[11/N] Fix extra warnings brought by clang-tidy-17 (#139599)

Follows #139385
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139599
Approved by: https://github.com/sraikund16

.clang-tidy

@@ -35,13 +35,14 @@ cppcoreguidelines-*,
 hicpp-exception-baseclass,
 hicpp-avoid-goto,
 misc-*,
+-misc-confusable-identifiers,
 -misc-const-correctness,
 -misc-include-cleaner,
 -misc-use-anonymous-namespace,
 -misc-unused-parameters,
 -misc-no-recursion,
 -misc-non-private-member-variables-in-classes,
--misc-confusable-identifiers,
+-misc-unused-using-decls,
 modernize-*,
 -modernize-macro-to-enum,
 -modernize-return-braced-init-list,

.lintrunner.toml

@@ -227,6 +227,9 @@ exclude_patterns = [
     '**/generated/**',
     '**/*pb.h',
     '**/*inl.h',
+    'aten/src/ATen/cpu/FlushDenormal.cpp',
+    'aten/src/ATen/cpu/Utils.cpp',
+    'aten/src/ATen/cpu/vml.h',
     'aten/src/ATen/CPUFixedAllocator.h',
     'aten/src/ATen/Parallel*.h',
     'c10/xpu/**/*.h',

aten/src/ATen/core/CachingHostAllocator.h

@@ -40,6 +40,7 @@ struct alignas(64) FreeBlockList {
 
 namespace {
   // Max cached block sizes: (1 << MAX_SIZE_INDEX) bytes
+  // NOLINTNEXTLINE(misc-definitions-in-headers)
   constexpr size_t MAX_SIZE_INDEX = 64;
 }
 

aten/src/ATen/cuda/CUDASparseBlas.h

@@ -12,6 +12,7 @@
 #include <ATen/cuda/CUDAContext.h>
 #include <ATen/cuda/CUDASparse.h>
 
+// NOLINTBEGIN(misc-misplaced-const)
 namespace at::cuda::sparse {
 
 #define CUSPARSE_CSRGEAM2_BUFFERSIZE_ARGTYPES(scalar_t)             \
@@ -316,3 +317,4 @@ void bsrsm2_solve<c10::complex<double>>(
 #endif // AT_USE_HIPSPARSE_TRIANGULAR_SOLVE
 
 } // namespace at::cuda::sparse
+// NOLINTEND(misc-misplaced-const)

aten/src/ATen/cuda/CUDASparseDescriptors.cpp

@@ -8,6 +8,7 @@
 namespace at::cuda::sparse {
 
 cusparseStatus_t destroyConstDnMat(const cusparseDnMatDescr* dnMatDescr) {
+  // NOLINTNEXTLINE(*const-cast)
   return cusparseDestroyDnMat(const_cast<cusparseDnMatDescr*>(dnMatDescr));
 }
 
@@ -83,6 +84,7 @@ cusparseDnMatDescr_t createRawDnMatDescriptor(const Tensor& input, int64_t batch
 #endif
 
   auto batch_stride = ndim > 2 && batch_offset >= 0 ? input_strides[ndim - 3] : 0;
+  // NOLINTNEXTLINE(*const-cast)
   void* data_ptr = is_const ? const_cast<void*>(input.const_data_ptr()) : input.data_ptr();
   void* values_ptr = static_cast<char*>(data_ptr) +
       batch_offset * batch_stride * input.itemsize();

aten/src/ATen/cudnn/AutocastRNN.cpp

@@ -18,7 +18,7 @@ Autocast wrapper for CuDNN RNNs (the weight reflattening needs special attention
 
