[新功能] 自HomePod 1.0以来pytorch的更新 about mlab

[gemfield]

• 3551bd31bef401a08bf407cfe466814f1fbe250c: [PyTorch] Lite interpreter with a backend delegate
• a0d977610421823619ca376a3f51457bf4c0be82: [JIT] Include conv3d in the conv-add-relu fusion
• NNAPI 系列
• 4757d4c077cb4d47793aae2fb89637bf8d3c0213: Don't allocate result Tensors in out xxxxxxx
• 34b46359e3cdca01f460fe8f4ce894868a2b4191: fix IValue Fix forwarding/move bug

[gemfield]

• 10abbb812a47bfabba4905b50dde02a511191522: Support tensor subclasses in Torchscript

[gemfield]

• 2564c0c88974bdfc13321c3d4fdffab454302c81: avoid CPU std::copysign segfault when compiling on arm64
• 305abde976d232494c6b5252111ce15523511ffe: Fix nvcc warnings
• 42486963b23868d5b423165aa56de2e5547d841e: Integrate NNC conv2d with fuser
• d2784c233bfc57a1d836d961694bcc8ec4ed45e4: Partially migrate sort from THC to ATen, replace the thrust path with cub
• bbd2b1bd3ca3e84901afd336194024fca5c44b1c: [quant][graphmode][fx] Add shape to nontensor op list
• torch/elastic stufff
• NNC stuff

[gemfield]

• 84a7ab250bd1db6ff72d8b0e01f046e9955da65f: Optimize constructing tensors from external data
• c6d9ca0c2b4bb124a902db00393372c0fc12b190: [reland]Replace AutoNonVariableTypeMode with InferenceMode in static runtime.
• 24f9a446c9ddb9442e41f254d0884d417e7fd76a: Fix wrong detection of depthwise conv on neon
• aceceb3d5c4f8b0632c01b75d6c51c08609cb131: Reland #50999 (Added pow() on CPU for float16 & bfloat16)
• 1688a5d31a740fe502232295ec22e49d376e4f2d: Cleanup since FEATURE_TORCH_MOBILE is always true.
• 88c06d9dfce27f49bc7cf0c308ba1cb0dabfc5b9: Add cuda device synchronization support in JIT
• f8d331b33ba6ebea20aee62f02dafe3ac9b69ea5: PyTorch Execution Graph Observers
• 81f181567a85dda92ba2ea24e0448ddebb9ea666: Add USE_MAGMA build flag
• 75b6644a4cc7d9c6848588f1804f98e46cc64ce1: Add USE_NUMPY define only if PyTorch is compiled with Numpy
• 9d3d169d2d1507aec7cca3ab67abb45f0059cfa0: Implement hardswish/hardsigmoid on MKLDNN tensors
• f26a6cb372569b957318c7f2a44c92428cc9686b: [quantization] Fix deepcopy on quantized ConvNd

[gemfield]

• 2c9972facf Tao Xu Mon Apr 19 13:51:37 2021 -0700 [iOS GPU][Kernel] Implement transpose in Metal shaders (#54522)
• 94406f77f6 Jerry Zhang Mon Apr 19 21:36:01 2021 -0700 [quant][graphmode][fx] Add support for keeping output quantized for list and dict (#56391)
• 1f0223d6bb M.L. Croci Thu Apr 22 07:42:11 2021 -0700 Fix bug in gaussian_nll_loss (#56469)
• c37095760d Liang Luo Fri Apr 23 14:11:09 2021 -0700 [torch distributed] Implementing all_gather_base (#56315)
• 58fcf77712 Ivan Yashchuk Sun Apr 25 01:16:04 2021 -0700 Port CPU torch.geqrf to ATen (#56249)
• 267b554b6f Peter Bell Sun Apr 25 20:27:25 2021 -0700 fx: Fix type_matches for Optional[List[int]] arguments (#56790)
• 7b31ba4708 Xiao Wang Sun Apr 25 22:50:21 2021 -0700 Fix cudnn ctc loss backward (#56639)

