[BE][5/5] fix typos in aten/ (aten/src/ATen/) (#157554)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/157554
Approved by: https://github.com/yewentao256, https://github.com/albanD
ghstack dependencies: #157553

.lintrunner.toml

@@ -1113,7 +1113,6 @@ exclude_patterns = [
     # These files are all grandfathered in, feel free to remove from this list
     # as necessary
     # NOTE: remove the patterns in the order they are listed
-    'aten/src/ATen/[a-mA-M]*/**',
     'test/**',
 ]
 init_command = [

aten/src/ATen/core/DimVector.h

@@ -3,7 +3,7 @@
 
 namespace at {
 
-// Re-declaring 'DimVector' type and size inside 'at' namespace.
+// Redeclaring 'DimVector' type and size inside 'at' namespace.
 // This is done to avoid modifying every use into their 'c10'
 // equivalent.
 

aten/src/ATen/core/GeneratorForPrivateuseone.cpp

@@ -16,7 +16,7 @@ _GeneratorRegister::_GeneratorRegister(const GeneratorFuncType& func) {
 
   TORCH_WARN_DEPRECATION(
       "REGISTER_GENERATOR_PRIVATEUSE1 is deprecated. \
-      Please derive PrivateUse1HooksInterface to implememt getNewGenerator instead.")
+      Please derive PrivateUse1HooksInterface to implement getNewGenerator instead.")
 
   TORCH_CHECK(
       !GetGeneratorPrivate().has_value(),

aten/src/ATen/core/IListRef.h

@@ -149,7 +149,7 @@
  *    First, keep in mind that we assume that boxed containers will
  *    have to deal with `IValue` (e.g. `c10::List`). In this context,
  *    what may be happening is that `IValue` doesn't store internally
- *    your type `T`. Instead, it constructs a type new `T` everytime
+ *    your type `T`. Instead, it constructs a type new `T` every time
  *    you try to get `T` for it (see `IListRef<at::OptinalTensorRef>`).
  */
 
@@ -186,7 +186,7 @@ class IListRef;
  * This macro is useful because it allows us to handle different
  * types (that correspond to different tags) to be implemented
  * only once. We can do it even when the implementation of the
- * different tags aren't syntatically the same, by dispatching
+ * different tags aren't syntactically the same, by dispatching
  * it to a function (e.g. `ImplT::<dispatch-function>(this_)`).
  */
 #define TORCH_ILISTREF_UNWRAP(TAG, BODY)                         \

aten/src/ATen/core/IListRef_inl.h

@@ -42,7 +42,7 @@ class IListRefTagImplBase<IListRefTag::Unboxed, T, ListElemT> {
   /*
    * We have these function (besides the `unwrap`s above) because the
    * implementation for both `IListRef::operator[]` and `IListRefIterator::operator*`
-   * weren't syntatically equal for the existing tags at the time
+   * weren't syntactically equal for the existing tags at the time
    * (`Unboxed` and `Boxed`).
    */
   static IListRefConstRef<T> front(const list_type& lst) {

aten/src/ATen/core/Variadic.h

@@ -12,7 +12,7 @@ namespace at {
 // in order.  This is most commonly used in autogenerated code,
 // where it is convenient to have a function that can uniformly
 // take arguments of different types.  If your arguments
-// are homogenous consider using a std::initializer_list instead.
+// are homogeneous consider using a std::initializer_list instead.
 //
 // For examples of this in use, see torch/csrc/utils/variadic.h
 template <typename F>

aten/src/ATen/core/dispatch/Dispatcher.cpp

@@ -111,7 +111,7 @@ void Dispatcher::waitForDef(const FunctionSchema& schema) {
   TORCH_INTERNAL_ASSERT(r,
     "Expected main interpreter to define ", schema.operator_name(),
     ", but this didn't happen within timeout.  Are you trying to load "
-    "different models in the same torchdeploy/multipy instance?  You "
+    "different models in the same torchdeploy/multipy instance?  You " // codespell:ignore
     "must warmup each interpreter identically, e.g., import all "
     "the same dependencies.");
 }
@@ -129,7 +129,7 @@ void Dispatcher::waitForImpl(const OperatorName& op_name, std::optional<c10::Dis
   TORCH_INTERNAL_ASSERT(r,
     "Expected main interpreter to implement ", dk, " for ", op_name,
     ", but this didn't happen within timeout.  Are you trying to load "
-    "different models in the same torchdeploy/multipy instance?  You "
+    "different models in the same torchdeploy/multipy instance?  You " // codespell:ignore
     "must warmup each interpreter identically, e.g., import all "
     "the same dependencies.");
 }
@@ -442,8 +442,8 @@ RegistrationHandleRAII Dispatcher::registerFallback(DispatchKey dispatchKey, Ker
 