 // To be registered for the "_cudnn_rnn(...)" schema.
 // _cudnn_rnn is autograd-exposed (test_autocast_cudnn_rnn in test_cuda.py includes a test to confirm)
-std::tuple<Tensor,Tensor,Tensor,Tensor,Tensor>
+static std::tuple<Tensor,Tensor,Tensor,Tensor,Tensor>
 _cudnn_rnn_cast_reflatten(const Tensor & input,
                           TensorList weight,
                           int64_t weight_stride0,

aten/src/ATen/cudnn/Descriptors.cpp

@@ -6,6 +6,7 @@
 #include <iostream>
 #include <sstream>
 
+// NOLINTBEGIN(*c-arrays*)
 namespace at::native {
 
 namespace {
@@ -101,7 +102,7 @@ std::ostream& operator<<(std::ostream & out, const TensorDescriptor& d) {
   int nbDims = 0;
   int dimA[CUDNN_DIM_MAX];
   int strideA[CUDNN_DIM_MAX];
-  cudnnDataType_t dtype;
+  cudnnDataType_t dtype{};
   cudnnGetTensorNdDescriptor(d.desc(), CUDNN_DIM_MAX, &dtype, &nbDims, dimA, strideA);
   out << "    type = " << cudnnTypeToString(dtype) << "\n";
   out << "    nbDims = " << nbDims << "\n";
@@ -143,7 +144,7 @@ void FilterDescriptor::set(const at::Tensor &t, const at::MemoryFormat memory_fo
     size[i] = (int) 1;
   }
   dim = std::max(dim, pad);
-  cudnnTensorFormat_t filter_format;
+  cudnnTensorFormat_t filter_format{};
   switch(memory_format) {
     case at::MemoryFormat::Contiguous:
       filter_format = CUDNN_TENSOR_NCHW;
@@ -155,7 +156,8 @@ void FilterDescriptor::set(const at::Tensor &t, const at::MemoryFormat memory_fo
     default:
       TORCH_INTERNAL_ASSERT(false, "unsupported memory_format for cuDNN filters");
   }
-  set(getDataType(t), (int) dim, size, filter_format);
+  // NOLINTNEXTLINE(*narrowing-conversions)
+  set(getDataType(t), static_cast<int64_t>(dim), size, filter_format);
 }
 
 std::string cudnnMemoryFormatToString(cudnnTensorFormat_t tformat) {
@@ -175,8 +177,8 @@ std::ostream& operator<<(std::ostream & out, const FilterDescriptor& d) {
   out << "FilterDescriptor " << static_cast<void*>(d.desc()) << "\n";
   int nbDims = 0;
   int dimA[CUDNN_DIM_MAX];
-  cudnnDataType_t dtype;
-  cudnnTensorFormat_t tformat;
+  cudnnDataType_t dtype{};
+  cudnnTensorFormat_t tformat{};
   cudnnGetFilterNdDescriptor(d.desc(), CUDNN_DIM_MAX, &dtype, &tformat, &nbDims, dimA);
   out << "    type = " << cudnnTypeToString(dtype) << "\n";
   out << "    tensor_format = " << cudnnMemoryFormatToString(tformat) << "\n";
@@ -193,3 +195,4 @@ std::ostream& operator<<(std::ostream & out, const FilterDescriptor& d) {
 void FilterDescriptor::print() { std::cout << *this; }
 
 }
+// NOLINTEND(*c-arrays*)

aten/src/ATen/cudnn/Descriptors.h

@@ -92,6 +92,7 @@ struct DescriptorDeleter {
 // initialized the first time you call set() or any other initializing
 // function.
 template <typename T, cudnnStatus_t (*ctor)(T**), cudnnStatus_t (*dtor)(T*)>
+// NOLINTNEXTLINE(bugprone-exception-escape)
 class TORCH_CUDA_CPP_API Descriptor {
  public:
   // TODO: Figure out why const-correctness doesn't work here
@@ -128,7 +129,7 @@ public:
   void set(const at::Tensor &t, cudnnRNNDataLayout_t layout, int maxSeqLength, int batchSize, int vectorSize, const int* seqLengthArray);
 private:
   void set(cudnnDataType_t dataType, cudnnRNNDataLayout_t layout, int maxSeqLength, int batchSize, int vectorSize, const int* seqLengthArray) {
-    AT_CUDNN_CHECK(cudnnSetRNNDataDescriptor(mut_desc(), dataType, layout, maxSeqLength, batchSize, vectorSize, seqLengthArray, NULL));
+    AT_CUDNN_CHECK(cudnnSetRNNDataDescriptor(mut_desc(), dataType, layout, maxSeqLength, batchSize, vectorSize, seqLengthArray, nullptr));
   }
 };
 
