From b2afdd787130b2c54a661bc3d570831ac9ec2636 Mon Sep 17 00:00:00 2001
From: weili10 <liwei386@huawei.com>
Date: Fri, 23 Jul 2021 18:20:45 +0800
Subject: [PATCH] =?UTF-8?q?=E5=90=8C=E6=AD=A5=2099ba430b04031e42d0284436f1?=
 =?UTF-8?q?3584a845d82b53=20=20Symeig=E7=AE=97=E5=AD=90=E9=80=82=E9=85=8D?=
 =?UTF-8?q?=20=20rrelu=5Fwith=5Fnoise=E9=80=82=E9=85=8D?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 patch/npu.patch                               | 385 ++++++++++--------
 .../src/ATen/native/native_functions.yaml     |  10 +
 .../npu/RreluWithNoiseBackwardKernelNpu.cpp   |  42 ++
 .../native/npu/RreluWithNoiseKernelNpu.cpp    | 105 +++++
 .../src/ATen/native/npu/SymeigKernelNpu.cpp   |  50 +++
 .../test_network_ops/test_rrelu_with_noise.py |  73 ++++
 .../test_rrelu_with_noise_backward.py         |  81 ++++
 test/test_npu/test_network_ops/test_symeig.py |  61 +++
 8 files changed, 632 insertions(+), 175 deletions(-)
 create mode 100644 src/aten/src/ATen/native/npu/RreluWithNoiseBackwardKernelNpu.cpp
 create mode 100644 src/aten/src/ATen/native/npu/RreluWithNoiseKernelNpu.cpp
 create mode 100644 src/aten/src/ATen/native/npu/SymeigKernelNpu.cpp
 create mode 100644 test/test_npu/test_network_ops/test_rrelu_with_noise.py
 create mode 100644 test/test_npu/test_network_ops/test_rrelu_with_noise_backward.py
 create mode 100644 test/test_npu/test_network_ops/test_symeig.py

diff --git a/patch/npu.patch b/patch/npu.patch
index 6dade37fdd..b6dbaf1791 100644
--- a/patch/npu.patch
+++ b/patch/npu.patch
@@ -1,6 +1,6 @@
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/CMakeLists.txt pytorch-develop/aten/CMakeLists.txt
 --- pytorch-v1.5.0/aten/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/CMakeLists.txt	2021-07-22 20:24:34.089995394 +0800
++++ pytorch-develop/aten/CMakeLists.txt	2021-07-23 18:20:43.601371780 +0800
 @@ -22,8 +22,10 @@
  set(ATen_CPU_INCLUDE)
  set(ATen_THIRD_PARTY_INCLUDE)
@@ -51,7 +51,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  set(ATen_CPU_DEPENDENCY_LIBS ${ATen_CPU_DEPENDENCY_LIBS} PARENT_SCOPE)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/CMakeLists.txt pytorch-develop/aten/src/ATen/CMakeLists.txt
 --- pytorch-v1.5.0/aten/src/ATen/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/CMakeLists.txt	2021-07-22 20:24:34.089995394 +0800
++++ pytorch-develop/aten/src/ATen/CMakeLists.txt	2021-07-23 18:20:43.605371924 +0800
 @@ -67,6 +67,9 @@
  FILE(GLOB native_quantized_h "native/quantized/*.h" "native/quantized/cpu/*.h")
  FILE(GLOB native_cpu_h "native/cpu/*.h")
@@ -129,7 +129,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  set(ATen_QUANTIZED_SRCS ${ATen_QUANTIZED_SRCS} PARENT_SCOPE)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/core/dispatch/DispatchTable.h pytorch-develop/aten/src/ATen/core/dispatch/DispatchTable.h
 --- pytorch-v1.5.0/aten/src/ATen/core/dispatch/DispatchTable.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/core/dispatch/DispatchTable.h	2021-07-22 20:24:34.097995681 +0800
++++ pytorch-develop/aten/src/ATen/core/dispatch/DispatchTable.h	2021-07-23 18:20:43.609372067 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -170,7 +170,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/function_wrapper.py pytorch-develop/aten/src/ATen/function_wrapper.py
 --- pytorch-v1.5.0/aten/src/ATen/function_wrapper.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/function_wrapper.py	2021-07-22 20:24:34.101995825 +0800
++++ pytorch-develop/aten/src/ATen/function_wrapper.py	2021-07-23 18:20:43.617372354 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -354,7 +354,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
          for option in declaration['options']:
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/gen.py pytorch-develop/aten/src/ATen/gen.py
 --- pytorch-v1.5.0/aten/src/ATen/gen.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/gen.py	2021-07-22 20:24:34.101995825 +0800
++++ pytorch-develop/aten/src/ATen/gen.py	2021-07-23 18:20:43.617372354 +0800
 @@ -1,3 +1,18 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -512,7 +512,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      generate_outputs()
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/cpu/Activation.cpp pytorch-develop/aten/src/ATen/native/cpu/Activation.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/cpu/Activation.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/cpu/Activation.cpp	2021-07-22 20:24:34.113996255 +0800
++++ pytorch-develop/aten/src/ATen/native/cpu/Activation.cpp	2021-07-23 18:20:43.629372785 +0800
 @@ -339,20 +339,20 @@
  
  void hardsigmoid_backward_kernel(TensorIterator& iter) {
@@ -540,7 +540,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    });
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/Memory.cpp pytorch-develop/aten/src/ATen/native/Memory.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/Memory.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/Memory.cpp	2021-07-22 20:24:34.105995968 +0800
++++ pytorch-develop/aten/src/ATen/native/Memory.cpp	2021-07-23 18:20:43.625372640 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -595,7 +595,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        detail::computeStorageSize(self.sizes(), self.strides()),
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/native_functions.yaml pytorch-develop/aten/src/ATen/native/native_functions.yaml
 --- pytorch-v1.5.0/aten/src/ATen/native/native_functions.yaml	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/native_functions.yaml	2021-07-22 20:24:34.125996685 +0800
++++ pytorch-develop/aten/src/ATen/native/native_functions.yaml	2021-07-23 18:20:43.645373358 +0800
 @@ -1,6 +1,5 @@
  # See README.md in this directory for more guidance
  
@@ -4626,7 +4626,16 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: lstsq.X(Tensor self, Tensor A, *, Tensor(a!) X, Tensor(b!) qr) -> (Tensor(a!) solution, Tensor(b!) QR)
    dispatch:
-@@ -4826,9 +5944,13 @@
+@@ -4753,6 +5871,8 @@
+   dispatch:
+     CPU: _symeig_helper_cpu
+     CUDA: _symeig_helper_cuda
++  npu_dispatch:
++    NPU: _symeig_helper_npu
+ 
+ - func: eig.e(Tensor self, bool eigenvectors=False, *, Tensor(a!) e, Tensor(b!) v) -> (Tensor(a!) eigenvalues, Tensor(b!) eigenvectors)
+   dispatch:
+@@ -4826,9 +5946,13 @@
      CUDA: legacy::cuda::_th_potri
  
  - func: qr.Q(Tensor self, bool some=True, *, Tensor(a!) Q, Tensor(b!) R) -> (Tensor(a!) Q, Tensor(b!) R)
@@ -4640,7 +4649,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _qr_helper(Tensor self, bool some) -> (Tensor, Tensor)
    variants: function
-@@ -4891,12 +6013,16 @@
+@@ -4891,12 +6015,16 @@
    dispatch:
      CPU: multinomial_out
      CUDA: multinomial_out
@@ -4657,7 +4666,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _multinomial_alias_setup(Tensor probs) -> (Tensor, Tensor)
    variants: function
-@@ -4947,6 +6073,8 @@
+@@ -4947,6 +6075,8 @@
    dispatch:
      CPU: erfinv
      CUDA: erfinv
@@ -4666,7 +4675,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: erfinv_(Tensor(a!) self) -> Tensor(a!)
    supports_named_tensor: True
-@@ -4954,26 +6082,36 @@
+@@ -4954,26 +6084,36 @@
    dispatch:
      CPU: _erfinv__cpu
      CUDA: _erfinv__cuda
@@ -4703,7 +4712,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: dist(Tensor self, Tensor other, Scalar p=2) -> Tensor
    use_c10_dispatcher: full
-@@ -4981,21 +6119,29 @@
+@@ -4981,21 +6121,29 @@
  
  - func: atan2.out(Tensor self, Tensor other, *, Tensor(a!) out) -> Tensor(a!)
    supports_named_tensor: True
@@ -4733,7 +4742,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: lerp.Scalar(Tensor self, Tensor end, Scalar weight) -> Tensor
    use_c10_dispatcher: full
-@@ -5003,6 +6149,8 @@
+@@ -5003,6 +6151,8 @@
    dispatch:
      CPU: lerp_cpu_scalar
      CUDA: lerp_cuda_scalar
@@ -4742,7 +4751,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: lerp.Tensor(Tensor self, Tensor end, Tensor weight) -> Tensor
    use_c10_dispatcher: full
-@@ -5010,6 +6158,8 @@
+@@ -5010,6 +6160,8 @@
    dispatch:
      CPU: lerp_cpu_tensor
      CUDA: lerp_cuda_tensor
@@ -4751,7 +4760,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: histc.out(Tensor self, int bins=100, Scalar min=0, Scalar max=0, *, Tensor(a!) out) -> Tensor(a!)
    dispatch:
-@@ -5027,6 +6177,8 @@
+@@ -5027,6 +6179,8 @@
    dispatch:
      CPU: fmod_out
      CUDA: legacy::cuda::_th_fmod_out
@@ -4760,7 +4769,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: fmod.Scalar(Tensor self, Scalar other) -> Tensor
    use_c10_dispatcher: full
-@@ -5034,11 +6186,15 @@
+@@ -5034,11 +6188,15 @@
    dispatch:
      CPU: fmod
      CUDA: legacy::cuda::_th_fmod
@@ -4776,7 +4785,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: fmod.Tensor(Tensor self, Tensor other) -> Tensor
    use_c10_dispatcher: full
-@@ -5046,11 +6202,15 @@
+@@ -5046,11 +6204,15 @@
    dispatch:
      CPU: fmod
      CUDA: legacy::cuda::_th_fmod
@@ -4792,7 +4801,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: remainder.Scalar(Tensor self, Scalar other) -> Tensor
    use_c10_dispatcher: full
-@@ -5058,11 +6218,15 @@
+@@ -5058,11 +6220,15 @@
    dispatch:
      CPU: remainder
      CUDA: remainder
@@ -4808,7 +4817,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: remainder.Tensor(Tensor self, Tensor other) -> Tensor
    use_c10_dispatcher: full
-@@ -5070,12 +6234,18 @@
+@@ -5070,12 +6236,18 @@
    dispatch:
      CPU: remainder
      CUDA: remainder
@@ -4827,7 +4836,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: min(Tensor self) -> Tensor
    use_c10_dispatcher: full
-@@ -5084,13 +6254,19 @@
+@@ -5084,13 +6256,19 @@
      CPU: min
      CUDA: legacy::cuda::_th_min
      QuantizedCPU: min_quant
@@ -4847,7 +4856,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: max(Tensor self) -> Tensor
    use_c10_dispatcher: full
-@@ -5099,6 +6275,8 @@
+@@ -5099,6 +6277,8 @@
      CPU: max
      CUDA: legacy::cuda::_th_max
      QuantizedCPU: max_quant
@@ -4856,7 +4865,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    supports_named_tensor: True
  
