You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

execution_tree.cc 13 kB

5 years ago
5 years ago
5 years ago
added python api based on cpp api 1st draft of python iterator Added Cifar10 and Cifar100 pybind port Change pybind to use IR for Skip and Manifest Signed-off-by: alex-yuyue <yue.yu1@huawei.com> DatasetNode as a base for all IR nodes namespace change Fix the namespace issue and make ut tests work Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Add VOCDataset !63 Added RandomDataset * Added RandomDataset add imagefolder ir Pybind switch: CelebA and UT !61 CLUE example with class definition * Merge branch 'python-api' of gitee.com:ezphlow/mindspore into clue_class_pybind * Passing testcases * Added CLUE, not working add ManifestDataset IR Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Update Coco & VOC & TFReader, Update clang-format, Reorder datasets_binding !69 Add Generator and move c_dataset.Iterator to dataset.Iterator * Add GeneratorDataset to c_dataset * Add GeneratorDataset to c_dataset !67 Moving c_datasets and adding sampler wrapper * Need to add create() method in datasets.py * migration from c_dataset to dataset part 1 !71 Fix indent error * Fix indentation error !72 Fix c_api tests cases * Fix c_api tests cases !73 Added CSV Dataset * Added CSVDataset pybind switch: Take and CelebA fixes !75 move c_dataset functionality to datasets * Fixed existing testcases * Added working clue and imagefolder * Added sampler conversion from pybind * Added sampler creation !77 Add Python API tree * Python API tree add minddataset TextFileDataset pybind Rename to skip test_concat.py and test_minddataset_exception.py !80 Add batch IR to python-api branch, most test cases work * staging III * staging, add pybind Enable more c_api take and CelebA tests; delete util_c_api !84 Schema changes in datasets.py * Schema changes !85 Remove input_indexes from sub-classes * remove input_index from each subclass !83 Remove C datasets * Removed c_dataset package * Remove c_datasets !82 pybind switch: shuffle * pybind switch: shuffle !86 Add build_vocab * Add build_vocab Rebase with upstream/master _shuffle conflict BatchNode error !88 Fix rebase problem * fix rebase problem Enable more unit tests; code typo/nit fixes !91 Fix python vocag hang * Fix python vocab hang !89 Added BucketBatchByLength Pybind switch * Added BucketBatchByLength Update and enable more tet_c_api_*.py tests !95 Add BuildSentencePeiceVocab * - Add BuildSentencePeiceVocab !96 Fix more tests * - Fix some tests - Enable more test_c_api_* - Add syncwait !99 pybind switch for device op * pybind switch for device op !93 Add getters to python API * Add getters to python API !101 Validate tree, error if graph * - Add sync wait !103 TFrecord/Random Datasets schema problem * - TfRecord/Random schem aproblem !102 Added filter pybind switch * Added Filter pybind switch !104 Fix num_samples * - TfRecord/Random schem aproblem !105 Fix to_device hang * Fix to_device hang !94 Adds Cache support for CLUE dataset * Added cache for all dataset ops * format change * Added CLUE cache support * Added Cache conversion Add save pybind fix compile err init modify concat_node !107 Fix some tests cases * Fix tests cases Enable and fix more tests !109 pybind switch for get dataset size * pybind_get_dataset_size some check-code fixes for pylint, cpplint and clang-format !113 Add callback * revert * dataset_sz 1 line * fix typo * get callback to work !114 Make Android compile clean * Make Android Compile Clean Fix build issues due to rebase !115 Fix more tests * Fix tests cases * !93 Add getters to python API fix test_profiling.py !116 fix get dataset size * fix get dataset size !117 GetColumnNames pybind switch * Added GetColumnNames pybind switch code-check fixes: clangformat, cppcheck, cpplint, pylint Delete duplicate test_c_api_*.py files; more lint fixes !121 Fix cpp tests * Remove extra call to getNext in cpp tests !122 Fix Schema with Generator * Fix Schema with Generator fix some cases of csv & mindrecord !124 fix tfrecord get_dataset_size and add some UTs * fix tfrecord get dataset size and add some ut for get_dataset_size !125 getter separation * Getter separation !126 Fix sampler.GetNumSamples * Fix sampler.GetNumSampler !127 Assign runtime getter to each get function * Assign runtime getter to each get function Fix compile issues !128 Match master code * Match master code !129 Cleanup DeviceOp/save code * Cleanup ToDevice/Save code !130 Add cache fix * Added cache fix for map and image folder !132 Fix testing team issues * Pass queue_name from python to C++ * Add Schema.from_json !131 Fix Cache op issues and delete de_pipeline * Roll back C++ change * Removed de_pipeline and passing all cache tests. * fixed cache tests !134 Cleanup datasets.py part1 * Cleanup dataset.py part1 !133 Updated validation for SentencePieceVocab.from_dataset * Added type_check for column names in SentencePieceVocab.from_dataset Rebase on master 181120 10:20 fix profiling temporary solution of catching stauts from Node.Build() !141 ToDevice Termination * ToDevice termination pylint fixes !137 Fix test team issues and add some corresponding tests * Fix test team issues and add some corresponding tests !138 TreeGetter changes to use OptPass * Getter changes to use OptPass (Zirui) Rebase fix !143 Fix cpplint issue * Fix cpplint issue pylint fixes in updated testcases !145 Reset exceptions testcase * reset exception test to master !146 Fix Check_Pylint Error * Fix Check_Pylint Error !147 fix android * fix android !148 ToDevice changes * Add ToDevice to the iterator List for cleanup at exit !149 Pylint issue * Add ToDevice to the iterator List for cleanup at exit !150 Pylint 2 * Add ToDevice to the iterator List for cleanup at exit !152 ExecutionTree error * ET destructor error !153 in getter_pass, only remove callback, without deleting map op * getter pass no longer removes map !156 early __del__ of iterator/to_device * early __del__ of iterator !155 