dont use trimean in exchange-strong

cwpearson · Aug 17, 2020 · d151a93 · d151a93
1 parent 79bdfcb
commit d151a93
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 17 deletions.
diff --git a/bin/CMakeLists.txt b/bin/CMakeLists.txt
@@ -75,7 +75,7 @@ add_executable(exchange-weak exchange_weak.cu)
 target_link_libraries(exchange-weak stencil::stencil)
 add_args(exchange-weak)
 
-add_executable(exchange-strong exchange_strong.cu statistics.cpp)
+add_executable(exchange-strong exchange_strong.cu)
 target_link_libraries(exchange-strong stencil::stencil)
 add_args(exchange-strong)
 

diff --git a/bin/exchange_strong.cu b/bin/exchange_strong.cu
@@ -5,8 +5,6 @@
 
 #include <nvToolsExt.h>
 
-#include "statistics.hpp"
-
 #include "argparse/argparse.hpp"
 #include "stencil/stencil.hpp"
 
@@ -79,7 +77,6 @@ int main(int argc, char **argv) {
     exit(EXIT_FAILURE);
   }
 
-
   MethodFlags methods = MethodFlags::None;
   if (useStaged) {
     methods = MethodFlags::CudaMpi;
@@ -141,19 +138,15 @@ int main(int argc, char **argv) {
 
     MPI_Barrier(MPI_COMM_WORLD);
 
-    Statistics stats;
-
     double elapsed = MPI_Wtime();
-
     for (int iter = 0; iter < nIters; ++iter) {
       if (0 == rank) {
         std::cerr << "exchange " << iter << "\n";
       }
       dd.exchange();
+      elapsed = MPI_Wtime() - elapsed;
     }
-    elapsed = MPI_Wtime() - elapsed;
     MPI_Allreduce(MPI_IN_PLACE, &elapsed, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
-    stats.insert(elapsed);
 
 #ifdef STENCIL_SETUP_STATS
     if (0 == rank) {
@@ -186,15 +179,16 @@ int main(int argc, char **argv) {
       // clang-format off
       // same as strong.cu
       // header should be
-      // bin,config,naive,x,y,z,s,MPI (B),Colocated (B),cudaMemcpyPeer (B),direct (B)iters,gpus,nodes,ranks,mpi_topo,node_gpus,trimean (s)
+      // bin,config,naive,x,y,z,s,MPI (B),Colocated (B),cudaMemcpyPeer (B),direct (B)iters,gpus,nodes,ranks,mpi_topo,node_gpus,exchange (s)
       // clang-format on
-      printf("exchange,%s,%d,%lu,%lu,%lu,%lu," // s
-             "%lu,%lu,%lu,%lu,"             // <- exchange bytes
-             "%d,%d,%d,%d,%e\n",
-             methodStr.c_str(), useNaivePlacement, x, y, z, x * y * z, dd.exchange_bytes_for_method(MethodFlags::CudaMpi),
-             dd.exchange_bytes_for_method(MethodFlags::CudaMpiColocated),
-             dd.exchange_bytes_for_method(MethodFlags::CudaMemcpyPeer),
-             dd.exchange_bytes_for_method(MethodFlags::CudaKernel), nIters, numSubdoms, numNodes, size, stats.trimean());
+      printf(
+          "exchange,%s,%d,%lu,%lu,%lu,%lu," // s
+          "%lu,%lu,%lu,%lu,"                // <- exchange bytes
+          "%d,%d,%d,%d,%e\n",
+          methodStr.c_str(), useNaivePlacement, x, y, z, x * y * z, dd.exchange_bytes_for_method(MethodFlags::CudaMpi),
+          dd.exchange_bytes_for_method(MethodFlags::CudaMpiColocated),
+          dd.exchange_bytes_for_method(MethodFlags::CudaMemcpyPeer),
+          dd.exchange_bytes_for_method(MethodFlags::CudaKernel), nIters, numSubdoms, numNodes, size, elapsed);
     }
 #endif // STENCIL_EXCHANGE_STATS