  - func: median(Tensor self) -> Tensor
-@@ -5107,12 +6285,16 @@
+@@ -5107,12 +6287,16 @@
    dispatch:
      CPU: median_cpu
      CUDA: median_cuda
@@ -4873,7 +4882,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: sort(Tensor self, int dim=-1, bool descending=False) -> (Tensor values, Tensor indices)
    variants: method, function
-@@ -5120,23 +6302,45 @@
+@@ -5120,23 +6304,45 @@
      CPU: legacy::cpu::_th_sort
      CUDA: legacy::cuda::_th_sort
      QuantizedCPU: sort_quant
@@ -4919,7 +4928,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: topk(Tensor self, int k, int dim=-1, bool largest=True, bool sorted=True) -> (Tensor values, Tensor indices)
    variants: method, function
-@@ -5144,11 +6348,15 @@
+@@ -5144,11 +6350,15 @@
      CPU: topk
      CUDA: topk
      QuantizedCPU: quantized_topk_cpu
@@ -4935,7 +4944,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: any(Tensor self) -> Tensor
    use_c10_dispatcher: full
-@@ -5159,11 +6367,15 @@
+@@ -5159,11 +6369,15 @@
      CUDA: any
      SparseCPU: any_sparse
      SparseCUDA: any_sparse
@@ -4951,7 +4960,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: renorm(Tensor self, Scalar p, int dim, Scalar maxnorm) -> Tensor
    use_c10_dispatcher: full
-@@ -5171,6 +6383,8 @@
+@@ -5171,6 +6385,8 @@
    dispatch:
      CPU: legacy::cpu::_th_renorm
      CUDA: legacy::cuda::_th_renorm
@@ -4960,7 +4969,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: unfold(Tensor(a) self, int dimension, int size, int step) -> Tensor(a)
    variants: method
-@@ -5178,6 +6392,8 @@
+@@ -5178,6 +6394,8 @@
    dispatch:
      CPU: unfold
      CUDA: unfold
@@ -4969,7 +4978,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: equal(Tensor self, Tensor other) -> bool
    use_c10_dispatcher: full
-@@ -5186,6 +6402,8 @@
+@@ -5186,6 +6404,8 @@
      CPU: legacy::cpu::_th_equal
      CUDA: legacy::cuda::_th_equal
      QuantizedCPU: quantized_equal
@@ -4978,7 +4987,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    supports_named_tensor: True
  
  - func: pow.Tensor_Tensor_out(Tensor self, Tensor exponent, *, Tensor(a!) out) -> Tensor(a!)
-@@ -5193,6 +6411,8 @@
+@@ -5193,6 +6413,8 @@
    dispatch:
      CPU: pow_out
      CUDA: pow_out
@@ -4987,7 +4996,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: pow.Tensor_Tensor(Tensor self, Tensor exponent) -> Tensor
    use_c10_dispatcher: full
-@@ -5201,12 +6421,16 @@
+@@ -5201,12 +6423,16 @@
    dispatch:
      CPU: pow
      CUDA: pow
@@ -5004,7 +5013,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: pow.Scalar(Scalar self, Tensor exponent) -> Tensor
    use_c10_dispatcher: full
-@@ -5214,6 +6438,8 @@
+@@ -5214,6 +6440,8 @@
    dispatch:
      CPU: pow
      CUDA: pow
@@ -5013,7 +5022,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: normal_(Tensor(a!) self, float mean=0, float std=1, *, Generator? generator=None) -> Tensor(a!)
    variants: method
-@@ -5221,40 +6447,58 @@
+@@ -5221,40 +6449,58 @@
      CPU: normal_cpu_
      CUDA: normal_cuda_
    supports_named_tensor: True
@@ -5072,7 +5081,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: alias(Tensor(a) self) -> Tensor(a)
    variants: method, function
-@@ -5265,43 +6509,59 @@
+@@ -5265,43 +6511,59 @@
    dispatch:
      CPU: legacy::cpu::_th_addr
      CUDA: legacy::cuda::_th_addr
@@ -5133,7 +5142,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _var(Tensor self, bool unbiased=True) -> Tensor
    use_c10_dispatcher: full
-@@ -5309,6 +6569,8 @@
+@@ -5309,6 +6571,8 @@
      CPU: legacy::cpu::_th_var
      CUDA: legacy::cuda::_th_var
    supports_named_tensor: True
@@ -5142,7 +5151,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _std(Tensor self, bool unbiased=True) -> Tensor
    use_c10_dispatcher: full
-@@ -5321,6 +6583,8 @@
+@@ -5321,6 +6585,8 @@
    variants: function
    dispatch:
      CUDA: _amp_non_finite_check_and_unscale_cuda_
@@ -5151,7 +5160,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _amp_update_scale(Tensor(a!) growth_tracker, Tensor current_scale, Tensor found_inf, float scale_growth_factor, float scale_backoff_factor, int growth_interval) -> Tensor
    variants: function
-@@ -5332,12 +6596,16 @@
+@@ -5332,12 +6598,16 @@
      CPU: _cat_cpu
      CUDA: cat_cuda
      QuantizedCPU: quantized_cat
@@ -5168,7 +5177,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _mode(Tensor self, int dim=-1, bool keepdim=False) -> (Tensor, Tensor)
    dispatch:
-@@ -5353,36 +6621,50 @@
+@@ -5353,36 +6623,50 @@
    dispatch:
      CPU: legacy::cpu::_th_max
      CUDA: legacy::cuda::_th_max
@@ -5219,7 +5228,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: mse_loss_backward(Tensor grad_output, Tensor self, Tensor target, int reduction) -> Tensor
    use_c10_dispatcher: full
-@@ -5390,23 +6672,33 @@
+@@ -5390,23 +6674,33 @@
    dispatch:
      CPU: mse_loss_backward
      CUDA: mse_loss_backward
@@ -5253,7 +5262,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: multi_margin_loss.out(Tensor self, Tensor target, Scalar p=1, Scalar margin=1, Tensor? weight=None, int reduction=Mean, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5434,22 +6726,30 @@
+@@ -5434,22 +6728,30 @@
  
  - func: multilabel_margin_loss.out(Tensor self, Tensor target, int reduction=Mean, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
@@ -5284,7 +5293,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: multilabel_margin_loss_backward.grad_input(Tensor grad_output, Tensor self, Tensor target, int reduction, Tensor is_target, *, Tensor(a!) grad_input) -> Tensor(a!)
    python_module: nn
-@@ -5466,97 +6766,137 @@
+@@ -5466,97 +6768,137 @@
  
  - func: nll_loss.out(Tensor self, Tensor target, Tensor? weight=None, int reduction=Mean, int ignore_index=-100, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
@@ -5422,7 +5431,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: elu.out(Tensor self, Scalar alpha=1, Scalar scale=1, Scalar input_scale=1, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5564,6 +6904,8 @@
+@@ -5564,6 +6906,8 @@
      CPU: elu_out
      CUDA: elu_out
      QuantizedCPU: quantized_elu_out
@@ -5431,7 +5440,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: elu(Tensor self, Scalar alpha=1, Scalar scale=1, Scalar input_scale=1) -> Tensor
    use_c10_dispatcher: full
-@@ -5572,16 +6914,22 @@
+@@ -5572,16 +6916,22 @@
      CPU: elu
      CUDA: elu
      QuantizedCPU: quantized_elu
@@ -5454,7 +5463,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: elu_(Tensor(a!) self, Scalar alpha=1, Scalar scale=1, Scalar input_scale=1) -> Tensor(a!)
    python_module: nn
-@@ -5589,12 +6937,16 @@
+@@ -5589,12 +6939,16 @@
      CPU: elu_
      CUDA: elu_
      QuantizedCPU: quantized_elu_
@@ -5471,7 +5480,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: glu(Tensor self, int dim=-1) -> Tensor
    use_c10_dispatcher: full
-@@ -5602,12 +6954,16 @@
+@@ -5602,12 +6956,16 @@
    dispatch:
      CPU: glu
      CUDA: legacy::cuda::_thnn_glu_forward
@@ -5488,7 +5497,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: glu_backward(Tensor grad_output, Tensor self, int dim) -> Tensor
    use_c10_dispatcher: full
-@@ -5615,20 +6971,30 @@
+@@ -5615,20 +6973,30 @@
    dispatch:
      CPU: glu_backward
      CUDA: legacy::cuda::_thnn_glu_backward
@@ -5519,7 +5528,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: hardtanh.out(Tensor self, Scalar min_val=-1, Scalar max_val=1, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5636,6 +7002,8 @@
+@@ -5636,6 +7004,8 @@
      CPU: hardtanh_out
      CUDA: hardtanh_out
      QuantizedCPU: quantized_hardtanh_out
@@ -5528,7 +5537,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: hardtanh(Tensor self, Scalar min_val=-1, Scalar max_val=1) -> Tensor
    use_c10_dispatcher: full
-@@ -5644,16 +7012,22 @@
+@@ -5644,16 +7014,22 @@
      CPU: hardtanh
      CUDA: hardtanh
      QuantizedCPU: quantized_hardtanh
@@ -5551,7 +5560,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: hardtanh_(Tensor(a!) self, Scalar min_val=-1, Scalar max_val=1) -> Tensor(a!)
    python_module: nn
-@@ -5661,6 +7035,8 @@
+@@ -5661,6 +7037,8 @@
      CPU: hardtanh_
      CUDA: hardtanh_
      QuantizedCPU: quantized_hardtanh_
@@ -5560,7 +5569,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: leaky_relu.out(Tensor self, Scalar negative_slope=0.01, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5668,6 +7044,8 @@
+@@ -5668,6 +7046,8 @@
      CPU: leaky_relu_out
      CUDA: leaky_relu_out
      QuantizedCPU: quantized_leaky_relu_out
@@ -5569,7 +5578,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: leaky_relu(Tensor self, Scalar negative_slope=0.01) -> Tensor
    use_c10_dispatcher: full
-@@ -5676,10 +7054,14 @@
+@@ -5676,10 +7056,14 @@
      CPU: leaky_relu
      CUDA: leaky_relu
      QuantizedCPU: quantized_leaky_relu
@@ -5584,7 +5593,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: leaky_relu_(Tensor(a!) self, Scalar negative_slope=0.01) -> Tensor(a!)
    python_module: nn
-@@ -5687,31 +7069,44 @@
+@@ -5687,31 +7071,44 @@
      CPU: leaky_relu_
      CUDA: leaky_relu_
      QuantizedCPU: quantized_leaky_relu_
@@ -5629,7 +5638,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: log_sigmoid_backward(Tensor grad_output, Tensor self, Tensor buffer) -> Tensor
    use_c10_dispatcher: full
-@@ -5719,6 +7114,8 @@
+@@ -5719,62 +7116,88 @@
    dispatch:
      CPU: log_sigmoid_backward_cpu
      CUDA: legacy::cuda::_thnn_log_sigmoid_backward
@@ -5638,7 +5647,33 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: rrelu_with_noise.out(Tensor self, Tensor noise, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5744,37 +7141,53 @@
+   dispatch:
+     CPU: rrelu_with_noise_out_cpu
+     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward_out
++  npu_dispatch:
++    NPU: rrelu_with_noise_out_npu
+ 
+ - func: rrelu_with_noise(Tensor self, Tensor noise, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor
+   python_module: nn
+   dispatch:
+     CPU: rrelu_with_noise_cpu
+     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward
++  npu_dispatch:
++    NPU: rrelu_with_noise_npu
+ 
+ - func: rrelu_with_noise_backward(Tensor grad_output, Tensor self, Tensor noise, Scalar lower, Scalar upper, bool training, bool self_is_result) -> Tensor
+   use_c10_dispatcher: full
+   python_module: nn
++  npu_dispatch:
++    NPU: rrelu_with_noise_backward_npu
+ 
+ - func: rrelu_with_noise_(Tensor(a!) self, Tensor noise, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor(a!)
+   python_module: nn
+   dispatch:
+     CPU: rrelu_with_noise_cpu_
+     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward_
++  npu_dispatch:
++    NPU: rrelu_with_noise_npu_
  
