[BE] Enforce missing override keyword (#104032)

This PR enables `-Winconsistent-missing-destructor-override` and `-Winconsistent-missing-override`
and fixes violations.

<!--
copilot:summary
-->
### <samp>🤖 Generated by Copilot at 47e904e</samp>

This pull request updates the code of various classes and operators in the `caffe2` and `aten` subdirectories to use the `override` specifier instead of the `virtual` keyword for destructors and other virtual functions that override a base class function. This improves the code readability, quality, and consistency with C++ best practices. It also modifies the `./CMakeLists.txt` file to enable warnings for these specifiers, but disable errors.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/104032
Approved by: https://github.com/malfet

CMakeLists.txt

@@ -817,6 +817,8 @@ if(NOT MSVC)
   append_cxx_flag_if_supported("-Wno-strict-aliasing" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wvla-extension" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wnewline-eof" CMAKE_CXX_FLAGS)
+  append_cxx_flag_if_supported("-Winconsistent-missing-override" CMAKE_CXX_FLAGS)
+  append_cxx_flag_if_supported("-Winconsistent-missing-destructor-override" CMAKE_CXX_FLAGS)
   if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
     string(APPEND CMAKE_CXX_FLAGS " -Wno-range-loop-analysis")
     string(APPEND CMAKE_CXX_FLAGS " -Wno-pass-failed")
@@ -861,6 +863,8 @@ if(NOT MSVC)
 
   append_cxx_flag_if_supported("-Wno-error=pedantic" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-error=old-style-cast" CMAKE_CXX_FLAGS)
+  append_cxx_flag_if_supported("-Wno-error=inconsistent-missing-override" CMAKE_CXX_FLAGS)
+  append_cxx_flag_if_supported("-Wno-error=inconsistent-missing-destructor-override" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wconstant-conversion" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-invalid-partial-specialization" CMAKE_CXX_FLAGS)
   append_cxx_flag_if_supported("-Wno-unused-private-field" CMAKE_CXX_FLAGS)

aten/src/ATen/test/cpu_rng_test.cpp

@@ -19,8 +19,7 @@ constexpr auto kCustomRNG = DispatchKey::CustomRNGKeyId;
 
 struct TestCPUGenerator : public c10::GeneratorImpl {
   TestCPUGenerator(uint64_t value) : GeneratorImpl{Device(DeviceType::CPU), DispatchKeySet(kCustomRNG)}, value_(value) { }
-  // NOLINTNEXTLINE(modernize-use-override)
-  ~TestCPUGenerator() = default;
+  ~TestCPUGenerator() override = default;
   uint32_t random() { return value_; }
   uint64_t random64() { return value_; }
   c10::optional<float> next_float_normal_sample() { return next_float_normal_sample_; }

caffe2/contrib/gloo/allgather_ops.h

@@ -44,7 +44,7 @@ class AllgatherOp final : public Operator<Context> {
     }
   }
 
-  virtual ~AllgatherOp() {}
+  ~AllgatherOp() override {}
 
   bool RunOnDevice() override {
     std::call_once(once_, [&] { initialize(); });

caffe2/contrib/gloo/allreduce_ops.h

@@ -32,7 +32,7 @@ class AllreduceOp final : public Operator<Context> {
     }
   }
 
-  virtual ~AllreduceOp() {}
+  ~AllreduceOp() override {}
 
   bool RunOnDevice() override {
     std::call_once(once_, [&] { initialize(); });

caffe2/contrib/gloo/barrier_ops.h

@@ -24,7 +24,7 @@ class BarrierOp final : public Operator<Context> {
     }
   }
 
-  virtual ~BarrierOp() {}
+  ~BarrierOp() override {}
 
   bool RunOnDevice() override {
     auto context = OperatorBase::Input<std::shared_ptr<::gloo::Context>>(0);

caffe2/contrib/gloo/broadcast_ops.h

@@ -29,7 +29,7 @@ class BroadcastOp final : public Operator<Context> {
     }
   }
 
-  virtual ~BroadcastOp() {}
+  ~BroadcastOp() override {}
 
   bool RunOnDevice() override {
     std::call_once(once_, [&] { initialize(); });

caffe2/contrib/gloo/common_world_ops.h

@@ -49,7 +49,7 @@ class CreateCommonWorld final : public Operator<Context> {
     initialize();
   }
 
-  virtual ~CreateCommonWorld() {
+  ~CreateCommonWorld() override {
   }
 
   CommonWorld rendezvousWithMPI() {
@@ -176,7 +176,7 @@ class CloneCommonWorld final : public Operator<Context> {
     }
   }
 
-  virtual ~CloneCommonWorld() {}
+  ~CloneCommonWorld() override {}
 
   bool RunOnDevice() override {
     try {

caffe2/contrib/gloo/reduce_scatter_ops.h

@@ -44,7 +44,7 @@ class ReduceScatterOp final : public Operator<Context> {
     }
   }
 
-  virtual ~ReduceScatterOp() {}
+  ~ReduceScatterOp() override {}
 
   bool RunOnDevice() override {
     std::call_once(once_, [&] { initialize(); });

caffe2/contrib/gloo/store_handler.h

@@ -12,7 +12,7 @@ class TORCH_API StoreHandlerWrapper : public ::gloo::rendezvous::Store {
  public:
   explicit StoreHandlerWrapper(StoreHandler& handler) : handler_(handler) {}
 
-  virtual ~StoreHandlerWrapper() {}
+  virtual ~StoreHandlerWrapper() override {}
 
   virtual void set(const std::string& key, const std::vector<char>& data)
       override;

caffe2/core/nomnigraph/include/nomnigraph/Representations/NeuralNet.h

@@ -168,7 +168,7 @@ class TORCH_API Tensor : public NeuralNetData {
     return D->getKind() == NNDataKind::Tensor;
   }
 
-  NeuralNetData* clone() {
+  NeuralNetData* clone() override {
     return new Tensor(name_);
   }
 
@@ -188,7 +188,7 @@ class TORCH_API Tensor : public NeuralNetData {
     name_ = name;
   }
 
-  ~Tensor() {}
+  ~Tensor() override {}
 
  private:
   std::string name_;

caffe2/core/operator.h

@@ -1157,7 +1157,7 @@ class Operator : public OperatorBase {
   template <class... Args>                                \
   explicit name(Args&&... args)                           \
       : Operator<Context>(std::forward<Args>(args)...) {} \
-  virtual ~name() noexcept {}
+  virtual ~name() noexcept override {}
 