[gemfield]

• f84f2063b4 Ivan Yashchuk Mon Apr 26 09:47:13 2021 -0700 Port CUDA torch.geqrf to ATen (#56251)
• d221be6fb4 Tao Xu Mon Apr 26 16:33:19 2021 -0700 [iOS GPU] Use thread buffer to store indices for transpose (#56706)
• 502c58ad84 Vasiliy Kuznetsov Mon Apr 26 16:58:35 2021 -0700 ns for fx:xxxxxxxxxx
• e909ad2dc4 Ansha Yu Mon Apr 26 17:25:26 2021 -0700 [static runtime] binding for aten::argmin_out (#56638)

[gemfield]

• dc8a8cea79 Pritam Damania Mon Apr 26 23:06:55 2021 -0700 Move caffe2 signal_handler to c10. (#56717)
• ab1457ad14 Edward Yang Tue Apr 27 07:33:43 2021 -0700 Remove C++17 only optional include (#56782)
• 5b3c0ae563 Shen Li Tue Apr 27 12:27:44 2021 -0700 Use a FutureFactoryRegistry to allow libtorch_cpu files to create CUDAFuture (#56984)
• 522dca4ab0 Eddie Yan Tue Apr 27 14:47:05 2021 -0700 Port topk from THC to ATen, migrate most of sort as well (#55392)
• fda8561944 Heitor Schueroff Thu Apr 29 08:11:10 2021 -0700 Adding vector_norm to the C++ API (#57055)
• 9486fc3229 Chen Lai Thu Apr 29 10:06:38 2021 -0700 [PyTorch][Edge] share readArchiveAndTensors between mobile and jit (#57098)
• e31b67f550 Michael Suo Thu Apr 29 22:28:08 2021 -0700 [torch/deploy] opt torch/csrc/depoy into autofromatting
• 731cc472c5 leslie-fang-intel Fri Apr 30 08:45:08 2021 -0700 refactor autocast to be extensible for devices (#57104)
• e62cdae469 Edvard Ghazaryan Fri Apr 30 10:48:38 2021 -0700 Static Runtime support for aten::matmul (#57291)
• f54aa85a6c Ivan Yashchuk Fri Apr 30 11:07:31 2021 -0700 Fix MAGMA qr for empty batched inputs (#56257)
• 3a777b6792 Scott Wolchok Fri Apr 30 11:19:41 2021 -0700 [PyTorch] Optimize intrusive_ptr(TTarget*) ctor (pybind) (#57053)
• c1a442248b Bin Bao Fri Apr 30 15:26:13 2021 -0700 [JIT] Enable conv-add-relu fusion as a part of frozen graph optimization (#56580)
• 293830bc19 Sameer Deshmukh Fri Apr 30 15:51:37 2021 -0700 Fix min() and max() for empty tensors (#52565)
• 3a5f85465b Jongsoo Park Fri Apr 30 16:34:48 2021 -0700 [pytorch] fewer cuda sync in unique by using cub instead of thrust (#57323)
• 47e9ec401a Horace He Fri Apr 30 17:33:39 2021 -0700 [nnc] ported some more ops + added vectors to argvalue (#56766)
• bd347012ec Mustafa Bal Fri Apr 30 17:50:40 2021 -0700 Added sm_75 support for CI Xenial CUDA 11.1 cuDNN 8 builds (#57320)
• 3c4d57c18b Jiakai Liu Fri Apr 30 19:06:08 2021 -0700 [pytorch][nnc] update external functions for mobile build (#56850)
• 0787d781c5 Richard Zou Wed May 5 07:35:04 2021 -0700 Fix compatibility problem with LSTMs and torch.save (#57558)
• a948e279ac Rohan Varma Wed May 5 10:12:20 2021 -0700 [c10d] Profiler support for nccl p2p collectives (#56427)
• 28cd04ea64 Akshit Khurana Wed May 5 13:51:59 2021 -0700 NNAPI: add flexible size support for conv2d (#57561)
• c4bb6a5781 Akshit Khurana Wed May 5 13:51:59 2021 -0700 NNAPI: flex size support for upsample_nearest2d op (#57563)
• 56211524a7 Horace He Wed May 5 17:08:16 2021 -0700 [NNC] ported over sum and softmax to new scheme (#56775)
• c27428b5e9 Horace He Wed May 5 20:53:25 2021 -0700 [nnc] ported conv2d lowering over (#56875)
• 86b061c80e Horace He Wed May 5 20:53:25 2021 -0700 [FX] Changes in order to move python key out of tree (#57427)
• ba78bf1363 Yi Huang (Symphony) Wed May 5 21:40:07 2021 -0700 [standaloneRunner] fix another GIL mutithreading issue exposed by torch::jit::toIValue() (#57688)
• 7870450706 Scott Wolchok Thu May 6 00:12:18 2021 -0700 [PyTorch] Use c10::ThreadLocal instead thread_local in record_function.cpp for specific GLIBCXX on Android (#57689)
• 95fbc158d4 Raghavan Raman Thu May 6 01:05:49 2021 -0700 [NNC] Add a method to compute conv without bias (#57512)
• 1f178de800 Raghavan Raman Thu May 6 01:05:49 2021 -0700 [NNC] Add support for computing conv with dynamic shapes (#57514)
• d82333e92a Jiakai Liu Thu May 6 03:22:59 2021 -0700 [pytorch][nnc] protocol classes to persist the context for compiled functions (#56851);
• b4a098f1fb Jiakai Liu Thu May 6 03:22:59 2021 -0700 [pytorch][nnc] mobile nnc backend skeleton (#56852);
• 59d794b2c3 Ivan Yashchuk Thu May 6 04:43:10 2021 -0700 Port CPU torch.ormqr to ATen (#57315);

