Update

[ghstack-poisoned]

Author: swolchok

kernels/portable/cpu/op_glu.cpp

@@ -41,7 +41,10 @@ Tensor& glu_out_tensor(
     Tensor& out) {
   const auto self_size = self.size(dim);
   ET_KERNEL_CHECK(
-      ctx, self.dim() <= kTensorDimensionLimit, InvalidArgument, out);
+      ctx,
+      self.dim() <= static_cast<ssize_t>(kTensorDimensionLimit),
+      InvalidArgument,
+      out);
   std::array<executorch::aten::SizesType, kTensorDimensionLimit> half_sizes;
   std::copy(self.sizes().begin(), self.sizes().end(), half_sizes.begin());
   half_sizes[dim] /= 2;
@@ -74,7 +77,7 @@ Tensor& glu_out_tensor(
     utils::apply_bitensor_elementwise_fn<
         CTYPE_COMPUTE,
         op_name,
-      utils::SupportedTensorDtypes::FLOATHBF16>(
+        utils::SupportedTensorDtypes::FLOATHBF16>(
         [](const auto val_a, const auto val_b) -> CTYPE_COMPUTE {
           // TODO: rewrite this to be vectorization-capable.
           const auto one = static_cast<decltype(val_a)>(1.0);

kernels/portable/cpu/util/broadcast_indexes_range.h

@@ -69,7 +69,8 @@ class BroadcastIndexesIterator {
         sizeof...(args) == kNumInputs && (std::is_same_v<Args, Tensor> && ...),
         "BroadcastIndexesIterator constructor requires kNumInputs input tensor"
         "arguments!");
-    if (support_noncontiguous_tensors || output_dim_or_zero_if_no_broadcasting_ != 0) {
+    if (support_noncontiguous_tensors ||
+        output_dim_or_zero_if_no_broadcasting_ != 0) {
       effective_input_broadcast_strides_ = {
           effective_input_broadcast_stride(output, args)...};
     }
@@ -261,8 +262,8 @@ class BroadcastIndexesIterator {
 template <std::size_t kNumInputs, bool support_noncontiguous_tensors = false>
 class BroadcastIndexesRange {
  public:
-  using iterator =
-      internal::BroadcastIndexesIterator<kNumInputs, support_noncontiguous_tensors>;
+  using iterator = internal::
+      BroadcastIndexesIterator<kNumInputs, support_noncontiguous_tensors>;
 
   template <typename... Args>
   BroadcastIndexesRange(const Tensor& output, const Args&... args)

kernels/portable/cpu/util/elementwise_util.h

@@ -56,7 +56,7 @@ namespace internal {
  * strides; normally, this is not strictly necessary because ExecuTorch
  * Tensors are contiguous.
  */
-struct SupportNoncontiguousTensors{
+struct SupportNoncontiguousTensors {
   explicit SupportNoncontiguousTensors() = default;
 };
 
@@ -245,7 +245,8 @@ inline void apply_elementwise_fn(
   apply_elementwise_fn_generic_impl<
       CTYPE_COMPUTE,
       op_name,
-      support_noncontiguous_tensors>(compute_fun, ctx, out, out_dtypes, inputs...);
+      support_noncontiguous_tensors>(
+      compute_fun, ctx, out, out_dtypes, inputs...);
 }
 
 /// DEPRECATED: prefer the variant with out_dtypes in the template argument.
@@ -343,8 +344,8 @@ inline void apply_bitensor_elementwise_fn(
   internal::apply_elementwise_fn<
       CTYPE_COMPUTE,
       op_name,
-    out_dtypes,
-    /*support_noncontiguous_tensors*/false>(
+      out_dtypes,
+      /*support_noncontiguous_tensors*/ false>(
       compute_fun,
       ctx,
       out,
@@ -369,8 +370,8 @@ inline void apply_bitensor_elementwise_fn(
   internal::apply_elementwise_fn<
       CTYPE_COMPUTE,
       op_name,
-    out_dtypes,
-    /*support_noncontiguous_tensors*/true>(
+      out_dtypes,
+      /*support_noncontiguous_tensors*/ true>(
       compute_fun,
       ctx,
       out,