 // Helpers to implement runtime op polymorphism. Often it's convenient to make
 // an op work on different input types (e.g. i32 vs i64 indices) or special-case

caffe2/core/qtensor_serialization.h

@@ -12,7 +12,7 @@ template <class Context>
 class QTensorSerializer : public BlobSerializerBase {
  public:
   QTensorSerializer() : context_() {}
-  ~QTensorSerializer() {}
+  ~QTensorSerializer() override {}
   /**
    * Serializes a Blob. Note that this blob has to contain QTensor<Context>.
    */

caffe2/distributed/file_store_handler.h

@@ -7,7 +7,7 @@ namespace caffe2 {
 class TORCH_API FileStoreHandler : public StoreHandler {
  public:
   explicit FileStoreHandler(const std::string& path, const std::string& prefix);
-  virtual ~FileStoreHandler();
+  ~FileStoreHandler() override;
 
   void set(const std::string& name, const std::string& data) override;
 

caffe2/ideep/operators/conv_op.cc

@@ -30,8 +30,8 @@ class IDEEPConvOp : public IDEEPConvPoolOpBase {
       algo_ = ialgo::convolution_winograd;
     }
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPConvOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPConvOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override {
     const auto& X = Input(INPUT_X);
@@ -199,8 +199,8 @@ class IDEEPConvFusionOp final : public IDEEPConvOp {
         CAFFE_THROW("Unsupported conv fusion type!");
     }
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPConvFusionOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPConvFusionOp() override {}
 };
 
 const char* kConvFusionDoc = R"DOC(
@@ -291,8 +291,8 @@ class IDEEPConvGradientOp final : public IDEEPConvPoolOpBase {
         "In order to backward propagate weights correctly, "
         "please set training_mode=1");
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPConvGradientOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPConvGradientOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override {
     const auto& X = Input(INPUT);

caffe2/ideep/operators/conv_pool_base_op.h

@@ -12,7 +12,7 @@ class IDEEPConvPoolOpBase : public ConvPoolOpBase<IDEEPContext> {
  public:
   IDEEPConvPoolOpBase(const OperatorDef& operator_def, Workspace* ws)
      : ConvPoolOpBase<IDEEPContext>(operator_def, ws) {}
-  virtual ~IDEEPConvPoolOpBase() {}
+  ~IDEEPConvPoolOpBase() override {}
 
   inline const ideep::tensor& Input(int index) {
     return OperatorBase::template Input<ideep::tensor>(index);

caffe2/ideep/operators/conv_transpose_unpool_base_op.h

@@ -116,7 +116,7 @@ class IDEEPConvTransposeUnpoolBase : public IDEEPOperator {
       CAFFE_ENFORCE_LE(adj_[dim], stride_[dim]);
     }
   }
-  virtual ~IDEEPConvTransposeUnpoolBase() {}
+  ~IDEEPConvTransposeUnpoolBase() override {}
 
   const ideep::tensor& Input(int index) {
     return OperatorBase::template Input<ideep::tensor>(index);

caffe2/ideep/operators/quantization/int8_add_op.cc

@@ -25,8 +25,8 @@ class IDEEPInt8SumReluOp final : public IDEEPOperator {
 
     Y_scales_ = ConvertScales({scale_});
   }
-  // NOLINTNEXTLINE(modernize-use-equals-default,modernize-use-override)
-  virtual ~IDEEPInt8SumReluOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8SumReluOp() override {}
 
   bool RunOnDevice() override {
     itensor temp_ten;

caffe2/ideep/operators/quantization/int8_conv_op.cc

@@ -28,8 +28,8 @@ class IDEEPInt8ConvOp : public IDEEPConvPoolOpBase {
     CAFFE_ENFORCE(zero_point_ == 128 || zero_point_ == 0);
     Y_scales_ = ConvertScales({scale_});
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8ConvOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8ConvOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override {
     const auto &X = Input(INPUT_X);
@@ -198,8 +198,8 @@ class IDEEPInt8ConvReluOp final : public IDEEPInt8ConvOp {
     attr_ = iattr::fuse_relu();
     fusion_type_ = FUSION_CONV_RELU;
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8ConvReluOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8ConvReluOp() override {}
 };
 
 class IDEEPInt8ConvSumOp final : public IDEEPInt8ConvOp {
@@ -213,8 +213,8 @@ class IDEEPInt8ConvSumOp final : public IDEEPInt8ConvOp {
     attr_ = iattr::fuse_sum();
     fusion_type_ = FUSION_CONV_SUM;
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8ConvSumOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8ConvSumOp() override {}
 };
 
 class IDEEPInt8ConvSumReluOp final : public IDEEPInt8ConvOp {
@@ -228,8 +228,8 @@ class IDEEPInt8ConvSumReluOp final : public IDEEPInt8ConvOp {
     attr_ = iattr::residual();
     fusion_type_ = FUSION_CONV_SUM_RELU;
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8ConvSumReluOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8ConvSumReluOp() override {}
 };
 
 REGISTER_IDEEP_OPERATOR_WITH_ENGINE(Int8Conv, DNNLOWP, IDEEPInt8ConvOp);

caffe2/ideep/operators/quantization/int8_dequantize_op.cc

@@ -18,8 +18,8 @@ class IDEEPInt8DequantizeOp final : public IDEEPOperator {
                                               static_cast<int>(iformat::nchw)));
     }
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8DequantizeOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8DequantizeOp() override {}
 
   bool RunOnDevice() override {
     const auto& X = Input(0);

caffe2/ideep/operators/quantization/int8_fully_connected_op.cc

@@ -26,8 +26,8 @@ public:
     }
     Y_scales_ = ConvertScales({scale_});
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8FullyConnectedOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8FullyConnectedOp() override {}
 
   bool RunOnDevice() override {
     const auto& X = Input(INPUT);

caffe2/ideep/operators/quantization/int8_quantize_op.cc

@@ -26,8 +26,8 @@ class IDEEPInt8QuantizeOp final : public IDEEPOperator {
     Y_data_type_ = zero_point_ == 0 ? idtype::u8 : idtype::s8;
     Y_scales_ = ConvertScales({scale_});
   }
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~IDEEPInt8QuantizeOp() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~IDEEPInt8QuantizeOp() override {}
 
   bool RunOnDevice() override {
     const auto& X = Input(0);

caffe2/ideep/utils/ideep_context.h

@@ -23,20 +23,20 @@ class IDEEPContext final : public BaseContext {
 
   ~IDEEPContext() noexcept override {}
 
-  inline void SwitchToDevice(int64_t /*stream_id*/) {}
+  inline void SwitchToDevice(int64_t /*stream_id*/) override {}
   using BaseContext::SwitchToDevice;
 
-  inline void WaitEvent(const Event& ev) {
+  inline void WaitEvent(const Event& ev) override {
     ev.Wait(IDEEP, this);
   }
 
-  inline void Record(Event* ev, const char* err_msg = nullptr) const {
+  inline void Record(Event* ev, const char* err_msg = nullptr) const override {
     CAFFE_ENFORCE(ev, "Event must not be null.");
     ev->Record(IDEEP, this, err_msg);
   }
 
 
-  inline void FinishDeviceComputation() {}
+  inline void FinishDeviceComputation() override {}
 
   inline rand_gen_type& RandGenerator() {
     if (!random_generator_.get()) {

caffe2/ideep/utils/ideep_operator.h

@@ -40,7 +40,7 @@ class IDEEPOperator : public OperatorBase {
         order_(StringToStorageOrder(
             OperatorBase::GetSingleArgument<string>("order", "NCHW"))) {
   }
-  virtual ~IDEEPOperator() {}
+  ~IDEEPOperator() override {}
 
   inline const ideep::tensor& Input(int index) {
     return OperatorBase::template Input<ideep::tensor>(index);
@@ -114,7 +114,7 @@ class IDEEPOperator : public OperatorBase {
 #define USE_SIMPLE_IDEEP_CTOR_DTOR(name)                                       \
   name(const OperatorDef& operator_def, Workspace* ws)                         \
       : IDEEPOperator(operator_def, ws) {}                                     \
-  virtual ~name() {}
+  ~name() override {}
 