[gemfield]

• aedcff7275 Rong Rong (AI Infra) Thu May 6 15:59:50 2021 -0700 fix codegen for lite_interpreter (#57761)
• 241c2f4496 Elias Ellison Thu May 6 16:03:30 2021 -0700 Add Gelu To NNC (#57753)
• 8b38458011 Zhengxu Chen Thu May 6 16:42:41 2021 -0700 [jit] Break interpreter.cpp into smaller files. (#56546)
• e5e095cbe4 Can Balioglu Thu May 6 19:50:38 2021 -0700 [torch/elastic] Rename etcd-/c10d-experimental to etcd-v2 and c10d (#57764)
• f2fdb61e2d Tao Xu Thu May 6 23:13:17 2021 -0700 [iOS GPU][Perf][1/n] Use aten::contiguous instead of permuting weights manually (#57664)
• 45012da298 Luca Wehrstedt Fri May 7 03:56:50 2021 -0700 Migrate from shared_ptr to intrusive_ptr for Future (#57636)
• 2901d2e694 Vasiliy Kuznetsov Fri May 7 08:39:37 2021 -0700 make quantizeable MHA work with torch.jit.script (#57774)

[gemfield]

• 5044d9dc51 Charles David Hernandez Fri May 7 12:22:54 2021 -0700 Fixing quantize_per_tensor on cuda (#57703)
• 96fce78ac4 Sicheng Stephen Jia Fri May 7 13:13:11 2021 -0700 [Vulkan] Add -Os flag to shader compilation (#57199)
• 8639fd104e Ilia Cherniavskii Fri May 7 21:22:17 2021 -0700 [profiler][kineto] Support for memory allocs/deallocs in the traces (#57835)
• 259d19a733 Raghavan Raman Sun May 9 22:05:41 2021 -0700 [JIT] Adding a concat optimization pass (#55474)
• 19706d91cd Sicheng Stephen Jia Mon May 10 12:38:12 2021 -0700 [vulkan] Add sigmoid activation functions (#57867)
• 0d564904b5 Tao Xu Tue May 11 04:46:52 2021 -0700 [iOS GPU][Perf][4/n] Reuse the same command buffer when copying results to CPU (#57667)
• 565550d89a Tao Xu Tue May 11 09:31:57 2021 -0700 [iOS GPU][perf][5/n] Replace std:vector with IntArrayRef and SmallVector (#57668)
• ce1a8620d9 Winston Smith Wed May 12 20:28:17 2021 -0700 Enabled roll & diag for BFloat16 dtype on CUDA (#57916)
• f9aa6b2432 Winston Smith Wed May 12 20:20:38 2021 -0700 Enable lerp for BFloat16 on CUDA (#57907)
• e6d8f45523 Winston Smith Wed May 12 20:19:25 2021 -0700 Enable ceil, floor, frac, round & trunc for BFloat16 on CUDA (#57910)
• c4a486f4b1 Winston Smith Wed May 12 20:18:04 2021 -0700 Enable atan2 & hypot for BFloat16 on CUDA (#57905)
• 047ae6b713 Ilia Cherniavskii Thu May 13 19:20:27 2021 -0700 [profiler][small] CUDA synchronize guard, minor fix (#58254)
• a8122062c0 Lunwen He Thu May 13 22:52:10 2021 -0700 [PyTorch Mobile]Add light version of RandomSampler (#58201)
• 72a90c3ea5 Stephen Jia Fri May 14 07:51:20 2021 -0700 [metal] Add reflection_pad2d for metal (#58263)
