gallo/atrip
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix bench building for cublas-parallel-atrip

Browse Source
This commit is contained in:
Alejandro Gallo 2022-09-30 14:34:54 +02:00
parent 399447131c
commit 1ca31f4929
2 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
bench/Makefile.am
View File

@ -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

4
bench/test-cublas-parallel-atrip.cxx
View File

@ -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]);