PR hygiene: remove accidental files; revert unrelated changes; keep only the intended test update

Author: naomiOvad

.gitignore

@@ -13,6 +13,7 @@
 #endif
 using namespace std;
 namespace onnxruntime {
+
 #define REGISTER_UNARY_ELEMENTWISE_KERNEL(x, sinceVersion)                            \
   ONNX_CPU_OPERATOR_TYPED_KERNEL(                                                     \
       x,                                                                              \
@@ -759,10 +760,14 @@ bool CommonFastReduceCopy(OpKernelContext* ctx, TensorShapeVector& input_axes, b
     } else {
       input_axes.clear();
     }
-    // noop_with_empty_axes is handled upstream by ApplyNoopEmptyAxesElementwise().
-    // Return false for clarity and to prevent unsafe memcpy fallback.
+
     if (input_axes.empty() && noop_with_empty_axes) {
-      return false;
+      const Tensor* input = ctx->Input<Tensor>(0);
+      auto* output = ctx->Output(0, input->Shape());
+      memcpy(output->MutableDataRaw(),
+             input->DataRaw(),
+             input->SizeInBytes());
+      return true;
     }
   }
   return false;
@@ -795,6 +800,7 @@ bool CommonFastReduceSwitch(OpKernelContext* ctx,
   fast_kind = OptimizeShapeForFastReduce(
       reduced_dims, input_axes.empty() ? axes_ : input_axes,
       fast_shape, output_shape, fast_axes, keepdims_ != 0, noop_with_empty_axes);
+
   if (which_fast_reduce != FastReduceKind::kNone) {
     if (IsFastReduceKindAvailable(fast_kind, which_fast_reduce)) {
       Tensor* output = ctx->Output(0, output_shape);
@@ -914,30 +920,6 @@ bool check_and_reduce_empty_set_input(OpKernelContext* ctx, const gsl::span<cons
   return true;
 }
 
-template <typename AGG>
-inline void ApplyNoopEmptyAxesElementwise(OpKernelContext* ctx) {
-  const Tensor* X = ctx->Input<Tensor>(0);
-  const auto& shape = X->Shape();
-  Tensor* Y = ctx->Output(0, shape);
-
-  if constexpr (!ReduceAggTraits<AGG>::kHasPreOp && !ReduceAggTraits<AGG>::kHasPostOp) {
-    std::memcpy(Y->MutableDataRaw(), X->DataRaw(), X->SizeInBytes());
-
-  } else {
-    using Tin = typename AGG::input_type;
-    using Tacc = typename AGG::value_type;
-    const Tin* x = X->Data<Tin>();
-    Tacc* y = Y->MutableData<Tacc>();
-    const int64_t n = shape.Size();
-
-    for (int64_t i = 0; i < n; ++i) {
-      AGG agg(1, x[i]);
-      agg.update(x[i]);
-      y[i] = agg.get_value();
-    }
-  }
-}
-
 template <typename AGG>
 void CommonReduce1Loop(OpKernelContext* ctx,
                        const gsl::span<const int64_t>& axes_, int64_t keepdims_,
@@ -946,11 +928,6 @@ void CommonReduce1Loop(OpKernelContext* ctx,
     return;
   }
 
-  if (axes_.empty() && noop_with_empty_axes) {
-    ApplyNoopEmptyAxesElementwise<AGG>(ctx);
-    return;
-  }
-
   FastReduceKind fast_kind;
   TensorShapeVector fast_shape;
   TensorShapeVector output_shape;
@@ -962,7 +939,6 @@ void CommonReduce1Loop(OpKernelContext* ctx,
 
   const Tensor* input = ctx->Input<Tensor>(0);
   Tensor* output = ctx->Output(0, output_shape);
-
   if (fast_kind == FastReduceKind::kEmpty) {
     const TensorShape& input_shape = input->Shape();
     if (input_shape.Size() == 1) {
@@ -989,11 +965,6 @@ void CommonReduce2Loops(OpKernelContext* ctx,
     return;
   }
 
-  if (axes_.empty() && noop_with_empty_axes) {
-    ApplyNoopEmptyAxesElementwise<AGG>(ctx);
-    return;
-  }
-
   FastReduceKind fast_kind;
   TensorShapeVector fast_shape, output_shape, fast_axes;
   if (CommonFastReduce<AGG>(ctx, axes_, keepdims_, noop_with_empty_axes,
@@ -1017,6 +988,7 @@ void CommonReduce2Loops(OpKernelContext* ctx,
     }
     return;
   }
+
   ResultsNoTransposePrepareForReduce last_results;
   NoTransposeReduce2Loops<AGG>(output, fast_shape, *input, fast_axes, ctx->GetOperatorThreadPool(), last_results);
 }

Issue#26432/chec_pytorch.py

@@ -784,42 +784,6 @@ class ReduceAggregatorLogSumExp : public ReduceAggregator<T, T> {
   }
 };
 
-template <typename AGG>
-struct ReduceAggTraits {
-  static constexpr bool kHasPreOp = false;
-  static constexpr bool kHasPostOp = false;
-};
-
-template <typename T>
-struct ReduceAggTraits<ReduceAggregatorL1<T>> {
-  static constexpr bool kHasPreOp = true;
-  static constexpr bool kHasPostOp = false;
-};
-
-template <typename T>
-struct ReduceAggTraits<ReduceAggregatorL2<T>> {
-  static constexpr bool kHasPreOp = true;
-  static constexpr bool kHasPostOp = true;
-};
-
-template <typename T, typename TVAL>
-struct ReduceAggTraits<ReduceAggregatorSumSquare<T, TVAL>> {
-  static constexpr bool kHasPreOp = true;
-  static constexpr bool kHasPostOp = false;
-};
-
-template <typename T>
-struct ReduceAggTraits<ReduceAggregatorLogSum<T>> {
-  static constexpr bool kHasPreOp = false;
-  static constexpr bool kHasPostOp = true;
-};
-
-template <typename T>
-struct ReduceAggTraits<ReduceAggregatorLogSumExp<T>> {
-  static constexpr bool kHasPreOp = true;
-  static constexpr bool kHasPostOp = true;
-};
-
 void NoTransposePrepareForReduce(const TensorShape& new_input_shape,
                                  gsl::span<const int64_t> reduced_axes,
                                  ResultsNoTransposePrepareForReduce& results);