  - func: softplus.out(Tensor self, Scalar beta=1, Scalar threshold=20, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
@@ -5692,7 +5727,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: adaptive_avg_pool2d.out(Tensor self, int[2] output_size, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -5782,9 +7195,13 @@
+@@ -5782,9 +7205,13 @@
      CPU: adaptive_avg_pool2d_out_cpu
      CUDA: adaptive_avg_pool2d_out_cuda
      MkldnnCPU: mkldnn_adaptive_avg_pool2d_out
@@ -5706,7 +5741,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: mkldnn_adaptive_avg_pool2d(Tensor self, int[2] output_size) -> Tensor
    dispatch:
-@@ -5796,6 +7213,8 @@
+@@ -5796,6 +7223,8 @@
      CPU: adaptive_avg_pool2d_cpu
      CUDA: adaptive_avg_pool2d_cuda
      QuantizedCPU: quantized_adaptive_avg_pool2d
@@ -5715,7 +5750,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: _adaptive_avg_pool2d_backward(Tensor grad_output, Tensor self) -> Tensor
    use_c10_dispatcher: full
-@@ -5803,24 +7222,32 @@
+@@ -5803,24 +7232,32 @@
    dispatch:
      CPU: adaptive_avg_pool2d_backward_cpu
      CUDA: adaptive_avg_pool2d_backward_cuda
@@ -5748,7 +5783,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: adaptive_avg_pool3d_backward(Tensor grad_output, Tensor self) -> Tensor
    use_c10_dispatcher: full
-@@ -5828,6 +7255,8 @@
+@@ -5828,6 +7265,8 @@
    dispatch:
      CPU: adaptive_avg_pool3d_backward_cpu
      CUDA: adaptive_avg_pool3d_backward_cuda
@@ -5757,7 +5792,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: adaptive_max_pool2d.out(Tensor self, int[2] output_size, *, Tensor(a!) out, Tensor(b!) indices) -> (Tensor(a!), Tensor(b!))
-@@ -5835,6 +7264,8 @@
+@@ -5835,6 +7274,8 @@
    dispatch:
      CPU: adaptive_max_pool2d_out_cpu
      CUDA: adaptive_max_pool2d_out_cuda
@@ -5766,7 +5801,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: adaptive_max_pool2d(Tensor self, int[2] output_size) -> (Tensor, Tensor)
-@@ -5842,12 +7273,16 @@
+@@ -5842,12 +7283,16 @@
    dispatch:
      CPU: adaptive_max_pool2d_cpu
      CUDA: adaptive_max_pool2d_cuda
@@ -5783,7 +5818,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: adaptive_max_pool2d_backward(Tensor grad_output, Tensor self, Tensor indices) -> Tensor
    use_c10_dispatcher: full
-@@ -5855,6 +7290,8 @@
+@@ -5855,6 +7300,8 @@
    dispatch:
      CPU: adaptive_max_pool2d_backward_cpu
      CUDA: adaptive_max_pool2d_backward_cuda
@@ -5792,7 +5827,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: adaptive_max_pool3d.out(Tensor self, int[3] output_size, *, Tensor(a!) out, Tensor(b!) indices) -> (Tensor(a!), Tensor(b!))
-@@ -5889,6 +7326,8 @@
+@@ -5889,6 +7336,8 @@
      CPU: avg_pool2d_out_cpu
      CUDA: avg_pool2d_out_cuda
      MkldnnCPU: mkldnn_avg_pool2d_out
@@ -5801,7 +5836,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: avg_pool2d(Tensor self, int[2] kernel_size, int[2] stride=[], int[2] padding=0, bool ceil_mode=False, bool count_include_pad=True, int? divisor_override=None) -> Tensor
    python_module: nn
-@@ -5897,24 +7336,32 @@
+@@ -5897,24 +7346,32 @@
      CUDA: avg_pool2d_cuda
      MkldnnCPU: mkldnn_avg_pool2d
      QuantizedCPU: quantized_avg_pool2d
@@ -5834,7 +5869,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: avg_pool3d(Tensor self, int[3] kernel_size, int[3] stride=[], int[3] padding=0, bool ceil_mode=False, bool count_include_pad=True, int? divisor_override=None) -> Tensor
    python_module: nn
-@@ -5922,18 +7369,24 @@
+@@ -5922,18 +7379,24 @@
      CPU: avg_pool3d_cpu
      CUDA: avg_pool3d_cuda
      QuantizedCPU: quantized_avg_pool3d
@@ -5859,7 +5894,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: fractional_max_pool2d.output(Tensor self, int[2] kernel_size, int[2] output_size, Tensor random_samples, *, Tensor(a!) output, Tensor(b!) indices) -> (Tensor(a!), Tensor(b!))
-@@ -5993,6 +7446,8 @@
+@@ -5993,6 +7456,8 @@
    dispatch:
      CPU: max_pool2d_with_indices_out_cpu
      CUDA: max_pool2d_with_indices_out_cuda
@@ -5868,7 +5903,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: max_pool2d_with_indices(Tensor self, int[2] kernel_size, int[2] stride=[], int[2] padding=0, int[2] dilation=1, bool ceil_mode=False) -> (Tensor, Tensor)
-@@ -6000,6 +7455,8 @@
+@@ -6000,6 +7465,8 @@
    dispatch:
      CPU: max_pool2d_with_indices_cpu
      CUDA: max_pool2d_with_indices_cuda
@@ -5877,7 +5912,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    supports_named_tensor: True
  
  - func: max_pool2d_with_indices_backward.grad_input(Tensor grad_output, Tensor self, int[2] kernel_size, int[2] stride, int[2] padding, int[2] dilation, bool ceil_mode, Tensor indices, *, Tensor(a!) grad_input) -> Tensor(a!)
-@@ -6007,12 +7464,16 @@
+@@ -6007,12 +7474,16 @@
    dispatch:
      CPU: max_pool2d_with_indices_backward_out_cpu
      CUDA: max_pool2d_with_indices_backward_out_cuda
@@ -5894,7 +5929,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: max_pool3d_with_indices.out(Tensor self, int[3] kernel_size, int[3] stride=[], int[3] padding=0, int[3] dilation=1, bool ceil_mode=False, *, Tensor(a!) out, Tensor(b!) indices) -> (Tensor(a!), Tensor(b!))
-@@ -6020,6 +7481,8 @@
+@@ -6020,6 +7491,8 @@
    dispatch:
      CPU: max_pool3d_with_indices_out_cpu
      CUDA: max_pool3d_with_indices_out_cuda
@@ -5903,7 +5938,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # Return: (Tensor output, Tensor indices)
  - func: max_pool3d_with_indices(Tensor self, int[3] kernel_size, int[3] stride=[], int[3] padding=0, int[3] dilation=1, bool ceil_mode=False) -> (Tensor, Tensor)
-@@ -6027,6 +7490,8 @@
+@@ -6027,6 +7500,8 @@
    dispatch:
      CPU: max_pool3d_with_indices_cpu
      CUDA: max_pool3d_with_indices_cuda
@@ -5912,7 +5947,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    supports_named_tensor: True
  
  - func: max_pool3d_with_indices_backward.grad_input(Tensor grad_output, Tensor self, int[3] kernel_size, int[3] stride, int[3] padding, int[3] dilation, bool ceil_mode, Tensor indices, *, Tensor(a!) grad_input) -> Tensor(a!)
-@@ -6034,12 +7499,17 @@
+@@ -6034,12 +7509,17 @@
    dispatch:
      CPU: max_pool3d_with_indices_backward_out_cpu
      CUDA: max_pool3d_with_indices_backward_out_cuda
@@ -5930,7 +5965,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: max_unpool2d.out(Tensor self, Tensor indices, int[2] output_size, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -6166,12 +7636,16 @@
+@@ -6166,12 +7646,16 @@
    dispatch:
      CPU: replication_pad2d_out_cpu
      CUDA: replication_pad2d_out_cuda
@@ -5947,7 +5982,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: replication_pad2d_backward.grad_input(Tensor grad_output, Tensor self, int[4] padding, *, Tensor(a!) grad_input) -> Tensor(a!)
    python_module: nn
-@@ -6214,12 +7688,16 @@
+@@ -6214,12 +7698,16 @@
    dispatch:
      CPU: upsample_linear1d_out_cpu
      CUDA: upsample_linear1d_out_cuda
@@ -5964,7 +5999,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: upsample_linear1d_backward.grad_input(Tensor grad_output, int[1] output_size, int[3] input_size, bool align_corners, float? scales=None, *, Tensor(a!) grad_input) -> Tensor(a!)
    python_module: nn
-@@ -6232,12 +7710,16 @@
+@@ -6232,12 +7720,16 @@
    dispatch:
      CPU: upsample_linear1d_backward_cpu
      CUDA: upsample_linear1d_backward_cuda
@@ -5981,7 +6016,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: upsample_bilinear2d(Tensor self, int[2] output_size, bool align_corners, float? scales_h=None, float? scales_w=None) -> Tensor
    python_module: nn
-@@ -6245,96 +7727,128 @@
+@@ -6245,96 +7737,128 @@
      CPU: upsample_bilinear2d_cpu
      CUDA: upsample_bilinear2d_cuda
      QuantizedCPU: quantized_upsample_bilinear2d_cpu
@@ -6110,7 +6145,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: upsample_nearest2d(Tensor self, int[2] output_size, float? scales_h=None, float? scales_w=None) -> Tensor
    python_module: nn
-@@ -6342,24 +7856,32 @@
+@@ -6342,24 +7866,32 @@
      CPU: upsample_nearest2d_cpu
      CUDA: upsample_nearest2d_cuda
      QuantizedCPU: quantized_upsample_nearest2d_cpu
@@ -6143,7 +6178,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: upsample_nearest3d(Tensor self, int[3] output_size, float? scales_d=None, float? scales_h=None, float? scales_w=None) -> Tensor
    python_module: nn
-@@ -6367,38 +7889,52 @@
+@@ -6367,38 +7899,52 @@
      CPU: upsample_nearest3d_cpu
      CUDA: upsample_nearest3d_cuda
      QuantizedCPU: quantized_upsample_nearest3d_cpu
@@ -6196,7 +6231,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  # What's a thnn_conv_ versus a slow_conv_?
  #
-@@ -6423,24 +7959,32 @@
+@@ -6423,24 +7969,32 @@
    dispatch:
      CPU: slow_conv_transpose2d_out_cpu
      CUDA: slow_conv_transpose2d_out_cuda
@@ -6229,7 +6264,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: slow_conv_transpose3d.out(Tensor self, Tensor weight, int[3] kernel_size, Tensor? bias=None, int[3] stride=1, int[3] padding=0, int[3] output_padding=0, int[3] dilation=1, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -6468,21 +8012,29 @@
+@@ -6468,21 +8022,29 @@
  
  - func: thnn_conv2d.out(Tensor self, Tensor weight, int[2] kernel_size, Tensor? bias=None, int[2] stride=1, int[2] padding=0, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
@@ -6259,7 +6294,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: thnn_conv2d_backward.grad_input(Tensor grad_output, Tensor self, Tensor weight, int[2] kernel_size, int[2] stride, int[2] padding, Tensor finput, Tensor fgrad_input, *, Tensor(a!)? grad_input, Tensor(b!)? grad_weight, Tensor(c!)? grad_bias) -> (Tensor(a!), Tensor(b!), Tensor(c!))
    python_module: nn
-@@ -6495,32 +8047,46 @@
+@@ -6495,32 +8057,46 @@
    dispatch:
      CPU: slow_conv2d_backward_cpu
      CUDA: legacy::cuda::_thnn_conv2d_backward
@@ -6306,7 +6341,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: slow_conv3d.out(Tensor self, Tensor weight, int[3] kernel_size, Tensor? bias=None, int[3] stride=1, int[3] padding=0, *, Tensor(a!) out) -> Tensor(a!)
    python_module: nn
-@@ -6553,12 +8119,16 @@
+@@ -6553,12 +8129,16 @@
    dispatch:
      CPU: slow_conv_dilated2d_cpu
      CUDA: slow_conv_dilated2d_cuda
@@ -6323,7 +6358,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
  - func: slow_conv_dilated3d(Tensor self, Tensor weight, int[3] kernel_size, Tensor? bias=None, int[3] stride=1, int[3] padding=0, int[3] dilation=1) -> Tensor
    python_module: nn
-@@ -6577,57 +8147,413 @@
+@@ -6577,57 +8157,413 @@
    dispatch:
      CPU: col2im_out_cpu
      CUDA: col2im_out_cuda
@@ -6740,7 +6775,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/8x8-dq-aarch64-neon.S pytorch-develop/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/8x8-dq-aarch64-neon.S
 --- pytorch-v1.5.0/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/8x8-dq-aarch64-neon.S	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/8x8-dq-aarch64-neon.S	2021-07-22 20:24:34.165998119 +0800
++++ pytorch-develop/aten/src/ATen/native/quantized/cpu/qnnpack/src/q8gemm/8x8-dq-aarch64-neon.S	2021-07-23 18:20:43.681374649 +0800
 @@ -659,14 +659,14 @@
  
