diff --git a/.github/issue_template.md b/.github/issue_template.md new file mode 100644 index 0000000000..4eaef77117 --- /dev/null +++ b/.github/issue_template.md @@ -0,0 +1,18 @@ + + +**Output of 'strings libarm_compute.so | grep arm_compute_version':** + +**Platform:** + +**Operating System:** + + + + +**Problem description:** diff --git a/README.md b/README.md index d9b497b5d6..471d28e834 100644 --- a/README.md +++ b/README.md @@ -1,9 +1,14 @@ -:warning: **Deprecation notice: QS8 and QS16 data types will be removed in the next release** (As far as we know nobody uses these data types, if you do or think they are useful please open an Issue or send us an email):warning: +:warning: **Deprecation notice: QS8 and QS16 data types will be removed in the 18.05 release** (As far as we know nobody uses these data types, if you do or think they are useful please open an Issue or send us an email):warning: Please report issues here: https://github.com/ARM-software/ComputeLibrary/issues **Make sure you are using the latest version of the library before opening an issue. Thanks** +News: + +We're hiring: [Senior Machine Learning C++ Software Engineer](https://careers.peopleclick.com/careerscp/client_arm/external/jobDetails.do?functionName=getJobDetail&jobPostId=36246&localeCode=en-us) +Come talk to us: [Gian Marco will be presenting his work at the EVS](https://www.embedded-vision.com/summit/even-faster-cnns-exploring-new-class-winograd-algorithms) + Related projects: - [Caffe on Compute Library](https://github.com/OAID/Caffe-HRT) @@ -12,6 +17,7 @@ Related projects: Documentation available here: +- [v18.03](https://arm-software.github.io/ComputeLibrary/v18.03/) - [v18.02](https://arm-software.github.io/ComputeLibrary/v18.02/) - [v18.01](https://arm-software.github.io/ComputeLibrary/v18.01/) - [v17.12](https://arm-software.github.io/ComputeLibrary/v17.12/) @@ -24,8 +30,10 @@ Documentation available here: Binaries available here: -- [v18.02-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.02-bin-linux.tar.gz) -- [v18.02-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.02-bin-android.tar.gz) +- [v18.03-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.03/arm_compute-v18.03-bin-linux.tar.gz) +- [v18.03-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.03/arm_compute-v18.03-bin-android.tar.gz) +- [v18.02-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.02/arm_compute-v18.02-bin-linux.tar.gz) +- [v18.02-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.02/arm_compute-v18.02-bin-android.tar.gz) - [v18.01](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.01-bin.tar.gz) - [v17.12](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.12/arm_compute-v17.12-bin.tar.gz) - [v17.10](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.10/arm_compute-v17.10-bin.tar.gz) diff --git a/SConscript b/SConscript index 5e05c64234..c5ace1c894 100644 --- a/SConscript +++ b/SConscript @@ -24,8 +24,8 @@ import os.path import re import subprocess -VERSION = "v18.02" -SONAME_VERSION="9.0.0" +VERSION = "v18.03" +SONAME_VERSION="10.0.0" Import('env') Import('vars') diff --git a/arm_compute/core/Dimensions.h b/arm_compute/core/Dimensions.h index 58ffd7ff3c..5319346866 100644 --- a/arm_compute/core/Dimensions.h +++ b/arm_compute/core/Dimensions.h @@ -50,7 +50,7 @@ class Dimensions */ template explicit Dimensions(Ts... dims) - : _id{ { dims... } }, _num_dimensions{ sizeof...(dims) } + : _id{ { static_cast(dims)... } }, _num_dimensions{ sizeof...(dims) } { } diff --git a/arm_compute/core/NEON/kernels/convolution/winograd/gemm.hpp b/arm_compute/core/NEON/kernels/convolution/winograd/gemm.hpp index 62a20c9eea..6e06db324c 100644 --- a/arm_compute/core/NEON/kernels/convolution/winograd/gemm.hpp +++ b/arm_compute/core/NEON/kernels/convolution/winograd/gemm.hpp @@ -65,11 +65,11 @@ inline void BlockedGemm( const int c_row_stride ) { // Array access methods - const auto A = [a, M, K, a_row_stride] (const int i, const int j) -> TIn { + const auto A = [a, a_row_stride] (const int i, const int j) -> TIn { return a[i*a_row_stride + j]; }; - const auto B = [b, K, N, b_row_stride] (const int i, const int j) -> TIn { + const auto B = [b, b_row_stride] (const int i, const int j) -> TIn { return b[i*b_row_stride + j]; }; diff --git a/arm_compute/core/utils/logging/Macros.h b/arm_compute/core/utils/logging/Macros.h index bc121e25eb..5593030261 100644 --- a/arm_compute/core/utils/logging/Macros.h +++ b/arm_compute/core/utils/logging/Macros.