Add cuda compile define and cmake setting

CMakeLists.txt

@@ -4,9 +4,14 @@ project(Aurora)
 set(CMAKE_CXX_STANDARD 14)
 set(CMAKE_INCLUDE_CURRENT_DIR ON)
 
+set(Aurora_USE_CUDA ON)
+
+if (Aurora_USE_CUDA)
 set(CMAKE_CUDA_COMPILER /usr/local/cuda/bin/nvcc)
 enable_language(CUDA)
 find_package(CUDAToolkit REQUIRED)
+add_definitions(-DUSE_CUDA)
+endif(Aurora_USE_CUDA)
 
 find_package (OpenMP REQUIRED)
 set(CMAKE_C_FLAGS  "${CMAKE_C_FLAGS}  -march=native")
@@ -30,13 +35,14 @@ target_include_directories(Aurora PUBLIC $<TARGET_PROPERTY:MKL::MKL,INTERFACE_IN
 target_link_libraries(Aurora PUBLIC $<LINK_ONLY:MKL::MKL>)
 target_link_libraries(Aurora PUBLIC OpenMP::OpenMP_CXX)
 target_link_libraries(Aurora PUBLIC matio)
-
+if (Aurora_USE_CUDA)
 target_include_directories(Aurora PRIVATE ./src /usr/local/cuda/include) 
 set_target_properties(Aurora PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
 target_compile_options(Aurora PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:
     -arch=sm_75
 >)
 target_link_libraries(Aurora PRIVATE ${CUDA_RUNTIME_LIBRARY} CUDA::cufft CUDA::cudart)
+endif(Aurora_USE_CUDA)
 
 find_package(GTest  REQUIRED)
 INCLUDE_DIRECTORIES(${GTEST_INCLUDE_DIRS})
@@ -55,11 +61,13 @@ target_link_libraries(Aurora_Test PUBLIC OpenMP::OpenMP_CXX)
 target_link_libraries(Aurora_Test PUBLIC matio)
 target_link_libraries(Aurora_Test PUBLIC ${GTEST_BOTH_LIBRARIES}  )
 
+if (Aurora_USE_CUDA)
 target_include_directories(Aurora_Test PRIVATE ./src /usr/local/cuda/include) 
 set_target_properties(Aurora_Test PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
 target_compile_options(Aurora_Test PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:
     -arch=sm_75
 >)
 target_link_libraries(Aurora_Test PRIVATE ${CUDA_RUNTIME_LIBRARY} CUDA::cufft CUDA::cudart)
+endif(Aurora_USE_CUDA)
 gtest_discover_tests(Aurora_Test )
 #target_link_libraries(CreateMatchedFilter PRIVATE TBB::tbb)

src/CudaMatrix.cpp

@@ -1,3 +1,4 @@
+#ifdef USE_CUDA
 #include "CudaMatrix.h"
 
 #include "Function.h"
@@ -218,6 +219,7 @@ CudaMatrix CudaMatrix::block(int aDim,int aBeginIndex, int aEndIndex) const
             return CudaMatrix::fromRawData(dataOutput,getDimSize(0),dimLength,getDimSize(2),getValueType());
         }
         case 2:
+        default:
         {
             int copySize = dimLength*sliceStride;
             cudaMemcpy(dataOutput, 
@@ -235,4 +237,6 @@ bool CudaMatrix::setBlockValue(int aDim,int aBeginIndx, int aEndIndex,float valu
         std::cerr<<"CudaMatrix block only support 1D-3D data!"<<std::endl;
         return false;
     }
-}
+    return true;
+}
+#endif // USE_CUDA

src/CudaMatrix.h

@@ -1,6 +1,6 @@
 #ifndef CUDAMATRIX_H
 #define CUDAMATRIX_H
-
+#ifdef USE_CUDA
 #include "Matrix.h"
 
 
@@ -235,5 +235,5 @@ namespace Aurora
         std::vector<int> mInfo;
     };
 }
-
+#endif // USE_CUDA
 #endif

src/Function.cpp

@@ -13,7 +13,10 @@
 #include <Eigen/Core>
 #include <Eigen/Eigen>
 #include <Eigen/Dense>
+
+#ifdef USE_CUDA
 #include <cuda_runtime.h>
+#endif
 
 namespace Aurora {
 
@@ -26,9 +29,10 @@ namespace Aurora {
     void free(void* ptr){
         mkl_free(ptr);
     }
-
+#ifdef USE_CUDA
     void gpuFree(void* ptr)
     {
         cudaFree(ptr);
     }
+#endif
 }

src/Matrix.cpp

@@ -17,7 +17,9 @@
 #include "Eigen/src/Core/Matrix.h"
 #include "Function.h"
 
+#ifdef USE_CUDA
 #include <cuda_runtime.h>
+#endif
 
 namespace Aurora{
     typedef void(*CalcFuncD)(const MKL_INT n, const float a[], const MKL_INT inca, const float b[],
@@ -394,6 +396,7 @@ namespace Aurora {
         return ret;
     }
 
+#ifdef USE_CUDA
     CudaMatrix Matrix::toDeviceMatrix() const
     {
         float* deviceData = nullptr;
@@ -401,7 +404,7 @@ namespace Aurora {
         cudaMemcpy(deviceData, mData.get(), sizeof(float) * getDataSize() * getValueType(), cudaMemcpyHostToDevice);
         return CudaMatrix::fromRawData(deviceData, mInfo[0], mInfo[1], mInfo[2], getValueType());
     }
-
+#endif
     Matrix Matrix::New(float *data, const Matrix &shapeMatrix) {
         return New(data,
                    shapeMatrix.getDimSize(0),