try peer

CUDA/cudapic.cu

@@ -27,7 +27,7 @@ int main(void) {
    init<<<blocks,threads>>>(darr,nloop);
    cudaDeviceSynchronize();
    cudaMemcpy(arr,darr,npts*sizeof(double),cudaMemcpyDeviceToHost);
-   float pi = 0;
+   double pi = 0;
    for (int i = 0; i < npts; ++i)
       pi += arr[i];
    auto tend = std::chrono::high_resolution_clock::now();