @ref GCGEMMMatrixAdditionKernel / @ref GCGEMMMatrixMultiplyKernel / @ref GCGEMM - @ref GCGEMMTranspose1xWKernel / @ref GCGEMMTranspose1xW - @ref GCIm2ColKernel - @ref GCNormalizationLayerKernel / @ref GCNormalizationLayer - @ref GCPixelWiseMultiplicationKernel / @ref GCPixelWiseMultiplication - @ref GCPoolingLayerKernel / @ref GCPoolingLayer - @ref GCLogits1DMaxKernel / @ref GCLogits1DShiftExpSumKernel / @ref GCLogits1DNormKernel / @ref GCSoftmaxLayer - @ref GCTransposeKernel / @ref GCTranspose - New NEON kernels / functions - arm_compute::NEGEMMLowpAArch64A53Kernel / arm_compute::NEGEMMLowpAArch64Kernel / arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore - arm_compute::NEHGEMMAArch64FP16Kernel - NEDepthwiseConvolutionLayer3x3Kernel / NEDepthwiseIm2ColKernel / NEGEMMMatrixVectorMultiplyKernel / NEDepthwiseVectorToTensorKernel / @ref NEDepthwiseConvolutionLayer - @ref NEGEMMLowpOffsetContributionKernel / @ref NEGEMMLowpMatrixAReductionKernel / @ref NEGEMMLowpMatrixBReductionKernel / @ref NEGEMMLowpMatrixMultiplyCore - @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint - NEWinogradLayer / NEWinogradLayerKernel - New OpenCL kernels / functions - @ref CLGEMMLowpOffsetContributionKernel / @ref CLGEMMLowpMatrixAReductionKernel / @ref CLGEMMLowpMatrixBReductionKernel / @ref CLGEMMLowpMatrixMultiplyCore - CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint - New graph nodes for NEON and OpenCL - graph::BranchLayer - graph::DepthConvertLayer - graph::DepthwiseConvolutionLayer - graph::DequantizationLayer - graph::FlattenLayer - graph::QuantizationLayer - graph::ReshapeLayer v17.10 Public maintenance release - Bug fixes: - Check the maximum local workgroup size supported by OpenCL devices - Minor documentation updates (Fixed instructions to build the examples) - Introduced a graph::GraphContext - Added a few new Graph nodes, support for branches and grouping. - Automatically enable cl_printf in debug builds - Fixed bare metal builds for armv7a - Added AlexNet and cartoon effect examples - 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) v17.09 Public major release - Experimental Graph support: initial implementation of a simple stream API to easily chain machine learning layers. - Memory Manager (@ref BlobLifetimeManager, @ref BlobMemoryPool, @ref ILifetimeManager, @ref IMemoryGroup, @ref IMemoryManager, @ref IMemoryPool, @ref IPoolManager, @ref MemoryManagerOnDemand, @ref PoolManager) - New validation and benchmark frameworks (Boost and Google frameworks replaced by homemade framework). - Most machine learning functions support both fixed point 8 and 16 bit (QS8, QS16) for both NEON and OpenCL. - New NEON kernels / functions: - arm_compute::NEGEMMAssemblyBaseKernel arm_compute::NEGEMMAArch64Kernel - @ref NEDequantizationLayerKernel / @ref NEDequantizationLayer - @ref NEFloorKernel / @ref NEFloor - @ref NEL2NormalizeLayerKernel / @ref NEL2NormalizeLayer - @ref NEQuantizationLayerKernel @ref NEMinMaxLayerKernel / @ref NEQuantizationLayer - @ref NEROIPoolingLayerKernel / @ref NEROIPoolingLayer - @ref NEReductionOperationKernel / @ref NEReductionOperation - @ref NEReshapeLayerKernel / @ref NEReshapeLayer - New OpenCL kernels / functions: - @ref CLDepthwiseConvolutionLayer3x3NCHWKernel @ref CLDepthwiseConvolutionLayer3x3NHWCKernel CLDepthwiseIm2ColKernel CLDepthwiseVectorToTensorKernel CLDepthwiseWeightsReshapeKernel / CLDepthwiseConvolutionLayer3x3 @ref CLDepthwiseConvolutionLayer CLDepthwiseSeparableConvolutionLayer - @ref CLDequantizationLayerKernel / @ref CLDequantizationLayer - @ref CLDirectConvolutionLayerKernel / @ref CLDirectConvolutionLayer - @ref CLFlattenLayer - @ref CLFloorKernel / @ref CLFloor - CLGEMMTranspose1xW - @ref CLGEMMMatrixVectorMultiplyKernel - @ref CLL2NormalizeLayerKernel / @ref CLL2NormalizeLayer - @ref CLQuantizationLayerKernel @ref CLMinMaxLayerKernel / @ref CLQuantizationLayer - @ref CLROIPoolingLayerKernel / @ref CLROIPoolingLayer - @ref CLReductionOperationKernel / @ref CLReductionOperation - @ref CLReshapeLayerKernel / @ref CLReshapeLayer v17.06 Public major release - Various bug fixes - Added support for fixed point 8 bit (QS8) to the various NEON machine learning kernels. - Added unit tests and benchmarks (AlexNet, LeNet) - Added support for sub tensors. - Added infrastructure to provide GPU specific optimisation for some OpenCL kernels. - Added @ref OMPScheduler (OpenMP) scheduler for NEON - Added @ref SingleThreadScheduler scheduler for NEON (For bare metal) - User can specify his own scheduler by implementing the @ref IScheduler interface. - New OpenCL kernels / functions: - @ref CLBatchNormalizationLayerKernel / @ref CLBatchNormalizationLayer - @ref CLDepthConcatenateLayerKernel / CLDepthConcatenateLayer - @ref CLHOGOrientationBinningKernel @ref CLHOGBlockNormalizationKernel, @ref CLHOGDetectorKernel / @ref CLHOGDescriptor @ref CLHOGDetector @ref CLHOGGradient @ref CLHOGMultiDetection - @ref CLLocallyConnectedMatrixMultiplyKernel / @ref CLLocallyConnectedLayer - @ref CLWeightsReshapeKernel / @ref CLConvolutionLayerReshapeWeights - New C++ kernels: - @ref CPPDetectionWindowNonMaximaSuppressionKernel - New NEON kernels / functions: - @ref NEBatchNormalizationLayerKernel / @ref NEBatchNormalizationLayer - @ref NEDepthConcatenateLayerKernel / NEDepthConcatenateLayer - @ref NEDirectConvolutionLayerKernel / @ref NEDirectConvolutionLayer - @ref NELocallyConnectedMatrixMultiplyKernel / @ref NELocallyConnectedLayer - @ref NEWeightsReshapeKernel / @ref NEConvolutionLayerReshapeWeights v17.05 Public bug fixes release - Various bug fixes - Remaining of the functions ported to use accurate padding. - Library does not link against OpenCL anymore (It uses dlopen / dlsym at runtime instead to determine whether or not OpenCL is available). - Added "free" method to allocator. - Minimum version of g++ required for armv7 Linux changed from 4.8 to 4.9 v17.04 Public bug fixes release The following functions have been ported to use the new accurate padding: - @ref CLColorConvertKernel - @ref CLEdgeNonMaxSuppressionKernel - @ref CLEdgeTraceKernel - @ref CLGaussianPyramidHorKernel - @ref CLGaussianPyramidVertKernel - @ref CLGradientKernel - @ref NEChannelCombineKernel - @ref NEFillArrayKernel - @ref NEGaussianPyramidHorKernel - @ref NEGaussianPyramidVertKernel - NEHarrisScoreFP16Kernel - @ref NEHarrisScoreKernel - @ref NEHOGDetectorKernel - @ref NELogits1DMaxKernel - NELogits1DShiftExpSumKernel - NELogits1DNormKernel - @ref NENonMaximaSuppression3x3FP16Kernel - @ref NENonMaximaSuppression3x3Kernel v17.03.1 First Major public release of the sources - Renamed the library to arm_compute - New CPP target introduced for C++ kernels shared between NEON and CL functions. - New padding calculation interface introduced and ported most kernels / functions to use it. - New OpenCL kernels / functions: - CLGEMMLowpMatrixMultiplyKernel / CLGEMMLowp - New NEON kernels / functions: - @ref NENormalizationLayerKernel / @ref NENormalizationLayer - @ref NETransposeKernel / @ref NETranspose - @ref NELogits1DMaxKernel, NELogits1DShiftExpSumKernel, NELogits1DNormKernel / @ref NESoftmaxLayer - @ref NEIm2ColKernel, @ref NECol2ImKernel, NEConvolutionLayerWeightsReshapeKernel / @ref NEConvolutionLayer - NEGEMMMatrixAccumulateBiasesKernel / @ref NEFullyConnectedLayer - @ref NEGEMMLowpMatrixMultiplyKernel / NEGEMMLowp v17.03 Sources preview - New OpenCL kernels / functions: - @ref CLGradientKernel, @ref CLEdgeNonMaxSuppressionKernel, @ref CLEdgeTraceKernel / @ref CLCannyEdge - GEMM refactoring + FP16 support: CLGEMMInterleave4x4Kernel, CLGEMMTranspose1xWKernel, @ref CLGEMMMatrixMultiplyKernel, CLGEMMMatrixAdditionKernel / @ref CLGEMM - CLGEMMMatrixAccumulateBiasesKernel / @ref CLFullyConnectedLayer - @ref CLTransposeKernel / @ref CLTranspose - @ref CLLKTrackerInitKernel, @ref CLLKTrackerStage0Kernel, @ref CLLKTrackerStage1Kernel, @ref CLLKTrackerFinalizeKernel / @ref CLOpticalFlow - @ref CLNormalizationLayerKernel / @ref CLNormalizationLayer - @ref CLLaplacianPyramid, @ref CLLaplacianReconstruct - New NEON