Address review comments Eric 1 * Added one liner fix to validators.py * roll back signature fix * lint fix * Eric Address comments 2 * C++ lint fix * Address comments Eric 1 !158 Review rework for dataset bindings - part 1 * Reorder nodes repeat and rename * Review rework for dataset bindings - part 1 !154 Fixing minor problems in the comments (datasets.py, python_tree_consumer.cc, iterators_bindings.cc, and iterators.py) * Fixing minor problems in the comments (datasets.py, python_tree_consum… !157 add replace none * Add replace_none to datasets.py, address comments in tests Trying to resolve copy Override the deepcopy method of deviceop Create_ir_tree method Create_ir_tree method 2 Create_ir_tree method 2 del to_device if already exists del to_device if already exists cache getters shapes and types Added yolov3 relaxation, to be rolled back Get shapes and types together bypass yolo NumWorkers for MapOp revert Yolo revert Thor Print more info Debug code: Update LOG INFO to LOG ERROR do not remove epochctrl for getter pass Remove repeat(1) pritn batch size add log to tree_consumer and device_queue op Revert PR 8744 Signed-off-by: alex-yuyue <yue.yu1@huawei.com> __del__ toDEvice __del__ toDevice2 !165 add ifndef ENABLE_ANDROID to device queue print * Add ifndef ENABLE_ANDROID to device queue print revert some changes !166 getter: get_data_info * getter: get_data_info !168 add back tree print * revert info to warnning in one log * add back the missed print tree log Release GIL in GetDataInfo
5 years ago
added python api based on cpp api 1st draft of python iterator Added Cifar10 and Cifar100 pybind port Change pybind to use IR for Skip and Manifest Signed-off-by: alex-yuyue <yue.yu1@huawei.com> DatasetNode as a base for all IR nodes namespace change Fix the namespace issue and make ut tests work Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Add VOCDataset !63 Added RandomDataset * Added RandomDataset add imagefolder ir Pybind switch: CelebA and UT !61 CLUE example with class definition * Merge branch 'python-api' of gitee.com:ezphlow/mindspore into clue_class_pybind * Passing testcases * Added CLUE, not working add ManifestDataset IR Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Update Coco & VOC & TFReader, Update clang-format, Reorder datasets_binding !69 Add Generator and move c_dataset.Iterator to dataset.Iterator * Add GeneratorDataset to c_dataset * Add GeneratorDataset to c_dataset !67 Moving c_datasets and adding sampler wrapper * Need to add create() method in datasets.py * migration from c_dataset to dataset part 1 !71 Fix indent error * Fix indentation error !72 Fix c_api tests cases * Fix c_api tests cases !73 Added CSV Dataset * Added CSVDataset pybind switch: Take and CelebA fixes !75 move c_dataset functionality to datasets * Fixed existing testcases * Added working clue and imagefolder * Added sampler conversion from pybind * Added sampler creation !77 Add Python API tree * Python API tree add minddataset TextFileDataset pybind Rename to skip test_concat.py and test_minddataset_exception.py !80 Add batch IR to python-api branch, most test cases work * staging III * staging, add pybind Enable more c_api take and CelebA tests; delete util_c_api !84 Schema changes in datasets.py * Schema changes !85 Remove input_indexes from sub-classes * remove input_index from each subclass !83 Remove C datasets * Removed c_dataset package * Remove c_datasets !82 pybind switch: shuffle * pybind switch: shuffle !86 Add build_vocab * Add build_vocab Rebase with upstream/master _shuffle conflict BatchNode error !88 Fix rebase problem * fix rebase problem Enable more unit tests; code typo/nit fixes !91 Fix python vocag hang * Fix python vocab hang !89 Added BucketBatchByLength Pybind switch * Added BucketBatchByLength Update and enable more tet_c_api_*.py tests !95 Add BuildSentencePeiceVocab * - Add BuildSentencePeiceVocab !96 Fix more tests * - Fix some tests - Enable more test_c_api_* - Add syncwait !99 pybind switch for device op * pybind switch for device op !93 Add getters to python API * Add getters to python API !101 Validate tree, error if graph * - Add sync wait !103 TFrecord/Random Datasets schema problem * - TfRecord/Random schem aproblem !102 Added filter pybind switch * Added Filter pybind switch !104 Fix num_samples * - TfRecord/Random schem aproblem !105 Fix to_device hang * Fix to_device hang !94 Adds Cache support for CLUE dataset * Added cache for all dataset ops * format change * Added CLUE cache support * Added Cache conversion Add save pybind fix compile err init modify concat_node !107 Fix some tests cases * Fix tests cases Enable and fix more tests !109 pybind switch for get dataset size * pybind_get_dataset_size some check-code fixes for pylint, cpplint and clang-format !113 Add callback * revert * dataset_sz 1 line * fix typo * get callback to work !114 Make Android compile clean * Make Android Compile Clean Fix build issues due to rebase !115 Fix more tests * Fix tests cases * !93 Add getters to python API fix test_profiling.py !116 fix get dataset size * fix get dataset size !117 GetColumnNames pybind switch * Added GetColumnNames pybind switch code-check fixes: clangformat, cppcheck, cpplint, pylint Delete duplicate test_c_api_*.py files; more lint fixes !121 Fix cpp tests * Remove extra call to getNext in cpp tests !122 Fix Schema with Generator * Fix Schema with Generator fix some cases of csv & mindrecord !124 fix tfrecord get_dataset_size and add some UTs * fix tfrecord get dataset size and add some ut for get_dataset_size !125 getter separation * Getter separation !126 Fix sampler.GetNumSamples * Fix sampler.GetNumSampler !127 Assign runtime getter to each get function * Assign runtime getter to each get function Fix compile issues !128 Match master code * Match master code !129 Cleanup DeviceOp/save code * Cleanup ToDevice/Save code !130 Add cache fix * Added cache fix for map and image folder !132 Fix testing team issues * Pass queue_name from