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017 ARM Limited. + * Copyright (c) 2017-2018 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -50,14 +50,16 @@ } \ } while(false) -#define ARM_COMPUTE_LOG_STREAM(logger_name, log_level, stream) \ - do \ - { \ - auto __logger = arm_compute::logging::LoggerRegistry::get().logger(logger_name); \ - if(__logger != nullptr) \ - { \ - __logger->log(log_level, static_cast(std::ostringstream() << stream).str()); \ - } \ +#define ARM_COMPUTE_LOG_STREAM(logger_name, log_level, stream) \ + do \ + { \ + auto __logger = arm_compute::logging::LoggerRegistry::get().logger(logger_name); \ + if(__logger != nullptr) \ + { \ + std::ostringstream s; \ + s << stream; \ + __logger->log(log_level, s.str()); \ + } \ } while(false) #else /* ARM_COMPUTE_LOGGING_ENABLED */ diff --git a/arm_compute/runtime/CL/CLTuner.h b/arm_compute/runtime/CL/CLTuner.h index 386994682d..1c71709a7a 100644 --- a/arm_compute/runtime/CL/CLTuner.h +++ b/arm_compute/runtime/CL/CLTuner.h @@ -37,26 +37,43 @@ class ICLKernel; class CLTuner : public ICLTuner { public: - /** Constructor */ - CLTuner(); + /** Constructor + * + * @param[in] tune_new_kernels Find the optimal local workgroup size for kernels which are not present in the table ? + * + */ + CLTuner(bool tune_new_kernels = true); /** Destructor */ ~CLTuner() = default; + /** Setter for tune_new_kernels option + * + * @param[in] tune_new_kernels Find the optimal local workgroup size for kernels which are not present in the table ? + */ + void set_tune_new_kernels(bool tune_new_kernels); + /** Tune kernels that are not in the LWS table + * + * @return True if tuning of new kernels is enabled. + */ + bool tune_new_kernels() const; + /** Manually add a LWS for a kernel + * + * @param[in] kernel_id Unique identifiant of the kernel + * @param[in] optimal_lws Optimal local workgroup size to use for the given kernel + */ + void add_lws_to_table(const std::string &kernel_id, cl::NDRange optimal_lws); /** Import LWS table * * @param[in] lws_table The unordered_map container to import */ void import_lws_table(const std::unordered_map &lws_table); - /** Export LWS table + /** Give read access to the LWS table * * return The lws table as unordered_map container */ - const std::unordered_map &export_lws_table(); - - // Inherited methods overridden: - void tune_kernel(ICLKernel &kernel) override; + const std::unordered_map &lws_table() const; /** Set the OpenCL kernel event * @@ -66,7 +83,28 @@ class CLTuner : public ICLTuner */ void set_cl_kernel_event(cl_event kernel_event); - std::function real_function; + std::function real_clEnqueueNDRangeKernel; + + /** Load the LWS table from file + * + * @param[in] filename Load the LWS table from this file.(Must exist) + */ + void load_from_file(const std::string &filename); + + /** Save the content of the LWS table to file + * + * @param[in] filename Save the LWS table to this file. (Content will be overwritten) + */ + void save_to_file(const std::string &filename) const; + + // Inherited methods overridden: + void tune_kernel(ICLKernel &kernel) override; + + /** Is the kernel_event set ? + * + * @return true if the kernel_event is set. + */ + bool kernel_event_is_set() const; private: /** Find optimal LWS using brute-force approach @@ -81,33 +119,7 @@ class CLTuner : public ICLTuner cl::CommandQueue _queue; cl::CommandQueue _queue_profiler; cl::Event _kernel_event; -}; - -/* Function to be used to intercept kernel enqueues and store their OpenCL Event */ -class Interceptor -{ -public: - explicit Interceptor(CLTuner &tuner); - - /** clEnqueueNDRangeKernel interface - * - * @param[in] command_queue A valid command-queue. The kernel will be queued for execution on the device associated with command_queue. - * @param[in] kernel A valid kernel object. The OpenCL context associated with kernel and command_queue must be the same. - * @param[in] work_dim The number of dimensions used to specify the global work-items and work-items in the work-group. work_dim must be greater than zero and less than or equal to CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS. - * @param[in] gwo Global-Workgroup-Offset. It can be used to specify an array of work_dim unsigned values that describe the offset used to calculate the global ID of a work-item. If global_work_offset is NULL, the global IDs start at offset (0, 0, ... 