Merge pull request #1720 from DARMA-tasking/1265-stats-replay-without…

…-collection

1265 lb data replay without collection

CMakeLists.txt

@@ -56,6 +56,7 @@ set(PROJECT_BIN_DIR     ${CMAKE_CURRENT_BINARY_DIR})
 set(PROJECT_BASE_DIR    ${CMAKE_CURRENT_SOURCE_DIR})
 set(PROJECT_LIB_DIR     ${CMAKE_CURRENT_SOURCE_DIR}/lib)
 set(PROJECT_EXAMPLE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/examples)
+set(PROJECT_TOOLS_DIR   ${CMAKE_CURRENT_SOURCE_DIR}/tools)
 
 # Import the linking macros for VT-related targets
 include(cmake/link_vt.cmake)
@@ -114,6 +115,25 @@ if (VT_BUILD_TESTS
   include(CTest)
 endif()
 
+#
+# Tools
+#
+option(vt_build_tools "Build VT tools" ON)
+
+if (vt_build_tools)
+  message(
+    STATUS
+    "VT: building tools"
+  )
+
+  add_custom_target(tools)
+  add_subdirectory(tools)
+else()
+  message(
+    STATUS "VT: NOT building tools because vt_build_tools is not set."
+  )
+endif()
+
 #
 # Examples
 #

scripts/check_license.sh

@@ -3,7 +3,7 @@
 path_to_vt=${1}
 cd "$path_to_vt" || exit 1
 
-for sub_dir in "src" "tests/unit" "tests/perf" "tutorial" "examples"
+for sub_dir in "src" "tests/unit" "tests/perf" "tutorial" "examples" "tools"
 do
   "$path_to_vt/scripts/add-license-perl.pl" "$path_to_vt/$sub_dir" "$path_to_vt/scripts/license-template"
 done

src/vt/configs/arguments/app_config.h

@@ -225,6 +225,7 @@ struct AppConfig {
   bool vt_debug_phase        = false;
   bool vt_debug_context      = false;
   bool vt_debug_epoch        = false;
+  bool vt_debug_replay       = false;
 
   bool vt_debug_print_flush = false;
 
@@ -386,6 +387,7 @@ struct AppConfig {
       | vt_debug_phase
       | vt_debug_context
       | vt_debug_epoch
+      | vt_debug_replay
 
       | vt_debug_print_flush
 

src/vt/configs/arguments/args.cc

@@ -374,6 +374,7 @@ void addDebugPrintArgs(CLI::App& app, AppConfig& appConfig) {
   auto dcp = "Enable debug_phase        = \"" debug_pp(phase)        "\"";
   auto ddp = "Enable debug_context      = \"" debug_pp(context)      "\"";
   auto dep = "Enable debug_epoch        = \"" debug_pp(epoch)        "\"";
+  auto dfp = "Enable debug_replay       = \"" debug_pp(replay)       "\"";
 
   auto r1 = app.add_option("--vt_debug_level",      appConfig.vt_debug_level,        rq);
 
@@ -410,6 +411,7 @@ void addDebugPrintArgs(CLI::App& app, AppConfig& appConfig) {
   auto dc = app.add_flag("--vt_debug_phase",        appConfig.vt_debug_phase,        dcp);
   auto dd = app.add_flag("--vt_debug_context",      appConfig.vt_debug_context,      ddp);
   auto de = app.add_flag("--vt_debug_epoch",        appConfig.vt_debug_epoch,        dep);
+  auto df = app.add_flag("--vt_debug_replay",       appConfig.vt_debug_replay,       dfp);
 
   auto debugGroup = "Debug Print Configuration (must be compile-time enabled)";
   r->group(debugGroup);
@@ -446,6 +448,7 @@ void addDebugPrintArgs(CLI::App& app, AppConfig& appConfig) {
   dc->group(debugGroup);
   dd->group(debugGroup);
   de->group(debugGroup);
+  df->group(debugGroup);
 
