[BE] fix typos in torchgen/ (#156083)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/156083
Approved by: https://github.com/jingsh
ghstack dependencies: #156079, #156082

.lintrunner.toml

@@ -1165,7 +1165,6 @@ exclude_patterns = [
     'scripts/**',
     'test/**',
     'torch/**',
-    'torchgen/**',
 ]
 init_command = [
     'python3',

torchgen/_autoheuristic/README.md

@@ -89,7 +89,7 @@ context = AHContext()
 context.add_feature("m", mat1.shape[0])
 context.add_feature("k", mat1.shape[1])
 
-# adding a categorical feture
+# adding a categorical feature
 context.add_feature("mat1_dtype", mat1.dtype, is_categorical=True)
 ```
 

torchgen/_autoheuristic/benchmark_utils.py

@@ -18,7 +18,7 @@ def transpose_tensors(p_transpose_both: float = 0.05) -> tuple[bool, bool]:
 
 def fits_in_memory(dtype: Any, m: int, k: int, n: int) -> Any:
     threshold_memory = torch.cuda.get_device_properties(0).total_memory / 4
-    # dividing by 4 beause we otherwise sometimes run out of memory, I assume because
+    # dividing by 4 because we otherwise sometimes run out of memory, I assume because
     # inductor creates copies of tensors for benchmarking?
     return dtype.itemsize * (m * k + k * n + m * n) < threshold_memory
 

torchgen/_autoheuristic/collect_data.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-# this script makes it easy parallize collecting data across using multiple GPUs
+# This script makes it easy to parallelize data collection across multiple GPUs
 
 # Check if tmux is installed
 if ! command -v tmux &> /dev/null; then

torchgen/_autoheuristic/mixed_mm/generate_heuristic_mixedmm.sh

@@ -12,7 +12,7 @@ MODE=$1
 # !!! SPECIFY THE GPUs THAT YOU WANT TO USE HERE !!!
 GPU_DEVICE_IDS="4,5"
 
-# !!! SPECIFY THE CONDA ENVIRONEMNT THAT YOU WANT TO BE ACTIVATED HERE !!!
+# !!! SPECIFY THE CONDA ENVIRONMENT THAT YOU WANT TO BE ACTIVATED HERE !!!
 CONDA_ENV=heuristic-pr
 
 NUM_SAMPLES=2000

torchgen/_autoheuristic/pad_mm/generate_heuristic_pad_mm.sh

@@ -12,7 +12,7 @@ MODE=$1
 # !!! SPECIFY THE GPUs THAT YOU WANT TO USE HERE !!!
 GPU_DEVICE_IDS="4,5"
 
-# !!! SPECIFY THE CONDA ENVIRONEMNT THAT YOU WANT TO BE ACTIVATED HERE !!!
+# !!! SPECIFY THE CONDA ENVIRONMENT THAT YOU WANT TO BE ACTIVATED HERE !!!
 CONDA_ENV=heuristic-pr
 
 NUM_SAMPLES=2000

torchgen/api/autograd.py

@@ -93,7 +93,7 @@ class ForwardDerivative:
     # This is only used by inplace operations
     required_original_self_value: bool
 
-    # If this formula is specified in derivatives.yaml or if we are re-using the
+    # If this formula is specified in derivatives.yaml or if we are reusing the
     # out of place formula for inplace
     is_reusing_outplace_formula: bool
 
@@ -632,7 +632,7 @@ def match_differentiability_info(
             info_dict = non_functional_info_by_signature[f_sig]
             # See https://github.com/pytorch/pytorch/pull/76320/files#r874816389
             assert not any(
-                any("self" in str(inpt.nctype.name) for inpt in info.all_saved_inputs)
+                any("self" in str(input.nctype.name) for input in info.all_saved_inputs)
                 for info in info_dict.values()
             ), f"""\
 Attempted to convert a derivative formula for a mutable operator
@@ -699,7 +699,7 @@ Attempted to convert a derivative formula for a mutable operator
                 #      we make sure that the original value of the input that is being modified inplace (self_p) is
                 #      not used in the formula. Note that the formula can use "original_self_p" here and that would
                 #      trigger a clone of the original input.
-                #    - If we are re-using the out of place formula (is_exact_match == False) then we replace every
+                #    - If we are reusing the out of place formula (is_exact_match == False) then we replace every
                 #      occurrence of self_p and self_t by original_self_p and original_self_t. These will be
                 #      populated by cloned version of the original input (either the clone done by the backward AD
                 #      logic if self is also used in a backward formula or a special clone that we add).

torchgen/api/cpp.py

@@ -127,7 +127,7 @@ def valuetype_type(
 
 
 # Translation of types occurring in JIT arguments to a C++ argument type.
-# If remove_non_owning_ref_types is set, we'll guarantee that the outputed CType is not a non-owning reference type.
+# If remove_non_owning_ref_types is set, we'll guarantee that the output CType is not a non-owning reference type.
 # For example, we'll return std::vector<int> instead of IntArrayRef.
 # See Note [translation from C++ reference to value types]
 def argumenttype_type(

torchgen/api/types/signatures.py

@@ -49,7 +49,7 @@ class CppSignature:
 
     # Is this a fallback C++ binding?  Fallback bindings are enabled by
     # manual_cpp_binding: True and are alternate, non-public API that
-    # lets manual C++ binding implementors access the binding that would
+    # lets manual C++ binding implementers access the binding that would
     # have been automatically generated
     fallback_binding: bool = False
 

torchgen/dest/native_functions.py

@@ -12,7 +12,7 @@ def torch_api_key_word_prefix(bankend_index: BackendIndex) -> str:
     if bankend_index.external:
         return ""
 
-    # Although Intel GPU ATen library is out-of-tree, it still utilizes torchgen to produce structrued
+    # Although Intel GPU ATen library is out-of-tree, it still utilizes torchgen to produce structured
     # kernels. Regarding these produced structured kernels, they should be visible for the Intel GPU ATen
     # library. Therefore, we need to add "TORCH_XPU_API" prefix to these structured kernels,
     # rather than "TORCH_API". Because the semantic of "TORCH_API" is "hidden" for out-of-tree backends.