      SUB x1, x1, 4
@@ -6766,7 +6801,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      CMP x1, 2
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/TensorCompare.cpp pytorch-develop/aten/src/ATen/native/TensorCompare.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/TensorCompare.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/TensorCompare.cpp	2021-07-22 20:24:34.109996111 +0800
++++ pytorch-develop/aten/src/ATen/native/TensorCompare.cpp	2021-07-23 18:20:43.629372785 +0800
 @@ -64,7 +64,7 @@
  
  Tensor isinf(const Tensor &self) {
@@ -6778,7 +6813,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    return AT_DISPATCH_FLOATING_TYPES_AND_HALF(self.scalar_type(), "isinf", [&]() {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/TensorFactories.cpp pytorch-develop/aten/src/ATen/native/TensorFactories.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/TensorFactories.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/TensorFactories.cpp	2021-07-22 20:24:34.109996111 +0800
++++ pytorch-develop/aten/src/ATen/native/TensorFactories.cpp	2021-07-23 18:20:43.629372785 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -6823,7 +6858,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/TensorProperties.cpp pytorch-develop/aten/src/ATen/native/TensorProperties.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/TensorProperties.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/TensorProperties.cpp	2021-07-22 20:24:34.109996111 +0800
++++ pytorch-develop/aten/src/ATen/native/TensorProperties.cpp	2021-07-23 18:20:43.629372785 +0800
 @@ -87,6 +87,7 @@
    if (self.is_contiguous(memory_format)) {
      return self;
@@ -6834,7 +6869,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        "preserve memory format is unsupported by the contiguous operator");
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native/UpSampleBicubic2d.cpp pytorch-develop/aten/src/ATen/native/UpSampleBicubic2d.cpp
 --- pytorch-v1.5.0/aten/src/ATen/native/UpSampleBicubic2d.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native/UpSampleBicubic2d.cpp	2021-07-22 20:24:34.113996255 +0800
++++ pytorch-develop/aten/src/ATen/native/UpSampleBicubic2d.cpp	2021-07-23 18:20:43.629372785 +0800
 @@ -26,7 +26,7 @@
          const scalar_t* in = &idata[output_y * input_width + output_x];
          scalar_t* out = &odata[output_y * output_width + output_x];
@@ -6846,7 +6881,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
            out += output_width * output_height;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/native_parse.py pytorch-develop/aten/src/ATen/native_parse.py
 --- pytorch-v1.5.0/aten/src/ATen/native_parse.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/native_parse.py	2021-07-22 20:24:34.177998550 +0800
++++ pytorch-develop/aten/src/ATen/native_parse.py	2021-07-23 18:20:43.697375223 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -6884,7 +6919,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
                  msg = '''Exception raised in processing function:
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/preprocess_declarations.py pytorch-develop/aten/src/ATen/preprocess_declarations.py
 --- pytorch-v1.5.0/aten/src/ATen/preprocess_declarations.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/preprocess_declarations.py	2021-07-22 20:24:34.177998550 +0800
++++ pytorch-develop/aten/src/ATen/preprocess_declarations.py	2021-07-23 18:20:43.697375223 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -6916,7 +6951,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/templates/TensorBody.h pytorch-develop/aten/src/ATen/templates/TensorBody.h
 --- pytorch-v1.5.0/aten/src/ATen/templates/TensorBody.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/templates/TensorBody.h	2021-07-22 20:24:34.177998550 +0800
++++ pytorch-develop/aten/src/ATen/templates/TensorBody.h	2021-07-23 18:20:43.697375223 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -6949,7 +6984,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/ATen/templates/TensorMethods.h pytorch-develop/aten/src/ATen/templates/TensorMethods.h
 --- pytorch-v1.5.0/aten/src/ATen/templates/TensorMethods.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/ATen/templates/TensorMethods.h	2021-07-22 20:24:34.177998550 +0800
++++ pytorch-develop/aten/src/ATen/templates/TensorMethods.h	2021-07-23 18:20:43.697375223 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -6983,7 +7018,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/TH/CMakeLists.txt pytorch-develop/aten/src/TH/CMakeLists.txt
 --- pytorch-v1.5.0/aten/src/TH/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/TH/CMakeLists.txt	2021-07-22 20:24:34.181998693 +0800
++++ pytorch-develop/aten/src/TH/CMakeLists.txt	2021-07-23 18:20:43.701375366 +0800
 @@ -48,6 +48,11 @@
    ${CMAKE_CURRENT_SOURCE_DIR}
  PARENT_SCOPE)
@@ -6998,7 +7033,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/TH/generic/THStorage.cpp pytorch-develop/aten/src/TH/generic/THStorage.cpp
 --- pytorch-v1.5.0/aten/src/TH/generic/THStorage.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/TH/generic/THStorage.cpp	2021-07-22 20:24:34.181998693 +0800
++++ pytorch-develop/aten/src/TH/generic/THStorage.cpp	2021-07-23 18:20:43.701375366 +0800
 @@ -1,9 +1,32 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7107,7 +7142,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/aten/src/TH/generic/THStorage.h pytorch-develop/aten/src/TH/generic/THStorage.h
 --- pytorch-v1.5.0/aten/src/TH/generic/THStorage.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/aten/src/TH/generic/THStorage.h	2021-07-22 20:24:34.181998693 +0800
++++ pytorch-develop/aten/src/TH/generic/THStorage.h	2021-07-23 18:20:43.701375366 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7146,7 +7181,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/CMakeLists.txt pytorch-develop/c10/CMakeLists.txt
 --- pytorch-v1.5.0/c10/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/CMakeLists.txt	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/CMakeLists.txt	2021-07-23 18:20:43.713375796 +0800
 @@ -63,6 +63,14 @@
    message(STATUS "don't use NUMA")
  endif()
@@ -7175,7 +7210,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    # not checked in
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/Backend.h pytorch-develop/c10/core/Backend.h
 --- pytorch-v1.5.0/c10/core/Backend.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/Backend.h	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/Backend.h	2021-07-23 18:20:43.713375796 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7270,7 +7305,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/Device.cpp pytorch-develop/c10/core/Device.cpp
 --- pytorch-v1.5.0/c10/core/Device.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/Device.cpp	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/Device.cpp	2021-07-23 18:20:43.713375796 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7310,7 +7345,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        types.begin(),
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/Device.h pytorch-develop/c10/core/Device.h
 --- pytorch-v1.5.0/c10/core/Device.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/Device.h	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/Device.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7345,7 +7380,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      return type_ == DeviceType::CPU;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/DeviceType.cpp pytorch-develop/c10/core/DeviceType.cpp
 --- pytorch-v1.5.0/c10/core/DeviceType.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/DeviceType.cpp	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/DeviceType.cpp	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7385,7 +7420,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        return false;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/DeviceType.h pytorch-develop/c10/core/DeviceType.h
 --- pytorch-v1.5.0/c10/core/DeviceType.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/DeviceType.h	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/DeviceType.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7428,7 +7463,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  constexpr DeviceType kXLA = DeviceType::XLA;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/DispatchKey.cpp pytorch-develop/c10/core/DispatchKey.cpp
 --- pytorch-v1.5.0/c10/core/DispatchKey.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/DispatchKey.cpp	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/DispatchKey.cpp	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7460,7 +7495,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      case DispatchKey::TESTING_ONLY_GenericModeTensorId:
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/DispatchKey.h pytorch-develop/c10/core/DispatchKey.h
 --- pytorch-v1.5.0/c10/core/DispatchKey.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/DispatchKey.h	2021-07-22 20:24:34.193999123 +0800
++++ pytorch-develop/c10/core/DispatchKey.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7492,7 +7527,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/Storage.h pytorch-develop/c10/core/Storage.h
 --- pytorch-v1.5.0/c10/core/Storage.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/Storage.h	2021-07-22 20:24:34.197999266 +0800
++++ pytorch-develop/c10/core/Storage.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7526,7 +7561,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  };
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/StorageImpl.h pytorch-develop/c10/core/StorageImpl.h
 --- pytorch-v1.5.0/c10/core/StorageImpl.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/StorageImpl.h	2021-07-22 20:24:34.197999266 +0800
++++ pytorch-develop/c10/core/StorageImpl.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,12 +1,39 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7583,7 +7618,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/TensorImpl.h pytorch-develop/c10/core/TensorImpl.h
 --- pytorch-v1.5.0/c10/core/TensorImpl.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/TensorImpl.h	2021-07-22 20:24:34.197999266 +0800
++++ pytorch-develop/c10/core/TensorImpl.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7653,7 +7688,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/core/TensorOptions.h pytorch-develop/c10/core/TensorOptions.h
 --- pytorch-v1.5.0/c10/core/TensorOptions.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/core/TensorOptions.h	2021-07-22 20:24:34.197999266 +0800
++++ pytorch-develop/c10/core/TensorOptions.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7694,7 +7729,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/c10/macros/Export.h pytorch-develop/c10/macros/Export.h
 --- pytorch-v1.5.0/c10/macros/Export.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/c10/macros/Export.h	2021-07-22 20:24:34.197999266 +0800
++++ pytorch-develop/c10/macros/Export.h	2021-07-23 18:20:43.717375940 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -7821,7 +7856,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/caffe2/CMakeLists.txt pytorch-develop/caffe2/CMakeLists.txt
 --- pytorch-v1.5.0/caffe2/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/caffe2/CMakeLists.txt	2021-07-22 20:24:34.205999553 +0800
++++ pytorch-develop/caffe2/CMakeLists.txt	2021-07-23 18:20:43.725376226 +0800
 @@ -32,6 +32,7 @@
    # Add source, includes, and libs to lists
    list(APPEND Caffe2_CPU_SRCS ${ATen_CPU_SRCS})
@@ -7968,7 +8003,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    # Call again since Caffe2_HIP_INCLUDE is extended with ATen include dirs.
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/.clang-format pytorch-develop/.clang-format
 --- pytorch-v1.5.0/.clang-format	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/.clang-format	2021-07-22 20:24:34.085995250 +0800
++++ pytorch-develop/.clang-format	2021-07-23 18:20:43.597371637 +0800
 @@ -84,5 +84,4 @@
  SpacesInSquareBrackets: false
  Standard:        Cpp11
@@ -7979,7 +8014,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/cmake/BuildVariables.cmake pytorch-develop/cmake/BuildVariables.cmake
 --- pytorch-v1.5.0/cmake/BuildVariables.cmake	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/cmake/BuildVariables.cmake	2021-07-22 20:24:34.318003569 +0800
++++ pytorch-develop/cmake/BuildVariables.cmake	2021-07-23 18:20:43.837380242 +0800
 @@ -11,6 +11,7 @@
  # CMakeLists.txt files under each folder respectively.
  set(Caffe2_CPU_SRCS)
@@ -8006,7 +8041,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  # symbols. However, if the lib is whole linked in caffe2 lib, we don't want
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/cmake/Codegen.cmake pytorch-develop/cmake/Codegen.cmake
 --- pytorch-v1.5.0/cmake/Codegen.cmake	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/cmake/Codegen.cmake	2021-07-22 20:24:34.318003569 +0800
++++ pytorch-develop/cmake/Codegen.cmake	2021-07-23 18:20:43.837380242 +0800
 @@ -191,13 +191,14 @@
    file(READ ${CMAKE_BINARY_DIR}/aten/src/ATen/generated_cpp.txt generated_cpp)
    file(READ ${CMAKE_BINARY_DIR}/aten/src/ATen/generated_cpp.txt-cuda cuda_generated_cpp)
@@ -8037,7 +8072,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  endif()
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/cmake/Dependencies.cmake pytorch-develop/cmake/Dependencies.cmake
 --- pytorch-v1.5.0/cmake/Dependencies.cmake	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/cmake/Dependencies.cmake	2021-07-22 20:24:34.318003569 +0800
++++ pytorch-develop/cmake/Dependencies.cmake	2021-07-23 18:20:43.837380242 +0800
 @@ -1509,6 +1509,13 @@
    ENDIF(NOT C_HAS_THREAD)
  endif()
@@ -8054,7 +8089,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  #
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/cmake/Summary.cmake pytorch-develop/cmake/Summary.cmake
 --- pytorch-v1.5.0/cmake/Summary.cmake	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/cmake/Summary.cmake	2021-07-22 20:24:34.318003569 +0800
++++ pytorch-develop/cmake/Summary.cmake	2021-07-23 18:20:43.837380242 +0800
 @@ -134,6 +134,7 @@
    if(NOT "${SELECTED_OP_LIST}" STREQUAL "")
      message(STATUS "  SELECTED_OP_LIST    : ${SELECTED_OP_LIST}")
@@ -8065,7 +8100,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  endfunction()
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/cmake/TorchConfig.cmake.in pytorch-develop/cmake/TorchConfig.cmake.in
 --- pytorch-v1.5.0/cmake/TorchConfig.cmake.in	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/cmake/TorchConfig.cmake.in	2021-07-22 20:24:34.318003569 +0800
++++ pytorch-develop/cmake/TorchConfig.cmake.in	2021-07-23 18:20:43.837380242 +0800
 @@ -112,6 +112,11 @@
    list(APPEND TORCH_LIBRARIES ${TORCH_CUDA_LIBRARIES})
  endif()
@@ -8080,7 +8115,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    set(TORCH_CXX_FLAGS "-D_GLIBCXX_USE_CXX11_ABI=@GLIBCXX_USE_CXX11_ABI@")
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/CMakeLists.txt pytorch-develop/CMakeLists.txt
 --- pytorch-v1.5.0/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/CMakeLists.txt	2021-07-22 20:24:34.085995250 +0800
++++ pytorch-develop/CMakeLists.txt	2021-07-23 18:20:43.597371637 +0800
 @@ -205,6 +205,10 @@
  option(USE_TBB "Use TBB" OFF)
  option(ONNX_ML "Enable traditional ONNX ML API." ON)
@@ -8147,7 +8182,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-missing-braces")
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/.dockerignore pytorch-develop/.dockerignore
 --- pytorch-v1.5.0/.dockerignore	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/.dockerignore	2021-07-22 20:24:34.085995250 +0800
++++ pytorch-develop/.dockerignore	2021-07-23 18:20:43.597371637 +0800
 @@ -1,257 +1 @@
 -# READ THIS BEFORE YOU REFACTOR ME
 -#
@@ -8423,7 +8458,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/requirements.txt pytorch-develop/requirements.txt
 --- pytorch-v1.5.0/requirements.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/requirements.txt	2021-07-22 20:24:34.338004286 +0800
++++ pytorch-develop/requirements.txt	2021-07-23 18:20:43.857380960 +0800
 @@ -4,4 +4,12 @@
  requests
  setuptools
@@ -8442,7 +8477,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/setup.py pytorch-develop/setup.py
 --- pytorch-v1.5.0/setup.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/setup.py	2021-07-22 20:24:34.338004286 +0800
++++ pytorch-develop/setup.py	2021-07-23 18:20:43.857380960 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -8541,7 +8576,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
                  'python/serialized_test/data/operator_test/*.zip',
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/derivatives.yaml pytorch-develop/tools/autograd/derivatives.yaml
 --- pytorch-v1.5.0/tools/autograd/derivatives.yaml	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/derivatives.yaml	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/derivatives.yaml	2021-07-23 18:20:44.997421834 +0800
 @@ -107,6 +107,10 @@
  #
  # NB: The parameter names here MUST be consistent with the parameter names
@@ -8657,7 +8692,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/dump_utils.py pytorch-develop/tools/autograd/dump_utils.py
 --- pytorch-v1.5.0/tools/autograd/dump_utils.py	1970-01-01 08:00:00.000000000 +0800
-+++ pytorch-develop/tools/autograd/dump_utils.py	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/dump_utils.py	2021-07-23 18:20:44.997421834 +0800
 @@ -0,0 +1,115 @@
