Change a comment to avoid the disturb to op benchmark ci. (#38148)

test=document_fix

paddle/fluid/operators/elementwise/elementwise_op_function.h

@@ -1566,7 +1566,7 @@ void ElemwiseExplicitGradCompute(const framework::ExecutionContext &ctx,
 // of broadcast, supporting both CPU and GPU.
 // - CPU implementation cannot support the case when x needs broadcast, thus
 //   this function need to be called with XxxFunctor and XxxInverseFunctor,
-//   like paddle/fluid/operators/elementwise/elementwise_add_op.h#L49 - L55.
+//   like AddFunctor and InverseAddFunctor.
 // - GPU implementation supports all the broadcast cases, thus there is no need
 //   to define and call with XxxInverseFunctor.
 // TODO(liuyiqun): optimize the CPU implementation to support all broadcast