0). - * @param[in] gws Global-Workgroup-Size. Points to an array of work_dim unsigned values that describe the number of global work-items in work_dim dimensions that will execute the kernel function. - * @param[in] lws Local-Workgroup-Size. Points to an array of work_dim unsigned values that describe the number of work-items that make up a work-group - * @param[in] num_events_in_wait_list Number of events in the waiting list - * @param[in] event_wait_list Event waiting list - * @param[in] event OpenCL kernel event - * - * @return the OpenCL status - */ - cl_int operator()(cl_command_queue command_queue, cl_kernel kernel, cl_uint work_dim, const size_t *gwo, const size_t *gws, const size_t *lws, cl_uint num_events_in_wait_list, - const cl_event *event_wait_list, cl_event *event); - -private: - CLTuner &_tuner; + bool _tune_new_kernels; }; } #endif /*__ARM_COMPUTE_CLTUNER_H__ */ diff --git a/docs/00_introduction.dox b/docs/00_introduction.dox index 6de2d0f0e3..eb6130bda5 100644 --- a/docs/00_introduction.dox +++ b/docs/00_introduction.dox @@ -1,3 +1,5 @@ +namespace arm_compute +{ /** @mainpage Introduction @tableofcontents @@ -26,8 +28,8 @@ For each release we provide some pre-built binaries of the library [here](https: These binaries have been built using the following toolchains: - Linux armv7a: gcc-linaro-arm-linux-gnueabihf-4.9-2014.07_linux - Linux arm64-v8a: gcc-linaro-4.9-2016.02-x86_64_aarch64-linux-gnu - - Android armv7a: clang++ / gnustl NDK r14 - - Android am64-v8a: clang++ / gnustl NDK r14 + - Android armv7a: clang++ / gnustl NDK r16b + - Android am64-v8a: clang++ / gnustl NDK r16b @warning Make sure to use a compatible toolchain to build your application or you will get some std::bad_alloc errors at runtime. @@ -108,6 +110,8 @@ You should have the following file organisation: │   │   └── OMPScheduler.h --> OpenMP scheduler (Alternative to the CPPScheduler) │ ├── Memory manager files (LifetimeManager, PoolManager, etc.) │   └── Basic implementations of the generic object interfaces (Array, Image, Tensor, etc.) + ├── data -> Contains test images and reference data dumps used by validation tests + ├── docs -> Contains Doxyfile and Doxygen sources used to generate the HTML pages in the documentation folder. ├── documentation │   ├── index.xhtml │   └── ... @@ -155,6 +159,8 @@ You should have the following file organisation: │ │ ├── Benchmark specific files │ │ ├── CL --> OpenCL benchmarking tests │ │ ├── GLES_COMPUTE --> GLES benchmarking tests + │   │ ├── fixtures + │ │ │ └── Fixtures to initialise and run the runtime Functions. │ │ └── NEON --> NEON benchmarking tests │   ├── datasets │ │ └── Datasets for all the validation / benchmark tests, layer configurations for various networks, etc. @@ -189,6 +195,14 @@ If there is more than one release in a month then an extra sequential number is @subsection S2_2_changelog Changelog +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. + v18.02 Public major release - Various NEON / OpenCL / GLES optimisations. - Various bug fixes. @@ -197,57 +211,57 @@ v18.02 Public major release - graph_mobilenet_qassym8 - graph_resnet - graph_squeezenet_v1_1 - - Renamed @ref arm_compute::CLConvolutionLayer into @ref arm_compute::CLGEMMConvolutionLayer and created a new @ref arm_compute::CLConvolutionLayer to select the fastest convolution method. - - Renamed @ref arm_compute::NEConvolutionLayer into @ref arm_compute::NEGEMMConvolutionLayer and created a new @ref arm_compute::NEConvolutionLayer to select the fastest convolution method. + - 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 arm_compute::CLActivationLayer - - @ref arm_compute::CLBatchNormalizationLayer + - @ref CLActivationLayer + - @ref CLBatchNormalizationLayer - Added QASYMM8 support to: - - @ref arm_compute::CLActivationLayer - - @ref arm_compute::CLDepthwiseConvolutionLayer - - @ref arm_compute::NEDepthwiseConvolutionLayer - - @ref arm_compute::NESoftmaxLayer + - @ref CLActivationLayer + - @ref CLDepthwiseConvolutionLayer + - @ref NEDepthwiseConvolutionLayer + - @ref NESoftmaxLayer - Added FP16 support to: - - @ref arm_compute::CLDepthwiseConvolutionLayer3x3 - - @ref arm_compute::CLDepthwiseConvolutionLayer - - Added broadcasting support to @ref arm_compute::NEArithmeticAddition / @ref arm_compute::CLArithmeticAddition / @ref arm_compute::CLPixelWiseMultiplication - - Added fused batched normalization and activation to @ref arm_compute::CLBatchNormalizationLayer and @ref arm_compute::NEBatchNormalizationLayer - - Added support for non-square pooling to @ref arm_compute::NEPoolingLayer and @ref arm_compute::CLPoolingLayer + - @ref 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: - - @ref arm_compute::CLDirectConvolutionLayerOutputStageKernel + - @ref CLDirectConvolutionLayerOutputStageKernel - New NEON kernels / functions - Added name() method to all kernels. - Added support for Winograd 5x5. - - @ref arm_compute::NEPermuteKernel / @ref arm_compute::NEPermute - - @ref arm_compute::NEWinogradLayerTransformInputKernel / @ref arm_compute::NEWinogradLayer - - @ref arm_compute::NEWinogradLayerTransformOutputKernel / @ref arm_compute::NEWinogradLayer - - @ref arm_compute::NEWinogradLayerTransformWeightsKernel / @ref arm_compute::NEWinogradLayer - - Renamed arm_compute::NEWinogradLayerKernel into @ref arm_compute::NEWinogradLayerBatchedGEMMKernel + - @ref NEPermuteKernel / @ref NEPermute + - @ref NEWinogradLayerTransformInputKernel / @ref NEWinogradLayer + - @ref