@@ -224,6 +225,7 @@ struct TORCH_CUDA_CPP_API SpatialTransformerDescriptor
   }
 };
 
+// NOLINTNEXTLINE(bugprone-exception-escape)
 struct TORCH_CUDA_CPP_API DropoutDescriptor
     : public Descriptor<
           cudnnDropoutStruct,
@@ -244,9 +246,8 @@ struct TORCH_CUDA_CPP_API DropoutDescriptor
   }
 
   // Restore a dropout descriptor given a dropout probability and existing RNG state.
-  void set(cudnnHandle_t handle, float dropout, at::Tensor state_) {
+  void set(cudnnHandle_t handle, float dropout, const at::Tensor& state) {
     TORCH_INTERNAL_ASSERT(dropout > 0, "dropout must be nonzero; otherwise call set_no_dropout");
-    state = state_;
     void *state_ptr = state.data_ptr();
     size_t state_size = state.size(0);
     // NB: The seed doesn't actually matter, so we give a dummy value

aten/src/ATen/cudnn/Types.cpp

@@ -5,7 +5,7 @@
 namespace at::native {
 
 cudnnDataType_t getCudnnDataTypeFromScalarType(const at::ScalarType dtype) {
-  if (dtype == c10::kQInt8) {
+  if (dtype == c10::kQInt8 || dtype == at::kChar) {
     return CUDNN_DATA_INT8;
   } else if (dtype == at::kFloat) {
     return CUDNN_DATA_FLOAT;
@@ -19,8 +19,6 @@ cudnnDataType_t getCudnnDataTypeFromScalarType(const at::ScalarType dtype) {
     return CUDNN_DATA_INT32;
   } else if (dtype == at::kByte) {
     return CUDNN_DATA_UINT8;
-  } else if (dtype == at::kChar) {
-    return CUDNN_DATA_INT8;
   }
   std::string msg("getCudnnDataTypeFromScalarType() not supported for ");
   msg += toString(dtype);

aten/src/ATen/native/nested/NestedTensorUtils.h

@@ -32,7 +32,7 @@ struct NestedTensorImpl;
 // The following functions are used to construct nested tensors from buffers and
 // metadata.
 
-inline at::Tensor wrap_buffer(at::Tensor buffer, at::Tensor nested_sizes) {
+inline at::Tensor wrap_buffer(const at::Tensor& buffer, const at::Tensor& nested_sizes) {
   TORCH_CHECK(
       buffer.dim() == 1,
       "Expected given buffer to be 1dim, but got ",
@@ -41,19 +41,19 @@ inline at::Tensor wrap_buffer(at::Tensor buffer, at::Tensor nested_sizes) {
   TORCH_CHECK(
       buffer.is_contiguous(), "Expected given buffer to be contiguous.");
   return at::detail::make_tensor<NestedTensorImpl>(
-      std::move(buffer), std::move(nested_sizes));
+      buffer, nested_sizes);
 }
 