kernels / functions: - @ref NEActivationLayerKernel / @ref NEActivationLayer - GEMM refactoring + FP16 support (Requires armv8.2 CPU): @ref NEGEMMInterleave4x4Kernel, @ref NEGEMMTranspose1xWKernel, @ref NEGEMMMatrixMultiplyKernel, @ref NEGEMMMatrixAdditionKernel / @ref NEGEMM - @ref NEPoolingLayerKernel / @ref NEPoolingLayer v17.02.1 Sources preview - New OpenCL kernels / functions: - CLLogits1DMaxKernel, CLLogits1DShiftExpSumKernel, @ref CLLogits1DNormKernel / @ref CLSoftmaxLayer - @ref CLPoolingLayerKernel / @ref CLPoolingLayer - @ref CLIm2ColKernel, @ref CLCol2ImKernel, CLConvolutionLayerWeightsReshapeKernel / @ref CLConvolutionLayer - @ref CLRemapKernel / @ref CLRemap - @ref CLGaussianPyramidHorKernel, @ref CLGaussianPyramidVertKernel / @ref CLGaussianPyramid, @ref CLGaussianPyramidHalf, @ref CLGaussianPyramidOrb - @ref CLMinMaxKernel, @ref CLMinMaxLocationKernel / @ref CLMinMaxLocation - @ref CLNonLinearFilterKernel / @ref CLNonLinearFilter - New NEON FP16 kernels (Requires armv8.2 CPU) - @ref NEAccumulateWeightedFP16Kernel - @ref NEBox3x3FP16Kernel - @ref NENonMaximaSuppression3x3FP16Kernel v17.02 Sources preview - New OpenCL kernels / functions: - @ref CLActivationLayerKernel / @ref CLActivationLayer - @ref CLChannelCombineKernel / @ref CLChannelCombine - @ref CLDerivativeKernel / @ref CLChannelExtract - @ref CLFastCornersKernel / @ref CLFastCorners - @ref CLMeanStdDevKernel / @ref CLMeanStdDev - New NEON kernels / functions: - HOG / SVM: @ref NEHOGOrientationBinningKernel, @ref NEHOGBlockNormalizationKernel, @ref NEHOGDetectorKernel, NEHOGNonMaximaSuppressionKernel / @ref NEHOGDescriptor, @ref NEHOGDetector, @ref NEHOGGradient, @ref NEHOGMultiDetection - @ref NENonLinearFilterKernel / @ref NENonLinearFilter - Introduced a CLScheduler to manage the default context and command queue used by the runtime library and create synchronisation events. - Switched all the kernels / functions to use tensors instead of images. - Updated documentation to include instructions to build the library from sources. v16.12 Binary preview release - Original release @section S3_how_to_build How to build the library and the examples @subsection S3_1_build_options Build options scons 2.3 or above is required to build the library. To see the build options available simply run ```scons -h```: debug: Debug (yes|no) default: False actual: False asserts: Enable asserts (this flag is forced to 1 for debug=1) (yes|no) default: False actual: False arch: Target Architecture (armv7a|arm64-v8a|arm64-v8.2-a|x86_32|x86_64) default: armv7a actual: armv7a os: Target OS (linux|android|bare_metal) default: linux actual: linux build: Build type (native|cross_compile|embed_only) default: cross_compile actual: cross_compile examples: Build example programs (yes|no) default: True actual: True Werror: Enable/disable the -Werror compilation flag (yes|no) default: True actual: True opencl: Enable OpenCL support (yes|no) default: True actual: True neon: Enable Neon support (yes|no) default: False actual: False gles_compute: Enable OpenGL ES Compute Shader support (yes|no) default: False actual: False embed_kernels: Embed OpenCL kernels and OpenGL ES compute shader in library binary (yes|no) default: True actual: True set_soname: Set the library's soname and shlibversion (requires SCons 2.4 or above) (yes|no) default: False actual: False openmp: Enable OpenMP backend (yes|no) default: False actual: False cppthreads: Enable C++11 threads backend (yes|no) default: True actual: True build_dir: Specify sub-folder for the build ( /path/to/build_dir ) default: . actual: . extra_cxx_flags: Extra CXX flags to be appended to the build command default: actual: pmu: Enable PMU counters (yes|no) default: False actual: False mali: Enable Mali hardware counters (yes|no) default: False actual: False validation_tests: Build validation test programs (yes|no) default: False actual: False benchmark_tests: Build benchmark test programs (yes|no) default: False actual: False @b debug / @b asserts: - With debug=1 