#include #include #include "kernel.h" #include "hostOnly.h" int main( void ) { int c; int *dev_c; cudaMalloc( (void**)&dev_c, sizeof(int) ); add<<<1,1>>>( 20, 14, dev_c ); cudaMemcpy( &c, dev_c, sizeof(int), cudaMemcpyDeviceToHost ); printf( "20 + 14 = %d\n", c ); cudaFree( dev_c ); hostOnly(); return 0; }