 +# Copyright (c) 2021 Huawei Technologies Co., Ltd
 +# All rights reserved.
@@ -8776,7 +8811,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 +]
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/gen_autograd_functions.py pytorch-develop/tools/autograd/gen_autograd_functions.py
 --- pytorch-v1.5.0/tools/autograd/gen_autograd_functions.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/gen_autograd_functions.py	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/gen_autograd_functions.py	2021-07-23 18:20:44.997421834 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2021 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -8962,7 +8997,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 +
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/gen_python_functions.py pytorch-develop/tools/autograd/gen_python_functions.py
 --- pytorch-v1.5.0/tools/autograd/gen_python_functions.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/gen_python_functions.py	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/gen_python_functions.py	2021-07-23 18:20:44.997421834 +0800
 @@ -1,3 +1,20 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -9004,7 +9039,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
              'value': argname,
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/gen_variable_type.py pytorch-develop/tools/autograd/gen_variable_type.py
 --- pytorch-v1.5.0/tools/autograd/gen_variable_type.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/gen_variable_type.py	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/gen_variable_type.py	2021-07-23 18:20:44.997421834 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2021 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -9177,7 +9212,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/templates/Functions.cpp pytorch-develop/tools/autograd/templates/Functions.cpp
 --- pytorch-v1.5.0/tools/autograd/templates/Functions.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/templates/Functions.cpp	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/templates/Functions.cpp	2021-07-23 18:20:45.001421978 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2021 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -9257,7 +9292,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    auto sparse = sparse_.coalesce();
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/templates/python_torch_functions.cpp pytorch-develop/tools/autograd/templates/python_torch_functions.cpp
 --- pytorch-v1.5.0/tools/autograd/templates/python_torch_functions.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/templates/python_torch_functions.cpp	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/templates/python_torch_functions.cpp	2021-07-23 18:20:45.001421978 +0800
 @@ -22,7 +22,7 @@
  #include "torch/csrc/autograd/generated/variable_factories.h"
  #include "torch/csrc/utils/structseq.h"
@@ -9341,7 +9376,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/templates/python_variable_methods.cpp pytorch-develop/tools/autograd/templates/python_variable_methods.cpp
 --- pytorch-v1.5.0/tools/autograd/templates/python_variable_methods.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/templates/python_variable_methods.cpp	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/templates/python_variable_methods.cpp	2021-07-23 18:20:45.001421978 +0800
 @@ -15,7 +15,13 @@
  #include "torch/csrc/cuda/Stream.h"
  #include "torch/csrc/cuda/Event.h"
@@ -9428,7 +9463,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    {"has_names", (PyCFunction)THPVariable_has_names, METH_NOARGS, NULL},
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/templates/VariableType.cpp pytorch-develop/tools/autograd/templates/VariableType.cpp
 --- pytorch-v1.5.0/tools/autograd/templates/VariableType.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/templates/VariableType.cpp	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/templates/VariableType.cpp	2021-07-23 18:20:45.001421978 +0800
 @@ -1,7 +1,27 @@
 +// Copyright (c) 2021 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -9459,7 +9494,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/autograd/templates/VariableType.h pytorch-develop/tools/autograd/templates/VariableType.h
 --- pytorch-v1.5.0/tools/autograd/templates/VariableType.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/autograd/templates/VariableType.h	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/autograd/templates/VariableType.h	2021-07-23 18:20:45.001421978 +0800
 @@ -1,3 +1,20 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -9491,7 +9526,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    const at::Tensor & unpack(const Tensor & t, const char * name, int pos);
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/tools/build_variables.bzl pytorch-develop/tools/build_variables.bzl
 --- pytorch-v1.5.0/tools/build_variables.bzl	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/tools/build_variables.bzl	2021-07-22 20:24:35.482045305 +0800
++++ pytorch-develop/tools/build_variables.bzl	2021-07-23 18:20:45.001421978 +0800
 @@ -46,6 +46,7 @@
      "torch/csrc/autograd/functions/utils.cpp",
      "torch/csrc/autograd/input_buffer.cpp",
@@ -9577,7 +9612,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -def grad(outputs: _TensorOrTensors, inputs: _TensorOrTensors, grad_outputs: Optional[_TensorOrTensors]=..., retain_graph: Optional[bool]=..., create_graph: bool=..., only_inputs: bool=..., allow_unused: bool=...) -> Tuple[Tensor, ...]: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/autograd/profiler.py pytorch-develop/torch/autograd/profiler.py
 --- pytorch-v1.5.0/torch/autograd/profiler.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/autograd/profiler.py	2021-07-22 20:24:35.486045448 +0800
++++ pytorch-develop/torch/autograd/profiler.py	2021-07-23 18:20:45.005422122 +0800
 @@ -1,8 +1,25 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -10050,7 +10085,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      return ''.join(result)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/CMakeLists.txt pytorch-develop/torch/CMakeLists.txt
 --- pytorch-v1.5.0/torch/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/CMakeLists.txt	2021-07-22 20:24:35.486045448 +0800
++++ pytorch-develop/torch/CMakeLists.txt	2021-07-23 18:20:45.001421978 +0800
 @@ -97,6 +97,7 @@
      ${TORCH_SRC_DIR}/csrc/tensor/python_tensor.cpp
      ${TORCH_SRC_DIR}/csrc/utils.cpp
@@ -10082,7 +10117,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  endif()
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/engine.cpp pytorch-develop/torch/csrc/autograd/engine.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/engine.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/engine.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/engine.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10205,7 +10240,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        auto event = c10::Event{c10::DeviceType::CUDA};
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/functions/tensor.cpp pytorch-develop/torch/csrc/autograd/functions/tensor.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/functions/tensor.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/functions/tensor.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/functions/tensor.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10237,7 +10272,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
            /*non_blocking=*/false,
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/init.cpp pytorch-develop/torch/csrc/autograd/init.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/init.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/init.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/init.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10280,7 +10315,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    m.def("_enable_profiler", enableProfiler);
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/input_buffer.cpp pytorch-develop/torch/csrc/autograd/input_buffer.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/input_buffer.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/input_buffer.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/input_buffer.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10332,7 +10367,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    auto& old_var = buffer[pos];
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/profiler.cpp pytorch-develop/torch/csrc/autograd/profiler.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/profiler.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/profiler.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/profiler.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10528,7 +10563,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  CUDAStubs::~CUDAStubs() = default;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/profiler.h pytorch-develop/torch/csrc/autograd/profiler.h
 --- pytorch-v1.5.0/torch/csrc/autograd/profiler.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/profiler.h	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/profiler.h	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10653,7 +10688,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/python_variable.cpp pytorch-develop/torch/csrc/autograd/python_variable.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/python_variable.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/python_variable.cpp	2021-07-22 20:24:35.502046021 +0800
++++ pytorch-develop/torch/csrc/autograd/python_variable.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10707,7 +10742,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    {"is_complex", (getter)THPVariable_is_complex, nullptr, nullptr, nullptr},
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/python_variable_indexing.cpp pytorch-develop/torch/csrc/autograd/python_variable_indexing.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/python_variable_indexing.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/python_variable_indexing.cpp	2021-07-22 20:24:35.502046021 +0800
++++ pytorch-develop/torch/csrc/autograd/python_variable_indexing.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10748,7 +10783,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/utils/wrap_outputs.h pytorch-develop/torch/csrc/autograd/utils/wrap_outputs.h
 --- pytorch-v1.5.0/torch/csrc/autograd/utils/wrap_outputs.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/utils/wrap_outputs.h	2021-07-22 20:24:35.502046021 +0800
++++ pytorch-develop/torch/csrc/autograd/utils/wrap_outputs.h	2021-07-23 18:20:45.017422551 +0800
 @@ -168,6 +168,45 @@
    return r.release();
  }
@@ -10797,7 +10832,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    if (!r) throw python_error();
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/autograd/VariableTypeManual.cpp pytorch-develop/torch/csrc/autograd/VariableTypeManual.cpp
 --- pytorch-v1.5.0/torch/csrc/autograd/VariableTypeManual.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/autograd/VariableTypeManual.cpp	2021-07-22 20:24:35.498045878 +0800
++++ pytorch-develop/torch/csrc/autograd/VariableTypeManual.cpp	2021-07-23 18:20:45.017422551 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10831,7 +10866,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    if (!t.defined()) {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/distributed/c10d/comm.cpp pytorch-develop/torch/csrc/distributed/c10d/comm.cpp
 --- pytorch-v1.5.0/torch/csrc/distributed/c10d/comm.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/distributed/c10d/comm.cpp	2021-07-22 20:24:35.502046021 +0800
++++ pytorch-develop/torch/csrc/distributed/c10d/comm.cpp	2021-07-23 18:20:45.021422695 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10937,7 +10972,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    while (!in_flight.empty()) {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/distributed/c10d/init.cpp pytorch-develop/torch/csrc/distributed/c10d/init.cpp
 --- pytorch-v1.5.0/torch/csrc/distributed/c10d/init.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/distributed/c10d/init.cpp	2021-07-22 20:24:35.502046021 +0800
++++ pytorch-develop/torch/csrc/distributed/c10d/init.cpp	2021-07-23 18:20:45.021422695 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -10994,7 +11029,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        .def("is_success", &::c10d::ProcessGroup::Work::isSuccess)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/distributed/c10d/reducer.cpp pytorch-develop/torch/csrc/distributed/c10d/reducer.cpp
 --- pytorch-v1.5.0/torch/csrc/distributed/c10d/reducer.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/distributed/c10d/reducer.cpp	2021-07-22 20:24:35.506046165 +0800
++++ pytorch-develop/torch/csrc/distributed/c10d/reducer.cpp	2021-07-23 18:20:45.021422695 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11119,7 +11154,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  }
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/DynamicTypes.cpp pytorch-develop/torch/csrc/DynamicTypes.cpp
 --- pytorch-v1.5.0/torch/csrc/DynamicTypes.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/DynamicTypes.cpp	2021-07-22 20:24:35.490045592 +0800
++++ pytorch-develop/torch/csrc/DynamicTypes.cpp	2021-07-23 18:20:45.005422122 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11168,7 +11203,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      return it->second;
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/Generator.cpp pytorch-develop/torch/csrc/Generator.cpp
 --- pytorch-v1.5.0/torch/csrc/Generator.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/Generator.cpp	2021-07-22 20:24:35.490045592 +0800
++++ pytorch-develop/torch/csrc/Generator.cpp	2021-07-23 18:20:45.009422265 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11236,7 +11271,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  #endif 
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/generic/serialization.cpp pytorch-develop/torch/csrc/generic/serialization.cpp
 --- pytorch-v1.5.0/torch/csrc/generic/serialization.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/generic/serialization.cpp	2021-07-22 20:24:35.506046165 +0800
++++ pytorch-develop/torch/csrc/generic/serialization.cpp	2021-07-23 18:20:45.025422838 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11336,7 +11371,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/generic/Storage.cpp pytorch-develop/torch/csrc/generic/Storage.cpp
 --- pytorch-v1.5.0/torch/csrc/generic/Storage.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/generic/Storage.cpp	2021-07-22 20:24:35.506046165 +0800
++++ pytorch-develop/torch/csrc/generic/Storage.cpp	2021-07-23 18:20:45.025422838 +0800
 @@ -1,7 +1,25 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11415,7 +11450,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
        for (Py_ssize_t i = 0; i < length; i++) {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/generic/StorageMethods.cpp pytorch-develop/torch/csrc/generic/StorageMethods.cpp
 --- pytorch-v1.5.0/torch/csrc/generic/StorageMethods.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/generic/StorageMethods.cpp	2021-07-22 20:24:35.506046165 +0800
++++ pytorch-develop/torch/csrc/generic/StorageMethods.cpp	2021-07-23 18:20:45.025422838 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11463,7 +11498,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    {"_write_file", (PyCFunction)THPStorage_(writeFile), METH_VARARGS, nullptr},
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/Module.cpp pytorch-develop/torch/csrc/Module.cpp
 --- pytorch-v1.5.0/torch/csrc/Module.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/Module.cpp	2021-07-22 20:24:35.490045592 +0800
++++ pytorch-develop/torch/csrc/Module.cpp	2021-07-23 18:20:45.009422265 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11607,7 +11642,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    auto set_module_attr = [&](const char* name, PyObject* v, bool incref = true) {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/tensor/python_tensor.cpp pytorch-develop/torch/csrc/tensor/python_tensor.cpp
 --- pytorch-v1.5.0/torch/csrc/tensor/python_tensor.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/tensor/python_tensor.cpp	2021-07-22 20:24:35.526046882 +0800
++++ pytorch-develop/torch/csrc/tensor/python_tensor.cpp	2021-07-23 18:20:45.045423556 +0800
 @@ -1,18 +1,35 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -11984,7 +12019,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 +} // namespace torch
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/init.cpp pytorch-develop/torch/csrc/utils/init.cpp
 --- pytorch-v1.5.0/torch/csrc/utils/init.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/init.cpp	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/init.cpp	2021-07-23 18:20:45.045423556 +0800
 @@ -1,6 +1,10 @@
  #include <ATen/core/ivalue.h>
  #include <torch/csrc/utils/init.h>
@@ -12072,7 +12107,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  } // namespace torch
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/init.h pytorch-develop/torch/csrc/utils/init.h
 --- pytorch-v1.5.0/torch/csrc/utils/init.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/init.h	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/init.h	2021-07-23 18:20:45.045423556 +0800
 @@ -8,4 +8,7 @@
  void initThroughputBenchmarkBindings(PyObject* module);
  