asserts are enabled, and the library is built with symbols and no optimisations enabled. - 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) - 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). @b arch: The x86_32 and x86_64 targets can only be used with neon=0 and opencl=1. @b os: Choose the operating system you are targeting: Linux, Android or bare metal. @note bare metal can only be used for NEON (not OpenCL), only static libraries get built and NEON's multi-threading support is disabled. @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. @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. 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. @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). @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) @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". @b set_soname: Do you want to build the versioned version of the library ? 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 -> libarm_compute_core.so.1.0.0 libarm_compute_core.so.1 -> libarm_compute_core.so.1.0.0 libarm_compute_core.so.1.0.0 @note This options is disabled by default as it requires SCons version 2.4 or above. @b extra_cxx_flags: Custom CXX flags which will be appended to the end of the build command. @b build_dir: Build the library in a subfolder of the "build" folder. (Allows to build several configurations in parallel). @b examples: Build or not the examples @b validation_tests: Enable the build of the validation suite. @b benchmark_tests: Enable the build of the benchmark tests @b pmu: Enable the PMU cycle counter to measure execution time in benchmark tests. (Your device needs to support it) @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) @b openmp Build in the OpenMP scheduler for NEON. @note Only works when building with g++ not clang++ @b cppthreads Build in the C++11 scheduler for NEON. @sa Scheduler::set @subsection S3_2_linux Building for Linux @subsubsection S3_2_1_library How to build the library ? For Linux, the library was successfully built and tested using the following Linaro GCC toolchain: - gcc-linaro-6.3.1-2017.05-x86_64_arm-linux-gnueabihf - gcc-linaro-6.3.1-2017.05-x86_64_aarch64-linux-gnu To cross-compile the library in debug mode, with NEON only support, for Linux 32bit: scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=linux arch=armv7a To cross-compile the library in asserts mode, with OpenCL only support, for Linux 64bit: scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=linux arch=arm64-v8a To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Linux 64bit: scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=0 gles_compute=1 embed_kernels=1 os=linux arch=arm64-v8a You can also compile the library natively on an ARM device by using build=native: scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=arm64-v8a build=native scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=native @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. For example on a 64bit Debian based system you would have to install g++-arm-linux-gnueabihf apt-get install g++-arm-linux-gnueabihf Then run scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a build=cross_compile or simply remove the build parameter as build=cross_compile is the default value: scons Werror=1 -j8 debug=0 neon=1 opencl=0 os=linux arch=armv7a @subsubsection S3_2_2_examples How to manually build the examples ? 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. @note The following command lines assume the arm_compute libraries 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 libraries 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. To cross compile a NEON example for Linux 32bit: 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 To cross compile a NEON example for Linux 64bit: 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 (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) To cross compile an OpenCL example for Linux 32bit: arm-linux-gnueabihf-g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL To cross compile an OpenCL example for Linux 64bit: aarch64-linux-gnu-g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -L. -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL To cross compile a GLES example for Linux 32bit: 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 To cross compile a GLES example for Linux 64bit: 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 (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) 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. i.e. to cross compile the "graph_lenet" example for Linux 32bit: arm-linux-gnueabihf-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet i.e. to cross compile the "graph_lenet" example for Linux 64bit: aarch64-linux-gnu-g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet (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) @note If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, arm_compute, arm_compute_core To compile natively (i.e directly on an ARM device) for NEON for Linux 32bit: g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -mfpu=neon -larm_compute -larm_compute_core -o neon_convolution To compile natively (i.e directly on an ARM device) for NEON for Linux 64bit: g++ examples/neon_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o neon_convolution (notice the only difference with the 32 bit command is that we don't need the -mfpu option) To compile natively (i.e directly on an ARM device) for OpenCL for Linux 32bit or Linux 64bit: g++ examples/cl_convolution.cpp utils/Utils.cpp -I. -Iinclude -std=c++11 -larm_compute -larm_compute_core -o cl_convolution -DARM_COMPUTE_CL To compile natively (i.e directly on an ARM device) for GLES for Linux 32bit or Linux 64bit: 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 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. i.e. to natively compile the "graph_lenet" example for Linux 32bit: g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -mfpu=neon -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet i.e. to natively compile the "graph_lenet" example for Linux 64bit: g++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -L. -larm_compute_graph -larm_compute -larm_compute_core -Wl,--allow-shlib-undefined -o graph_lenet (notice the only difference with the 32 bit command is that we don't need the -mfpu option) @note If compiling using static libraries, this order must be followed when linking: arm_compute_graph_static, arm_compute, arm_compute_core @note These two commands assume libarm_compute.so is available in your library path, if not add the path to it using -L (e.g. -Llib/linux-arm64-v8a-neon-cl-asserts/) @note You might need to export the path to OpenCL library as well in your LD_LIBRARY_PATH if Compute Library was built with OpenCL enabled. To run the built executable simply run: LD_LIBRARY_PATH=build ./neon_convolution or LD_LIBRARY_PATH=build ./cl_convolution @note Examples accept different types of arguments, to find out what they are run the example with \a --help as an argument. If no arguments are specified then random values will be used to execute the graph. For example: LD_LIBRARY_PATH=. ./graph_lenet --help Below is a list of the common parameters among the graph examples : @snippet utils/CommonGraphOptions.h Common graph examples parameters @subsection S3_3_android Building for Android For Android, the library was successfully built and tested using Google's standalone toolchains: - clang++ from NDK r18b for armv7a - clang++ from NDK r18b for arm64-v8a - clang++ from NDK r18b for arm64-v8.2-a with FP16 support Here is a guide to create your Android standalone toolchains from the NDK - Download the NDK r18b from here: https://developer.android.com/ndk/downloads/index.html to directory $NDK - Make sure you have Python 2.7 installed on your machine. - Generate the 32 and/or 64 toolchains by running the following commands to your toolchain dirctory $MY_TOOLCHAINS: $NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-ndk-r18b --stl