 // TODO: Figure out if we need a non-moving wrap_buffer()
 inline at::Tensor wrap_buffer(
-    at::Tensor buffer,
+    const at::Tensor& buffer,
     at::Tensor nested_sizes,
     at::Tensor nested_strides,
     at::Tensor storage_offsets) {
   TORCH_INTERNAL_ASSERT_DEBUG_ONLY(
       buffer.is_contiguous(), "Given buffer must be contiguous.");
   return at::detail::make_tensor<NestedTensorImpl>(
-      std::move(buffer),
+      buffer,
       std::move(nested_sizes),
       std::move(nested_strides),
       std::move(storage_offsets));
@@ -95,9 +95,9 @@ inline at::Tensor create_nested_view_tensor(
   return at::detail::make_tensor<NestedTensorImpl>(
       c10::TensorImpl::VIEW,
       base,
-      nested_sizes,
-      nested_strides,
-      storage_offsets);
+      std::move(nested_sizes),
+      std::move(nested_strides),
+      std::move(storage_offsets));
 }
 //  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

aten/src/ATen/templates/TensorBody.h

@@ -195,7 +195,7 @@ class TORCH_API Tensor: public TensorBase {
   //
   // TODO: temporarily disabled
 
-  Tensor& operator=(const TensorBase& x) & {
+  Tensor& operator=(const TensorBase& x) & noexcept {
     impl_ = x.getIntrusivePtr();
     return *this;
   }
@@ -204,7 +204,7 @@ class TORCH_API Tensor: public TensorBase {
     return *this;
   }
 
-  Tensor& operator=(const Tensor &x) & {
+  Tensor& operator=(const Tensor &x) & noexcept {
     return operator=(static_cast<const TensorBase&>(x));
   }
   Tensor& operator=(Tensor &&x) & noexcept {

c10/util/intrusive_ptr.h

@@ -664,15 +664,17 @@ struct MaybeOwnedTraits<c10::intrusive_ptr<T>> {
     toDestroy.release();
   }
 
-  static const owned_type& referenceFromBorrow(const borrow_type& borrow) {
+  static const owned_type& referenceFromBorrow(
+      const borrow_type& borrow) noexcept {
     return borrow;
   }
 
-  static const owned_type* pointerFromBorrow(const borrow_type& borrow) {
+  static const owned_type* pointerFromBorrow(
+      const borrow_type& borrow) noexcept {
     return &borrow;
   }
 
-  static bool debugBorrowIsValid(const borrow_type& /*borrow*/) {
+  static bool debugBorrowIsValid(const borrow_type& /*borrow*/) noexcept {
     return true;
   }
 };

tools/onnx/templates/rules.h.in

@@ -1,4 +1,5 @@
 #pragma once
+#include <cstdint>
 
 /**
 ${generated_comment}

torch/csrc/autograd/python_autograd.h

@@ -1,5 +1,6 @@
 #ifndef THP_AUTOGRAD_H
 #define THP_AUTOGRAD_H
+#include <torch/csrc/utils/pythoncapi_compat.h>
 
 PyObject* THPAutograd_initExtension(PyObject* _unused, PyObject* unused);
 void THPAutograd_initFunctions();

torch/csrc/autograd/python_fft_functions.h

@@ -1,4 +1,5 @@
 #pragma once
+#include <torch/csrc/utils/pythoncapi_compat.h>
 
 namespace torch::autograd {
 

torch/csrc/autograd/python_linalg_functions.h

@@ -1,4 +1,5 @@
 #pragma once
+#include <torch/csrc/utils/pythoncapi_compat.h>
 
 namespace torch::autograd {
 

torch/csrc/autograd/python_sparse_functions.h

@@ -1,4 +1,5 @@
 #pragma once
+#include <torch/csrc/utils/pythoncapi_compat.h>
 
 namespace torch::autograd {
 

torch/csrc/autograd/python_special_functions.h

@@ -1,5 +1,5 @@
 #pragma once
-
+#include <torch/csrc/utils/pythoncapi_compat.h>
 namespace torch::autograd {
 
 void initSpecialFunctions(PyObject* module);

torch/csrc/cuda/Module.h

@@ -1,5 +1,6 @@
 #ifndef THCP_CUDA_MODULE_INC
 #define THCP_CUDA_MODULE_INC
+#include <torch/csrc/utils/pythoncapi_compat.h>
 