NEWinogradLayerTransformOutputKernel / @ref NEWinogradLayer + - @ref NEWinogradLayerTransformWeightsKernel / @ref NEWinogradLayer + - Renamed NEWinogradLayerKernel into @ref NEWinogradLayerBatchedGEMMKernel - New GLES kernels / functions: - - @ref arm_compute::GCTensorShiftKernel / @ref arm_compute::GCTensorShift + - @ref GCTensorShiftKernel / @ref GCTensorShift v18.01 Public maintenance release - Various bug fixes - Added some of the missing validate() methods - - Added @ref arm_compute::CLDeconvolutionLayerUpsampleKernel / @ref arm_compute::CLDeconvolutionLayer @ref arm_compute::CLDeconvolutionLayerUpsample - - Added @ref arm_compute::CLPermuteKernel / @ref arm_compute::CLPermute + - 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 arm_compute::GCArithmeticAdditionKernel / @ref arm_compute::GCArithmeticAddition - - Added @ref arm_compute::GCDepthwiseConvolutionLayer3x3Kernel / @ref arm_compute::GCDepthwiseConvolutionLayer3x3 - - Added @ref arm_compute::GCNormalizePlanarYUVLayerKernel / @ref arm_compute::GCNormalizePlanarYUVLayer - - Added @ref arm_compute::GCScaleKernel / @ref arm_compute::GCScale - - Added @ref arm_compute::GCWeightsReshapeKernel / @ref arm_compute::GCConvolutionLayer + - 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 arm_compute::GCCol2ImKernel - - @ref arm_compute::GCGEMMInterleave4x4Kernel - - @ref arm_compute::GCGEMMTranspose1xWKernel - - @ref arm_compute::GCIm2ColKernel - - Refactored NEON Winograd (arm_compute::NEWinogradLayerKernel) - - Added @ref arm_compute::NEDirectConvolutionLayerOutputStageKernel + - @ref GCCol2ImKernel + - @ref GCGEMMInterleave4x4Kernel + - @ref GCGEMMTranspose1xWKernel + - @ref GCIm2ColKernel + - Refactored NEON Winograd (NEWinogradLayerKernel) + - Added @ref NEDirectConvolutionLayerOutputStageKernel - Added QASYMM8 support to the following NEON kernels: - - @ref arm_compute::NEDepthwiseConvolutionLayer3x3Kernel - - @ref arm_compute::NEFillBorderKernel - - @ref arm_compute::NEPoolingLayerKernel + - @ref NEDepthwiseConvolutionLayer3x3Kernel + - @ref NEFillBorderKernel + - @ref NEPoolingLayerKernel - Added new examples: - graph_cl_mobilenet_qasymm8.cpp - graph_inception_v3.cpp @@ -268,52 +282,52 @@ v17.12 Public major release - Added new kernels / functions for GLES compute - New OpenGL ES kernels / functions - - @ref arm_compute::GCAbsoluteDifferenceKernel / @ref arm_compute::GCAbsoluteDifference - - @ref arm_compute::GCActivationLayerKernel / @ref arm_compute::GCActivationLayer - - @ref arm_compute::GCBatchNormalizationLayerKernel / @ref arm_compute::GCBatchNormalizationLayer - - @ref arm_compute::GCCol2ImKernel - - @ref arm_compute::GCDepthConcatenateLayerKernel / @ref arm_compute::GCDepthConcatenateLayer - - @ref arm_compute::GCDirectConvolutionLayerKernel / @ref arm_compute::GCDirectConvolutionLayer - - @ref arm_compute::GCDropoutLayerKernel / @ref arm_compute::GCDropoutLayer - - @ref arm_compute::GCFillBorderKernel / @ref arm_compute::GCFillBorder - - @ref arm_compute::GCGEMMInterleave4x4Kernel / @ref arm_compute::GCGEMMInterleave4x4 - - @ref arm_compute::GCGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::GCGEMMMatrixAdditionKernel / @ref arm_compute::GCGEMMMatrixMultiplyKernel / @ref arm_compute::GCGEMM - - @ref arm_compute::GCGEMMTranspose1xWKernel / @ref arm_compute::GCGEMMTranspose1xW - - @ref arm_compute::GCIm2ColKernel - - @ref arm_compute::GCNormalizationLayerKernel / @ref arm_compute::GCNormalizationLayer - - @ref arm_compute::GCPixelWiseMultiplicationKernel / @ref arm_compute::GCPixelWiseMultiplication - - @ref arm_compute::GCPoolingLayerKernel / @ref arm_compute::GCPoolingLayer - - @ref arm_compute::GCLogits1DMaxKernel / @ref arm_compute::GCLogits1DShiftExpSumKernel / @ref arm_compute::GCLogits1DNormKernel / @ref arm_compute::GCSoftmaxLayer - - @ref arm_compute::GCTransposeKernel / @ref arm_compute::GCTranspose + - @ref GCAbsoluteDifferenceKernel / @ref GCAbsoluteDifference + - @ref GCActivationLayerKernel / @ref GCActivationLayer + - @ref GCBatchNormalizationLayerKernel / @ref GCBatchNormalizationLayer + - @ref GCCol2ImKernel + - @ref GCDepthConcatenateLayerKernel / @ref 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 - - @ref arm_compute::NEGEMMLowpAArch64A53Kernel / @ref arm_compute::NEGEMMLowpAArch64Kernel / @ref arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / @ref arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore - - @ref arm_compute::NEHGEMMAArch64FP16Kernel - - @ref arm_compute::NEDepthwiseConvolutionLayer3x3Kernel / @ref arm_compute::NEDepthwiseIm2ColKernel / @ref arm_compute::NEGEMMMatrixVectorMultiplyKernel / @ref arm_compute::NEDepthwiseVectorToTensorKernel / @ref arm_compute::NEDepthwiseConvolutionLayer - - @ref arm_compute::NEGEMMLowpOffsetContributionKernel / @ref