libc++ --api 21 $NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-android-ndk-r18b --stl libc++ --api 21 @attention We used to use gnustl but as of NDK r17 it is deprecated so we switched to libc++ @note Make sure to add the toolchains to your PATH: export PATH=$PATH:$MY_TOOLCHAINS/aarch64-linux-android-ndk-r18b/bin:$MY_TOOLCHAINS/arm-linux-android-ndk-r18b/bin @subsubsection S3_3_1_library How to build the library ? To cross-compile the library in debug mode, with NEON only support, for Android 32bit: CXX=clang++ CC=clang scons Werror=1 -j8 debug=1 neon=1 opencl=0 os=android arch=armv7a To cross-compile the library in asserts mode, with OpenCL only support, for Android 64bit: CXX=clang++ CC=clang scons Werror=1 -j8 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=android arch=arm64-v8a To cross-compile the library in asserts mode, with GLES_COMPUTE only support, for Android 64bit: 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 @subsubsection S3_3_2_examples How to manually build the examples ? 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. @note The following command lines assume the arm_compute libraries 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 libraries 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. Once you've got your Android standalone toolchain built and added to your path you can do the following: To cross compile a NEON example: #32 bit: 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 #64 bit: 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 To cross compile an OpenCL example: #32 bit: 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 -DARM_COMPUTE_CL #64 bit: 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 -DARM_COMPUTE_CL To cross compile a GLES example: #32 bit: 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 #64 bit: 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 To cross compile the examples with the Graph API, such as graph_lenet.cpp, you need to link the library arm_compute_graph also. #32 bit: arm-linux-androideabi-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -Wl,--whole-archive -larm_compute_graph-static -Wl,--no-whole-archive -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_arm -static-libstdc++ -pie -DARM_COMPUTE_CL #64 bit: aarch64-linux-android-clang++ examples/graph_lenet.cpp utils/Utils.cpp utils/GraphUtils.cpp utils/CommonGraphOptions.cpp -I. -Iinclude -std=c++11 -Wl,--whole-archive -larm_compute_graph-static -Wl,--no-whole-archive -larm_compute-static -larm_compute_core-static -L. -o graph_lenet_aarch64 -static-libstdc++ -pie -DARM_COMPUTE_CL @note Due to some issues in older versions of the Mali OpenCL DDK (<= r13p0), we recommend to link arm_compute statically on Android. @note When linked statically the arm_compute_graph library currently needs the --whole-archive linker flag in order to work properly Then you need to do is upload the executable and the shared library to the device using ADB: adb push neon_convolution_arm /data/local/tmp/ adb push cl_convolution_arm /data/local/tmp/ adb push gc_absdiff_arm /data/local/tmp/ adb shell chmod 777 -R /data/local/tmp/ And finally to run the example: adb shell /data/local/tmp/neon_convolution_arm adb shell /data/local/tmp/cl_convolution_arm adb shell /data/local/tmp/gc_absdiff_arm For 64bit: adb push neon_convolution_aarch64 /data/local/tmp/ adb push cl_convolution_aarch64 /data/local/tmp/ adb push gc_absdiff_aarch64 /data/local/tmp/ adb shell chmod 777 -R /data/local/tmp/ And finally to run the example: adb shell /data/local/tmp/neon_convolution_aarch64 adb shell /data/local/tmp/cl_convolution_aarch64 adb shell /data/local/tmp/gc_absdiff_aarch64 @note Examples accept different types of arguments, to find out what they are run the example with \a --help as an argument. If no arguments are specified then random values will be used to execute the graph. For example: adb shell /data/local/tmp/graph_lenet --help In this case the first argument of LeNet (like all the graph examples) is the target (i.e 0 to run on NEON, 1 to run on OpenCL if available, 2 to run on OpenCL using the CLTuner), the second argument is the path to the folder containing the npy files for the weights and finally the third argument is the number of batches to run. @subsection S3_4_bare_metal Building for bare metal For bare metal, the library was successfully built using linaro's latest (gcc-linaro-6.3.1-2017.05) bare metal toolchains: - arm-eabi for armv7a - aarch64-elf for arm64-v8a Download linaro for armv7a and arm64-v8a. @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 @subsubsection S3_4_1_library How to build the library ? To cross-compile the library with NEON support for baremetal arm64-v8a: 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 @subsubsection S3_4_2_examples How to manually build the examples ? 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 here. @subsection S3_5_windows_host Building on a Windows host system Using `scons` directly from the Windows command line is known to cause problems. The reason seems to be that if `scons` 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. @subsubsection S3_5_1_ubuntu_on_windows Bash on Ubuntu on Windows The best and easiest option is to use Ubuntu on Windows. This feature is still marked as *beta* and thus might not be available. However, if it is building the library is as simple as opening a *Bash on Ubuntu on Windows* shell and following the general guidelines given above. @subsubsection S3_5_2_cygwin Cygwin If the Windows subsystem for Linux is not available Cygwin can be used to install and run `scons`, the minimum Cygwin version must be 3.0.7 or later. In addition to the default packages installed by Cygwin `scons` has to be selected in the installer. (`git` 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 GCC cross-compilers 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. @subsection S3_6_cl_requirements OpenCL DDK Requirements @subsubsection S3_6_1_cl_hard_requirements Hard Requirements Compute Library requires OpenCL 1.1 and above with support of non uniform workgroup sizes, which is officially supported in the Mali OpenCL DDK r8p0 and above as an extension (respective extension flag is \a -cl-arm-non-uniform-work-group-size). Enabling 16-bit floating point calculations require \a cl_khr_fp16 extension to be supported. All Mali GPUs with compute capabilities have native support for half precision floating points. Use of @ref CLMeanStdDev function requires 64-bit atomics support, thus \a cl_khr_int64_base_atomics should be supported in order to use. @subsubsection S3_6_2_cl_performance_requirements Performance improvements Integer dot product built-in function extensions (and therefore optimized kernels) are available with Mali OpenCL DDK r22p0 and above for the following GPUs : G71, G76. The relevant extensions are \a cl_arm_integer_dot_product_int8, \a cl_arm_integer_dot_product_accumulate_int8 and \a cl_arm_integer_dot_product_accumulate_int16. OpenCL kernel level debugging can be simplified with the use of printf, this requires the \a cl_arm_printf extension to be supported. SVM allocations are supported for all the underlying allocations in Compute Library. To enable this OpenCL 2.0 and above is a requirement. @subsection S3_7_cl_tuner OpenCL Tuner The OpenCL tuner, a.k.a. CLTuner, is a module of Arm Compute Library that can improve the performance of the OpenCL kernels tuning the Local-Workgroup-Size (LWS). The optimal LWS for each unique OpenCL kernel configuration is stored in a table. This table can be either imported or exported from/to a file. The OpenCL tuner runs the same OpenCL kernel for a range of local workgroup sizes and keeps the local workgroup size of the fastest run to use in subsequent calls to the kernel. It supports three modes of tuning with different trade-offs between the time taken to tune and the kernel execution time achieved using the best LWS found. In the Exhaustive mode, it searches all the supported values of LWS. This mode takes the