diff --git a/onnxruntime/core/providers/cuda/cu_inc/binary_elementwise_impl.cuh b/onnxruntime/core/providers/cuda/cu_inc/binary_elementwise_impl.cuh
index 5824c516af..72f0d3652d 100644
--- a/onnxruntime/core/providers/cuda/cu_inc/binary_elementwise_impl.cuh
+++ b/onnxruntime/core/providers/cuda/cu_inc/binary_elementwise_impl.cuh
@@ -272,7 +272,7 @@ void BinaryElementWiseImpl(
           output_rank_or_simple_broadcast,
           *lhs_padded_strides,
           lhs_data,
-          *rhs_padded_strides,
+          TArray<int64_t>(), // rhs is not computed, so no need to deference rhs_padded_strides
           rhs_data,
           *fdm_output_strides,
           output_data,
@@ -281,7 +281,7 @@ void BinaryElementWiseImpl(
     else
       _BinaryElementWise<T, T1, T2, FuncT, false, true, GridDim::maxThreadsPerBlock, GridDim::maxElementsPerThread><<<blocksPerGrid, GridDim::maxThreadsPerBlock, 0>>>(
           output_rank_or_simple_broadcast,
-          *lhs_padded_strides,
+          TArray<int64_t>(), // lhs is not computed, so no need to deference lhs_padded_strides
           lhs_data,
           *rhs_padded_strides,
           rhs_data,
diff --git a/onnxruntime/test/optimizer/cse_test.cc b/onnxruntime/test/optimizer/cse_test.cc
index 8859959517..3ad91954cc 100644
--- a/onnxruntime/test/optimizer/cse_test.cc
+++ b/onnxruntime/test/optimizer/cse_test.cc
@@ -222,6 +222,9 @@ TEST(CseTests, Subgraph) {
     }
   }
 
+  ASSERT_NE(if_true_graph, nullptr);
+  ASSERT_NE(if_false_graph, nullptr);
+
   output_names = GetSortedNames(if_true_graph->GetOutputs());
   ASSERT_EQ(output_names, (std::vector<std::string>{"Result1", "Result2", "Result3"}));