arm_compute::NEGEMMLowpMatrixAReductionKernel / @ref arm_compute::NEGEMMLowpMatrixBReductionKernel / @ref arm_compute::NEGEMMLowpMatrixMultiplyCore - - @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint - - @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref arm_compute::NEGEMMLowpQuantizeDownInt32ToUint8Scale - - @ref arm_compute::NEWinogradLayer / arm_compute::NEWinogradLayerKernel + - @ref NEGEMMLowpAArch64A53Kernel / @ref NEGEMMLowpAArch64Kernel / @ref NEGEMMLowpAArch64V8P4Kernel / NEGEMMInterleavedBlockedKernel / @ref NEGEMMLowpAssemblyMatrixMultiplyCore + - @ref NEHGEMMAArch64FP16Kernel + - @ref NEDepthwiseConvolutionLayer3x3Kernel / @ref NEDepthwiseIm2ColKernel / @ref NEGEMMMatrixVectorMultiplyKernel / @ref NEDepthwiseVectorToTensorKernel / @ref NEDepthwiseConvolutionLayer + - @ref NEGEMMLowpOffsetContributionKernel / @ref NEGEMMLowpMatrixAReductionKernel / @ref NEGEMMLowpMatrixBReductionKernel / @ref NEGEMMLowpMatrixMultiplyCore + - @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint + - @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref NEGEMMLowpQuantizeDownInt32ToUint8Scale + - @ref NEWinogradLayer / NEWinogradLayerKernel - New OpenCL kernels / functions - - @ref arm_compute::CLGEMMLowpOffsetContributionKernel / @ref arm_compute::CLGEMMLowpMatrixAReductionKernel / @ref arm_compute::CLGEMMLowpMatrixBReductionKernel / @ref arm_compute::CLGEMMLowpMatrixMultiplyCore - - @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint - - @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref arm_compute::CLGEMMLowpQuantizeDownInt32ToUint8Scale + - @ref CLGEMMLowpOffsetContributionKernel / @ref CLGEMMLowpMatrixAReductionKernel / @ref CLGEMMLowpMatrixBReductionKernel / @ref CLGEMMLowpMatrixMultiplyCore + - @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint + - @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel / @ref CLGEMMLowpQuantizeDownInt32ToUint8Scale - New graph nodes for NEON and OpenCL - - @ref arm_compute::graph::BranchLayer - - @ref arm_compute::graph::DepthConvertLayer - - @ref arm_compute::graph::DepthwiseConvolutionLayer - - @ref arm_compute::graph::DequantizationLayer - - @ref arm_compute::graph::FlattenLayer - - @ref arm_compute::graph::QuantizationLayer - - @ref arm_compute::graph::ReshapeLayer + - @ref graph::BranchLayer + - @ref graph::DepthConvertLayer + - @ref graph::DepthwiseConvolutionLayer + - @ref graph::DequantizationLayer + - @ref graph::FlattenLayer + - @ref graph::QuantizationLayer + - @ref 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 arm_compute::graph::GraphContext + - 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 @@ -322,32 +336,32 @@ v17.10 Public maintenance release v17.09 Public major release - Experimental Graph support: initial implementation of a simple stream API to easily chain machine learning layers. - - 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) + - 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: - - @ref arm_compute::NEGEMMAssemblyBaseKernel @ref arm_compute::NEGEMMAArch64Kernel - - @ref arm_compute::NEDequantizationLayerKernel / @ref arm_compute::NEDequantizationLayer - - @ref arm_compute::NEFloorKernel / @ref arm_compute::NEFloor - - @ref arm_compute::NEL2NormalizeLayerKernel / @ref arm_compute::NEL2NormalizeLayer - - @ref arm_compute::NEQuantizationLayerKernel @ref arm_compute::NEMinMaxLayerKernel / @ref arm_compute::NEQuantizationLayer - - @ref arm_compute::NEROIPoolingLayerKernel / @ref arm_compute::NEROIPoolingLayer - - @ref arm_compute::NEReductionOperationKernel / @ref arm_compute::NEReductionOperation - - @ref arm_compute::NEReshapeLayerKernel / @ref arm_compute::NEReshapeLayer + - @ref NEGEMMAssemblyBaseKernel @ref 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 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 - - @ref arm_compute::CLDequantizationLayerKernel / @ref arm_compute::CLDequantizationLayer - - @ref arm_compute::CLDirectConvolutionLayerKernel / @ref arm_compute::CLDirectConvolutionLayer - - @ref arm_compute::CLFlattenLayer - - @ref arm_compute::CLFloorKernel / @ref arm_compute::CLFloor - - @ref arm_compute::CLGEMMTranspose1xW - - @ref arm_compute::CLGEMMMatrixVectorMultiplyKernel - - @ref arm_compute::CLL2NormalizeLayerKernel / @ref arm_compute::CLL2NormalizeLayer - - @ref arm_compute::CLQuantizationLayerKernel @ref arm_compute::CLMinMaxLayerKernel / @ref arm_compute::CLQuantizationLayer - - @ref arm_compute::CLROIPoolingLayerKernel / @ref arm_compute::CLROIPoolingLayer - - @ref arm_compute::CLReductionOperationKernel / @ref arm_compute::CLReductionOperation - - @ref arm_compute::CLReshapeLayerKernel / @ref arm_compute::CLReshapeLayer + - @ref CLDepthwiseConvolutionLayer3x3Kernel @ref CLDepthwiseIm2ColKernel @ref CLDepthwiseVectorToTensorKernel @ref