   auto idx = getDispatchTableIndexForDispatchKey(dispatchKey);
   TORCH_CHECK(idx >= 0 && static_cast<uint64_t>(idx) < backendFallbackKernels_.size(), "idx=", idx);
-  // NB: Perserve BC for registering fallback for AutogradPrivateUse1 multiple time,
-  // refer to https://github.com/pytorch/pytorch/issues/163979 for more informations.
+  // NB: Preserve BC for registering fallback for AutogradPrivateUse1 multiple time,
+  // refer to https://github.com/pytorch/pytorch/issues/163979 for more information.
   TORCH_CHECK(
       dispatchKey == DispatchKey::AutogradPrivateUse1 ||
           !backendFallbackKernels_[idx].kernel.isValid(),

aten/src/ATen/core/dispatch/Dispatcher.h

@@ -222,7 +222,8 @@ class TORCH_API Dispatcher final {
     return backendFallbackKernels_[dispatch_ix].kernel.isValid();
   }
 
-  // Used by torchdeploy/multipy for multiple interpreters racing.
+  // Used by torchdeploy/multipy for multiple  // codespell:ignore: multipy
+  // interpreters racing.
   void waitForDef(const FunctionSchema& schema);
   void waitForImpl(
       const OperatorName& op_name,
@@ -414,7 +415,7 @@ class TORCH_API Dispatcher final {
   std::unique_ptr<detail::RegistrationListenerList> listeners_;
 
   // This condition variable gets notified whenever we add a new def/impl to the
-  // dispatch table.  This is primarily used by multipy/torchdeploy, when
+  // dispatch table.  This is primarily used by multiply/torchdeploy, when
   // we have multiple interpreters trying to register to the dispatch table.
   // In this situation, whenever the non-primary interpreter would have tried
   // to register to the dispatch table, instead it will check to see if the

aten/src/ATen/core/ivalue_inl.h

@@ -992,7 +992,7 @@ struct C10_EXPORT ivalue::Future final : c10::intrusive_ptr_target {
     std::unique_lock<std::mutex> lock(mutex_);
     if (completed_) {
       // This should be rare and shouldn't cause log spew. Its important to
-      // log errors and thats why we have this log here.
+      // log errors and that's why we have this log here.
       std::string msg = c10::str(
           "Skipping setting following error on the Future since "
           "it is already marked completed (this is not necessarily "

aten/src/ATen/core/jit_type.h

@@ -887,7 +887,7 @@ struct TORCH_API ListType
   // this function will return the global singleton type pointer
   // the type List<T>.
   // The extra "identifier" argument is needed because we have multiple container types
-  // that all re-use this function (List<T>, array<T, N>, etc.)
+  // that all reuse this function (List<T>, array<T, N>, etc.)
   static TypePtr get(const std::string& identifier, TypePtr inner);
 
   // common cast List[Tensor]
@@ -985,7 +985,7 @@ struct TORCH_API DictType : public SharedType {
   // this function will return the global singleton type pointer
   // the type List<T>.
   // The extra "identifier" argument is needed because we have multiple container types
-  // that all re-use this function (Dict<K, V> and unordered_map<K, V>)
+  // that all reuse this function (Dict<K, V> and unordered_map<K, V>)
   static TypePtr get(const std::string& identifier, TypePtr key, TypePtr val);
 
  private:

aten/src/ATen/cpu/vec/vec512/vec512_float8.h

@@ -498,8 +498,8 @@ static inline Vectorized<T> binary_fp8_op_as_fp32(
 
 // Refer to
 // https://github.com/pytorch/pytorch/pull/153364#discussion_r2086509353 FP8 +,
-// -, *, /, planed to be deleted in the future and here is just to make compiler
-// happy
+// -, *, /, planned to be deleted in the future and here is just to make
+// compiler happy
 Vectorized<Float8_e4m3fn> inline operator+(
     const Vectorized<Float8_e4m3fn>& a,
     const Vectorized<Float8_e4m3fn>& b) {
@@ -585,8 +585,8 @@ class Vectorized<Float8_e5m2> : public Vectorizedf8<Float8_e5m2> {
 