 // Convert zero_point scales to min_max scales
 // NOTE:

caffe2/mpi/mpi_ops.h

@@ -31,7 +31,7 @@ class MPIBroadcastOp final : public Operator<Context> {
   MPIBroadcastOp(const OperatorDef& operator_def, Workspace* ws)
       : Operator<Context>(operator_def, ws),
         root_(OperatorBase::template GetSingleArgument<int>("root", 0)) {}
-  ~MPIBroadcastOp() {}
+  ~MPIBroadcastOp() override {}
 
   bool RunOnDevice() override {
     MPI_Comm comm = OperatorBase::Input<MPICommonWorldWrapper>(0).comm();
@@ -65,7 +65,7 @@ class MPIReduceOp final : public Operator<Context> {
   MPIReduceOp(const OperatorDef& operator_def, Workspace* ws)
       : Operator<Context>(operator_def, ws),
         root_(OperatorBase::template GetSingleArgument<int>("root", 0)) {}
-  ~MPIReduceOp() {}
+  ~MPIReduceOp() override {}
 
   bool RunOnDevice() override {
     MPI_Comm comm = OperatorBase::Input<MPICommonWorldWrapper>(0).comm();

caffe2/operators/batch_gather_ops.h

@@ -50,7 +50,7 @@ class BatchGatherGradientOp final : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...),
         OP_SINGLE_ARG(int, "axis", axis_, 1),
         OP_SINGLE_ARG(bool, "match_outer", match_outer_, false) {}
-  virtual ~BatchGatherGradientOp() noexcept {}
+  ~BatchGatherGradientOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int32_t, int64_t>>::call(

caffe2/operators/batch_permutation_op.h

@@ -19,7 +19,7 @@ class BatchPermutationOp final : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {}
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
-  bool RunOnDevice();
+  bool RunOnDevice() override;
 };
 
 template <typename T, class Context>
@@ -29,7 +29,7 @@ class BatchPermutationGradientOp final : public Operator<Context> {
       : Operator<Context>(def, ws) {}
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
-  bool RunOnDevice();
+  bool RunOnDevice() override;
 };
 
 } // namespace caffe2

caffe2/operators/batch_sparse_to_dense_op.h

@@ -125,7 +125,7 @@ class BatchDenseToSparseOp : public Operator<Context> {
   template <class... Args>
   explicit BatchDenseToSparseOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  bool RunOnDevice() {
+  bool RunOnDevice() override{
     return DispatchHelper<TensorTypes<int32_t, int64_t>>::call(
         this, Input(LENGTHS));
   }

caffe2/operators/box_with_nms_limit_op.h

@@ -58,7 +58,7 @@ class BoxWithNMSLimitOp final : public Operator<Context> {
     input_scores_fg_cls_starting_id_ = (int)input_boxes_include_bg_cls_;
   }
 
-  ~BoxWithNMSLimitOp() {}
+  ~BoxWithNMSLimitOp() override {}
 
   bool RunOnDevice() override {
     if (InputSize() > 2) {

caffe2/operators/channel_backprop_stats_op.h

@@ -14,7 +14,7 @@ class ChannelBackpropStatsOp : public Operator<Context> {
   template <class... Args>
   explicit ChannelBackpropStatsOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~ChannelBackpropStatsOp() {}
+  ~ChannelBackpropStatsOp() override {}
 
   bool RunOnDevice() override {
     return true;

caffe2/operators/conv_op.h

@@ -29,7 +29,7 @@ class ConvOp final : public ConvPoolOpBase<Context> {
       createSharedBuffer<Context>(ws_);
     }
   }
-  ~ConvOp() {}
+  ~ConvOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;
@@ -79,7 +79,7 @@ class ConvGradientOp final : public ConvPoolOpBase<Context> {
          std::is_same<Context, CPUContext>::value),
         "Group convolution only supports NCHW order or CPUContext right now.");
   }
-  ~ConvGradientOp() {}
+  ~ConvGradientOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;

caffe2/operators/conv_pool_op_base.h

@@ -651,7 +651,7 @@ class ConvPoolOpBase : public Operator<Context> {
     return TensorInferenceForSchema(def, in, in[1].dims(img_ndim));
   }
 
-  virtual ~ConvPoolOpBase() {}
+  ~ConvPoolOpBase() override {}
 
  protected:
   LegacyPadding legacy_pad_;

caffe2/operators/conv_transpose_unpool_op_base.h

@@ -208,7 +208,7 @@ class ConvTransposeUnpoolBase : public Operator<Context> {
     CAFFE_THROW("Not implemented");
   }
 
-  virtual ~ConvTransposeUnpoolBase() {}
+  ~ConvTransposeUnpoolBase() override {}
 
  protected:
   // Accessors for 2D conv params.

caffe2/operators/elementwise_logical_ops.h

@@ -139,7 +139,7 @@ class IsMemberOfOp final : public Operator<Context> {
         CAFFE_THROW("Unexpected 'dtype' argument value: ", dtype);
     }
   }
-  virtual ~IsMemberOfOp() noexcept {}
+  ~IsMemberOfOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<

