fix header setting in cmake

scripts/run.sh

@@ -3,5 +3,5 @@
 ./build/test/unit-test --gtest_filter=GPU.maximum
 ./build/test/unit-test --gtest_filter=GPU.minimum
 ./build/test/unit-test --gtest_filter=GPU.multiple*
-#./build/test/unit-test --gtest_filter=GPU.abs
+##./build/test/unit-test --gtest_filter=GPU.abs
 #./build/test/unit-test --gtest_filter=GPU.dot*

src/ngraph/CMakeLists.txt

@@ -194,14 +194,12 @@ if (NGRAPH_CPU_ENABLE AND LLVM_INCLUDE_DIR AND
             "NGRAPH_TBB_ENABLE"
         )
     endif()
-    set_source_files_properties(codegen/compiler.cpp PROPERTIES COMPILE_DEFINITIONS "${HEADER_SEARCH_DEFINES}")
     set(NGRAPH_CPU_DEBUGINFO_ENABLE 0 CACHE STRING "Enable debuginfo in the CPU backend")
 
     # GPU backend current requires CPU because they share compiler.cpp,
     # and compiler.cpp requires MKLDNN
     if(NGRAPH_GPU_ENABLE)
         include_directories(SYSTEM ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIR})
-        message(STATUS "cuda ---------------  ${CUDA_INCLUDE_DIRS}")
         # Add sources for the GPU backend
         # and all its dependencies
         set(SRC ${SRC}
@@ -215,9 +213,11 @@ if (NGRAPH_CPU_ENABLE AND LLVM_INCLUDE_DIR AND
             runtime/gpu/gpu_tensor_view_wrapper.cpp
             runtime/gpu/gpu_util.cpp
             )
-        set_property(SOURCE codegen/compiler.cpp APPEND_STRING PROPERTY COMPILE_DEFINITIONS
-        "CUDA_HEADER_PATHS=\"${CUDA_INCLUDE_DIRS}\";")
+        set(HEADER_SEARCH_DEFINES ${HEADER_SEARCH_DEFINES}
+            "CUDA_HEADER_PATHS=\"${CUDA_INCLUDE_DIRS}\""
+        )
     endif()
+    set_source_files_properties(codegen/compiler.cpp PROPERTIES COMPILE_DEFINITIONS "${HEADER_SEARCH_DEFINES}")
 
 endif()
 

src/ngraph/codegen/compiler.cpp

@@ -65,7 +65,7 @@
 #error "This source file interfaces with LLVM and Clang and must be compiled with RTTI disabled"
 #endif
 
-//#define USE_BUILTIN
+#define USE_BUILTIN
 
 using namespace clang;
 using namespace llvm;
@@ -360,7 +360,6 @@ void codegen::StaticCompiler::configure_search_path()
     add_header_search_path(CLANG_BUILTIN_HEADERS_PATH);
     add_header_search_path("/usr/include/x86_64-linux-gnu");
     add_header_search_path("/usr/include");
-    add_header_search_path("/usr/local/cuda/include");
 
     // Search for headers in
     //    /usr/include/x86_64-linux-gnu/c++/N.N

src/ngraph/runtime/gpu/gpu_emitter.cpp

@@ -50,26 +50,26 @@ using namespace std;
 using namespace ngraph;
 
 
-#define NVRTC_SAFE_CALL(x) \ 
-do { \ 
-    nvrtcResult result = x; \ 
-        if (result != NVRTC_SUCCESS) { \ 
-            std::cerr << "\nerror: " #x " failed with error " \ 
+#define NVRTC_SAFE_CALL(x) \
+do { \
+    nvrtcResult result = x; \
+        if (result != NVRTC_SUCCESS) { \
+            std::cerr << "\nerror: " #x " failed with error " \
                 << nvrtcGetErrorString(result) << '\n'; \
-                exit(1); \ 
+                exit(1); \
+        } \
+} while(0)
+
+#define CUDA_SAFE_CALL(x) \
+do { \
+    CUresult result = x; \
+        if (result != CUDA_SUCCESS) { \
+            const char *msg; \
+                cuGetErrorName(result, &msg); \
+                std::cerr << "\nerror: " #x " failed with error " \
+                << msg << '\n'; \
+                exit(1); \
         } \
-} while(0) 
-
-#define CUDA_SAFE_CALL(x) \ 
-do { \ 
-    CUresult result = x; \ 
-        if (result != CUDA_SUCCESS) { \ 
-            const char *msg; \ 
-                cuGetErrorName(result, &msg); \ 
-                std::cerr << "\nerror: " #x " failed with error " \ 
-                << msg << '\n'; \ 
-                exit(1); \ 
-        } \ 
 } while(0)
 
 
@@ -151,7 +151,7 @@ void runtime::gpu::GPU_Emitter::EmitDot(codegen::CodeWriter& writer,
         const vector<runtime::gpu::GPU_TensorViewWrapper>& out)
 {
     writer << " // " << n->get_name() << "\n  return;\n";
-    /*
+    
        const Shape& arg0_shape = args[0].get_shape();
        const Shape& arg1_shape = args[1].get_shape();
        if (arg0_shape.empty() || arg1_shape.empty())
@@ -256,7 +256,7 @@ else
 {
     // General ND Call?
 }
-*/
+
 }
 
 void runtime::gpu::GPU_Emitter::EmitDivide(codegen::CodeWriter& writer,