bda/gpu-mpi-tests
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add iteration loop

Browse Source
This commit is contained in:
Bryce Allen
2022-10-24 18:43:12 +00:00
parent 2309afb2ab
commit 37a97f24dd
1 changed files with 22 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
mpi_stencil2d_gt.cc
View File

@@ -225,6 +225,8 @@ int main(int argc, char** argv)
// Note: domain will be n_global x n_global plus ghost points in one dimension // Note: domain will be n_global x n_global plus ghost points in one dimension
int n_global = 8 * 1024; int n_global = 8 * 1024;
bool stage_host = false; bool stage_host = false;
int n_iter = 100;
int n_warmup = 5;
if (argc > 1) { if (argc > 1) {
n_global = std::atoi(argv[1]) * 1024; n_global = std::atoi(argv[1]) * 1024;
@@ -234,6 +236,9 @@ int main(int argc, char** argv)
stage_host = true; stage_host = true;
} }
} }
if (argc > 3) {
n_iter = std::atoi(argv[3]);
}
int n_sten = 5; int n_sten = 5;
int n_bnd = (n_sten - 1) / 2; int n_bnd = (n_sten - 1) / 2;
@@ -280,7 +285,8 @@ int main(int argc, char** argv)
auto fn_dzdy = [](double x, double y) { return 2 * x; }; auto fn_dzdy = [](double x, double y) { return 2 * x; };
struct timespec start, end; struct timespec start, end;
double seconds = 0.0; double iter_time = 0.0;
double total_time = 0.0;
double x_start = world_rank * lx_local; double x_start = world_rank * lx_local;
for (int i = 0; i < n_local; i++) { for (int i = 0; i < n_local; i++) {
@@ -315,16 +321,22 @@ int main(int argc, char** argv)
gt::copy(h_z, d_z); gt::copy(h_z, d_z);
// gt::synchronize(); // gt::synchronize();
for (int i = 0; i < n_warmup + n_iter; i++) {
clock_gettime(CLOCK_MONOTONIC, &start); clock_gettime(CLOCK_MONOTONIC, &start);
boundary_exchange_y(MPI_COMM_WORLD, world_size, world_rank, d_z, n_bnd, stage_host); boundary_exchange_y(MPI_COMM_WORLD, world_size, world_rank, d_z, n_bnd, stage_host);
// gt::synchronize();
clock_gettime(CLOCK_MONOTONIC, &end); clock_gettime(CLOCK_MONOTONIC, &end);
seconds = iter_time =
((end.tv_sec - start.tv_sec) + (end.tv_nsec - start.tv_nsec) * 1.0e-9); ((end.tv_sec - start.tv_sec) + (end.tv_nsec - start.tv_nsec) * 1.0e-9);
printf("%d/%d exchange time %0.4f\n", world_rank, world_size, seconds);
if (i >= n_warmup) {
total_time += iter_time;
}
// do some calculation, to try to more closely simulate what happens in GENE
d_dzdy_numeric = stencil2d_1d_5(d_z, stencil5) * scale; d_dzdy_numeric = stencil2d_1d_5(d_z, stencil5) * scale;
// gt::synchronize(); gt::synchronize();
}
printf("%d/%d exchange time %0.4f\n", world_rank, world_size, total_time / n_iter);
gt::copy(d_dzdy_numeric, h_dzdy_numeric); gt::copy(d_dzdy_numeric, h_dzdy_numeric);
// gt::synchronize(); // gt::synchronize();