python to C++ * Add Schema.from_json !131 Fix Cache op issues and delete de_pipeline * Roll back C++ change * Removed de_pipeline and passing all cache tests. * fixed cache tests !134 Cleanup datasets.py part1 * Cleanup dataset.py part1 !133 Updated validation for SentencePieceVocab.from_dataset * Added type_check for column names in SentencePieceVocab.from_dataset Rebase on master 181120 10:20 fix profiling temporary solution of catching stauts from Node.Build() !141 ToDevice Termination * ToDevice termination pylint fixes !137 Fix test team issues and add some corresponding tests * Fix test team issues and add some corresponding tests !138 TreeGetter changes to use OptPass * Getter changes to use OptPass (Zirui) Rebase fix !143 Fix cpplint issue * Fix cpplint issue pylint fixes in updated testcases !145 Reset exceptions testcase * reset exception test to master !146 Fix Check_Pylint Error * Fix Check_Pylint Error !147 fix android * fix android !148 ToDevice changes * Add ToDevice to the iterator List for cleanup at exit !149 Pylint issue * Add ToDevice to the iterator List for cleanup at exit !150 Pylint 2 * Add ToDevice to the iterator List for cleanup at exit !152 ExecutionTree error * ET destructor error !153 in getter_pass, only remove callback, without deleting map op * getter pass no longer removes map !156 early __del__ of iterator/to_device * early __del__ of iterator !155 Address review comments Eric 1 * Added one liner fix to validators.py * roll back signature fix * lint fix * Eric Address comments 2 * C++ lint fix * Address comments Eric 1 !158 Review rework for dataset bindings - part 1 * Reorder nodes repeat and rename * Review rework for dataset bindings - part 1 !154 Fixing minor problems in the comments (datasets.py, python_tree_consumer.cc, iterators_bindings.cc, and iterators.py) * Fixing minor problems in the comments (datasets.py, python_tree_consum… !157 add replace none * Add replace_none to datasets.py, address comments in tests Trying to resolve copy Override the deepcopy method of deviceop Create_ir_tree method Create_ir_tree method 2 Create_ir_tree method 2 del to_device if already exists del to_device if already exists cache getters shapes and types Added yolov3 relaxation, to be rolled back Get shapes and types together bypass yolo NumWorkers for MapOp revert Yolo revert Thor Print more info Debug code: Update LOG INFO to LOG ERROR do not remove epochctrl for getter pass Remove repeat(1) pritn batch size add log to tree_consumer and device_queue op Revert PR 8744 Signed-off-by: alex-yuyue <yue.yu1@huawei.com> __del__ toDEvice __del__ toDevice2 !165 add ifndef ENABLE_ANDROID to device queue print * Add ifndef ENABLE_ANDROID to device queue print revert some changes !166 getter: get_data_info * getter: get_data_info !168 add back tree print * revert info to warnning in one log * add back the missed print tree log Release GIL in GetDataInfo
5 years ago
added python api based on cpp api 1st draft of python iterator Added Cifar10 and Cifar100 pybind port Change pybind to use IR for Skip and Manifest Signed-off-by: alex-yuyue <yue.yu1@huawei.com> DatasetNode as a base for all IR nodes namespace change Fix the namespace issue and make ut tests work Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Add VOCDataset !63 Added RandomDataset * Added RandomDataset add imagefolder ir Pybind switch: CelebA and UT !61 CLUE example with class definition * Merge branch 'python-api' of gitee.com:ezphlow/mindspore into clue_class_pybind * Passing testcases * Added CLUE, not working add ManifestDataset IR Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Update Coco & VOC & TFReader, Update clang-format, Reorder datasets_binding !69 Add Generator and move c_dataset.Iterator to dataset.Iterator * Add GeneratorDataset to c_dataset * Add GeneratorDataset to c_dataset !67 Moving c_datasets and adding sampler wrapper * Need to add create() method in datasets.py * migration from c_dataset to dataset part 1 !71 Fix indent error * Fix indentation error !72 Fix c_api tests cases * Fix c_api tests cases !73 Added CSV Dataset * Added CSVDataset pybind switch: Take and CelebA fixes !75 move c_dataset functionality to datasets * Fixed existing testcases * Added working clue and imagefolder * Added sampler conversion from pybind * Added sampler creation !77 Add Python API tree * Python API tree add minddataset TextFileDataset pybind Rename to skip test_concat.py and test_minddataset_exception.py !80 Add batch IR to python-api branch, most test cases work * staging III * staging, add pybind Enable more c_api take and CelebA tests; delete util_c_api !84 Schema changes in datasets.py * Schema changes !85 Remove input_indexes from sub-classes * remove input_index from each subclass !83 Remove C datasets * Removed c_dataset package * Remove c_datasets !82 pybind switch: shuffle * pybind switch: shuffle !86 Add build_vocab * Add build_vocab Rebase with upstream/master _shuffle conflict BatchNode error !88 Fix rebase problem * fix rebase problem Enable more unit tests; code typo/nit fixes !91 Fix python vocag hang * Fix python vocab hang !89 Added BucketBatchByLength Pybind switch * Added BucketBatchByLength Update and enable more tet_c_api_*.py tests !95 Add BuildSentencePeiceVocab * - Add BuildSentencePeiceVocab !96 Fix more tests * - Fix some tests - Enable more test_c_api_* - Add syncwait !99 pybind switch for device op * pybind switch for device op !93 Add getters to python API * Add getters to python API !101 Validate tree, error if graph * - Add sync wait !103 TFrecord/Random Datasets schema problem * - TfRecord/Random schem aproblem !102 Added filter pybind switch * Added Filter pybind switch !104 Fix num_samples * - TfRecord/Random schem aproblem !105 Fix to_device hang * Fix to_device hang !94 Adds Cache support for CLUE dataset * Added cache for all dataset ops * format change * Added CLUE cache support * Added Cache conversion Add save pybind fix compile err init modify concat_node !107 Fix some tests cases * Fix tests cases Enable and fix more tests !109 pybind switch for get dataset size * pybind_get_dataset_size some check-code fixes for pylint, cpplint and clang-format !113 Add callback * revert * dataset_sz 1 line * fix typo * get callback to work !114 Make Android compile clean * Make Android Compile Clean Fix build issues due to rebase !115 Fix more tests * Fix tests cases * !93 Add getters to python API fix test_profiling.py !116 fix get dataset size * fix get dataset size !117 GetColumnNames pybind switch * Added GetColumnNames pybind switch code-check fixes: clangformat, cppcheck, cpplint, pylint Delete duplicate test_c_api_*.py files; more lint fixes !121 Fix cpp tests * Remove extra call to getNext in cpp tests !122 Fix Schema with Generator * Fix Schema with Generator fix some cases of csv & mindrecord !124 fix tfrecord get_dataset_size and add some UTs * fix tfrecord get dataset size and add some ut for get_dataset_size !125 getter separation * Getter separation !126 Fix sampler.GetNumSamples * Fix sampler.GetNumSampler !127 Assign runtime getter to each get function * Assign runtime getter to each get function Fix compile issues !128 Match master code * Match master code !129 Cleanup DeviceOp/save code * Cleanup ToDevice/Save code !130 Add cache fix * Added cache fix for map and image folder !132 Fix testing team issues * Pass queue_name from python to C++ * Add Schema.from_json !131 Fix Cache op issues and delete de_pipeline * Roll back C++ change * Removed de_pipeline and passing all cache tests. * fixed cache tests !134 Cleanup datasets.py part1 * Cleanup dataset.py part1 !133 Updated validation for SentencePieceVocab.from_dataset * Added type_check for column names in SentencePieceVocab.from_dataset Rebase on master 181120 10:20 fix profiling temporary solution of catching stauts from Node.Build() !141 ToDevice Termination * ToDevice termination pylint fixes !137 Fix test team issues and add some corresponding tests * Fix test team issues and add some corresponding tests !138 TreeGetter changes to use OptPass * Getter changes to use OptPass (Zirui) Rebase fix !143 Fix cpplint issue * Fix cpplint issue pylint fixes in updated testcases !145 Reset exceptions testcase * reset exception test to master !146 Fix Check_Pylint Error * Fix Check_Pylint Error !147 fix android * fix android !148 ToDevice changes * Add ToDevice to the iterator List for cleanup at exit !149 Pylint issue * Add ToDevice to the iterator List for cleanup at exit !150 Pylint 2 * Add ToDevice to the iterator List for cleanup at exit !152 ExecutionTree error * ET destructor error !153 in getter_pass, only remove callback, without deleting map op * getter pass no longer removes map !156 early __del__ of iterator/to_device * early __del__ of iterator !155 Address review comments Eric 1 * Added one liner fix to validators.py * roll back signature fix * lint fix * Eric Address comments 2 * C++ lint fix * Address comments Eric 1 !158 Review rework for dataset bindings - part 1 * Reorder nodes repeat and rename * Review rework for dataset bindings - part 1 !154 Fixing minor problems in the comments (datasets.py, python_tree_consumer.cc, iterators_bindings.cc, and iterators.py) * Fixing minor problems in the comments (datasets.py, python_tree_consum… !157 add replace none * Add replace_none to datasets.py, address comments in tests Trying to resolve copy Override the deepcopy method of deviceop Create_ir_tree method Create_ir_tree method 2 Create_ir_tree method 2 del to_device if already exists del to_device if already exists cache getters shapes and types Added yolov3 relaxation, to be rolled back Get shapes and types together bypass yolo NumWorkers for MapOp revert Yolo revert Thor Print more info Debug code: Update LOG INFO to LOG ERROR do not remove epochctrl for getter pass Remove repeat(1) pritn batch size add log to tree_consumer and device_queue op Revert PR 8744 Signed-off-by: alex-yuyue <yue.yu1@huawei.com> __del__ toDEvice __del__ toDevice2 !165 add ifndef ENABLE_ANDROID to device queue print * Add ifndef ENABLE_ANDROID to device queue print revert some changes !166 getter: get_data_info * getter: get_data_info !168 add back tree print * revert info to warnning in one log * add back the missed print tree log Release GIL in GetDataInfo
5 years ago
added python api based on cpp api 1st draft of python iterator Added Cifar10 and Cifar100 pybind port Change pybind to use IR for Skip and Manifest Signed-off-by: alex-yuyue <yue.yu1@huawei.com> DatasetNode as a base for all IR nodes namespace change Fix the namespace issue and make ut tests work Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Add VOCDataset !63 Added RandomDataset * Added RandomDataset add imagefolder ir Pybind switch: CelebA and UT !61 CLUE example with class definition * Merge branch 'python-api' of gitee.com:ezphlow/mindspore into clue_class_pybind * Passing testcases * Added CLUE, not working add ManifestDataset IR Signed-off-by: alex-yuyue <yue.yu1@huawei.com> Update Coco & VOC & TFReader, Update clang-format, Reorder datasets_binding !69 Add Generator and move c_dataset.Iterator to dataset.Iterator * Add GeneratorDataset to c_dataset * Add GeneratorDataset to c_dataset !67 Moving c_datasets and adding sampler wrapper * Need to add create() method in datasets.py * migration from c_dataset to dataset part 1 !71 Fix indent error * Fix indentation error !72 Fix c_api tests cases * Fix c_api tests cases !73 Added CSV Dataset * Added CSVDataset pybind switch: Take and CelebA fixes !75 move c_dataset functionality to datasets * Fixed existing testcases * Added working clue and imagefolder * Added sampler conversion from pybind * Added sampler creation !77 Add Python API tree * Python API tree add minddataset TextFileDataset pybind Rename to skip test_concat.py and test_minddataset_exception.py !80 Add batch IR to python-api branch, most test cases work * staging III * staging, add pybind Enable more c_api take