• d11dbf511 BowenBao Fri May 14 09:49:58 2021 -0700 [ONNX] Support index_add_ function. (#56867) (#57830)
• 73d51406fa Lunwen He Fri May 14 12:53:18 2021 -0700 [PyTorch Mobile]Move train related files to their own folder (#58205)
• c29e6d37e8 Sicheng Stephen Jia Mon May 17 12:03:20 2021 -0700 [Vulkan] Switch to Image2D for Convolution biases (#57201)
• 50f9a1812e David Reiss Mon May 17 12:50:24 2021 -0700 Enable NNAPI in internal build (#58324)
• 0c3db1cb33 Chen Lai Mon May 17 14:11:46 2021 -0700 [Pytorch] Build lite interpreter as default for Android
• 3d12ab452e BowenBao Mon May 17 14:49:26 2021 -0700 [ONNX] Fix split export in opset13 (#56277) (#57605)
• b5a834a739 Chen Lai Mon May 17 22:34:37 2021 -0700 [Pytorch] Build lite interpreter as default for iOS
• 954d39ba38 Hao Lu Wed May 19 12:35:22 2021 -0700 [ATen][Quant] Pass at::Tensor by reference (#58284)
• 8a3d9962e0 Winston Smith Thu May 20 14:51:32 2021 -0700 Enable ceil, floor, frac, round & trunc for BFloat16 on CUDA (#57910)
• 6edd49a8e8 H1Gdev Thu May 20 15:43:34 2021 -0700 [Android]Removed dependency with AppCompat. (#58527)
• 4668d09ca6 Jerry Zhang Thu May 20 16:02:35 2021 -0700 [quant][graphmode][fx] Quantize the output of statically quantized fp16 op in QuantizeHandler (#58445)
• f879e70fc1 Jerry Zhang Thu May 20 16:23:13 2021 -0700 [quant][fx][graphmode][refactor] Factor out generate_qconfig_map to qconfig_utils.py (#58453)
• 618be18a41 johnlu Thu May 20 17:00:38 2021 -0700 Enable the quantization on XPU devices (#54857)
• b6dcdeacc9 Jerry Zhang Thu May 20 17:22:38 2021 -0700 [quant][graphmode][fx] Move qat_swap_modules outside of Quantizer (#58454)
• 0ede83db7a leslie-fang-intel Thu May 20 17:45:18 2021 -0700 enable torch.cpu.amp.autocast (#57386)
• 6d97a80dd2 Jordan Fix Thu May 20 21:24:54 2021 -0700 [fx][graph_drawer] Improve graph drawer coloring and tensor_meta handling (#58699)
• d88d321ee3 Tugsbayasgalan (Tugsuu) Manlaibaatar Thu May 20 22:59:12 2021 -0700 More robust slicing logic for nn.ModuleList (#58361)
• e574c2c025 Angela Yi Fri May 21 15:19:05 2021 -0700 [quant][fx] Validate qconfig_dict keys (#58566)
• 51b7224f8f Stephen Jia Mon May 24 14:13:06 2021 -0700 [vulkan] Add max_pool2d op (#58806)
• f29e75c4dc Jerry Zhang Mon May 24 14:54:16 2021 -0700 [reland][quant][fx][graphmode][refactor] Remove qconfig_map from Quantizer (#58455) (#58756)
• 09a8f22bf9 Adnios Tue May 25 10:34:50 2021 -0700 Add mish activation function (#58648)
• 8a28bbeeb9 Natalia Gimelshein Tue May 25 10:43:22 2021 -0700 various TensorIterator speed improvements (#58810)

