Merge pull request #1113 from DARMA-tasking/1111-over-2gib-send-bug

1111 Fix for large message sends
DARMA-tasking · Nov 13, 2020 · 9dd92a0 · 9dd92a0
2 parents f0c4480 + 90c9305
commit 9dd92a0
Show file tree

Hide file tree

Showing 17 changed files with 919 additions and 228 deletions.
diff --git a/src/vt/configs/arguments/app_config.h b/src/vt/configs/arguments/app_config.h
@@ -161,6 +161,7 @@ struct AppConfig {
   bool vt_diag_csv_base_units = false;
 
   bool vt_pause = false;
+  std::size_t vt_max_mpi_send_size = 1ull << 30;
 
   bool vt_debug_all          = false;
   bool vt_debug_verbose      = false;

diff --git a/src/vt/configs/arguments/args.cc b/src/vt/configs/arguments/args.cc
@@ -485,6 +485,18 @@ void ArgConfig::addConfigFileArgs(CLI::App& app) {
   a2->group(configGroup);
 }
 
+void ArgConfig::addRuntimeArgs(CLI::App& app) {
+  auto max_size = "Maximum MPI send size (causes larger messages to be split "
+                  "into multiple MPI sends)";
+
+  auto a1 = app.add_option(
+    "--vt_max_mpi_send_size", config_.vt_max_mpi_send_size, max_size, true
+  );
+
+  auto configRuntime = "Runtime";
+  a1->group(configRuntime);
+}
+
 class VtFormatter : public CLI::Formatter {
 public:
   std::string make_usage(const CLI::App *, std::string name) const override {
@@ -537,6 +549,7 @@ std::tuple<int, std::string> ArgConfig::parse(int& argc, char**& argv) {
   addUserArgs(app);
   addSchedulerArgs(app);
   addConfigFileArgs(app);
+  addRuntimeArgs(app);
 
   std::tuple<int, std::string> result = parseArguments(app, /*out*/ argc, /*out*/ argv);
   if (std::get<0>(result) not_eq -1) {

diff --git a/src/vt/configs/arguments/args.h b/src/vt/configs/arguments/args.h
@@ -96,6 +96,7 @@ struct ArgConfig : runtime::component::Component<ArgConfig> {
   void addUserArgs(CLI::App& app);
   void addSchedulerArgs(CLI::App& app);
   void addConfigFileArgs(CLI::App& app);
+  void addRuntimeArgs(CLI::App& app);
 
   void postParseTransform();
 

diff --git a/src/vt/configs/types/types_type.h b/src/vt/configs/types/types_type.h
@@ -103,7 +103,7 @@ using UniqueIndexBitType      = uint64_t;
 /// Used for hold an identifier for a group
 using GroupType               = uint64_t;
 /// Used for hold the size of a message
-using MsgSizeType             = int32_t;
+using MsgSizeType             = int64_t;
 /// Used for hold a phase for load balancing
 using PhaseType               = uint64_t;
 /// Used for hold a sub-phase for load balancing