// ------------------------------------------------------------------
// definition of a macro instruction that checks if a CUDA function
// was successull or not. If the execution of the function resulted
// in some error we display it and stop the program
// ------------------------------------------------------------------
#define cume_check(value) {     \
  cudaError_t err = value; \
  if (err != cudaSuccess) {       \
    cerr << endl; \
    cerr << "============================================\n"; \
    cerr << "Error: " << cudaGetErrorString(err) << " at line "; \
    cerr << __LINE__ << " in file " <<  __FILE__;   \
    cerr <<  endl; \
    exit(EXIT_FAILURE); \
  } \
}

// ------------------------------------------------------------------
// Same as cuda_check but for kernel. This macro instruction is used
// after the execution of the kernel (see the macros KERNEL_EXECUTE_NR
// and KERNEL_EXECUTE_WR in cume_kernel.h)
// ------------------------------------------------------------------
#define cume_check_kernel() { \
  cudaError_t err = cudaGetLastError(); \
  if (err != cudaSuccess)  { \
    cerr << endl; \
    cerr << "============================================\n"; \
    cerr << "Kernel Error: " << cudaGetErrorString(err) << " at line "; \
    cerr << __LINE__ << " in file " <<  __FILE__;   \
    cerr <<  endl; \
    exit(EXIT_FAILURE); \
  } \
}


// check that cudaMalloc was executed successfully
cume_check( cudaMalloc((void**) &x_gpu, SIZE * sizeof(float)) );

// check that kernel was executed successfully
sum<<< grid, block >>>(x_gpu, y_gpu, z_gpu, SIZE);
cume_check_kernel();