@@ -12083,7 +12118,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  } // namespace torch
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/python_arg_parser.h pytorch-develop/torch/csrc/utils/python_arg_parser.h
 --- pytorch-v1.5.0/torch/csrc/utils/python_arg_parser.h	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/python_arg_parser.h	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/python_arg_parser.h	2021-07-23 18:20:45.045423556 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -12118,7 +12153,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    return at::Device(device_str);
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/tensor_layouts.cpp pytorch-develop/torch/csrc/utils/tensor_layouts.cpp
 --- pytorch-v1.5.0/torch/csrc/utils/tensor_layouts.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/tensor_layouts.cpp	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/tensor_layouts.cpp	2021-07-23 18:20:45.049423699 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -12149,7 +12184,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    registerLayoutObject((THPLayout*)strided_layout, at::Backend::QuantizedCPU);
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/tensor_new.cpp pytorch-develop/torch/csrc/utils/tensor_new.cpp
 --- pytorch-v1.5.0/torch/csrc/utils/tensor_new.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/tensor_new.cpp	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/tensor_new.cpp	2021-07-23 18:20:45.049423699 +0800
 @@ -1,3 +1,19 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -12285,7 +12320,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    } else if(expected_layout == c10::kSparse) {
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/csrc/utils/tensor_types.cpp pytorch-develop/torch/csrc/utils/tensor_types.cpp
 --- pytorch-v1.5.0/torch/csrc/utils/tensor_types.cpp	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/csrc/utils/tensor_types.cpp	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/csrc/utils/tensor_types.cpp	2021-07-23 18:20:45.049423699 +0800
 @@ -1,58 +1,91 @@
 +// Copyright (c) 2020 Huawei Technologies Co., Ltd
 +// Copyright (c) 2019, Facebook CORPORATION. 
@@ -12498,7 +12533,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -def get_rng_state(): ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/distributed/distributed_c10d.py pytorch-develop/torch/distributed/distributed_c10d.py
 --- pytorch-v1.5.0/torch/distributed/distributed_c10d.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/distributed/distributed_c10d.py	2021-07-22 20:24:35.530047026 +0800
++++ pytorch-develop/torch/distributed/distributed_c10d.py	2021-07-23 18:20:45.049423699 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -12579,7 +12614,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/__init__.py pytorch-develop/torch/__init__.py
 --- pytorch-v1.5.0/torch/__init__.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/__init__.py	2021-07-22 20:24:35.486045448 +0800
++++ pytorch-develop/torch/__init__.py	2021-07-23 18:20:45.001421978 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -12622,7 +12657,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 \ No newline at end of file
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/lib/c10d/CMakeLists.txt pytorch-develop/torch/lib/c10d/CMakeLists.txt
 --- pytorch-v1.5.0/torch/lib/c10d/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/lib/c10d/CMakeLists.txt	2021-07-22 20:24:35.534047169 +0800
++++ pytorch-develop/torch/lib/c10d/CMakeLists.txt	2021-07-23 18:20:45.053423842 +0800
 @@ -28,6 +28,10 @@
    option(USE_C10D_NCCL "USE C10D NCCL" ON)
  endif()
@@ -12675,7 +12710,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
    copy_header(ProcessGroupMPI.hpp)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/lib/libshm/CMakeLists.txt pytorch-develop/torch/lib/libshm/CMakeLists.txt
 --- pytorch-v1.5.0/torch/lib/libshm/CMakeLists.txt	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/lib/libshm/CMakeLists.txt	2021-07-22 20:24:35.538047312 +0800
++++ pytorch-develop/torch/lib/libshm/CMakeLists.txt	2021-07-23 18:20:45.057423986 +0800
 @@ -37,8 +37,11 @@
  SET_TARGET_PROPERTIES(shm PROPERTIES
    PREFIX "lib"
@@ -12732,7 +12767,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -_maybe_indices_t = _scalar_or_tuple_2_t[Tensor]
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/nn/functional.py pytorch-develop/torch/nn/functional.py
 --- pytorch-v1.5.0/torch/nn/functional.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/nn/functional.py	2021-07-22 20:24:35.538047312 +0800
++++ pytorch-develop/torch/nn/functional.py	2021-07-23 18:20:45.057423986 +0800
 @@ -1611,7 +1611,7 @@
      else:
          output = input.matmul(weight.t())
@@ -12755,7 +12790,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -from . import parallel as parallel
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/nn/modules/batchnorm.py pytorch-develop/torch/nn/modules/batchnorm.py
 --- pytorch-v1.5.0/torch/nn/modules/batchnorm.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/nn/modules/batchnorm.py	2021-07-22 20:24:35.542047456 +0800
++++ pytorch-develop/torch/nn/modules/batchnorm.py	2021-07-23 18:20:45.057423986 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -12787,7 +12822,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
              self.register_parameter('running_var', None)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/nn/modules/module.py pytorch-develop/torch/nn/modules/module.py
 --- pytorch-v1.5.0/torch/nn/modules/module.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/nn/modules/module.py	2021-07-22 20:24:35.542047456 +0800
++++ pytorch-develop/torch/nn/modules/module.py	2021-07-23 18:20:45.061424129 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -12930,7 +12965,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
                  return t.to(device, dtype if t.is_floating_point() else None, non_blocking, memory_format=convert_to_format)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/nn/modules/normalization.py pytorch-develop/torch/nn/modules/normalization.py
 --- pytorch-v1.5.0/torch/nn/modules/normalization.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/nn/modules/normalization.py	2021-07-22 20:24:35.542047456 +0800
++++ pytorch-develop/torch/nn/modules/normalization.py	2021-07-23 18:20:45.061424129 +0800
 @@ -128,13 +128,14 @@
      """
      __constants__ = ['normalized_shape', 'eps', 'elementwise_affine']
@@ -12999,7 +13034,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -                  module_kwargs: Optional[Any] = ...) -> Tensor: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/nn/parallel/distributed.py pytorch-develop/torch/nn/parallel/distributed.py
 --- pytorch-v1.5.0/torch/nn/parallel/distributed.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/nn/parallel/distributed.py	2021-07-22 20:24:35.542047456 +0800
++++ pytorch-develop/torch/nn/parallel/distributed.py	2021-07-23 18:20:45.061424129 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -13350,7 +13385,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -def remove_weight_norm(module: T_module, name: str = ...) -> T_module: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/onnx/symbolic_opset9.py pytorch-develop/torch/onnx/symbolic_opset9.py
 --- pytorch-v1.5.0/torch/onnx/symbolic_opset9.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/onnx/symbolic_opset9.py	2021-07-22 20:24:35.546047600 +0800
++++ pytorch-develop/torch/onnx/symbolic_opset9.py	2021-07-23 18:20:45.065424272 +0800
 @@ -1621,14 +1621,23 @@
          slices = [sym_help._slice_helper(g, w, axes=[0], starts=[x * n], ends=[y * n]) for x, y in intervals]
          return g.op('Concat', *slices, axis_i=0)
@@ -13428,7 +13463,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -    def __init__(self, params: _params_t, lr: float=..., lr_decay: float=..., weight_decay: float=..., initial_accumulator_value: float=...,  eps: float=...) -> None: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/optim/adamax.py pytorch-develop/torch/optim/adamax.py
 --- pytorch-v1.5.0/torch/optim/adamax.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/optim/adamax.py	2021-07-22 20:24:35.546047600 +0800
++++ pytorch-develop/torch/optim/adamax.py	2021-07-23 18:20:45.065424272 +0800
 @@ -80,8 +80,8 @@
                      exp_inf.mul_(beta2).unsqueeze(0),
                      grad.abs().add_(eps).unsqueeze_(0)
@@ -13605,7 +13640,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -    def __init__(self, params: _params_t, lr: float=..., betas: Tuple[float, float]=..., eps: float=...) -> None: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/serialization.py pytorch-develop/torch/serialization.py
 --- pytorch-v1.5.0/torch/serialization.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/serialization.py	2021-07-22 20:24:35.550047743 +0800
++++ pytorch-develop/torch/serialization.py	2021-07-23 18:20:45.065424272 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -13689,7 +13724,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  def location_tag(storage):
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/storage.py pytorch-develop/torch/storage.py
 --- pytorch-v1.5.0/torch/storage.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/storage.py	2021-07-22 20:24:35.550047743 +0800
++++ pytorch-develop/torch/storage.py	2021-07-23 18:20:45.065424272 +0800
 @@ -7,6 +7,7 @@
  
