diff --git a/src/CUDAStream.cu b/src/CUDAStream.cu index 163a3f1..9b1b36d 100644 --- a/src/CUDAStream.cu +++ b/src/CUDAStream.cu @@ -11,7 +11,7 @@ __global__ void copy_kernel(const T * a, T * c) template void CUDAStream::copy() { - copy<<<1024, 1024>>>(a, c); + copy_kernel<<<1024, 1024>>>(d_a, d_c); } template diff --git a/src/CUDAStream.h b/src/CUDAStream.h index ab9ac38..9348b29 100644 --- a/src/CUDAStream.h +++ b/src/CUDAStream.h @@ -5,16 +5,17 @@ template class CUDAStream : public Stream { - public: - void copy(); - void add(); - void mul(); - void triad(); - private: // Device side pointers to arrays T *d_a; T *d_b; T *d_c; + + public: + void copy(); + void add(); + void mul(); + void triad(); + }; diff --git a/src/main.cpp b/src/main.cpp index 9e22b89..76c3da4 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -16,6 +16,7 @@ int main(int argc, char *argv[]) Stream *stream; stream = new CUDAStream(); + stream->copy(); delete[] stream;