Add methods to copy data between host and device

src/CUDAStream.cu

@@ -23,6 +23,23 @@ CUDAStream<T>::CUDAStream(const unsigned int ARRAY_SIZE)
   check_error();
 }
 
+
+template <class T>
+void CUDAStream<T>::write_arrays(const std::vector<T>& a, const std::vector<T>& b, const std::vector<T>& c)
+{
+  // Copy host memory to device
+  cudaMemcpy(d_a, a.data(), a.size()*sizeof(T), cudaMemcpyHostToDevice);
+  check_error();
+  cudaMemcpy(d_b, b.data(), b.size()*sizeof(T), cudaMemcpyHostToDevice);
+  check_error();
+  cudaMemcpy(d_c, c.data(), c.size()*sizeof(T), cudaMemcpyHostToDevice);
+  check_error();
+}
+
+template <class T>
+void CUDAStream<T>::read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c)
+{
+}
 template <typename T>
 __global__ void copy_kernel(const T * a, T * c)
 {
@@ -34,6 +51,9 @@ template <class T>
 void CUDAStream<T>::copy()
 {
   copy_kernel<<<1024, 1024>>>(d_a, d_c);
+  check_error();
+  cudaDeviceSynchronize();
+  check_error();
 }
 
 template <class T>

src/CUDAStream.h

@@ -22,4 +22,8 @@ class CUDAStream : public Stream<T>
     void mul();
     void triad();
 
+    void write_arrays(const std::vector<T>& a, const std::vector<T>& b, const std::vector<T>& c);
+    void read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c);
+
 };
+

src/Stream.h

@@ -14,7 +14,9 @@ class Stream
     virtual void add() = 0;
     virtual void triad() = 0;
 
-
+    // Copy memory between host and device
+    virtual void write_arrays(const std::vector<T>& a, const std::vector<T>& b, const std::vector<T>& c) = 0;
+    virtual void read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c) = 0;
 
     // Implementation specific device functions
     static std::vector<int> getDeviceList();

src/main.cpp

@@ -29,6 +29,8 @@ int main(int argc, char *argv[])
   // Use the CUDA implementation
   stream = new CUDAStream<double>(ARRAY_SIZE);
 
+  stream->write_arrays(a, b, c);
+
   stream->copy();
 
   delete[] stream;