   auto dbq = "Always flush VT runtime prints";
   auto eb  = app.add_flag("--vt_debug_print_flush", appConfig.vt_debug_print_flush, dbq);

src/vt/configs/debug/debug_config.h

@@ -80,7 +80,8 @@ enum CatEnum : uint64_t {
   phase        = 1ull<<28,
   context      = 1ull<<29,
   epoch        = 1ull<<30,
-  temperedwmin = 1ull<<31
+  temperedwmin = 1ull<<31,
+  replay       = 1ull<<32
 };
 
 enum CtxEnum : uint64_t {
@@ -138,6 +139,7 @@ vt_option_category_pretty_print(reduce,       "reduce")
 vt_option_category_pretty_print(rdma,         "RDMA")
 vt_option_category_pretty_print(rdma_channel, "RDMA Channel")
 vt_option_category_pretty_print(rdma_state,   "RDMA State")
+vt_option_category_pretty_print(replay,       "replay")
 vt_option_category_pretty_print(runtime,      "runtime")
 vt_option_category_pretty_print(scatter,      "scatter")
 vt_option_category_pretty_print(serial_msg,   "serialized-msg")

src/vt/phase/phase_manager.cc

@@ -301,7 +301,7 @@ void PhaseManager::printSummary(vrt::collection::lb::PhaseInfo* last_phase_info)
       phase,
       "phase={}, duration={}, rank_max_compute_time={}, rank_avg_compute_time={}, imbalance={:.3f}, "
       "grain_max_time={}, migration count={}, lb_name={}\n",
-      cur_phase_,
+      last_phase_info->phase,
       total_time,
       TimeType(last_phase_info->max_load),
       TimeType(last_phase_info->avg_load),
@@ -313,7 +313,7 @@ void PhaseManager::printSummary(vrt::collection::lb::PhaseInfo* last_phase_info)
     // vt_print(
     //   phase,
     //   "POST phase={}, total time={}, max_load={}, avg_load={}, imbalance={:.3f}, migration count={}\n",
-    //   cur_phase_,
+    //   last_phase_info->phase,
     //   total_time,
     //   TimeType(last_phase_info->max_load_post_lb),
     //   TimeType(last_phase_info->avg_load_post_lb),
@@ -336,7 +336,7 @@ void PhaseManager::printSummary(vrt::collection::lb::PhaseInfo* last_phase_info)
       auto percent_improvement = compute_percent_improvement(
         last_phase_info->max_load, last_phase_info->avg_load
       );
-      if (percent_improvement > 3.0 and cur_phase_ > 0) {
+      if (percent_improvement > 3.0 and last_phase_info->phase > 0) {
         if (grain_percent_improvement < 0.5) {
           // grain size is blocking improvement
           vt_print(
@@ -395,7 +395,7 @@ void PhaseManager::printSummary(vrt::collection::lb::PhaseInfo* last_phase_info)
           }
         }
       }
-    } else if (cur_phase_ == 0) {
+    } else if (last_phase_info->phase == 0) {
        // ran the lb on a phase that may have included initialization costs
        vt_print(
          phase,

src/vt/runtime/runtime_banner.cc

@@ -904,6 +904,7 @@ void Runtime::printStartupBanner() {
   vt_runtime_debug_warn_compile(phase)
   vt_runtime_debug_warn_compile(context)
   vt_runtime_debug_warn_compile(epoch)
+  vt_runtime_debug_warn_compile(replay)
 
   auto arg_str = [](std::vector<char*> const& args) -> std::string {
     std::stringstream ss;

src/vt/vrt/collection/balance/lb_invoke/lb_manager.h

@@ -261,6 +261,10 @@ struct LBManager : runtime::component::Component<LBManager> {
 
   void statsHandler(std::vector<balance::LoadData> const& in_stat_vec);
 
+  lb::PhaseInfo *getPhaseInfo() { return last_phase_info_.get(); }
+
+  void setComputingBeforeLBStats(bool before_lb) { before_lb_stats_ = before_lb; }
+
 private:
   bool isCollectiveComm(elm::CommCategory cat) const;