and CelebA tests; delete util_c_api !84 Schema changes in datasets.py * Schema changes !85 Remove input_indexes from sub-classes * remove input_index from each subclass !83 Remove C datasets * Removed c_dataset package * Remove c_datasets !82 pybind switch: shuffle * pybind switch: shuffle !86 Add build_vocab * Add build_vocab Rebase with upstream/master _shuffle conflict BatchNode error !88 Fix rebase problem * fix rebase problem Enable more unit tests; code typo/nit fixes !91 Fix python vocag hang * Fix python vocab hang !89 Added BucketBatchByLength Pybind switch * Added BucketBatchByLength Update and enable more tet_c_api_*.py tests !95 Add BuildSentencePeiceVocab * - Add BuildSentencePeiceVocab !96 Fix more tests * - Fix some tests - Enable more test_c_api_* - Add syncwait !99 pybind switch for device op * pybind switch for device op !93 Add getters to python API * Add getters to python API !101 Validate tree, error if graph * - Add sync wait !103 TFrecord/Random Datasets schema problem * - TfRecord/Random schem aproblem !102 Added filter pybind switch * Added Filter pybind switch !104 Fix num_samples * - TfRecord/Random schem aproblem !105 Fix to_device hang * Fix to_device hang !94 Adds Cache support for CLUE dataset * Added cache for all dataset ops * format change * Added CLUE cache support * Added Cache conversion Add save pybind fix compile err init modify concat_node !107 Fix some tests cases * Fix tests cases Enable and fix more tests !109 pybind switch for get dataset size * pybind_get_dataset_size some check-code fixes for pylint, cpplint and clang-format !113 Add callback * revert * dataset_sz 1 line * fix typo * get callback to work !114 Make Android compile clean * Make Android Compile Clean Fix build issues due to rebase !115 Fix more tests * Fix tests cases * !93 Add getters to python API fix test_profiling.py !116 fix get dataset size * fix get dataset size !117 GetColumnNames pybind switch * Added GetColumnNames pybind switch code-check fixes: clangformat, cppcheck, cpplint, pylint Delete duplicate test_c_api_*.py files; more lint fixes !121 Fix cpp tests * Remove extra call to getNext in cpp tests !122 Fix Schema with Generator * Fix Schema with Generator fix some cases of csv & mindrecord !124 fix tfrecord get_dataset_size and add some UTs * fix tfrecord get dataset size and add some ut for get_dataset_size !125 getter separation * Getter separation !126 Fix sampler.GetNumSamples * Fix sampler.GetNumSampler !127 Assign runtime getter to each get function * Assign runtime getter to each get function Fix compile issues !128 Match master code * Match master code !129 Cleanup DeviceOp/save code * Cleanup ToDevice/Save code !130 Add cache fix * Added cache fix for map and image folder !132 Fix testing team issues * Pass queue_name from python to C++ * Add Schema.from_json !131 Fix Cache op issues and delete de_pipeline * Roll back C++ change * Removed de_pipeline and passing all cache tests. * fixed cache tests !134 Cleanup datasets.py part1 * Cleanup dataset.py part1 !133 Updated validation for SentencePieceVocab.from_dataset * Added type_check for column names in SentencePieceVocab.from_dataset Rebase on master 181120 10:20 fix profiling temporary solution of catching stauts from Node.Build() !141 ToDevice Termination * ToDevice termination pylint fixes !137 Fix test team issues and add some corresponding tests * Fix test team issues and add some corresponding tests !138 TreeGetter changes to use OptPass * Getter changes to use OptPass (Zirui) Rebase fix !143 Fix cpplint issue * Fix cpplint issue pylint fixes in updated testcases !145 Reset exceptions testcase * reset exception test to master !146 Fix Check_Pylint Error * Fix Check_Pylint Error !147 fix android * fix android !148 ToDevice changes * Add ToDevice to the iterator List for cleanup at exit !149 Pylint issue * Add ToDevice to the iterator List for cleanup at exit !150 Pylint 2 * Add ToDevice to the iterator List for cleanup at exit !152 ExecutionTree error * ET destructor error !153 in getter_pass, only remove callback, without deleting map op * getter pass no longer removes map !156 early __del__ of iterator/to_device * early __del__ of iterator !155 Address review comments Eric 1 * Added one liner fix to validators.py * roll back signature fix * lint fix * Eric Address comments 2 * C++ lint fix * Address comments Eric 1 !158 Review rework for dataset bindings - part 1 * Reorder nodes repeat and rename * Review rework for dataset bindings - part 1 !154 Fixing minor problems in the comments (datasets.py, python_tree_consumer.cc, iterators_bindings.cc, and iterators.py) * Fixing minor problems in the comments (datasets.py, python_tree_consum… !157 add replace none * Add replace_none to datasets.py, address comments in tests Trying to resolve copy Override the deepcopy method of deviceop Create_ir_tree method Create_ir_tree method 2 Create_ir_tree method 2 del to_device if already exists del to_device if already exists cache getters shapes and types Added yolov3 relaxation, to be rolled back Get shapes and types together bypass yolo NumWorkers for MapOp revert Yolo revert Thor Print more info Debug code: Update LOG INFO to LOG ERROR do not remove epochctrl for getter pass Remove repeat(1) pritn batch size add log to tree_consumer and device_queue op Revert PR 8744 Signed-off-by: alex-yuyue <yue.yu1@huawei.com> __del__ toDEvice __del__ toDevice2 !165 add ifndef ENABLE_ANDROID to device queue print * Add ifndef ENABLE_ANDROID to device queue print revert some changes !166 getter: get_data_info * getter: get_data_info !168 add back tree print * revert info to warnning in one log * add back the missed print tree log Release GIL in GetDataInfo
5 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365
  1. /**
  2. * Copyright 2019 Huawei Technologies Co., Ltd
  3. *
  4. * Licensed under the Apache License, Version 2.0 (the "License");
  5. * you may not use this file except in compliance with the License.