[gemfield]

• 7179e7ea7b Nikita Shulga Tue May 25 15:08:56 2021 -0700 [CMake] Prefer third_party/pybind11 by default (#58951)
• 1bd22e28b3 Xiang Gao Wed May 26 16:47:50 2021 -0700 BFloat16 support for torch.sort (#58196)
• 95b1bc1009 Peter Bell Thu May 27 10:05:40 2021 -0700 Migrate nonzero from TH to ATen (CPU) (#58811)
• 4e543d017f Eddie Yan Thu May 27 18:34:37 2021 -0700 Move remaining *Sort* in THC to ATen (#58953)
• 9c83e4160d David Reiss Thu May 27 20:46:51 2021 -0700 Use some c10::ThreadLocal to avoid crashes on old Android toolchains (#59017)
• ab372ba510 Tao Xu Fri May 28 12:14:24 2021 -0700 [iOS GPU] Add debug information to track memory allocation exception (#59112)
• 27009d6129 Mikhail Zolotukhin Sat May 29 20:35:25 2021 -0700 [TensorExpr] Add NNC lowerings for aten::view, aten::reshape and aten::expand_as. (#59157)
• 06399d441d Jack Montgomery Tue Jun 1 11:39:44 2021 -0700 Create EngineHolder for serializing and running TRT Engines with PyTorch
• e4b2684331 Jerry Zhang Tue Jun 1 13:42:23 2021 -0700 [quant][graphmode][fx][refactor] Remove patterns from Quantizer class (#59033)
• 83892c1861 Jerry Zhang Tue Jun 1 13:24:32 2021 -0700 [quant][graphmode][fx][refactor] Remove node_name_to_scope from Quantizer (#59032)
• 3826f7e8e0 Jerry Zhang Tue Jun 1 13:00:20 2021 -0700 [quant][graphmode][fx][refactor] Remove quantized_graph from Quantizer (#59031)
• 1b4586ee20 Jerry Zhang Tue Jun 1 12:41:10 2021 -0700 [quant][gx][graphmode][refactor] Remove modules from Quantizer (#59030)
• a0ce8da26e Andrew McCollum Tue Jun 1 14:12:49 2021 -0700 Fix DistributedSampler mem usage on large datasets (#51841)
• 599f5058cf neginraoof Wed Jun 2 08:26:05 2021 -0700 [ONNX] Update ONNX to rel-1.9 (#55889) (#57080)
• 87a25e09f4 Jerry Zhang Wed Jun 2 08:49:28 2021 -0700 [quant][graphmode][fx][refactor] Remove _convert from Quantizer class (#59042)
• 0528325b5f Tao Xu Wed Jun 2 11:52:02 2021 -0700 [iOS GPU] Raise the minimum OS support version to 11.0 (#59310)
• f8f06e7099 Tao Xu Wed Jun 2 11:46:06 2021 -0700 [iOS GPU] Fix the OSS macos build (#59102)
• 99f2000a99 Peter Bell Wed Jun 2 12:22:14 2021 -0700 Migrate nonzero from TH to ATen (CPU) (#59149)
• 6408cbd918 Peter Bell Thu Jun 3 11:42:14 2021 -0700 Migrate renorm to ATen (CPU and CUDA) (#59250)
• 6627c00e63 Hao Lu Thu Jun 3 19:16:49 2021 -0700 [Static Runtime] Fix bug in quantized::linear wrapper (#59407)
• b3ee645cbf Peter Bell Fri Jun 4 11:56:46 2021 -0700 Migrate _th_std_var to ATen (#59258)
• 390fe74944 Peter Bell Sun Jun 6 02:17:18 2021 -0700 Migrate torch.lstsq to ATen (#59400)
• 8c852de54d Dhruv Matani Sun Jun 6 13:14:28 2021 -0700 [PyTorch Edge] Remove legacy and kineto profilers from mobile build (#58730)
• 501320ed81 Jiakai Liu Mon Jun 7 19:36:23 2021 -0700 [pytorch] deprecate default_op_deps.yaml (#59573)