  class _StorageBase(object):
@@ -13709,7 +13744,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
          else:
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/tensor.py pytorch-develop/torch/tensor.py
 --- pytorch-v1.5.0/torch/tensor.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/tensor.py	2021-07-22 20:24:35.550047743 +0800
++++ pytorch-develop/torch/tensor.py	2021-07-23 18:20:45.065424272 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -13771,7 +13806,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      def __reversed__(self):
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/_tensor_str.py pytorch-develop/torch/_tensor_str.py
 --- pytorch-v1.5.0/torch/_tensor_str.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/_tensor_str.py	2021-07-22 20:24:35.486045448 +0800
++++ pytorch-develop/torch/_tensor_str.py	2021-07-23 18:20:45.005422122 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -13825,7 +13860,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
      has_default_dtype = self.dtype in (torch.get_default_dtype(), torch.int64, torch.bool)
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/utils/data/dataloader.py pytorch-develop/torch/utils/data/dataloader.py
 --- pytorch-v1.5.0/torch/utils/data/dataloader.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/utils/data/dataloader.py	2021-07-22 20:24:35.554047887 +0800
++++ pytorch-develop/torch/utils/data/dataloader.py	2021-07-23 18:20:45.069424416 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -14034,7 +14069,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -    def __init__(self, sampler: Sampler[int], batch_size: int, drop_last: bool) -> None: ...
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/utils/data/_utils/pin_memory.py pytorch-develop/torch/utils/data/_utils/pin_memory.py
 --- pytorch-v1.5.0/torch/utils/data/_utils/pin_memory.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/utils/data/_utils/pin_memory.py	2021-07-22 20:24:35.550047743 +0800
++++ pytorch-develop/torch/utils/data/_utils/pin_memory.py	2021-07-23 18:20:45.069424416 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
@@ -14095,7 +14130,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
 -
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/utils/__init__.py pytorch-develop/torch/utils/__init__.py
 --- pytorch-v1.5.0/torch/utils/__init__.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/utils/__init__.py	2021-07-22 20:24:35.550047743 +0800
++++ pytorch-develop/torch/utils/__init__.py	2021-07-23 18:20:45.069424416 +0800
 @@ -1,6 +1,7 @@
  from __future__ import absolute_import, division, print_function, unicode_literals
  
@@ -14106,7 +14141,7 @@ diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=
  def set_module(obj, mod):
 diff -Nur '--exclude=.git*' '--exclude=.jenkins' '--exclude=android' '--exclude=OWNERS' '--exclude=third_party' '--exclude=README*' -Nur pytorch-v1.5.0/torch/_utils.py pytorch-develop/torch/_utils.py
 --- pytorch-v1.5.0/torch/_utils.py	2021-04-10 18:39:32.000000000 +0800
-+++ pytorch-develop/torch/_utils.py	2021-07-22 20:24:35.486045448 +0800
++++ pytorch-develop/torch/_utils.py	2021-07-23 18:20:45.005422122 +0800
 @@ -1,3 +1,19 @@
 +# Copyright (c) 2020 Huawei Technologies Co., Ltd
 +# Copyright (c) 2019, Facebook CORPORATION. 
diff --git a/src/aten/src/ATen/native/native_functions.yaml b/src/aten/src/ATen/native/native_functions.yaml
index 88248a9249..ebd9756a6e 100644
--- a/src/aten/src/ATen/native/native_functions.yaml
+++ b/src/aten/src/ATen/native/native_functions.yaml
@@ -5871,6 +5871,8 @@
   dispatch:
     CPU: _symeig_helper_cpu
     CUDA: _symeig_helper_cuda
+  npu_dispatch:
+    NPU: _symeig_helper_npu
 
 - func: eig.e(Tensor self, bool eigenvectors=False, *, Tensor(a!) e, Tensor(b!) v) -> (Tensor(a!) eigenvalues, Tensor(b!) eigenvectors)
   dispatch:
@@ -7122,22 +7124,30 @@
   dispatch:
     CPU: rrelu_with_noise_out_cpu
     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward_out
+  npu_dispatch:
+    NPU: rrelu_with_noise_out_npu
 
 - func: rrelu_with_noise(Tensor self, Tensor noise, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor
   python_module: nn
   dispatch:
     CPU: rrelu_with_noise_cpu
     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward
+  npu_dispatch:
+    NPU: rrelu_with_noise_npu
 
 - func: rrelu_with_noise_backward(Tensor grad_output, Tensor self, Tensor noise, Scalar lower, Scalar upper, bool training, bool self_is_result) -> Tensor
   use_c10_dispatcher: full
   python_module: nn
+  npu_dispatch:
+    NPU: rrelu_with_noise_backward_npu
 
 - func: rrelu_with_noise_(Tensor(a!) self, Tensor noise, Scalar lower=0.125, Scalar upper=0.3333333333333333, bool training=False, Generator? generator=None) -> Tensor(a!)
   python_module: nn
   dispatch:
     CPU: rrelu_with_noise_cpu_
     CUDA: legacy::cuda::_thnn_rrelu_with_noise_forward_
+  npu_dispatch:
+    NPU: rrelu_with_noise_npu_
 