  6. * You may obtain a copy of the License at
  7. *
  8. * http://www.apache.org/licenses/LICENSE-2.0
  9. *
  10. * Unless required by applicable law or agreed to in writing, software
  11. * distributed under the License is distributed on an "AS IS" BASIS,
  12. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13. * See the License for the specific language governing permissions and
  14. * limitations under the License.
  15. */
  16. #include "minddata/dataset/engine/execution_tree.h"
  17. #include <iostream>
  18. #include <string>
  19. #include <utility>
  20. #include "minddata/dataset/engine/datasetops/dataset_op.h"
  21. #include "minddata/dataset/engine/datasetops/shuffle_op.h"
  22. #include "minddata/dataset/engine/datasetops/device_queue_op.h"
  23. #include "minddata/dataset/util/task_manager.h"
  24. #include "minddata/dataset/engine/opt/pass.h"
  25. #include "minddata/dataset/engine/opt/pre/removal_pass.h"
  26. #ifndef ENABLE_ANDROID
  27. #include "minddata/dataset/engine/opt/pre/cache_transform_pass.h"
  28. #include "minddata/dataset/engine/opt/post/repeat_pass.h"
  29. #include "minddata/dataset/engine/opt/pre/cache_error_pass.h"
  30. #include "mindspore/ccsrc/minddata/dataset/engine/opt/optional/tensor_op_fusion_pass.h"
  31. #endif
  32. #include "minddata/dataset/engine/opt/pre/epoch_injection_pass.h"
  33. #include "minddata/dataset/engine/perf/profiling.h"
  34. #include "minddata/dataset/engine/perf/monitor.h"
  35. namespace mindspore {
  36. namespace dataset {
  37. // Constructor
  38. ExecutionTree::ExecutionTree() : id_count_(0), pre_pass_override_(nullptr) {
  39. tg_ = std::make_unique<TaskGroup>();
  40. tree_state_ = kDeTStateInit;
  41. prepare_flags_ = kDePrepNone;
  42. profiling_manager_ = std::make_unique<ProfilingManager>(this);
  43. optimize_ = common::GetEnv("OPTIMIZE") == "true" ? true : false;
  44. }
  45. // Destructor
  46. ExecutionTree::~ExecutionTree() {
  47. #ifdef ENABLE_TDTQUE
  48. DeviceQueueOp *op = dynamic_cast<DeviceQueueOp *>(root_.get());
  49. if (op != nullptr) {
  50. op->StopWaiting();
  51. }
  52. #endif
  53. (void)tg_->ServiceStop();
  54. }
  55. // Associates a DatasetOp with this tree. This assigns a valid node id to the operator and
  56. // provides it with a link to the tree. A node cannot form any relationships (parent/child) with
  57. // other nodes unless they are associated with the same tree.
  58. Status ExecutionTree::AssociateNode(const std::shared_ptr<DatasetOp> &op) {
  59. // If we are already a part of the tree, no-op
  60. if (op->tree_ == this) {
  61. return Status::OK();
  62. }
  63. if (tree_state_ != kDeTStateInit && tree_state_ != kDeTStateBuilding && tree_state_ != kDeTStatePrepare) {
  64. std::string err_msg =
  65. "Invalid tree state for adding a node. Current state: " + std::to_string(static_cast<int>(tree_state_)) +
  66. " Expected states: " + std::to_string(static_cast<int>(kDeTStateInit)) + " or " +
  67. std::to_string(static_cast<int>(kDeTStateBuilding)) + " or " + std::to_string(static_cast<int>(kDeTStatePrepare));
  68. RETURN_STATUS_UNEXPECTED(err_msg);
  69. }
  70. // Enter the building state if we were not already there
  71. tree_state_ = kDeTStateBuilding;
  72. // Assign an id to the operator
  73. op->set_id(id_count_);
  74. id_count_++;
  75. // Assign our tree into the op so that each op has a link back to the tree
  76. op->set_tree(this);
  77. return Status::OK();
  78. }
  79. // Sets the root node of the tree
  80. Status ExecutionTree::AssignRoot(const std::shared_ptr<DatasetOp> &op) {
  81. // Tree must be in building state before we can assign root to it
  82. if (tree_state_ != kDeTStateBuilding) {
  83. std::string err_msg =
  84. "Invalid tree state for assigning a root node. Current state: " + std::to_string(static_cast<int>(tree_state_)) +
  85. " Expected state: " + std::to_string(static_cast<int>(kDeTStateBuilding));
  86. RETURN_STATUS_UNEXPECTED(err_msg);
  87. }
  88. // If they didn't already call AssociateNode for this node before calling AssignRoot,
  89. // then do so now.
