Optimize CPrecisionRecallLayer #533
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com>
yekatkov
pushed a commit
to yekatkov/neoml
that referenced
this pull request
Jan 27, 2022
Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru>
yekatkov
added a commit
that referenced
this pull request
Feb 8, 2022
* Fix arguments const-ness in ME convoutions (#508) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Add sample of differential evolution (#512) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix comments in diff evolution sample (#513) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Generate classifiable data in CRandomProblemImpl (#514) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix memory leak LAMB solver test (#516) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix MKL's condition in CCpuMathEngine::VectorPower (#517) * Fix MKL's condition in CCpuMathEngine::VectorPower Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix Linux build Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix CUDA warnings (#518) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Distributed exception (#519) * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * assert if no openmp Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * fix nccl DllLoader Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * fix nccl DllLoader Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * inc cuda build Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix NeoOnnx compilation on some of MSVC compilers (#521) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Vulkan math engine (Android build) (#522) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use overflow-safe softplus to CBinaryFocalLossLayer (#523) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Android build (#524) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Also use __ANDROID__ macro for android detection (#525) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix integer detection in CCpuMathEngine::VectorPower for x86 (#526) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Mark CDnn::DeleteLayerImpl as final (#527) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove duplicate Platforms.h (#528) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove ternary operator over constant value (#529) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove all mentions of GetVector from interfaces (#530) * Remove all mentions of GetVector from interfaces Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Add missing override Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.29.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Upgrade NeoMLTest version (#531) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Clarify blob dimensions in C++ comments for CFullyConnectedLayer and CBaseConvLayer (#532) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Optimize CPrecisionRecallLayer (#533) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Update IsoDataClustering.cpp (#467) It is possible to simplify the condition for an equivalent one. Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.30.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix build (#534) * Fix build Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix warning Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.31.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Using alias templates (#535) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Distributed training (#538) * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * save distributed net Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Properly using of `override` keyword (#537) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> * Properly using of `override` keyword Signed-off-by: lordnn <lordnn@yahoo.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Move to c++14 (#539) * Move to Android NDK r23 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Allow PackageName_ROOT environment variables for CMake Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Moving to c++14 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Disable link_openmp NeoMathEngineAvx on Darwin Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix GCC/Clang compilation with FineObjects Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Java wrapper compilation (#542) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.32.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.33.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.34.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Switch to Android NDK 23 and Gradle 6.1.1 (plugin 4.0.1) (#543) * Upgrade gradle to 6.1.1 (plugin 4.0.1) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Upgrade other projects to gradle 6.1.1 (plugin 4.0.1) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Move to NDK 23 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added some const-ness. Use `log1p` if possible. Some cleanup. (#540) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> * Properly using of `override` keyword Signed-off-by: lordnn <lordnn@yahoo.com> * Added some const-ness. Use `log1p` is possible. Some cleanup. Signed-off-by: lordnn <lordnn@yahoo.com> * Revert `log1p` for Metal Signed-off-by: lordnn <lordnn@yahoo.com> * Added `explicit` for `CMultiThreadDistributedCommunicator` Ctor Signed-off-by: lordnn <lordnn@yahoo.com> * More const-ness. Some variables scope reduced. Signed-off-by: lordnn <lordnn@yahoo.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Upgrade NeoMLTest version (#544) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 1.0.45.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added mutex for function initialization guard. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Code refactoring. Removed duplicated code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added Sigmoid and Exp functions. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Big refactoring. A lot of common codes were moved to template functions. Xbyak::Generator functions was redefined in order to operate with std::vectors in sequence. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use initYmmVecRange everywhere. Insert code of calculation of exponent directly into sigmoid code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm works Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * FastLstm was moved to class CLstmLayer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Comments refactoring. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Restore right version. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed GPU builds. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build warning. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Android build (#524) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 1.0.45.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added mutex for function initialization guard. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Code refactoring. Removed duplicated code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added Sigmoid and Exp functions. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Big refactoring. A lot of common codes were moved to template functions. Xbyak::Generator functions was redefined in order to operate with std::vectors in sequence. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use initYmmVecRange everywhere. Insert code of calculation of exponent directly into sigmoid code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm works Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * FastLstm was moved to class CLstmLayer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Comments refactoring. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Restore right version. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed GPU builds. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build warning. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed signoff. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed PR. Disable FastLstm for backward and learning Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed PR comments Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Replace OMP functions to nonOMP Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed IOS build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Refactor LstmDesc reallocation Signed-off-by: Yury Katkov <ykatkov@yandex.ru> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: valinurovdenis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Co-authored-by: BuildTech <BuildTech@abbyy.com> Co-authored-by: Timur <35902086+TimurIskahov@users.noreply.github.com> Co-authored-by: lordnn <lordnn@yahoo.com>
nismohl2
pushed a commit
to nismohl2/neoml
that referenced
this pull request
Apr 19, 2022