 - func: softplus.out(Tensor self, Scalar beta=1, Scalar threshold=20, *, Tensor(a!) out) -> Tensor(a!)
   python_module: nn
diff --git a/src/aten/src/ATen/native/npu/RreluWithNoiseBackwardKernelNpu.cpp b/src/aten/src/ATen/native/npu/RreluWithNoiseBackwardKernelNpu.cpp
new file mode 100644
index 0000000000..28c4bf1d25
--- /dev/null
+++ b/src/aten/src/ATen/native/npu/RreluWithNoiseBackwardKernelNpu.cpp
@@ -0,0 +1,42 @@
+// Copyright (c) 2020 Huawei Technologies Co., Ltd
+// Copyright (c) 2019, Facebook CORPORATION.
+// All rights reserved.
+//
+// Licensed under the BSD 3-Clause License  (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ATen/native/npu/utils/OpAdapter.h"
+
+namespace at {
+namespace native {
+using namespace at::native::npu;
+
+Tensor rrelu_with_noise_backward_npu(
+    const Tensor& grad_output,
+    const Tensor& self_or_result,
+    const Tensor& noise,
+    Scalar lower,
+    Scalar upper,
+    bool training,
+    bool is_result) {
+  auto folat_lower = lower.toFloat();
+  auto float_upper = upper.toFloat();
+  if (training && (float_upper - folat_lower > 1E-6)) {
+    return grad_output.mul(noise);
+  } else {
+    Scalar negative_slope = (folat_lower + float_upper) / 2;
+    return at::leaky_relu_backward(grad_output, self_or_result, negative_slope, is_result);
+  }
+}
+
+} // namespace native
+} // namespace at
diff --git a/src/aten/src/ATen/native/npu/RreluWithNoiseKernelNpu.cpp b/src/aten/src/ATen/native/npu/RreluWithNoiseKernelNpu.cpp
new file mode 100644
index 0000000000..870a66d039
--- /dev/null
+++ b/src/aten/src/ATen/native/npu/RreluWithNoiseKernelNpu.cpp
@@ -0,0 +1,105 @@
+// Copyright (c) 2020 Huawei Technologies Co., Ltd
+// Copyright (c) 2019, Facebook CORPORATION.
+// All rights reserved.
+//
+// Licensed under the BSD 3-Clause License  (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ATen/native/npu/utils/OpAdapter.h"
+#include <ATen/core/DistributionsHelper.h>
+
+namespace at {
+namespace native {
+using namespace at::native::npu;
+
+
+void _rrelu_with_noise_train(
+    Tensor& output,
+    const Tensor& input,
+    const Tensor& noise,
+    Scalar lower_,
+    Scalar upper_,
+    Generator* generator) {
+  float lower = lower_.toFloat();
+  float upper = upper_.toFloat();
+  auto shape = output.sizes();
+  auto noise_shape = noise.sizes();
+  Tensor tmp_tensor = output.contiguous();
+  Tensor output_data = tmp_tensor.reshape({output.numel()});
+  Tensor input_data = input.reshape({input.numel()});
+  Tensor tmp_noise = noise;
+  tmp_noise = tmp_noise.reshape({tmp_noise.numel()});
+  auto gen = at::get_generator_or_default<CPUGenerator>(generator, detail::getDefaultCPUGenerator());
+
+  for (int64_t i = 0; i < input.numel(); i++) {
+    if (input_data[i].item().toFloat() <= 0) {
+      at::uniform_real_distribution<double> uniform(lower, upper);
+      const float r = uniform(gen);
+      output_data[i] = input_data[i] * r;
+      tmp_noise[i] = r;
+    } else {
+      tmp_noise[i] = 1;
+      output_data[i] = input_data[i];
+    }
+  }
+  if (!output.is_contiguous()) {
+    output.copy_(tmp_tensor);
+  }
+  tmp_noise.reshape(noise_shape);
+  noise.copy_(tmp_noise);
+  output.reshape(shape);
+}
+
+Tensor rrelu_with_noise_npu(
+    const Tensor& self,
+    const Tensor& noise,
+    Scalar lower,
+    Scalar upper,
+    bool training,
+    Generator* generator) {
+  auto output = at::empty_like(self, LEGACY_CONTIGUOUS_MEMORY_FORMAT);
+  return rrelu_with_noise_out_npu(output, self, noise, lower, upper, training, generator);
+}
+
+Tensor& rrelu_with_noise_npu_(
+    Tensor& self,
+    const Tensor& noise,
+    Scalar lower,
+    Scalar upper,
+    bool training,
+    Generator* generator) {
+  return rrelu_with_noise_out_npu(self, self, noise, lower, upper, training, generator);
+}
+
+Tensor& rrelu_with_noise_out_npu(
+    Tensor& output,
+    const Tensor& self,
+    const Tensor& noise,
+    Scalar lower,
+    Scalar upper,
+    bool training,
+    Generator* generator) {
+  
+  if (training) {
+    _rrelu_with_noise_train(output, self.contiguous(), noise, lower, upper, generator);
+    return output;
+  } else {
+    auto float_lower = lower.toFloat();
+    auto float_upper = upper.toFloat();
+    Scalar negative_slope = (float_lower + float_upper) / 2;
+    return at::leaky_relu_out(output, self, negative_slope);
+  }
+}
+
+
+} // namespace native
+} // namespace at
diff --git a/src/aten/src/ATen/native/npu/SymeigKernelNpu.cpp b/src/aten/src/ATen/native/npu/SymeigKernelNpu.cpp
new file mode 100644
index 0000000000..f15d9e3b7c
--- /dev/null
+++ b/src/aten/src/ATen/native/npu/SymeigKernelNpu.cpp
@@ -0,0 +1,50 @@
+// Copyright (c) 2020 Huawei Technologies Co., Ltd
+// Copyright (c) 2019, Facebook CORPORATION. 
+// All rights reserved.
+//
+// Licensed under the BSD 3-Clause License  (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ATen/native/npu/utils/OpAdapter.h"
+#include<ATen/NamedTensorUtils.h>
+
+namespace at {
+namespace native {
+using namespace at::native::npu;
+
+std::tuple<Tensor, Tensor> _symeig_helper_npu(const Tensor& self, bool eigenvectors, bool upper) {
+  auto self_sizes = self.sizes().vec();
+  self_sizes.pop_back();
+  auto eigvals = at::empty(self_sizes, self.options());
+
+  if (self.numel() == 0) {
+    return std::tuple<Tensor, Tensor>(eigvals, at::empty_like(self, LEGACY_CONTIGUOUS_MEMORY_FORMAT));
+  }
+
+  auto self_working_copy = self.clone();
+  OpCommand cmd;
+  cmd.Name("SelfAdjointEig")
+      .Input(self)
+      .Output(eigvals)
+      .Output(self_working_copy)
+      .Attr("compute_v", true)
+      .Run();
+
+  if (eigenvectors) {
+    return std::tuple<Tensor, Tensor>(eigvals, self_working_copy);
+  } else {
+    return std::tuple<Tensor, Tensor>(eigvals, at::empty({0}, self.options()));
+  }
+}
+
+} // namespace native
+} // namespace at
diff --git a/test/test_npu/test_network_ops/test_rrelu_with_noise.py b/test/test_npu/test_network_ops/test_rrelu_with_noise.py
new file mode 100644
index 0000000000..959aca5a37
--- /dev/null
+++ b/test/test_npu/test_network_ops/test_rrelu_with_noise.py
@@ -0,0 +1,73 @@
+# Copyright (c) 2020, Huawei Technologies.
+# Copyright (c) 2019, Facebook CORPORATION.
+# All rights reserved.
+#
+# Licensed under the BSD 3-Clause License  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import numpy as np
+import sys
+import copy
+from common_utils import TestCase, run_tests
+from common_device_type import dtypes, instantiate_device_type_tests
+from util_test import create_common_tensor
+
+class TestRreluWithNoise(TestCase):
+    def cpu_op_exec(self, input1, input2):
+        output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+        output = output.numpy()
+        return output
+
+    def npu_op_exec(self, input1, input2):
+        output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+        output = output.to("cpu")
+        output = output.numpy()
+        return output
+
+    def test_leaky_relu_shape_format(self, device):
+        shape_format = [
+               [ [np.float32, 0, (1, 6, 4)], [np.float32, 0, (1, 4, 8)], [np.float32, 0, (1, 6, 8)]],
+               [ [np.float32, 3, (2, 4, 5)], [np.float32, 3, (2, 5, 10)],  [np.float32, 3, (2, 4, 10)]]
+               ]
+        for item in shape_format:
+            cpu_input1, npu_input1 = create_common_tensor(item[0], 1, 100)
+            cpu_input2, npu_input2 = create_common_tensor(item[0], 1, 100)
+            cpu_output = self.cpu_op_exec(cpu_input1, cpu_input2)
+            npu_output = self.npu_op_exec(npu_input1, npu_input2)
+            self.assertRtolEqual(cpu_output, npu_output)
+
+    def test_leaky_relu_shape_format_fp16(self, device):
+        shape_format = [
+               [ [np.float16, 0, (1, 6, 4)]],
+               [ [np.float16, 3, (2, 4, 5)]]
+               ]
+        def cpu_op_exec_fp16(input1, input2):
+            input1 = input1.to(torch.float32)
+            input2 = input2.to(torch.float32)
+            output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+            output = output.numpy()
+            output = output.astype(np.float16)
+            return output
+        for item in shape_format:
+            cpu_input1, npu_input1 = create_common_tensor(item[0], 1, 100)
+            cpu_input2, npu_input2 = create_common_tensor(item[0], 1, 100)
+            cpu_output = cpu_op_exec_fp16(cpu_input1, cpu_input2)
+            npu_output = self.npu_op_exec(npu_input1, npu_input2)
+            self.assertRtolEqual(cpu_output, npu_output)
+
+
+instantiate_device_type_tests(TestRreluWithNoise, globals(), except_for="cpu")
+if __name__ == "__main__":
+    run_tests()
+
+
diff --git a/test/test_npu/test_network_ops/test_rrelu_with_noise_backward.py b/test/test_npu/test_network_ops/test_rrelu_with_noise_backward.py
new file mode 100644
index 0000000000..df8a7a0ebe
--- /dev/null
+++ b/test/test_npu/test_network_ops/test_rrelu_with_noise_backward.py
@@ -0,0 +1,81 @@
+# Copyright (c) 2020, Huawei Technologies.
+# Copyright (c) 2019, Facebook CORPORATION.
+# All rights reserved.
+#
+# Licensed under the BSD 3-Clause License  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+import numpy as np
+import sys
+import copy
+from common_utils import TestCase, run_tests
+from common_device_type import dtypes, instantiate_device_type_tests
+from util_test import create_common_tensor
+
+class TestRreluWithNoiseBackward(TestCase):
+    def cpu_op_exec(self, input1, input2):
+        input1.requires_grad_(True)
+        output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+        output.backward(torch.ones_like(input1))
+        res = input1.grad
+        res = res.numpy()
+        return res
+
+    def npu_op_exec(self, input1, input2):
+        input1.requires_grad_(True)
+        output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+        output.backward(torch.ones_like(input1))
+        res = input1.grad.to("cpu")
+        res = res.numpy()
+        return res
+
+    def test_leaky_relu_shape_format(self, device):
+        shape_format = [
+               [ [np.float32, 0, (1, 6, 4)], [np.float32, 0, (1, 4, 8)], [np.float32, 0, (1, 6, 8)]],
+               [ [np.float32, 3, (2, 4, 5)], [np.float32, 3, (2, 5, 10)],  [np.float32, 3, (2, 4, 10)]]
+               ]
+        for item in shape_format:
+            cpu_input1, npu_input1 = create_common_tensor(item[0], 1, 100)
+            cpu_input2, npu_input2 = create_common_tensor(item[0], 1, 100)
+            cpu_output = self.cpu_op_exec(cpu_input1, cpu_input2)
+            npu_output = self.npu_op_exec(npu_input1, npu_input2)
+            self.assertRtolEqual(cpu_output, npu_output)
+
+    def test_leaky_relu_shape_format_fp16(self, device):
+        shape_format = [
+               [ [np.float16, 0, (1, 6, 4)]],
+               [ [np.float16, 3, (2, 4, 5)]]
+               ]
+        def cpu_op_exec_fp16(input1, input2):
+            input1 = input1.to(torch.float32)
+            input2 = input2.to(torch.float32)
+            input1.requires_grad_(True)
+            output = torch._C._nn.rrelu_with_noise(input1, input2, 0.1, 0.3)
+            output.backward(torch.ones_like(input1))
+            res = input1.grad
+            res = res.numpy()
+            res = res.astype(np.float16)
+            return res
+        for item in shape_format:
+            cpu_input1, npu_input1 = create_common_tensor(item[0], 1, 100)
+            cpu_input2, npu_input2 = create_common_tensor(item[0], 1, 100)
+            cpu_output = cpu_op_exec_fp16(cpu_input1, cpu_input2)
+            npu_output = self.npu_op_exec(npu_input1, npu_input2)
+            self.assertRtolEqual(cpu_output, npu_output)
+
+
+instantiate_device_type_tests(TestRreluWithNoiseBackward, globals(), except_for="cpu")
+if __name__ == "__main__":
+    run_tests()
+
+
diff --git a/test/test_npu/test_network_ops/test_symeig.py b/test/test_npu/test_network_ops/test_symeig.py
new file mode 100644
index 0000000000..4394d42b12
--- /dev/null
+++ b/test/test_npu/test_network_ops/test_symeig.py
@@ -0,0 +1,61 @@
+# Copyright (c) 2020 Huawei Technologies Co., Ltd
+# Copyright (c) 2019, Facebook CORPORATION. 
+# All rights reserved.
+#
+# Licensed under the BSD 3-Clause License  (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# https://opensource.org/licenses/BSD-3-Clause
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+from common_utils import TestCase, run_tests
+from common_device_type import instantiate_device_type_tests
+
+class TestSymeig(TestCase):
+    def op_exec(self, input1, eigenvectorsflag):
+        npu_input = input1.npu()
+        en, vn = torch.symeig(npu_input, eigenvectors = eigenvectorsflag)
+        if eigenvectorsflag:
+            ret = torch.matmul(vn, torch.matmul(en.diag_embed(), vn.transpose(-2, -1)))
+            self.assertRtolEqual(ret.cpu(), input1, prec = 1e-3)
+        else:
+            e, v = torch.symeig(input1, eigenvectors = eigenvectorsflag)
+            self.assertEqual(e, en.cpu())
+            self.assertEqual(v, vn.cpu())
+    
+    def case_exec(self, input1):
+        input1 = input1 + input1.transpose(-2, -1)
+        self.op_exec(input1, False)
+        self.op_exec(input1, True)
+
+    def test_symeig_null(self, device):
+        a = torch.randn(0, 0)
+        self.op_exec(a, False)
+        self.op_exec(a, True)
+
+    def test_symeig_2d(self, device):
+        a = torch.randn(5, 5, dtype = torch.float32)
+        self.case_exec(a)
+
+    def test_symeig_3d(self, device):
+        a = torch.randn(10, 5, 5, dtype = torch.float32)
+        self.case_exec(a)
+
+    def test_symeig_4d(self, device):
+        a = torch.randn(10, 3, 5, 5, dtype = torch.float32)
+        self.case_exec(a)
+
+    def test_symeig_5d(self, device):
+        a = torch.randn(2, 10, 3, 5, 5, dtype = torch.float32)
+        self.case_exec(a)
+
+instantiate_device_type_tests(TestSymeig, globals(), except_for="cpu")
+if __name__ == "__main__":
+    run_tests()
-- 
Gitee