  90. if (op->operator_id_ == DatasetOp::kInvalidOperatorId) {
  91. RETURN_IF_NOT_OK(this->AssociateNode(op));
  92. }
  93. // Then add it as the root.
  94. root_ = op;
  95. return Status::OK();
  96. }
  97. // A print method typically used for debugging
  98. void ExecutionTree::Print(std::ostream &out, const std::shared_ptr<DatasetOp> &op) const {
  99. out << "Execution tree summary:\n"
  100. << "-----------------------\n";
  101. this->PrintNode(out, op == nullptr ? root_ : op, "", true, false);
  102. out << "\nExecution tree operator details:\n"
  103. << "--------------------------------\n";
  104. this->PrintNode(out, op == nullptr ? root_ : op, "", true, true);
  105. }
  106. // A helper functions for doing the recursive printing
  107. void ExecutionTree::PrintNode(std::ostream &out, const std::shared_ptr<DatasetOp> &dataset_op, std::string indent,
  108. bool last, bool detailed) const {
  109. // Decide which printer to use based on detailed arg.
  110. if (!detailed) {
  111. out << indent << "+- " << *dataset_op;
  112. indent += (last ? " " : "| ");
  113. } else {
  114. dataset_op->Print(out, detailed);
  115. }
  116. // Descend to children
  117. for (int32_t i = 0; i < dataset_op->child_.size(); ++i) {
  118. this->PrintNode(out, dataset_op->child_[i], indent, (i == (dataset_op->child_.size() - 1)), detailed);
  119. }
  120. }
  121. // Start the execution of the tree
  122. Status ExecutionTree::Launch() {
  123. // opencv limit too many threads
  124. #ifndef ENABLE_ANDROID
  125. #if !defined(_WIN32) && !defined(_WIN64)
  126. int32_t thread_num = get_nprocs();
  127. if (thread_num == 0) {
  128. std::string err_msg = "Invalid thread number.";
  129. RETURN_STATUS_UNEXPECTED(err_msg);
  130. }
  131. if (thread_num > 8)
  132. cv::setNumThreads(8);
  133. else
  134. cv::setNumThreads(thread_num);
  135. #endif
  136. #endif
  137. // Tree must be built and prepared before it can be launched!
  138. if (tree_state_ != kDeTStateReady) {
  139. std::string err_msg =
  140. "Invalid tree state for launching tree. Current state: " + std::to_string(static_cast<int>(tree_state_)) +
  141. " Expected state: " + std::to_string(static_cast<int>(kDeTStateReady));
  142. RETURN_STATUS_UNEXPECTED(err_msg);
  143. }
  144. // Profiling infrastructures need to be initialized before Op launching
  145. if (profiling_manager_->IsProfilingEnable()) {
  146. // Setup profiling manager
  147. RETURN_IF_NOT_OK(profiling_manager_->Initialize());
  148. // Launch Monitor Thread
  149. RETURN_IF_NOT_OK(profiling_manager_->LaunchMonitor());
  150. }
  151. std::ostringstream ss;
  152. ss << *this;
  153. MS_LOG(DEBUG) << "Printing the tree before launch tasks:\n" << ss.str();
  154. for (auto itr = this->begin(); itr != this->end(); ++itr) {
  155. // An inlined operator is one that has an output connector size of 0, and it does not
  156. // require a thread to execute. Instead, the work of this operator is executed inlined
  157. // from the tree node directly above it (or in the case of a root node, it runs from within
  158. // the launching tree/user thread. Do not exec any thread for an inlined op.
  159. itr->state_ = DatasetOp::OpState::kDeOpRunning;
  160. if (!itr->inlined()) {
  161. RETURN_IF_NOT_OK(tg_->CreateAsyncTask(itr->NameWithID(), std::ref(*itr)));
  162. // Set the state of the Operator as running. This only matters in Leaf ops, CacheOp and TakeOp
  163. }
  164. }
  165. tree_state_ = kDeTStateExecuting;
  166. return Status::OK();
  167. }
  168. // A function that traverse the tree in postorder then save the results in nodes
  169. void ExecutionTree::Iterator::PostOrderTraverse(const std::shared_ptr<DatasetOp> &node) {
  170. if (node == nullptr) {
  171. return;
  172. }
  173. for (int32_t i = 0; i < node->child_.size(); ++i) {
  174. PostOrderTraverse(node->child_[i]);
  175. }
  176. nodes_.push_back(node);
  177. }
  178. ExecutionTree::Iterator::Iterator(const std::shared_ptr<DatasetOp> &root) : ind_(0) {
  179. // post-order traverse the tree, if root is null, it return
  180. PostOrderTraverse(root);
  181. nodes_.emplace_back(nullptr);
  182. }
  183. // Given the number of workers, launches the worker entry function for each. Essentially a
  184. // wrapper for the TaskGroup handling that is stored inside the execution tree.
  185. Status ExecutionTree::LaunchWorkers(int32_t num_workers, std::function<Status(uint32_t)> func, std::string name) {
  186. // Launch the workers
  187. for (int32_t i = 0; i < num_workers; ++i) {
  188. RETURN_IF_NOT_OK(tg_->CreateAsyncTask(name, std::bind(func, i)));
  189. }
  190. return Status::OK();
  191. }
  192. // The driver of the prepare phase of the execution tree.
  193. // Prepare phase consists of three sub phases
  194. //
  195. // 1. PrepareTreePreAction()
  196. // Compulsory transformation/action pre optimization.
  197. // For example, CacheOp Insertion
  198. //
  199. // 2. Optimize()
  200. // Optimization transformation/action, optional
  201. // For example, MapOp Fusion
  202. //
  203. // 3. PrepareTreePostAction()
  204. // Compulsory transformation/action post optimization.