caffe2/operators/feature_maps_ops.h

@@ -17,7 +17,7 @@ class MergeDenseFeatureTensorsOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {
     featureIDs_ = this->template GetRepeatedArgument<int64_t>("feature_ids");
   }
-  virtual ~MergeDenseFeatureTensorsOp() noexcept {}
+  ~MergeDenseFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -83,7 +83,7 @@ class MergeSingleScalarFeatureTensorsOp : public Operator<Context> {
     numInputs_ = InputSize() / kNumTensorsPerInput;
     featureIDs_ = this->template GetRepeatedArgument<int64_t>("feature_ids");
   }
-  virtual ~MergeSingleScalarFeatureTensorsOp() noexcept {}
+  ~MergeSingleScalarFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -148,7 +148,7 @@ class MergeSingleScalarFeatureTensorsGradientOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {
     numFeatureInputs_ = InputSize() - 1; // Everything other than values_grad
   }
-  virtual ~MergeSingleScalarFeatureTensorsGradientOp() noexcept {}
+  ~MergeSingleScalarFeatureTensorsGradientOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -198,7 +198,7 @@ class MergeSingleListFeatureTensorsOp : public Operator<Context> {
     inValuesOffset_.resize(numInputs_);
     featureIDs_ = this->template GetRepeatedArgument<int64_t>("feature_ids");
   }
-  virtual ~MergeSingleListFeatureTensorsOp() noexcept {}
+  ~MergeSingleListFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -284,7 +284,7 @@ class MergeSingleListOrMapFeatureTensorsGradientOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {
     numFeatureInputs_ = (InputSize() - 1) / kNumTensorsPerInput;
   }
-  virtual ~MergeSingleListOrMapFeatureTensorsGradientOp() noexcept {}
+  ~MergeSingleListOrMapFeatureTensorsGradientOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -352,7 +352,7 @@ class MergeSingleMapFeatureTensorsOp : public Operator<Context> {
     inValuesOffset_.resize(numInputs_);
     featureIDs_ = this->template GetRepeatedArgument<int64_t>("feature_ids");
   }
-  virtual ~MergeSingleMapFeatureTensorsOp() noexcept {}
+  ~MergeSingleMapFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -454,7 +454,7 @@ class MergeMultiScalarFeatureTensorsOp : public Operator<Context> {
     numInputs_ = InputSize() / kNumTensorsPerInput;
     inKeysOffset_.resize(numInputs_);
   }
-  virtual ~MergeMultiScalarFeatureTensorsOp() noexcept {}
+  ~MergeMultiScalarFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -526,7 +526,7 @@ class MergeMultiScalarFeatureTensorsGradientOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {
     numFeatureInputs_ = (InputSize() - 1) / kNumTensorsPerInput;
   }
-  virtual ~MergeMultiScalarFeatureTensorsGradientOp() noexcept {}
+  ~MergeMultiScalarFeatureTensorsGradientOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -588,7 +588,7 @@ class MergeMultiListFeatureTensorsOp : public Operator<Context> {
     inKeysOffset_.resize(numInputs_);
     inValuesValuesOffset_.resize(numInputs_);
   }
-  virtual ~MergeMultiListFeatureTensorsOp() noexcept {}
+  ~MergeMultiListFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -680,7 +680,7 @@ class MergeMultiMapFeatureTensorsOp : public Operator<Context> {
     inKeysOffset_.resize(numInputs_);
     inValuesValuesOffset_.resize(numInputs_);
   }
-  virtual ~MergeMultiMapFeatureTensorsOp() noexcept {}
+  ~MergeMultiMapFeatureTensorsOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<
@@ -786,7 +786,7 @@ class MergeMultiListOrMapFeatureTensorsGradientOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {
     numFeatureInputs_ = (InputSize() - 1) / kNumTensorsPerInput;
   }
-  virtual ~MergeMultiListOrMapFeatureTensorsGradientOp() noexcept {}
+  ~MergeMultiListOrMapFeatureTensorsGradientOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<

caffe2/operators/filler_op.h

@@ -49,7 +49,7 @@ class FillerOp : public Operator<Context> {
     }
   }
 
-  virtual ~FillerOp() {}
+  ~FillerOp() override {}
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
   bool RunOnDevice() override {

caffe2/operators/find_op.h

@@ -21,7 +21,7 @@ class FindOp final : public Operator<Context> {
   USE_OPERATOR_CONTEXT_FUNCTIONS;
   USE_DISPATCH_HELPER;
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int, long>>::call(this, Input(0));
   }
 

caffe2/operators/fully_connected_op.h

@@ -24,7 +24,7 @@ class FullyConnectedOp final : public Operator<Context> {
         axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)),
         float16_compute_(
             this->template GetSingleArgument<bool>("float16_compute", false)) {}
-  ~FullyConnectedOp() {}
+  ~FullyConnectedOp() override {}
 
   template <
       typename T_X,
@@ -175,7 +175,7 @@ class FullyConnectedGradientOp : public Operator<Context> {
         axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)),
         float16_compute_(
             this->template GetSingleArgument<bool>("float16_compute", false)) {}
-  ~FullyConnectedGradientOp() {}
+  ~FullyConnectedGradientOp() override {}
 
   template <
       typename T_X,

caffe2/operators/fused_rowwise_random_quantization_ops.h

@@ -45,7 +45,7 @@ class FloatToFusedRandRowwiseQuantizedOp : public Operator<Context> {
     }
   }
 
-  ~FloatToFusedRandRowwiseQuantizedOp() {
+  ~FloatToFusedRandRowwiseQuantizedOp() override {
     if (random_) {
 #ifdef FUSED_ROWWISE_RANDOM_QUANTIZATION_USE_MKL
       int status = vslDeleteStream(&vslStream_);

caffe2/operators/gather_op.h

@@ -194,7 +194,7 @@ class GatherOp : public Operator<Context> {
     }
   }
 
-  virtual ~GatherOp() noexcept {}
+  ~GatherOp() noexcept override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int32_t, int64_t>>::call(

caffe2/operators/generate_proposals_op.h

@@ -103,7 +103,7 @@ class GenerateProposalsOp final : public Operator<Context> {
         legacy_plus_one_(
             this->template GetSingleArgument<bool>("legacy_plus_one", true)) {}
 
-  ~GenerateProposalsOp() {}
+  ~GenerateProposalsOp() override {}
 
   bool RunOnDevice() override;
 

caffe2/operators/glu_op.h

@@ -15,7 +15,7 @@ class GluOp final : public Operator<Context> {
 
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     auto& X = Input(0);
 
     vector<int64_t> Yshape;

caffe2/operators/half_float_ops.h

@@ -37,7 +37,7 @@ class Float16ConstantFillOp : public Operator<CPUContext> {
         shape_(this->template GetRepeatedArgument<int64_t>("shape")) {}
 
   USE_OPERATOR_FUNCTIONS(CPUContext);
-  virtual ~Float16ConstantFillOp() {}
+  ~Float16ConstantFillOp() override {}
 
   bool RunOnDevice() override;
 
@@ -68,7 +68,7 @@ class Float16UniformFillOp : public Operator<Context> {
   }
 
   USE_OPERATOR_CONTEXT_FUNCTIONS;
-  virtual ~Float16UniformFillOp() {}
+  ~Float16UniformFillOp() override {}
 
   bool RunOnDevice() override;
 

caffe2/operators/instance_norm_op.h

@@ -27,7 +27,7 @@ class InstanceNormOp final : public Operator<Context> {
         "order should be either \"NCHW\" or \"NHWC\".");
   }
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     const auto& X = Input(INPUT);
     const auto& gamma = Input(SCALE);
     const auto& beta = Input(BIAS);
@@ -144,7 +144,7 @@ class InstanceNormGradientOp final : public Operator<Context> {
         "order should be either \"NCHW\" or \"NHWC\".");
   }
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     const auto& X = Input(INPUT);
     const auto& gamma = Input(SCALE);
     const auto& dY = Input(OUTPUT_GRAD);

caffe2/operators/length_split_op.h

@@ -29,7 +29,7 @@ class LengthsSplitOp final : public Operator<Context> {
           "`n_split` must contain a positive value for defined behavior.");
     }
   }
-  ~LengthsSplitOp() {}
+  ~LengthsSplitOp() override {}
 
   bool RunOnDevice() override {
     const auto& L = Input(0);

caffe2/operators/lengths_reducer_ops.h

@@ -32,7 +32,7 @@ class CPUSparseLengthsReductionOp : public Operator<CPUContext> {
         !(USE_WEIGHT & USE_MEAN), "Cannot both specify weight and mean.");
   }
 
-  ~CPUSparseLengthsReductionOp() {}
+  ~CPUSparseLengthsReductionOp() override {}
 