 PyObject* THCPModule_getDevice_wrap(PyObject* self);
 PyObject* THCPModule_setDevice_wrap(PyObject* self, PyObject* arg);

torch/csrc/cuda/nccl.cpp

@@ -273,6 +273,7 @@ struct NcclCommList {
         devices.data()));
   }
   NcclCommList(NcclCommList&& foo) = default;
+  // NOLINTNEXTLINE(bugprone-exception-escape)
   ~NcclCommList() {
     if (comms) {
       for (const auto i : c10::irange(ndevices)) {
@@ -457,6 +458,7 @@ AutoNcclGroup::AutoNcclGroup(ncclComm_t comm, bool comm_nonblocking)
 #endif
 }
 
+// NOLINTNEXTLINE(bugprone-exception-escape)
 AutoNcclGroup::~AutoNcclGroup() noexcept(false) {
 #if defined(NCCL_MAJOR) && (NCCL_MAJOR >= 2)
   if (comm_nonblocking_ && comm_ != nullptr) {

torch/csrc/cuda/shared/cudnn.cpp

@@ -4,7 +4,6 @@
 #if defined(USE_CUDNN) || defined(USE_ROCM)
 #include <torch/csrc/utils/pybind.h>
 
-#include <array>
 #include <tuple>
 
 namespace {
@@ -22,7 +21,7 @@ version_tuple getCompileVersion() {
 
 version_tuple getRuntimeVersion() {
 #ifndef USE_STATIC_CUDNN
-  int major, minor, patch;
+  int major = 0, minor = 0, patch = 0;
   cudnnGetProperty(MAJOR_VERSION, &major);
   cudnnGetProperty(MINOR_VERSION, &minor);
   cudnnGetProperty(PATCH_LEVEL, &patch);

torch/csrc/profiler/collection.h

@@ -378,7 +378,7 @@ struct TORCH_API Result : public std::enable_shared_from_this<Result> {
   }
 
   template <typename T, typename Fn>
-  void visit_if_base(Fn&& fn) const {
+  void visit_if_base(const Fn& fn) const {
     visit([&](const auto& extra_fields) {
       using extra_fields_t = typename std::remove_cv_t<
           typename std::remove_reference_t<decltype(extra_fields)>>;

torch/csrc/profiler/kineto_client_interface.h

@@ -6,6 +6,6 @@
 namespace torch {
 
 // declare global_kineto_init for libtorch_cpu.so to call
-TORCH_API void global_kineto_init(void);
+TORCH_API void global_kineto_init();
 
 } // namespace torch

torch/csrc/profiler/kineto_shim.cpp

@@ -222,7 +222,7 @@ bool collectivesProfilerExists() {
 
 #ifdef USE_KINETO
 static const std::string setTraceID(const std::string& trace_id) {
-  if (trace_id == "") {
+  if (trace_id.empty()) {
     return "";
   }
   std::stringstream configss;

torch/csrc/profiler/orchestration/observer.cpp

@@ -39,7 +39,7 @@ ProfilerConfig::ProfilerConfig(
     bool with_flops,
     bool with_modules,
     ExperimentalConfig experimental_config,
-    const std::string& trace_id)
+    std::string trace_id)
     : state{state},
       experimental_config{std::move(experimental_config)},
       report_input_shapes{report_input_shapes},
@@ -47,7 +47,7 @@ ProfilerConfig::ProfilerConfig(
       with_stack{with_stack},
       with_flops{with_flops},
       with_modules{with_modules},
-      trace_id{trace_id} {}
+      trace_id{std::move(trace_id)} {}
 
 bool ProfilerConfig::disabled() const {
   return state == torch::profiler::impl::ProfilerState::Disabled;

torch/csrc/profiler/orchestration/observer.h

@@ -104,7 +104,7 @@ struct TORCH_API ProfilerConfig {
       bool with_flops = false,
       bool with_modules = false,
       ExperimentalConfig experimental_config = ExperimentalConfig(),
-      const std::string& trace_id = "");
+      std::string trace_id = "");
 
   bool disabled() const;
   bool global() const;