Merge pull request #2137 from DARMA-tasking/2130-spec-for-skipped-pha…

…ses-2 #2130: Add spec for skipped and identical phases
DARMA-tasking · May 16, 2023 · 522945b · 522945b
2 parents fd7501a + 526a6ed
commit 522945b
Show file tree

Hide file tree

Showing 5 changed files with 346 additions and 1 deletion.
diff --git a/scripts/JSON_data_files_validator.py b/scripts/JSON_data_files_validator.py
@@ -53,6 +53,17 @@ def _get_valid_schema(self) -> Schema:
                         'rank': int,
                         'num_nodes': int,
                     },
+                    Optional('phases'): {
+                        'count': int,
+                        'skipped': {
+                            'list': [int],
+                            'range': [[int]],
+                        },
+                        'identical_to_previous': {
+                            'list': [int],
+                            'range': [[int]],
+                        },
+                    },
                 },
                 'phases': [
                     {

diff --git a/src/vt/vrt/collection/balance/lb_data_holder.cc b/src/vt/vrt/collection/balance/lb_data_holder.cc
@@ -68,6 +68,63 @@ void LBDataHolder::outputEntity(nlohmann::json& j, ElementIDStruct const& id) co
   }
 }
 
+std::unique_ptr<nlohmann::json> LBDataHolder::metadataToJson() const {
+  // Find last element of a range for which it's values are incremented by 1
+  auto find_last_of_range = [](
+                              const std::set<PhaseType>& notEmptySet,
+                              std::set<PhaseType>::iterator fromIt) {
+    vtAssert(!notEmptySet.empty(), "Input Set must be not empty");
+
+    do {
+      auto next = std::next(fromIt);
+      // end of a set or range
+      if (next == notEmptySet.end() || *fromIt + 1 != *next) {
+        break;
+      }
+      ++fromIt;
+    } while (fromIt != notEmptySet.end());
+
+    return fromIt;
+  };
+
+  nlohmann::json j;
+  j["count"] = count_;
+
+  // Generate list and ranges of skipped phases
+  std::set<PhaseType> skipped_list;
+  std::vector<std::pair<PhaseType, PhaseType>> skipped_ranges;
+  for (auto it = skipped_phases_.begin(); it != skipped_phases_.end(); it++) {
+    auto endOfRange = find_last_of_range(skipped_phases_, it);
+    if (it == endOfRange) {
+      skipped_list.insert(*it);
+    } else {
+      skipped_ranges.emplace_back(*it, *endOfRange);
+      it = endOfRange;
+    }
+  }
+
+  // Generate list and ranges of identical phases
+  std::set<PhaseType> identical_list;
+  std::vector<std::pair<PhaseType, PhaseType>> identical_ranges;
+  for (auto it = identical_phases_.begin(); it != identical_phases_.end();
+       it++) {
+    auto endOfRange = find_last_of_range(identical_phases_, it);
+    if (it == endOfRange) {
+      identical_list.insert(*it);
+    } else {
+      identical_ranges.emplace_back(*it, *endOfRange);
+      it = endOfRange;
+    }
+  }
+
+  // Save metadata
+  j["skipped"]["list"] = skipped_list;
+  j["skipped"]["range"] = skipped_ranges;
+  j["identical_to_previous"]["list"] = identical_list;
+  j["identical_to_previous"]["range"] = identical_ranges;
+  return std::make_unique<nlohmann::json>(std::move(j));
+}
+
 std::unique_ptr<nlohmann::json> LBDataHolder::toJson(PhaseType phase) const {
   using json = nlohmann::json;
 
@@ -164,9 +221,14 @@ std::unique_ptr<nlohmann::json> LBDataHolder::toJson(PhaseType phase) const {
   return std::make_unique<json>(std::move(j));
 }
 
-LBDataHolder::LBDataHolder(nlohmann::json const& j) {
+LBDataHolder::LBDataHolder(nlohmann::json const& j)
+  : count_(0)
+{
   auto this_node = theContext()->getNode();
 
+  // read metadata for skipped and identical phases
+  readMetadata(j);
+
   auto phases = j["phases"];
   if (phases.is_array()) {
     for (auto const& phase : phases) {
@@ -303,15 +365,64 @@ LBDataHolder::LBDataHolder(nlohmann::json const& j) {
     }
   }
 
