Fix MEM flag for CUDA, resolves #163

src/cuda/CUDAStream.cu

@@ -42,7 +42,13 @@ CUDAStream<T>::CUDAStream(const int ARRAY_SIZE, const int device_index)
   // Print out device information
   std::cout << "Using CUDA device " << getDeviceName(device_index) << std::endl;
   std::cout << "Driver: " << getDeviceDriver(device_index) << std::endl;
-
+#if defined(MANAGED)
+  std::cout << "Memory: MANAGED" << std::endl;
+#elif defined(PAGEFAULT)
+  std::cout << "Memory: PAGEFAULT" << std::endl;
+#else
+  std::cout << "Memory: DEFAULT" << std::endl;
+#endif
   array_size = ARRAY_SIZE;
 
   // Allocate the host array for partial sums for dot kernels

src/cuda/CUDAStream.h

@@ -13,13 +13,7 @@
 
 #include "Stream.h"
 
-#if defined(PAGEFAULT)
-  #define IMPLEMENTATION_STRING "CUDA - Page Fault"
-#elif defined(MANAGED)
-  #define IMPLEMENTATION_STRING "CUDA - Managed Memory"
-#else
 #define IMPLEMENTATION_STRING "CUDA"
-#endif
 
 #define TBSIZE 1024
 #define DOT_NUM_BLOCKS 1024

src/cuda/model.cmake

@@ -29,7 +29,7 @@ macro(setup)
     endif()
 
     enable_language(CUDA)
-    register_definitions(MEM=${MEM})
+    register_definitions(${MEM})
 
     # add -forward-unknown-to-host-compiler for compatibility reasons
     set(CMAKE_CUDA_FLAGS ${CMAKE_CUDA_FLAGS} "-forward-unknown-to-host-compiler -arch=${CUDA_ARCH}" ${CUDA_EXTRA_FLAGS})