   // Currently, we support float and at::Half inputs for input data type, and
   // int32_t and int64_t for the index type.
@@ -287,7 +287,7 @@ class TTSparseLengthsSumOp final : public Operator<Context> {
     }
   }
 
-  ~TTSparseLengthsSumOp() {}
+  ~TTSparseLengthsSumOp() override {}
 
   void Ind2Sub(int64_t* out_factor_index, const int64_t* indices, int len) {
     // TODO: vectorization
@@ -492,7 +492,7 @@ class TTSparseLengthsSumGradientOp final : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...) {}
   bool RunOnDevice() override;
 
-  ~TTSparseLengthsSumGradientOp() {}
+  ~TTSparseLengthsSumGradientOp() override {}
 };
 
 // implement the gradient op for TTLengthSumGradient op

caffe2/operators/locally_connected_op.h

@@ -26,7 +26,7 @@ class LocallyConnectedOp final : public ConvPoolOpBase<Context> {
         "Group locally connected only supports NCHW order right now.");
   }
 
-  ~LocallyConnectedOp() = default;
+  ~LocallyConnectedOp() override = default;
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;
@@ -81,7 +81,7 @@ class LocallyConnectedGradientOp final : public ConvPoolOpBase<Context> {
         "Group locally connected only supports NCHW order right now.");
   }
 
-  ~LocallyConnectedGradientOp() = default;
+  ~LocallyConnectedGradientOp() override = default;
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;

caffe2/operators/logit_op.h

@@ -32,7 +32,7 @@ class LogitGradientOp final : public Operator<Context> {
   explicit LogitGradientOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...),
         eps_(this->template GetSingleArgument<float>("eps", 1e-6f)) {}
-  ~LogitGradientOp() {}
+  ~LogitGradientOp() override {}
 
   bool RunOnDevice() override;
 

caffe2/operators/map_ops.h

@@ -52,7 +52,7 @@ class CreateMapOp final : public Operator<Context> {
   template <class... Args>
   explicit CreateMapOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~CreateMapOp() {}
+  ~CreateMapOp() override {}
 
   bool RunOnDevice() override {
     TensorProto::DataType key_dtype = static_cast<TensorProto::DataType>(
@@ -104,7 +104,7 @@ class KeyValueToMapOp final : public Operator<Context> {
   template <class... Args>
   explicit KeyValueToMapOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~KeyValueToMapOp() {}
+  ~KeyValueToMapOp() override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int32_t, int64_t>>::call(
@@ -157,7 +157,7 @@ class MapToKeyValueOp final : public Operator<Context> {
   template <class... Args>
   explicit MapToKeyValueOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~MapToKeyValueOp() {}
+  ~MapToKeyValueOp() override {}
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<

caffe2/operators/matmul_op.h

@@ -18,7 +18,7 @@ class MatMulOp final : public Operator<Context> {
         axis_b_(this->template GetSingleArgument<int>("axis_b", 1)),
         trans_a_(this->template GetSingleArgument<int>("trans_a", 0)),
         trans_b_(this->template GetSingleArgument<int>("trans_b", 0)) {}
-  ~MatMulOp() {}
+  ~MatMulOp() override {}
 
   bool RunOnDevice() override {
     const auto& A = Input(0);

caffe2/operators/minmax_ops.h

@@ -122,7 +122,7 @@ class MaxGradientOp final : public SelectGradientOpBase<T, Context> {
   explicit MaxGradientOp(Args&&... args)
       : SelectGradientOpBase<T, Context>(std::forward<Args>(args)...) {}
 
-  ~MaxGradientOp() = default;
+  ~MaxGradientOp() override = default;
 };
 
 template <typename T, class Context>
@@ -132,7 +132,7 @@ class MinGradientOp final : public SelectGradientOpBase<T, Context> {
   explicit MinGradientOp(Args&&... args)
       : SelectGradientOpBase<T, Context>(std::forward<Args>(args)...) {}
 
-  ~MinGradientOp() = default;
+  ~MinGradientOp() override = default;
 };
 
 } // namespace caffe2

caffe2/operators/numpy_tile_op.h

@@ -18,7 +18,7 @@ class NumpyTileOp : public Operator<Context> {
   template <class... Args>
   explicit NumpyTileOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~NumpyTileOp() {}
+  ~NumpyTileOp() override {}
 
   bool RunOnDevice() override {
     const auto& input = Input(0);

caffe2/operators/onnx_while_op.h

@@ -47,7 +47,7 @@ class ONNXWhileOp final : public Operator<Context> {
 
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int, bool, long>>::call(this, Input(1));
   }
 

caffe2/operators/pack_segments.h

@@ -37,7 +37,7 @@ class PackSegmentsOp final : public Operator<Context> {
     }
   }
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<int, long>>::call(this, Input(LENGTHS));
   }
 

caffe2/operators/pad_op.h

@@ -42,7 +42,7 @@ class PadImageOp final : public ConvPoolOpBase<Context> {
     // output size.
     kernel_.assign(pads_.size() / 2, 1);
   }
-  ~PadImageOp() {}
+  ~PadImageOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;
@@ -78,7 +78,7 @@ class PadImageGradientOp final : public ConvPoolOpBase<Context> {
     // output size.
     kernel_.assign(pads_.size() / 2, 1);
   }
-  ~PadImageGradientOp() {}
+  ~PadImageGradientOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;

caffe2/operators/pool_op.h

@@ -33,7 +33,7 @@ class PoolOp final : public ConvPoolOpBase<Context> {
     }
   }
 
-  ~PoolOp() = default;
+  ~PoolOp() override = default;
 
   bool RunOnDeviceWithOrderNCHW() override {
     const auto& X = Input(0);
@@ -112,7 +112,7 @@ class PoolGradientOp final : public ConvPoolOpBase<Context> {
   explicit PoolGradientOp(Args&&... args)
       : ConvPoolOpBase<Context>(std::forward<Args>(args)...), functor_(*this) {}
 
-  ~PoolGradientOp() = default;
+  ~PoolGradientOp() override = default;
 
   bool RunOnDeviceWithOrderNCHW() override {
     const auto& X = Input(0);

caffe2/operators/prefetch_op.h

@@ -35,7 +35,7 @@ class PrefetchOperator : public OperatorBase {
     context_.SwitchToDevice();
   }
 
-  virtual ~PrefetchOperator() noexcept {
+  ~PrefetchOperator() noexcept override {
     CHECK(finalize_ || !prefetch_thread_.get())
         << "YOU MADE A PROGRAMMING ERROR: derived class of PrefetchOperator "
            "should call Finalize() in its destructor so the prefetching "

caffe2/operators/quant_decode_op.h

@@ -108,7 +108,7 @@ class QuantDecodeOp final : public Operator<CPUContext> {
   explicit QuantDecodeOp(Args&&... args)
       : Operator<CPUContext>(std::forward<Args>(args)...) {}
 