CLDepthwiseWeightsReshapeKernel / @ref CLDepthwiseConvolutionLayer3x3 @ref CLDepthwiseConvolutionLayer @ref CLDepthwiseSeparableConvolutionLayer + - @ref CLDequantizationLayerKernel / @ref CLDequantizationLayer + - @ref CLDirectConvolutionLayerKernel / @ref CLDirectConvolutionLayer + - @ref CLFlattenLayer + - @ref CLFloorKernel / @ref CLFloor + - @ref 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 @@ -355,23 +369,23 @@ v17.06 Public major release - 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 arm_compute::OMPScheduler (OpenMP) scheduler for NEON - - Added @ref arm_compute::SingleThreadScheduler scheduler for NEON (For bare metal) - - User can specify his own scheduler by implementing the @ref arm_compute::IScheduler interface. + - 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 arm_compute::CLBatchNormalizationLayerKernel / @ref arm_compute::CLBatchNormalizationLayer - - @ref arm_compute::CLDepthConcatenateLayerKernel / @ref arm_compute::CLDepthConcatenateLayer - - @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 - - @ref arm_compute::CLLocallyConnectedMatrixMultiplyKernel / @ref arm_compute::CLLocallyConnectedLayer - - @ref arm_compute::CLWeightsReshapeKernel / @ref arm_compute::CLConvolutionLayerReshapeWeights + - @ref CLBatchNormalizationLayerKernel / @ref CLBatchNormalizationLayer + - @ref CLDepthConcatenateLayerKernel / @ref 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 arm_compute::CPPDetectionWindowNonMaximaSuppressionKernel + - @ref CPPDetectionWindowNonMaximaSuppressionKernel - New NEON kernels / functions: - - @ref arm_compute::NEBatchNormalizationLayerKernel / @ref arm_compute::NEBatchNormalizationLayer - - @ref arm_compute::NEDepthConcatenateLayerKernel / @ref arm_compute::NEDepthConcatenateLayer - - @ref arm_compute::NEDirectConvolutionLayerKernel / @ref arm_compute::NEDirectConvolutionLayer - - @ref arm_compute::NELocallyConnectedMatrixMultiplyKernel / @ref arm_compute::NELocallyConnectedLayer - - @ref arm_compute::NEWeightsReshapeKernel / @ref arm_compute::NEConvolutionLayerReshapeWeights + - @ref NEBatchNormalizationLayerKernel / @ref NEBatchNormalizationLayer + - @ref NEDepthConcatenateLayerKernel / @ref NEDepthConcatenateLayer + - @ref NEDirectConvolutionLayerKernel / @ref NEDirectConvolutionLayer + - @ref NELocallyConnectedMatrixMultiplyKernel / @ref NELocallyConnectedLayer + - @ref NEWeightsReshapeKernel / @ref NEConvolutionLayerReshapeWeights v17.05 Public bug fixes release - Various bug fixes @@ -383,77 +397,77 @@ v17.05 Public bug fixes release v17.04 Public bug fixes release The following functions have been ported to use the new accurate padding: - - @ref arm_compute::CLColorConvertKernel - - @ref arm_compute::CLEdgeNonMaxSuppressionKernel - - @ref arm_compute::CLEdgeTraceKernel - - @ref arm_compute::CLGaussianPyramidHorKernel - - @ref arm_compute::CLGaussianPyramidVertKernel - - @ref arm_compute::CLGradientKernel - - @ref arm_compute::NEChannelCombineKernel - - @ref arm_compute::NEFillArrayKernel - - @ref arm_compute::NEGaussianPyramidHorKernel - - @ref arm_compute::NEGaussianPyramidVertKernel - - @ref arm_compute::NEHarrisScoreFP16Kernel - - @ref arm_compute::NEHarrisScoreKernel - - @ref arm_compute::NEHOGDetectorKernel - - @ref arm_compute::NELogits1DMaxKernel - - arm_compute::NELogits1DShiftExpSumKernel - - arm_compute::NELogits1DNormKernel - - @ref arm_compute::NENonMaximaSuppression3x3FP16Kernel - - @ref arm_compute::NENonMaximaSuppression3x3Kernel + - @ref CLColorConvertKernel + - @ref CLEdgeNonMaxSuppressionKernel + - @ref CLEdgeTraceKernel + - @ref CLGaussianPyramidHorKernel + - @ref CLGaussianPyramidVertKernel + - @ref CLGradientKernel + - @ref NEChannelCombineKernel + - @ref NEFillArrayKernel + - @ref NEGaussianPyramidHorKernel + - @ref NEGaussianPyramidVertKernel + - @ref 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: - - @ref arm_compute::CLGEMMLowpMatrixMultiplyKernel / arm_compute::CLGEMMLowp + - @ref CLGEMMLowpMatrixMultiplyKernel / CLGEMMLowp - New NEON kernels / functions: - - @ref arm_compute::NENormalizationLayerKernel / @ref arm_compute::NENormalizationLayer - - @ref arm_compute::NETransposeKernel / @ref arm_compute::NETranspose - - @ref arm_compute::NELogits1DMaxKernel, arm_compute::NELogits1DShiftExpSumKernel, arm_compute::NELogits1DNormKernel / @ref arm_compute::NESoftmaxLayer - - @ref arm_compute::NEIm2ColKernel, @ref arm_compute::NECol2ImKernel, arm_compute::NEConvolutionLayerWeightsReshapeKernel / @ref arm_compute::NEConvolutionLayer - - @ref arm_compute::NEGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::NEFullyConnectedLayer - - @ref arm_compute::NEGEMMLowpMatrixMultiplyKernel / arm_compute::NEGEMMLowp + - @ref NENormalizationLayerKernel / @ref NENormalizationLayer + - @ref NETransposeKernel / @ref NETranspose + - @ref NELogits1DMaxKernel, NELogits1DShiftExpSumKernel, NELogits1DNormKernel / @ref NESoftmaxLayer + - @ref NEIm2ColKernel, @ref NECol2ImKernel, NEConvolutionLayerWeightsReshapeKernel / @ref NEConvolutionLayer + - @ref NEGEMMMatrixAccumulateBiasesKernel / @ref NEFullyConnectedLayer + - @ref NEGEMMLowpMatrixMultiplyKernel / NEGEMMLowp v17.03 Sources preview - New OpenCL kernels / functions: - - @ref arm_compute::CLGradientKernel, @ref arm_compute::CLEdgeNonMaxSuppressionKernel, @ref arm_compute::CLEdgeTraceKernel / @ref arm_compute::CLCannyEdge - - GEMM refactoring + FP16 support: @ref arm_compute::CLGEMMInterleave4x4Kernel, @ref arm_compute::CLGEMMTranspose1xWKernel, @ref arm_compute::CLGEMMMatrixMultiplyKernel, @ref arm_compute::CLGEMMMatrixAdditionKernel / @ref arm_compute::CLGEMM - - @ref arm_compute::CLGEMMMatrixAccumulateBiasesKernel / @ref arm_compute::CLFullyConnectedLayer - - @ref arm_compute::CLTransposeKernel / @ref arm_compute::CLTranspose - - @ref arm_compute::CLLKTrackerInitKernel, @ref arm_compute::CLLKTrackerStage0Kernel, @ref arm_compute::CLLKTrackerStage1Kernel, @ref arm_compute::CLLKTrackerFinalizeKernel / @ref arm_compute::CLOpticalFlow - - @ref arm_compute::CLNormalizationLayerKernel / @ref arm_compute::CLNormalizationLayer - - @ref arm_compute::CLLaplacianPyramid, @ref arm_compute::CLLaplacianReconstruct + - @ref CLGradientKernel, @ref CLEdgeNonMaxSuppressionKernel, @ref CLEdgeTraceKernel / @ref CLCannyEdge + - GEMM refactoring + FP16 support: @ref CLGEMMInterleave4x4Kernel, @ref CLGEMMTranspose1xWKernel, @ref CLGEMMMatrixMultiplyKernel, @ref CLGEMMMatrixAdditionKernel / @ref CLGEMM + - @ref 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 arm_compute::NEActivationLayerKernel / @ref arm_compute::NEActivationLayer - - 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 - - @ref arm_compute::NEPoolingLayerKernel / @ref arm_compute::NEPoolingLayer + - @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: - - @ref arm_compute::CLLogits1DMaxKernel, @ref arm_compute::CLLogits1DShiftExpSumKernel, @ref arm_compute::CLLogits1DNormKernel / @ref arm_compute::CLSoftmaxLayer - - @ref arm_compute::CLPoolingLayerKernel / @ref arm_compute::CLPoolingLayer - - @ref arm_compute::CLIm2ColKernel, @ref arm_compute::CLCol2ImKernel, arm_compute::CLConvolutionLayerWeightsReshapeKernel / @ref arm_compute::CLConvolutionLayer - - @ref arm_compute::CLRemapKernel / @ref arm_compute::CLRemap - - @ref arm_compute::CLGaussianPyramidHorKernel, @ref arm_compute::CLGaussianPyramidVertKernel / @ref arm_compute::CLGaussianPyramid, @ref arm_compute::CLGaussianPyramidHalf, @ref arm_compute::CLGaussianPyramidOrb - - @ref arm_compute::CLMinMaxKernel, @ref arm_compute::CLMinMaxLocationKernel / @ref arm_compute::CLMinMaxLocation - - @ref arm_compute::CLNonLinearFilterKernel / @ref arm_compute::CLNonLinearFilter + - @ref CLLogits1DMaxKernel, @ref 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 arm_compute::NEAccumulateWeightedFP16Kernel - - @ref arm_compute::NEBox3x3FP16Kernel - - @ref arm_compute::NENonMaximaSuppression3x3FP16Kernel + - @ref NEAccumulateWeightedFP16Kernel + - @ref NEBox3x3FP16Kernel + - @ref NENonMaximaSuppression3x3FP16Kernel v17.02 Sources preview - New OpenCL kernels / functions: - - @ref arm_compute::CLActivationLayerKernel / @ref arm_compute::CLActivationLayer - - @ref arm_compute::CLChannelCombineKernel / @ref arm_compute::CLChannelCombine - - @ref arm_compute::CLDerivativeKernel / @ref arm_compute::CLChannelExtract - - @ref arm_compute::CLFastCornersKernel / @ref arm_compute::CLFastCorners - - @ref arm_compute::CLMeanStdDevKernel / @ref arm_compute::CLMeanStdDev + - @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 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 - - @ref arm_compute::NENonLinearFilterKernel / @ref arm_compute::NENonLinearFilter + - 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. @@ -600,7 +614,7 @@ Example: @b cppthreads Build in the C++11 scheduler for NEON. -@sa arm_compute::Scheduler::set +@sa Scheduler::set @subsection S3_2_linux Building for Linux @@ -741,21 +755,35 @@ or LD_LIBRARY_PATH=build ./cl_convolution +@note Examples accept different types of arguments, to find out what they are run the example without any argument and the help will be displayed at the beginning of the run. + +For example: + LD_LIBRARY_PATH=. ./graph_lenet + + ./graph_lenet + + Usage: ./graph_lenet [target] [path_to_data] [batches] + + No data folder provided: using random values + + Test passed + +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_3_android Building for Android For Android, the library was successfully built and tested using Google's standalone toolchains: - - NDK r14 arm-linux-androideabi-4.9 for armv7a (clang++) - - NDK