  205. // For example, repeatOp inlining
  206. //
  207. // @return Status - The error code return
  208. Status ExecutionTree::Prepare(int32_t num_epochs) {
  209. num_epochs_ = num_epochs;
  210. // Pre optimization compulsory transformation
  211. RETURN_IF_NOT_OK(this->PrepareTreePreAction());
  212. // If optional optimizations are enabled
  213. if (optimize_) {
  214. RETURN_IF_NOT_OK(this->Optimize());
  215. }
  216. // Post optimization compulsory transformation
  217. RETURN_IF_NOT_OK(this->PrepareTreePostAction());
  218. // Existing transformation implementation, will be removed later
  219. RETURN_IF_NOT_OK(this->PrepareDeprecated());
  220. return Status::OK();
  221. }
  222. Status ExecutionTree::PrepareTreePreAction() {
  223. bool modified = false;
  224. std::vector<std::unique_ptr<Pass>> pre_actions;
  225. // Construct pre actions
  226. #ifndef ENABLE_ANDROID
  227. pre_actions.push_back(std::make_unique<CacheErrorPass>());
  228. #endif
  229. pre_actions.push_back(std::make_unique<EpochInjectionPass>());
  230. pre_actions.push_back(std::make_unique<RemovalPass>());
  231. #ifndef ENABLE_ANDROID
  232. pre_actions.push_back(std::make_unique<CacheTransformPass>());
  233. #endif
  234. // this offers a way to override the preset optimization pass with customized ones
  235. // this is used when certain nodes are removed for tree getters
  236. if (pre_pass_override_) {
  237. MS_LOG(INFO) << "Default pre optimization passes is being overridden,"
  238. << " number of passes before the override:" << pre_actions.size() << ".";
  239. pre_actions = pre_pass_override_(std::move(pre_actions));
  240. }
  241. MS_LOG(INFO) << "Running " << pre_actions.size() << " pre pass loops.";
  242. // Apply pre action passes
  243. for (auto &pass : pre_actions) {
  244. RETURN_IF_NOT_OK(pass->Run(this, &modified));
  245. }
  246. MS_LOG(INFO) << "Pre passes complete.";
  247. return Status::OK();
  248. }
  249. Status ExecutionTree::PrepareTreePostAction() {
  250. // The tree is ready to be prepared.
  251. tree_state_ = kDeTStatePrepare;
  252. bool modified = false;
  253. OptPass post_actions;
  254. // Construct pre actions
  255. MS_LOG(INFO) << "Running post pass loops.";
  256. #ifndef ENABLE_ANDROID
  257. post_actions.push_back(std::make_unique<RepeatPass>());
  258. #endif
  259. // Apply post action passes
  260. for (auto &pass : post_actions) {
  261. RETURN_IF_NOT_OK(pass->Run(this, &modified));
  262. }
  263. MS_LOG(INFO) << "Post passes complete.";
  264. return Status::OK();
  265. }
  266. Status ExecutionTree::Optimize() {
  267. // Vector of optimizations, currently only 1, add more as necessary
  268. OptPass optimizations;
  269. #ifndef ENABLE_ANDROID
  270. optimizations.push_back(std::make_unique<TensorOpFusionPass>());
  271. #endif
  272. // vector of flags for each optimization
  273. std::vector<bool> modified(optimizations.size(), false);
  274. for (auto i = 0; i < optimizations.size(); i++) {
  275. auto m = false;
  276. optimizations[i]->Run(this, &m);
  277. modified[i] = m;
  278. }
  279. return Status::OK();
  280. }
  281. // The driver of the prepare phase of the execution tree. The prepare phase will recursively
  282. // walk the tree to perform modifications to the tree or specific nodes within the tree to get
  283. // it ready for execution.
  284. //
  285. // This driver is deprecated.
  286. Status ExecutionTree::PrepareDeprecated() {
  287. // Tree must be in pending prepare state before we can assign root to it
  288. if (tree_state_ != kDeTStatePrepare) {
  289. std::string err_msg =
  290. "Invalid tree state for preparing the tree. Current state: " + std::to_string(static_cast<int>(tree_state_)) +
  291. " Expected state: " + std::to_string(static_cast<int>(kDeTStatePrepare));
  292. RETURN_STATUS_UNEXPECTED(err_msg);
  293. }
  294. if (root_ == nullptr) {
  295. RETURN_STATUS_UNEXPECTED("Please assign one operator as the root of this tree.");
  296. }
  297. // Start the recursive prepare
  298. RETURN_IF_NOT_OK(this->PrepareNode(root_));
  299. tree_state_ = kDeTStateReady;
  300. return Status::OK();
  301. }
  302. // Recursive function used during prepare phase to visit a node and drive any pre- and post-
  303. // node actions during a tree walk.
  304. Status ExecutionTree::PrepareNode(const std::shared_ptr<DatasetOp> &dataset_op) {
  305. // execute PreAction
  306. RETURN_IF_NOT_OK(dataset_op->PrepareNodePreAction());
  307. // Before going down into children, make any prepare flags updates based on this operator.
  308. uint32_t op_prep_flags = dataset_op->PrepareFlags();
  309. BitSet(&prepare_flags_, op_prep_flags);
  310. // Now, descend to children
  311. for (const auto &i : dataset_op->child_) {
  312. RETURN_IF_NOT_OK(this->PrepareNode(i));
  313. }
  314. // No more children, now we execute any prepare actions before going back up the
  315. // the tree on recursive function
  316. RETURN_IF_NOT_OK(dataset_op->PrepareNodePostAction());
  317. // Then clear the flags from this op now that we have prepared it.
  318. BitClear(&prepare_flags_, op_prep_flags);
  319. return Status::OK();
  320. }
  321. } // namespace dataset
  322. } // namespace mindspore