longest time to tune and is the most likely to find the optimal LWS. Normal mode searches a subset of LWS values to yield a good approximation of the optimal LWS. It takes less time to tune than Exhaustive mode. Rapid mode takes the shortest time to tune and finds an LWS value that is at least as good or better than the default LWS value. The mode affects only the search for the optimal LWS and has no effect when the LWS value is imported from a file. In order for the performance numbers to be meaningful you must disable the GPU power management and set it to a fixed frequency for the entire duration of the tuning phase. If you wish to know more about LWS and the important role on improving the GPU cache utilization, we suggest having a look at the presentation "Even Faster CNNs: Exploring the New Class of Winograd Algorithms available at the following link: https://www.embedded-vision.com/platinum-members/arm/embedded-vision-training/videos/pages/may-2018-embedded-vision-summit-iodice Tuning a network from scratch can be long and affect considerably the execution time for the first run of your network. It is recommended for this reason to store the CLTuner's result in a file to amortize this time when you either re-use the same network or the functions with the same configurations. The tuning is performed only once for each OpenCL kernel. CLTuner looks for the optimal LWS for each unique OpenCL kernel configuration. Since a function (i.e. Convolution Layer, Pooling Layer, Fully Connected Layer ...) can be called multiple times but with different parameters, we associate an "id" (called "config_id") to each kernel to distinguish the unique configurations. #Example: 2 unique Matrix Multiply configurations @code{.cpp} TensorShape a0 = TensorShape(32,32); TensorShape b0 = TensorShape(32,32); TensorShape c0 = TensorShape(32,32); TensorShape a1 = TensorShape(64,64); TensorShape b1 = TensorShape(64,64); TensorShape c1 = TensorShape(64,64); Tensor a0_tensor; Tensor b0_tensor; Tensor c0_tensor; Tensor a1_tensor; Tensor b1_tensor; Tensor c1_tensor; a0_tensor.allocator()->init(TensorInfo(a0, 1, DataType::F32)); b0_tensor.allocator()->init(TensorInfo(b0, 1, DataType::F32)); c0_tensor.allocator()->init(TensorInfo(c0, 1, DataType::F32)); a1_tensor.allocator()->init(TensorInfo(a1, 1, DataType::F32)); b1_tensor.allocator()->init(TensorInfo(b1, 1, DataType::F32)); c1_tensor.allocator()->init(TensorInfo(c1 1, DataType::F32)); CLGEMM gemm0; CLGEMM gemm1; // Configuration 0 gemm0.configure(&a0, &b0, nullptr, &c0, 1.0f, 0.0f); // Configuration 1 gemm1.configure(&a1, &b1, nullptr, &c1, 1.0f, 0.0f); @endcode @subsubsection S3_7_1_cl_tuner_how_to How to use it All the graph examples in the Compute Library's folder "examples" and the arm_compute_benchmark accept an argument to enable the OpenCL tuner and an argument to export/import the LWS values to/from a file #Enable CL tuner ./graph_mobilenet --enable-tuner –-target=CL ./arm_compute_benchmark --enable-tuner #Export/Import to/from a file ./graph_mobilenet --enable-tuner --target=CL --tuner-file=acl_tuner.csv ./arm_compute_benchmark --enable-tuner --tuner-file=acl_tuner.csv If you are importing the CLTuner'results from a file, the new tuned LWS values will be appended to it. Either you are benchmarking the graph examples or the test cases in the arm_compute_benchmark remember to: -# Disable the power management -# Keep the GPU frequency constant -# Run multiple times the network (i.e. 10). If you are not using the graph API or the benchmark infrastructure you will need to manually pass a CLTuner object to CLScheduler before configuring any function. @code{.cpp} CLTuner tuner; // Setup Scheduler CLScheduler::get().default_init(&tuner); @endcode After the first run, the CLTuner's results can be exported to a file using the method "save_to_file()". - tuner.save_to_file("results.csv"); This file can be also imported using the method "load_from_file("results.csv")". - tuner.load_from_file("results.csv"); */ } // namespace arm_compute