r14 aarch64-linux-android-4.9 for arm64-v8a (clang++) + - clang++ from NDK r16b for armv7a + - clang++ from NDK r16b for arm64-v8a Here is a guide to create your Android standalone toolchains from the NDK -- Download the NDK r14 from here: https://developer.android.com/ndk/downloads/index.html +- Download the NDK r16b from here: https://developer.android.com/ndk/downloads/index.html - Make sure you have Python 2 installed on your machine. - Generate the 32 and/or 64 toolchains by running the following commands: - - $NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-4.9 --stl gnustl --api 21 - $NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-androideabi-4.9 --stl gnustl --api 21 + $NDK/build/tools/make_standalone_toolchain.py --arch arm64 --install-dir $MY_TOOLCHAINS/aarch64-linux-android-ndk-r16b --stl gnustl --api 21 + $NDK/build/tools/make_standalone_toolchain.py --arch arm --install-dir $MY_TOOLCHAINS/arm-linux-android-ndk-r16b --stl gnustl --api 21 @attention Due to some NDK issues make sure you use clang++ & gnustl @@ -843,6 +871,21 @@ And finally to run the example: 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 without any argument and the help will be displayed at the beginning of the run. + +For example: + adb shell /data/local/tmp/graph_lenet + + /data/local/tmp/graph_lenet + + Usage: /data/local/tmp/graph_lenet [target] [path_to_data] [batches] + + No data folder provided: using random values + + Test passed + +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 linaros's latest (gcc-linaro-6.3.1-2017.05) bare metal toolchains: @@ -933,3 +976,4 @@ To cross-compile the stub OpenGLES and EGL libraries simply run: aarch64-linux-gnu-gcc -o libEGL.so -Iinclude/linux opengles-3.1-stubs/EGL.c -fPIC -shared aarch64-linux-gnu-gcc -o libGLESv2.so -Iinclude/linux opengles-3.1-stubs/GLESv2.c -fPIC -shared */ +} // namespace arm_compute diff --git a/docs/01_library.dox b/docs/01_library.dox index 20d057c2c9..e3f673df82 100644 --- a/docs/01_library.dox +++ b/docs/01_library.dox @@ -366,5 +366,22 @@ mm->finalize(); // Finalize memory manager (Object lifetime check conv1.run(); conv2.run(); @endcode + +@section S4_8_opencl_tuner OpenCL Tuner + +OpenCL kernels when dispatched to the GPU take two arguments: +- The Global Workgroup Size (GWS): That's the number of times to run an OpenCL kernel to process all the elements we want to process. +- The Local Workgroup Size (LWS): That's the number of elements we want to run in parallel on a GPU core at a given point in time. + +The LWS can be required by an algorithm (For example if it contains memory barriers or uses local memory) but it can also be used for performance reasons to tweak the performance of a kernel: the execution time of the overall kernel might vary significantly depending on how the GWS is broken down. + +However, there is no universal rule regarding which LWS is best for a given kernel, so instead we created the @ref CLTuner. + +When the @ref CLTuner is enabled ( Target = 2 for the graph examples), the first time an OpenCL kernel is executed the Compute Library will try to run it with a variety of LWS values and will remember which one performed best for subsequent runs. At the end of the run the @ref graph::Graph will try to save these tuning parameters to a file. + +However this process takes quite a lot of time, which is why it cannot be enabled all the time. + +But, when the @ref CLTuner is disabled ( Target = 1 for the graph examples), the @ref graph::Graph will try to reload the file containing the tuning parameters, then for each executed kernel the Compute Library will use the fine tuned LWS if it was present in the file or use a default LWS value if it's not. + */ } // namespace arm_compute diff --git a/docs/Doxyfile b/docs/Doxyfile index 744a9df4c5..6fa2570d84 100644 --- a/docs/Doxyfile +++ b/docs/Doxyfile @@ -38,7 +38,7 @@ PROJECT_NAME = "Compute Library" # could be handy for archiving the generated documentation or if some version # control system is used. -PROJECT_NUMBER = 18.02 +PROJECT_NUMBER = 18.03 # Using the PROJECT_BRIEF tag one can provide an optional one line description # for a project that appears at the top of each page and should give viewer a diff --git a/documentation/00__introduction_8dox.xhtml b/documentation/00__introduction_8dox.xhtml index c049e25bc5..972c157add 100644 --- a/documentation/00__introduction_8dox.xhtml +++ b/documentation/00__introduction_8dox.xhtml @@ -40,7 +40,7 @@
Compute Library -  18.02 +  18.03
@@ -113,17 +113,26 @@ $(document).ready(function(){initNavTree('00__introduction_8dox.xhtml','');});
+
+Namespaces
docs/00_introduction.dox File Reference
+ + + + + +

+Namespaces

 arm_compute
 This file contains all available output stages for GEMMLowp on OpenCL.