diff --git a/Tests/CudaOnly/SeparateCompilation/file4.cu b/Tests/CudaOnly/SeparateCompilation/file4.cu
index f2ef8e7ef4..f2e5e6ec91 100644
--- a/Tests/CudaOnly/SeparateCompilation/file4.cu
+++ b/Tests/CudaOnly/SeparateCompilation/file4.cu
@@ -4,17 +4,24 @@
 result_type __device__ file1_func(int x);
 result_type_dynamic __device__ file2_func(int x);
 
-static __global__ void file4_kernel(result_type& r, int x)
+static __global__ void file4_kernel(result_type* r, int x)
 {
   // call static_func which is a method that is defined in the
   // static library that is always out of date
-  r = file1_func(x);
+  *r = file1_func(x);
   result_type_dynamic rd = file2_func(x);
 }
 
 EXPORT int file4_launch_kernel(int x)
 {
-  result_type r;
+  result_type* r;
+  cudaMallocManaged(&r, sizeof(result_type));
+
   file4_kernel<<<1, 1>>>(r, x);
-  return r.sum;
+  cudaDeviceSynchronize();
+
+  auto sum = r->sum;
+  cudaFree(r);
+
+  return sum;
 }
diff --git a/Tests/CudaOnly/SeparateCompilation/file5.cu b/Tests/CudaOnly/SeparateCompilation/file5.cu
index 9b2c92a1d6..007152160b 100644
--- a/Tests/CudaOnly/SeparateCompilation/file5.cu
+++ b/Tests/CudaOnly/SeparateCompilation/file5.cu
@@ -4,17 +4,24 @@
 result_type __device__ file1_func(int x);
 result_type_dynamic __device__ file2_func(int x);
 
-static __global__ void file5_kernel(result_type& r, int x)
+static __global__ void file5_kernel(result_type *r, int x)
 {
   // call static_func which is a method that is defined in the
   // static library that is always out of date
-  r = file1_func(x);
+  *r = file1_func(x);
   result_type_dynamic rd = file2_func(x);
 }
 
 EXPORT int file5_launch_kernel(int x)
 {
-  result_type r;
+  result_type* r;
+  cudaMallocManaged(&r, sizeof(result_type));
+
   file5_kernel<<<1, 1>>>(r, x);
-  return r.sum;
+  cudaDeviceSynchronize();
+
+  auto sum = r->sum;
+  cudaFree(r);
+
+  return sum;
 }