 // Refer to
 // https://github.com/pytorch/pytorch/pull/153364#discussion_r2086509353 FP8 +,
-// -, *, /, planed to be deleted in the future and here is just to make compiler
-// happy
+// -, *, /, planned to be deleted in the future and here is just to make
+// compiler happy
 Vectorized<Float8_e5m2> inline operator+(
     const Vectorized<Float8_e5m2>& a,
     const Vectorized<Float8_e5m2>& b) {

aten/src/ATen/cuda/CUDAGreenContext.cpp

@@ -7,7 +7,7 @@
 #define HAS_CUDA_GREEN_CONTEXT() 1
 #else
 #define HAS_CUDA_GREEN_CONTEXT() 0
-// Suppress unsued private field warnings as this class is not supposed to be called
+// Suppress unused private field warnings as this class is not supposed to be called
 C10_DIAGNOSTIC_PUSH_AND_IGNORED_IF_DEFINED("-Wunused-private-field")
 #endif
 

aten/src/ATen/cuda/CUDASparseDescriptors.cpp

@@ -179,7 +179,7 @@ CuSparseSpMatCsrDescriptor::CuSparseSpMatCsrDescriptor(const Tensor& input, int6
           batch_offset * values_batch_stride * values.itemsize(),
       index_type, // data type of row offsets index
       index_type, // data type of col indices
-      CUSPARSE_INDEX_BASE_ZERO, // base index of row offset and col indes
+      CUSPARSE_INDEX_BASE_ZERO, // base index of row offset and col index
       value_type // data type of values
       ));
 

aten/src/ATen/cuda/CachingHostAllocator.h

@@ -10,7 +10,7 @@ namespace at::cuda {
 //
 // A caching allocator for CUDA host allocations (pinned memory).
 //
-// This provides a drop-in replacement for THCudaHostAllocator, which re-uses
+// This provides a drop-in replacement for THCudaHostAllocator, which reuses
 // freed pinned (page-locked) memory allocations. This avoids device
 // synchronizations due to cudaFreeHost calls.
 //
@@ -26,7 +26,7 @@ inline TORCH_CUDA_CPP_API at::HostAllocator* getCachingHostAllocator() {
 }
 
 // Records an event in the specified stream. The allocation corresponding to the
-// input `ptr`/`ctx` will not be re-used until the event has occurred.
+// input `ptr`/`ctx` will not be reused until the event has occurred.
 C10_DEPRECATED_MESSAGE(
   "at::cuda::CachingHostAllocator_recordEvent(...) is deprecated. Please use at::getHostAllocator(at::kCUDA)->record_event(...) instead.")
 inline TORCH_CUDA_CPP_API bool CachingHostAllocator_recordEvent(

aten/src/ATen/cuda/detail/TensorInfo.cuh

@@ -93,7 +93,7 @@ struct IndexToOffset {
   }
 };
 
-// Uses dynamic (runtime) instead of static (compiletime) dims
+// Uses dynamic (runtime) instead of static (compile time) dims
 template <typename T, typename IndexType>
 struct IndexToOffset<T, IndexType, -1> {
   static inline __host__ __device__ IndexType get(

aten/src/ATen/cuda/jiterator.cu

@@ -32,7 +32,7 @@ static inline void launch_jitted_vectorized_kernel_dynamic(
 
   // Different kernels are compiled depending on what we're vectorizing up to (1, 2 or 4 elements)
   //   fn_ptr is set to the appropriate function based on the vec size and GPU used
-  // TODO: Memory use can probably be optimized by re-using kernels across GPUs with
+  // TODO: Memory use can probably be optimized by reusing kernels across GPUs with
   //   the same compute capability
 
   std::string f_inputs_type_str = at::cuda::jit::typeName(common_dtype);

aten/src/ATen/functorch/LegacyVmapTransforms.h

@@ -143,7 +143,7 @@ struct TORCH_API VmapPhysicalView {
   // mapping a physical tensor to a new logical tensor (BatchedTensor)
   VmapPhysicalToLogicalMap getPhysicalToLogicalMap() const;
 
-  // Maps a logical shape to a physical shape by pre-pending the batch
+  // Maps a logical shape to a physical shape by prepending the batch
   // sizes to the logical shape.
   VmapDimVector getPhysicalShape(IntArrayRef logical_shape) const;
   SymDimVector getPhysicalShape(c10::SymIntArrayRef logical_shape) const;

aten/src/ATen/functorch/TensorWrapper.h

@@ -27,7 +27,7 @@ namespace at::functorch {
 //
 // There are alternative designs we could have chosen (e.g. each grad transform
 // stores a weak map of Tensor -> AutogradMeta); the benefit of the TensorWrapper
-// design is that we can re-use existing VariableType kernels (i.e. Autograd kernels)
+// design is that we can reuse existing VariableType kernels (i.e. Autograd kernels)
 // without much modification. Since a TensorWrapper looks like a regular Tensor,
 // the VariableType kernel can pull out the AutogradMeta struct from where it
 // expects and extend the autograd graph