+  if (!count_) {
+    count_ = node_data_.size();
+  }
+
   // @todo: implement subphase communication de-serialization, no use for it
   // right now, so it will be ignored
 }
 
+void LBDataHolder::readMetadata(nlohmann::json const& j) {
+  auto metadata = j["metadata"];
+  if (metadata.find("phases") != metadata.end()) {
+    auto phases = metadata["phases"];
+    // load count
+    vtAssertExpr(phases["count"].is_number());
+    count_ = phases["count"];
+    // load all skipped phases
+    auto sl = phases["skipped"]["list"];
+    if(sl.is_array()) {
+      for(PhaseType skipped : sl) {
+        skipped_phases_.insert(skipped);
+      }
+    }
+    auto sr = phases["skipped"]["range"];
+    if(sr.is_array()) {
+      for(auto const& pair : sr) {
+        vtAssertExpr(pair.is_array());
+        for(PhaseType i = pair[0]; i <= pair[1]; i++){
+          skipped_phases_.insert(i);
+        }
+      }
+    }
+    // load all identical phases
+    auto il = phases["identical_to_previous"]["list"];
+    if(il.is_array()) {
+      for(PhaseType identical : il) {
+        identical_phases_.insert(identical);
+      }
+    }
+    auto ir = phases["identical_to_previous"]["range"];
+    if(ir.is_array()) {
+      for(auto const& pair : ir) {
+        vtAssertExpr(pair.is_array());
+        for(PhaseType i = pair[0]; i <= pair[1]; i++){
+          identical_phases_.insert(i);
+        }
+      }
+    }
+  }
+}
+
 void LBDataHolder::clear() {
   node_comm_.clear();
   node_data_.clear();
   node_subphase_comm_.clear();
   node_idx_.clear();
+  count_ = 0;
+  skipped_phases_.clear();
+  identical_phases_.clear();
 }
 
 }}}} /* end namespace vt::vrt::collection::balance */
diff --git a/src/vt/vrt/collection/balance/lb_data_holder.h b/src/vt/vrt/collection/balance/lb_data_holder.h
@@ -78,6 +78,9 @@ struct LBDataHolder {
     s | node_subphase_comm_;
     s | user_defined_json_;
     s | node_idx_;
+    s | count_;
+    s | skipped_phases_;
+    s | identical_phases_;
   }
 
   /**
@@ -89,6 +92,13 @@ struct LBDataHolder {
    */
   std::unique_ptr<nlohmann::json> toJson(PhaseType phase) const;
 
+  /**
+   * \brief Output a LB phase's metdadata to JSON
+   *
+   * \return the json data structure
+   */
+  std::unique_ptr<nlohmann::json> metadataToJson() const;
+
   /**
    * \brief Clear all LB data
    */
@@ -103,6 +113,13 @@ struct LBDataHolder {
    */
   void outputEntity(nlohmann::json& j, ElementIDStruct const& elm_id) const;
 
+  /**
+   * \brief Read the LB phase's metadata
+   *
+   * \param[in] j the json
+   */
+  void readMetadata(nlohmann::json const& j);
+
 public:
   /// Node timings for each local object
   std::unordered_map<PhaseType, LoadMapType> node_data_;
@@ -118,6 +135,12 @@ struct LBDataHolder {
   std::unordered_map<ElementIDStruct, std::tuple<VirtualProxyType, std::vector<uint64_t>>> node_idx_;
   /// Map from id to objgroup proxy
   std::unordered_map<ElementIDStruct, ObjGroupProxyType> node_objgroup_;
+  // Number of all phases including skipped and identical
+  PhaseType count_;
+  // Set of phases that are skipped
+  std::set<PhaseType> skipped_phases_;
+  // Set of phases which are identical to previous
+  std::set<PhaseType> identical_phases_;
 };
 
 }}}} /* end namespace vt::vrt::collection::balance */

diff --git a/src/vt/vrt/collection/balance/node_lb_data.cc b/src/vt/vrt/collection/balance/node_lb_data.cc
@@ -205,6 +205,10 @@ void NodeLBData::createLBDataFile() {
     }
     metadata["type"] = "LBDatafile";
     metadata["rank"] = theContext()->getNode();
+    auto phasesMetadata = lb_data_->metadataToJson();
+    if(phasesMetadata) {
+       metadata["phases"] = *phasesMetadata;
+    }
     lb_data_writer_ = std::make_unique<JSONAppender>(
       "phases", metadata, file_name, compress
     );