-  ~QuantDecodeOp() {}
+  ~QuantDecodeOp() override {}
 
   bool RunOnDevice() override {
     CAFFE_ENFORCE_GT(InputSize(), 1);
@@ -144,7 +144,7 @@ class QuantDecodeGradientOp final : public Operator<CPUContext> {
   template <class... Args>
   explicit QuantDecodeGradientOp(Args&&... args)
       : Operator<CPUContext>(std::forward<Args>(args)...) {}
-  ~QuantDecodeGradientOp() {}
+  ~QuantDecodeGradientOp() override {}
 
   bool RunOnDevice() override {
     // Inputs: 1 codebook, n tensors of codes, and n corresponding gradients.

caffe2/operators/quantized/int8_add_op.h

@@ -18,7 +18,7 @@ class Int8AddOp final : public Operator<CPUContext> {
   explicit Int8AddOp(const OperatorDef& operator_def, Workspace* ws)
       : Operator<CPUContext>(operator_def, ws), ws_(ws) {}
 
-  ~Int8AddOp() {
+  ~Int8AddOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_average_pool_op.h

@@ -23,7 +23,7 @@ class Int8AveragePoolOp final : public ConvPoolOpBase<CPUContext> {
         this->order_ == StorageOrder::NHWC, "Int8 only supports NHWC order.");
   }
 
-  ~Int8AveragePoolOp() {
+  ~Int8AveragePoolOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_channel_shuffle_op.h

@@ -25,7 +25,7 @@ class Int8ChannelShuffleOp final : public ConvPoolOpBase<CPUContext> {
         "Int8ChannelShuffleOp only supports NHWC order");
   }
 
-  ~Int8ChannelShuffleOp() {
+  ~Int8ChannelShuffleOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_conv_op.h

@@ -27,7 +27,7 @@ class Int8ConvOp final : public ConvPoolOpBase<CPUContext> {
     createSharedBuffer<CPUContext>(ws_);
   }
 
-  ~Int8ConvOp() {
+  ~Int8ConvOp() override {
     if (this->qnnpackObject_ != nullptr) {
       qnnp_delete_operator(this->qnnpackObject_);
       this->qnnpackObject_ = nullptr;

caffe2/operators/quantized/int8_conv_transpose_op.h

@@ -26,7 +26,7 @@ class Int8ConvTransposeOp final : public ConvTransposeUnpoolBase<CPUContext> {
     createSharedBuffer<CPUContext>(ws_);
   }
 
-  ~Int8ConvTransposeOp() {
+  ~Int8ConvTransposeOp() override {
     if (this->qnnpackObject_ != nullptr) {
       qnnp_delete_operator(this->qnnpackObject_);
       this->qnnpackObject_ = nullptr;

caffe2/operators/quantized/int8_fc_op.h

@@ -20,7 +20,7 @@ class Int8FCOp final : public Operator<CPUContext> {
     createSharedBuffer<CPUContext>(ws_);
   }
 
-  ~Int8FCOp() {
+  ~Int8FCOp() override {
     if (this->qnnpackObject_ != nullptr) {
       qnnp_delete_operator(this->qnnpackObject_);
       this->qnnpackObject_ = nullptr;

caffe2/operators/quantized/int8_leaky_relu_op.h

@@ -25,7 +25,7 @@ class Int8LeakyReluOp final : public Operator<CPUContext> {
 #endif
   }
 
-  ~Int8LeakyReluOp() {
+  ~Int8LeakyReluOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_max_pool_op.h

@@ -23,7 +23,7 @@ class Int8MaxPoolOp final : public ConvPoolOpBase<CPUContext> {
         this->order_ == StorageOrder::NHWC, "Int8 only supports NHWC order.");
   }
 
-  ~Int8MaxPoolOp() {
+  ~Int8MaxPoolOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_relu_op.h

@@ -21,7 +21,7 @@ class Int8ReluOp final : public Operator<CPUContext> {
 #endif
       }
 
-  ~Int8ReluOp() {
+  ~Int8ReluOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_sigmoid_op.h

@@ -20,7 +20,7 @@ class Int8SigmoidOp final : public Operator<CPUContext> {
 #endif
       }
 
-  ~Int8SigmoidOp() {
+  ~Int8SigmoidOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/quantized/int8_softmax_op.h

@@ -21,7 +21,7 @@ class Int8SoftmaxOp final : public Operator<CPUContext> {
 #endif
       }
 
-  ~Int8SoftmaxOp() {
+  ~Int8SoftmaxOp() override {
     if (this->qnnpackOperator_ != nullptr) {
       qnnp_delete_operator(this->qnnpackOperator_);
       this->qnnpackOperator_ = nullptr;

caffe2/operators/reduce_front_back_max_ops.h

@@ -20,7 +20,7 @@ class MaxReduceDimsOp final : public Operator<Context> {
 
   USE_OPERATOR_CONTEXT_FUNCTIONS;
 
-  bool RunOnDevice() {
+  bool RunOnDevice() override {
     auto& X = Input(0);
 
     CAFFE_ENFORCE(

caffe2/operators/reduction_ops.h

@@ -27,7 +27,7 @@ class SumElementsOp : public Operator<Context> {
   explicit SumElementsOp(const c10::FunctionSchema& schema, std::vector<c10::IValue> inputs, std::vector<c10::IValue*> outputs, bool average)
       : Operator<Context>(schema, std::move(inputs), std::move(outputs)), average_(average) {}
 #endif
-  ~SumElementsOp() {}
+  ~SumElementsOp() override {}
 
   bool RunOnDevice() override {
     auto& X = Input(0);
@@ -62,7 +62,7 @@ class SumElementsIntOp : public Operator<Context> {
   template <class... Args>
   explicit SumElementsIntOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~SumElementsIntOp() {}
+  ~SumElementsIntOp() override {}
 
   bool RunOnDevice() override {
     auto& X = Input(0);
@@ -95,7 +95,7 @@ class SumElementsGradientOp : public Operator<Context> {
   explicit SumElementsGradientOp(const c10::FunctionSchema& schema, std::vector<c10::IValue> inputs, std::vector<c10::IValue*> outputs, bool average)
       : Operator<Context>(schema, std::move(inputs), std::move(outputs)), average_(average) {}
 #endif
-  ~SumElementsGradientOp() {}
+  ~SumElementsGradientOp() override {}
 
   bool RunOnDevice() override;
 

caffe2/operators/rnn/recurrent_network_executor.h

@@ -486,7 +486,7 @@ class TORCH_API ThreadedRecurrentNetworkExecutor : public RecurrentNetworkExecut
       : RecurrentNetworkExecutorBase(step_net_def, recurrent_input_map, timestep_blob),
         failed_(false) {}
 
-  ~ThreadedRecurrentNetworkExecutor() {
+  ~ThreadedRecurrentNetworkExecutor() override {
     task_queue_.NoMoreJobs();
     VLOG(1) << "Joining workers.";
     for (auto& worker : workers_) {

caffe2/operators/spatial_batch_norm_op.h

@@ -40,7 +40,7 @@ class SpatialBNOp : public Operator<Context> {
     CAFFE_ENFORCE_LE(momentum_, 1);
   }
 
-  virtual ~SpatialBNOp() = default;
+  ~SpatialBNOp() override = default;
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<float>>::call(this, Input(0));
@@ -305,7 +305,7 @@ class SpatialBNGradientOp : public Operator<Context> {
     CAFFE_ENFORCE_EQ(OutputSize(), 3);
   }
 
