Commit 586dae19 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Benchmarks: used cudaDeviceSynchronize() instead of cudaThreadsSynchronize()

parent dfc6c6d2
Loading
Loading
Loading
Loading
+5 −2
Original line number Diff line number Diff line
@@ -84,11 +84,14 @@ benchmarkCuda( const int & loops,

        reset();

        // Compute again on CUDA, with explicit synchronization of threads
        // Compute again on CUDA, with explicit synchronization
#ifdef HAVE_CUDA
        cudaDeviceSynchronize();
        timerCudaSync.start();
        computeCuda();
        cudaThreadSynchronize();
        cudaDeviceSynchronize();
        timerCudaSync.stop();
#endif

        reset();
    }