* Fix arguments const-ness in ME convoutions (neoml-lib#508) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Add sample of differential evolution (neoml-lib#512) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix comments in diff evolution sample (neoml-lib#513) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Generate classifiable data in CRandomProblemImpl (neoml-lib#514) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix memory leak LAMB solver test (neoml-lib#516) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix MKL's condition in CCpuMathEngine::VectorPower (neoml-lib#517) * Fix MKL's condition in CCpuMathEngine::VectorPower Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix Linux build Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix CUDA warnings (neoml-lib#518) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Distributed exception (neoml-lib#519) * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed exceptions Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * openmp threads Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * assert if no openmp Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * fix nccl DllLoader Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * fix nccl DllLoader Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * inc cuda build Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix NeoOnnx compilation on some of MSVC compilers (neoml-lib#521) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Vulkan math engine (Android build) (neoml-lib#522) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use overflow-safe softplus to CBinaryFocalLossLayer (neoml-lib#523) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Android build (neoml-lib#524) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Also use __ANDROID__ macro for android detection (neoml-lib#525) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix integer detection in CCpuMathEngine::VectorPower for x86 (neoml-lib#526) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Mark CDnn::DeleteLayerImpl as final (neoml-lib#527) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove duplicate Platforms.h (neoml-lib#528) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove ternary operator over constant value (neoml-lib#529) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Remove all mentions of GetVector from interfaces (neoml-lib#530) * Remove all mentions of GetVector from interfaces Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Add missing override Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.29.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Upgrade NeoMLTest version (neoml-lib#531) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Clarify blob dimensions in C++ comments for CFullyConnectedLayer and CBaseConvLayer (neoml-lib#532) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Optimize CPrecisionRecallLayer (neoml-lib#533) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Update IsoDataClustering.cpp (neoml-lib#467) It is possible to simplify the condition for an equivalent one. Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.30.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix build (neoml-lib#534) * Fix build Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix warning Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.31.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Using alias templates (neoml-lib#535) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Distributed training (neoml-lib#538) * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * distributed init Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * python wrapper for DistributedTraining Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> * save distributed net Signed-off-by: Valinurov Denis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Properly using of `override` keyword (neoml-lib#537) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> * Properly using of `override` keyword Signed-off-by: lordnn <lordnn@yahoo.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Move to c++14 (neoml-lib#539) * Move to Android NDK r23 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Allow PackageName_ROOT environment variables for CMake Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Moving to c++14 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Disable link_openmp NeoMathEngineAvx on Darwin Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Fix GCC/Clang compilation with FineObjects Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Java wrapper compilation (neoml-lib#542) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.32.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.33.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 2.0.34.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Switch to Android NDK 23 and Gradle 6.1.1 (plugin 4.0.1) (neoml-lib#543) * Upgrade gradle to 6.1.1 (plugin 4.0.1) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Upgrade other projects to gradle 6.1.1 (plugin 4.0.1) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> * Move to NDK 23 Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added some const-ness. Use `log1p` if possible. Some cleanup. (neoml-lib#540) * Using alias templates Signed-off-by: lordnn <lordnn@yahoo.com> * More aliases Signed-off-by: lordnn <lordnn@yahoo.com> * Properly using of `override` keyword Signed-off-by: lordnn <lordnn@yahoo.com> * Added some const-ness. Use `log1p` is possible. Some cleanup. Signed-off-by: lordnn <lordnn@yahoo.com> * Revert `log1p` for Metal Signed-off-by: lordnn <lordnn@yahoo.com> * Added `explicit` for `CMultiThreadDistributedCommunicator` Ctor Signed-off-by: lordnn <lordnn@yahoo.com> * More const-ness. Some variables scope reduced. Signed-off-by: lordnn <lordnn@yahoo.com> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Upgrade NeoMLTest version (neoml-lib#544) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 1.0.45.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added mutex for function initialization guard. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Code refactoring. Removed duplicated code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added Sigmoid and Exp functions. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Big refactoring. A lot of common codes were moved to template functions. Xbyak::Generator functions was redefined in order to operate with std::vectors in sequence. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use initYmmVecRange everywhere. Insert code of calculation of exponent directly into sigmoid code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm works Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * FastLstm was moved to class CLstmLayer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Comments refactoring. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Restore right version. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed GPU builds. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build warning. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix warning in Android build (neoml-lib#524) Signed-off-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Build Framework (NeoML-master 1.0.45.0): Incrementing version number. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added mutex for function initialization guard. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Code refactoring. Removed duplicated code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added Sigmoid and Exp functions. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Big refactoring. A lot of common codes were moved to template functions. Xbyak::Generator functions was redefined in order to operate with std::vectors in sequence. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Use initYmmVecRange everywhere. Insert code of calculation of exponent directly into sigmoid code. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm works Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * FastLstm was moved to class CLstmLayer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Comments refactoring. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Copied FastLstm from Lstm Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added fast lstm layer. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * CFastLstm MVP (not tested yet) Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * First successful build. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added reverse mode. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Lstm was moved into MathEngine. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added tanh function and LSTM Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Added implementation of JIT tanh Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Before debugging Lstm jit function Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Moved initialization code of fast ltsm into reshape. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Non-simd fastLstm became multithread. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed Linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Restore right version. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed GPU builds. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed build warning. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed signoff. Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed PR. Disable FastLstm for backward and learning Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fix linux build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed PR comments Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Replace OMP functions to nonOMP Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Fixed IOS build Signed-off-by: Yury Katkov <ykatkov@yandex.ru> * Refactor LstmDesc reallocation Signed-off-by: Yury Katkov <ykatkov@yandex.ru> Co-authored-by: Valeriy Fedyunin <valery.fedyunin@abbyy.com> Co-authored-by: valinurovdenis <denis.valinurow@yandex.ru> Co-authored-by: Stanislav Angeliuk <59917951+SAngeliuk@users.noreply.github.com> Co-authored-by: BuildTech <BuildTech@abbyy.com> Co-authored-by: Timur <35902086+TimurIskahov@users.noreply.github.com> Co-authored-by: lordnn <lordnn@yahoo.com> Signed-off-by: Michael Zatsepin <michael.zatsepin@abbyy.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fix #456
Signed-off-by: Valeriy Fedyunin valery.fedyunin@abbyy.com