torchgen/dest/register_dispatch_key.py

@@ -764,7 +764,7 @@ resize_out(out, sizes, strides, options);
         # we generate CompositeExplicitAutogradNonFunctional implementations of functional and inplace
         # based on the out implementation.  But in fact, out is definable by
         # functional too (just not very efficiently), and this is honestly the
-        # MORE likely situation for a backend implementor.  How do we pick?
+        # MORE likely situation for a backend implementer.  How do we pick?
         # Well, taking a page from Haskell type classes and default methods,
         # we could conceivably register a circular definition (out in terms
         # of functional, and functional in terms of out) and just require
@@ -777,7 +777,7 @@ resize_out(out, sizes, strides, options);
             and f.func.kind() is SchemaKind.out
         ):
             # Never generate a default implementation for out, that's what you
-            # have to define as a backend implementor
+            # have to define as a backend implementer
             return None
 
         # Note [Direct dispatch bindings]

torchgen/dest/ufunc.py

@@ -42,7 +42,7 @@ if TYPE_CHECKING:
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #
 
 # NB: not bothering to generate dispatch stub forward declaration in header,
-# we can just paste it whereever necessary
+# we can just paste it wherever necessary
 
 # TODO: use BackendIndex
 # dispatch_key: DispatchKey  # only CPU/CUDA right now

torchgen/gen.py

@@ -506,7 +506,7 @@ def static_dispatch(
 ) -> str:
     """
     For a given `NativeFunction`, find out the corresponding backend and dispatch to it. If more than one
-    backends exsit, fallback to static dispatch by determining dispatch key from inputs.
+    backends exist, fallback to static dispatch by determining dispatch key from inputs.
     Arguments:
         sig: A CppSignature or DispatcherSignature for this native function we want to use.
         f: NativeFunction to generate static dispatch.
@@ -2611,7 +2611,7 @@ def gen_source_files(
     #     but they could theoretically be called from user code (I added these kernels for completeness,
     #     since the ops are part of the public API).
     # (2) A derivative formula for every {view}_copy operator
-    #     {view}_copy operators can re-use the same derivative formulas as their {view} op counterparts,
+    #     {view}_copy operators can reuse the same derivative formulas as their {view} op counterparts,
     #     so rather than stamping all of the entries out in derivatives.yaml,
     #     we codegen them in.
     #     This is similar to how autograd codegen doesn't require inplace ops to have a derivatives.yaml entry.

torchgen/gen_functionalization_type.py

@@ -198,7 +198,7 @@ def is_tensor_like(a: Argument | TensorOptionsArguments | SelfArgument) -> bool:
 # We need to wrap / unwrap various arguments from the op in the functionalization kernels.
 # Some op schemas include non-owning types though (like TensorList),
 # and when we unwrap them we expect to get out an owning type!.
-# We also return a lambda that tells you how to conver the non-owning type argument into the owning type.
+# We also return a lambda that tells you how to convert the non-owning type argument into the owning type.
 def get_owning_type(t: CType) -> tuple[CType, Callable[[str], str]]:
     if t == BaseCType(tensorListT):
         return VectorCType(BaseCType(tensorT)), lambda x: f"{x}.vec()"
@@ -441,7 +441,7 @@ def emit_view_functionalization_body(
       // This function adds the above view meta to the current tensor and replays them off the base,
       // mutating the size/stride info of the current FunctionalTensorWrapper.
       // Because of this, we need to make sure to run the reference shape function above,
-      // BEFORE doing this (otherwise we'll end up runnin the reference function using the wrong sizes/strides)
+      // BEFORE doing this (otherwise we'll end up running the reference function using the wrong sizes/strides)
       at::functionalization::impl::mutate_view_meta({view_tensor_name}, view_meta);
       // See  Note [Propagating strides in the functionalization pass]
       // XLA/LTC don't implement the logic to propagate strides correctly, so we need to rely

torchgen/gen_schema_utils.py

@@ -47,7 +47,7 @@ class TypeGen:
             all_base_tys = [TypeGen.from_example(x) for x in obj]
             if len(set(all_base_tys)) > 1:
                 raise RuntimeError(
-                    f"Cannot generate schema for a seqeunce of args of heterogeneous types: {all_base_tys}. "
+                    f"Cannot generate schema for a sequence of args of heterogeneous types: {all_base_tys}. "
                     "Consider unpacking the argument and give proper names to them if possible "
                     "instead of using *args."
                 )

torchgen/model.py

@@ -593,7 +593,7 @@ class NativeFunction:
     has_composite_explicit_autograd_non_functional_kernel: bool
 
     # Tags are used to describe semantic information about (groups of) operators,
-    # That aren't easily inferrable directly from the operator's schema.
+    # That aren't easily inferable directly from the operator's schema.
     tags: set[str]
 
     # NB: The benefit of defining a dataclass is that we automatically get

torchgen/shape_functions/gen_jit_shape_functions.py

@@ -102,7 +102,7 @@ def gen_serialized_decompisitions() -> str:
     output_strs.append(curr_str)
 
     final_output = ""
-    # Windows compiler doesnt correctly handle adjacent
+    # Windows compiler doesn't correctly handle adjacent
     # string literals
     for output_str in output_strs:
         start = '+ std::string(R"=====('