-  virtual ~SpatialBNGradientOp() = default;
+  ~SpatialBNGradientOp() override = default;
 
   bool RunOnDevice() override {
     return DispatchHelper<TensorTypes<float>>::call(this, Input(0));

caffe2/operators/summarize_op.h

@@ -37,7 +37,7 @@ class SummarizeOp final : public Operator<Context> {
           log_file_->rdstate());
     }
   }
-  ~SummarizeOp() {
+  ~SummarizeOp() override {
     if (to_file_)
       log_file_->close();
   }

caffe2/operators/tensor_protos_db_input.h

@@ -16,7 +16,7 @@ class TensorProtosDBInput final : public PrefetchOperator<Context> {
   using OperatorBase::OutputSize;
   using PrefetchOperator<Context>::prefetch_thread_;
   explicit TensorProtosDBInput(const OperatorDef& operator_def, Workspace* ws);
-  ~TensorProtosDBInput() {
+  ~TensorProtosDBInput() override {
     PrefetchOperator<Context>::Finalize();
   }
 

caffe2/operators/text_file_reader_utils.h

@@ -107,7 +107,7 @@ class TORCH_API BufferedTokenizer {
 class TORCH_API FileReader : public StringProvider {
  public:
   explicit FileReader(const std::string& path, size_t bufferSize = 65536);
-  ~FileReader();
+  ~FileReader() override;
   void operator()(CharRange& range) override;
   void reset() override;
 

caffe2/operators/top_k.h

@@ -19,7 +19,7 @@ class TopKOp : public Operator<Context> {
         OP_SINGLE_ARG(int, "axis", axis_, -1) {
   }
 
-  ~TopKOp() {}
+  ~TopKOp() override {}
 
   bool RunOnDevice() override;
 
@@ -38,7 +38,7 @@ class TopKGradientOp : public Operator<Context> {
       : Operator<Context>(std::forward<Args>(args)...),
         OP_SINGLE_ARG(int, "axis", axis_, -1) {}
 
-  ~TopKGradientOp() {}
+  ~TopKGradientOp() override {}
 
   bool RunOnDevice() override;
 

caffe2/operators/tt_linear_op.h

@@ -25,7 +25,7 @@ class TTLinearOp final : public Operator<Context> {
         out_sizes_(this->template GetRepeatedArgument<int>("out_sizes")),
         tt_ranks_(this->template GetRepeatedArgument<int>("tt_ranks")),
         Y_temp_(unique_ptr<Blob>(new Blob())) {}
-  ~TTLinearOp() {}
+  ~TTLinearOp() override {}
 
   bool RunOnDevice() override {
     const auto& X = Input(0); // Input array
@@ -180,7 +180,7 @@ class TTLinearGradientOp : public Operator<Context> {
   template <class... Args>
   explicit TTLinearGradientOp(Args&&... args)
       : Operator<Context>(std::forward<Args>(args)...) {}
-  ~TTLinearGradientOp() {}
+  ~TTLinearGradientOp() override {}
 
   bool RunOnDevice() override {
     return false;

caffe2/operators/utility_ops.h

@@ -648,7 +648,7 @@ template <class Context>
 class ScatterAssignOp : public Operator<Context> {
  public:
   USE_OPERATOR_CONTEXT_FUNCTIONS;
-  virtual ~ScatterAssignOp() {}
+  ~ScatterAssignOp() override {}
 
   template <class... Args>
   explicit ScatterAssignOp(Args&&... args)

caffe2/opt/converter.cc

@@ -173,8 +173,8 @@ class ConvTransposeConverter : public Converter {
   }
   // Does not override default converter to OperatorDef
 
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~ConvTransposeConverter() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  ~ConvTransposeConverter() override {}
 };
 
 REGISTER_CONVERTER(Conv, ConvConverter);

caffe2/opt/converter.h

@@ -70,7 +70,7 @@ C10_DECLARE_REGISTRY(ConverterRegistry, Converter);
         const OperatorDef& op) override {                                     \
       return std::make_unique<nom::repr::opName>();                     \
     }                                                                         \
-    virtual ~opName##Converter() {}                                           \
+    ~opName##Converter() override {}                                           \
   };
 
 } // namespace caffe2

caffe2/opt/onnxifi_op.h

@@ -181,7 +181,7 @@ class OnnxifiOp final : public Operator<Context> {
     buildBackendAndGraph(ws, property_pointers, onnx_model_str);
   }
 
-  ~OnnxifiOp() {
+  ~OnnxifiOp() override {
     backend_graph_shared_ptr_.reset();
     backend_graph_map_ptr_->remove(op_id_string_);
 #ifdef ONNXIFI_ENABLE_EXT

caffe2/quantization/server/activation_distribution_observer.h

@@ -14,7 +14,7 @@ namespace caffe2 {
 class OutputMinMaxObserver final : public ObserverBase<OperatorBase> {
  public:
   explicit OutputMinMaxObserver(OperatorBase* op);
-  ~OutputMinMaxObserver();
+  ~OutputMinMaxObserver() override;
 
   struct TensorInfo {
     explicit TensorInfo(const std::string& name)
@@ -65,7 +65,7 @@ class OutputMinMaxNetObserver final : public NetObserver {
       const std::string& out_file_name,
       int dump_freq = -1,
       string delimiter = " ");
-  ~OutputMinMaxNetObserver();
+  ~OutputMinMaxNetObserver() override;
 
  private:
   void Stop() override;
@@ -140,7 +140,7 @@ class HistogramNetObserver final : public NetObserver {
       bool mul_nets = false,
       string op_filter = "",
       string delimiter = " ");
-  ~HistogramNetObserver();
+  ~HistogramNetObserver() override;
   void DumpHistogramFile() {
     DumpAndReset_(out_file_name_, false);
   }
@@ -174,7 +174,7 @@ class OutputColumnMaxHistogramNetObserver final : public NetObserver {
       int dump_freq = -1,
       bool mul_nets = false,
       string delimiter = " ");
-  ~OutputColumnMaxHistogramNetObserver();
+  ~OutputColumnMaxHistogramNetObserver() override;
   void DumpOutputColumnMaxHistogramFile() {
     DumpAndReset_(out_file_name_, false);
   }

caffe2/quantization/server/conv_dnnlowp_op.h

@@ -19,7 +19,7 @@ class ConvDNNLowPOp : public ConvPoolDNNLowPOpBase<T, ConvFp32Op> {
   USE_CONV_POOL_BASE_FUNCTIONS(CPUContext);
   USE_CONV_POOL_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ConvFp32Op);
   ConvDNNLowPOp(const OperatorDef& operator_def, Workspace* ws);
-  virtual ~ConvDNNLowPOp();
+  ~ConvDNNLowPOp() override;
 
  protected:
   bool RunOnDeviceWithOrderNCHW() override;

caffe2/quantization/server/conv_relu_op.h

@@ -20,7 +20,7 @@ class ConvReluOp final : public ConvPoolOpBase<Context> {
       TORCH_CHECK_NOTNULL(local_output_blobs_.back());
     }
   }
-  ~ConvReluOp() {}
+  ~ConvReluOp() override {}
 
   bool RunOnDeviceWithOrderNCHW() override;
   bool RunOnDeviceWithOrderNHWC() override;

caffe2/quantization/server/fb_fc_packed_op.h

@@ -70,7 +70,7 @@ class FbFCPackedOperator final : public Operator<Context> {
       : Operator<Context>(operator_def, ws),
         axis_(this->template GetSingleArgument<int32_t>("axis", 1)),
         axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)) {}
-  ~FbFCPackedOperator() {}
+  ~FbFCPackedOperator() override {}
 