[gemfield]

• 67b8e6410d Tao Xu Tue Jun 8 14:40:57 2021 -0700 [OSS] Add podspec for libtorch-lite (#59638)
• 9d1d799034 Hariom Narang Mon Jun 21 16:09:35 2021 -0700 Added API to change logging levels for JIT (#58821)
• 729f7cd52f Saketh Are Tue Jun 22 10:04:51 2021 -0700 Implement histogram operator on CPU (#58780)
• 567e6d3a87 Winston Smith Tue Jun 22 10:25:27 2021 -0700 Remove Caffe2 thread-pool leak warning (#60318)
• 0d2a936176 Ilqar Ramazanli Tue Jun 22 10:37:31 2021 -0700 To add Rectified Adam Algorithm to Optimizers (#58968)
• 31a884987d Peter Bell Tue Jun 22 10:53:55 2021 -0700 Remove some TH includes from ATen (#60323)
• 99ca2c5b4b Thomas J. Fan Tue Jun 22 12:02:12 2021 -0700 Migrates nll_loss_backward from TH to Aten (CUDA) (#60299)
• e8690dacb2 Ilqar Ramazanli Wed Jun 23 08:20:06 2021 -0700 To add Nesterov Adam Algorithm to Optimizers (#59009)
• 1120a1b92e Supriya Rao Wed Jun 23 15:53:55 2021 -0700 [quant][fx][fix] QAT with object_type in qconfig (#60555)
• a016150163 Luca Wehrstedt Thu Jun 24 12:37:29 2021 -0700 Move torch/lib/c10d to torch/csrc/distributed/c10d (#60543)
• d586248544 Peter Bell Thu Jun 24 19:39:36 2021 -0700 Migrate THStorage_resizeBytes to ATen (CPU) (#60324)
• 4a7d281119 Peter Bell Thu Jun 24 19:39:36 2021 -0700 Migrate THAllocator to ATen (#60325)
• 42c8439b6e Peter Bell Thu Jun 24 19:39:36 2021 -0700 TH: Clean up dead code (#60655)
• 9b94aa5356 Angela Yi Mon Jun 28 15:20:48 2021 -0700 [quant][fx][fix] Fused modules with object_type in qconfig (#60779)
• 287c0ab170 Malay Bag Tue Jun 29 23:04:07 2021 -0700 [FX] Add requires_grad to TensorMetadata (#60972)