diff --git a/bench/Makefile.am b/bench/Makefile.am
index 7588f08..90b2338 100644
--- a/bench/Makefile.am
+++ b/bench/Makefile.am
@@ -29,4 +29,7 @@ test_cublas_parallel_atrip_SOURCES = test-cublas-parallel-atrip.cxx
 
 bin_PROGRAMS += test-cuda-sanity
 test_cuda_sanity_SOURCES = test-cuda-sanity.cxx
+
+bin_PROGRAMS += test-cublas-parallel
+test_cublas_parallel_SOURCES = test-cublas-parallel.cxx
 endif
diff --git a/bench/test-cublas-parallel-atrip.cxx b/bench/test-cublas-parallel-atrip.cxx
index 93b28f1..d6629df 100644
--- a/bench/test-cublas-parallel-atrip.cxx
+++ b/bench/test-cublas-parallel-atrip.cxx
@@ -158,8 +158,8 @@ int main(int  argc, char** argv) {
       MPI_Barrier(MPI_COMM_WORLD);
       timings["memcpy"].start();
       for (size_t _s = 0; _s < slices_d.size(); _s++) {
-        // for (size_t _b = 0; _b < slices_h[_s].size(); _b++) {
-        for (size_t _b = 0; _b < 1 ; _b++) {
+        for (size_t _b = 0; _b < slices_h[_s].size(); _b++) {
+        // for (size_t _b = 0; _b < 1 ; _b++) {
           auto device = (*slices_d[_s])[_b];
           auto host   = slices_h[_s][_b].data();
           cuMemcpyHtoD(device, host, sizes[_s]);