   // template on X, B, and Y.
   template <typename T_X, typename T_B, typename T_Y>

caffe2/quantization/server/fbgemm_fp16_pack_op.h

@@ -37,7 +37,7 @@ class FbGemmPackOp final : public Operator<Context> {
         axis_(this->template GetSingleArgument<int32_t>("axis_w", 1)),
         no_packing_(
             this->template GetSingleArgument<int32_t>("no_packing", 0)) {}
-  ~FbGemmPackOp() {}
+  ~FbGemmPackOp() override {}
 
   bool RunOnDevice() override {
     const auto& X = Input(0);

caffe2/quantization/server/fully_connected_fake_lowp_op.h

@@ -58,7 +58,7 @@ class FullyConnectedFakeLowpFPOp final : public Operator<Context> {
         axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)),
         float16_compute_(
             this->template GetSingleArgument<bool>("float16_compute", false)) {}
-  ~FullyConnectedFakeLowpFPOp() {}
+  ~FullyConnectedFakeLowpFPOp() override {}
 
   template <
       typename T_X,
@@ -104,7 +104,7 @@ class FullyConnectedGradientFakeLowpFPOp : public Operator<Context> {
         axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)),
         float16_compute_(
             this->template GetSingleArgument<bool>("float16_compute", false)) {}
-  ~FullyConnectedGradientFakeLowpFPOp() {}
+  ~FullyConnectedGradientFakeLowpFPOp() override {}
 
   template <
       typename T_X,

caffe2/quantization/server/lstm_unit_dnnlowp_op.h

@@ -14,7 +14,7 @@ class LSTMUnitDNNLowPOp final : public LSTMUnitOp<CPUContext> {
 
  public:
   LSTMUnitDNNLowPOp(const OperatorDef& operator_def, Workspace* ws);
-  ~LSTMUnitDNNLowPOp();
+  ~LSTMUnitDNNLowPOp() override;
   bool RunOnDevice() override;
 
  private:

caffe2/quantization/server/utility_dnnlowp_ops.h

@@ -30,7 +30,7 @@ class GatherDNNLowPOp final : public GatherOp<CPUContext> {
 
  public:
   GatherDNNLowPOp(const OperatorDef& operator_def, Workspace* ws);
-  ~GatherDNNLowPOp();
+  ~GatherDNNLowPOp() override;
   bool RunOnDevice() override;
 
   template <typename Index>

caffe2/queue/blobs_queue_db.h

@@ -42,7 +42,7 @@ class BlobsQueueDBCursor : public Cursor {
     CAFFE_ENFORCE(value_blob_index_ >= 0, "value_blob_index < 0");
   }
 
-  virtual ~BlobsQueueDBCursor() {}
+  ~BlobsQueueDBCursor() override {}
 
   void Seek(const string& /* unused */) override {
     CAFFE_THROW("Seek is not supported.");
@@ -121,7 +121,7 @@ class BlobsQueueDB : public DB {
     LOG(INFO) << "BlobsQueueDB constructed";
   }
 
-  virtual ~BlobsQueueDB() {
+  ~BlobsQueueDB() override {
     Close();
   }
 

test/cpp/api/dataloader.cpp

@@ -616,8 +616,7 @@ struct UnCopyableDataset : public datasets::Dataset<UnCopyableDataset> {
   UnCopyableDataset(UnCopyableDataset&&) = default;
   UnCopyableDataset& operator=(UnCopyableDataset&&) = default;
 
-  // NOLINTNEXTLINE(modernize-use-override)
-  ~UnCopyableDataset() = default;
+  ~UnCopyableDataset() override = default;
 
   Example<> get(size_t index) override {
     return {

test/cpp/jit/test_backend_compiler_lib.cpp

@@ -68,8 +68,7 @@ class BackendWithCompiler : public PyTorchBackendInterface {
   // Constructor.
   // NOLINTNEXTLINE(modernize-use-equals-default)
   explicit BackendWithCompiler() {}
-  // NOLINTNEXTLINE(modernize-use-override)
-  virtual ~BackendWithCompiler() = default;
+  virtual ~BackendWithCompiler() override = default;
 
   bool is_available() override {
     return true;

test/cpp/jit/test_backend_lib.cpp

@@ -14,8 +14,7 @@ class TestBackend : public PyTorchBackendInterface {
   // Constructor.
   // NOLINTNEXTLINE(modernize-use-equals-default)
   explicit TestBackend() {}
-  // NOLINTNEXTLINE(modernize-use-override)
-  virtual ~TestBackend() = default;
+  virtual ~TestBackend() override = default;
 
   bool is_available() override {
     return isAvailable;

test/cpp/jit/test_misc.cpp

@@ -1335,8 +1335,8 @@ class TestThreadLocalDebugInfo : public c10::DebugInfoBase {
     model_id_ = model_id;
   }
 
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  virtual ~TestThreadLocalDebugInfo() {}
+  // NOLINTNEXTLINE(modernize-use-equals-default)
+  virtual ~TestThreadLocalDebugInfo() override {}
 
  private:
   int model_id_ = 0;

torch/csrc/distributed/c10d/ProcessGroupMPI.hpp

@@ -118,7 +118,7 @@ class TORCH_API ProcessGroupMPI : public Backend {
         const c10::optional<std::vector<at::Tensor>>& inputTensors =
             c10::nullopt);
 
-    virtual ~AsyncWork();
+    ~AsyncWork() override;
 
     bool isCompleted() override;
 
@@ -144,7 +144,7 @@ class TORCH_API ProcessGroupMPI : public Backend {
   // Constructor will spawn up the worker thread loop
   explicit ProcessGroupMPI(int rank, int size, MPI_Comm pgComm);
 
-  virtual ~ProcessGroupMPI();
+  ~ProcessGroupMPI() override;
 
   // Abort the MPI program, needs to be called when exception is detected
   void abort();

torch/csrc/jit/backends/xnnpack/xnnpack_backend_lib.cpp

@@ -27,8 +27,7 @@ class XNNPackBackend : public PyTorchBackendInterface {
   // Constructor.
   // NOLINTNEXTLINE(modernize-use-equals-default)
   explicit XNNPackBackend() {}
-  // NOLINTNEXTLINE(modernize-use-override)
-  virtual ~XNNPackBackend() = default;
+  virtual ~XNNPackBackend() override = default;
 
   bool is_available() override {
     return xnn_status_success == xnn_initialize(/*allocator=*/nullptr);