One critical bug fixed in distributed GNP, where the reference tree was grabbed multiple times.

2011-01-15 05:42:35 +00:00
parent c568e20283
commit 7fd2447b4e
6 changed files with 63 additions and 21 deletions
@@ -9,7 +9,6 @@
 #ifndef CORE_GNP_DISTRIBUTED_DUALTREE_DFS_DEV_H
 #define CORE_GNP_DISTRIBUTED_DUALTREE_DFS_DEV_H

-#include <boost/bind.hpp>
 #include <boost/mpi.hpp>
 #include "core/gnp/distributed_dualtree_dfs.h"
 #include "core/gnp/dualtree_dfs_dev.h"
@@ -126,8 +125,15 @@ void DistributedDualtreeDfs<DistributedProblemType>::ReduceScatter_(
          for(typename std::map<int, int>::const_iterator it =
                sub_engine.unpruned_reference_nodes().begin();
              it != sub_engine.unpruned_reference_nodes().end(); it++) {
-            receive_requests[i].push_back(
-              std::pair<int, int>(it->first, it->second));
+
+            // This operation might be less than ideal, so change it
+            // later.
+            std::pair<int, int> new_pair(it->first, it->second);
+            if(std::find(
+                  receive_requests[i].begin(), receive_requests[i].end(),
+                  new_pair) == receive_requests[i].end()) {
+              receive_requests[i].push_back(new_pair);
+            }
          }
          new_computation_frontier[i].insert(
            new_computation_frontier[i].end(),
@@ -284,4 +284,5 @@ int main(int argc, char *argv[]) {
    core::optimization::TrustRegionSearchMethod::STEIHAUG);

  printf("All tests passed!");
+  return 0;
 }
@@ -591,6 +591,14 @@ class DistributedTreeBuilder {
        printf(
          "Took %g seconds to read in the distributed tree.\n",
          distributed_table_index_timer.elapsed());
+        printf(
+          "The following is the distribution of points among all MPI "
+          "processes.\n");
+        for(int i = 0; i < world.size(); i++) {
+          printf(
+            "Process %d has %d points.\n", i,
+            distributed_table_->local_n_entries(i));
+        }
      }
    }
 };
@@ -16,8 +16,8 @@
 namespace core {
 namespace table {
 extern core::table::MemoryMappedFile *global_m_file_;
-};
-};
+}
+}

 namespace core {
 namespace parallel {
@@ -38,7 +38,7 @@ class TableExchange {

    std::vector<int *> new_from_old_cache_;

-    std::vector< SubTableListType > received_subtables;
+    std::vector< SubTableListType > received_subtables_;

  public:

@@ -60,18 +60,18 @@ class TableExchange {
    SubTableType &FindSubTable(int process_id, int begin, int count) {

      // Naive search, but probably should use a STL map here...
-      for(unsigned int i = 0; i < received_subtables[process_id].size(); i++) {
+      for(unsigned int i = 0; i < received_subtables_[process_id].size(); i++) {
        if(
-          received_subtables[
+          received_subtables_[
            process_id][i].table()->get_tree()->begin() == begin &&
-          received_subtables[
+          received_subtables_[
            process_id][i].table()->get_tree()->count() == count) {
-          return received_subtables[process_id][i];
+          return received_subtables_[process_id][i];
        }
      }

      // The code should not get to this point.
-      return received_subtables[process_id][0];
+      return received_subtables_[process_id][0];
    }

    void Init(
@@ -150,16 +150,16 @@ class TableExchange {
      }

      // Clear the received subtables and resize.
-      received_subtables.resize(0);
-      received_subtables.resize(world.size());
+      received_subtables_.resize(0);
+      received_subtables_.resize(world.size());
      for(unsigned int j = 0; j < receive_requests.size(); j++) {
        for(unsigned int i = 0; i < receive_requests[j].size(); i++) {
-          received_subtables[j].push_back(
+          received_subtables_[j].push_back(
            j, point_cache_[j], old_from_new_cache_[j], new_from_old_cache_[j],
            max_num_levels_to_serialize);
        }
      }
-      boost::mpi::all_to_all(world, send_subtables, received_subtables);
+      boost::mpi::all_to_all(world, send_subtables, received_subtables_);

      // Clear the receive requests so that it can be used in the next
      // iteration.
@@ -169,7 +169,7 @@ class TableExchange {
      return false;
    }
 };
-};
-};
+}
+}

 #endif
@@ -360,7 +360,7 @@ class SubTable {
      tree_ = table_in->get_tree_offset_ptr();
    }
 };
-};
-};
+}
+}

 #endif
@@ -1,4 +1,7 @@
 /** @file sub_table_list.h
+ *
+ *  An abstract class to maintain a list of subtables to aid in the
+ *  all-to-all exchange.
 *
 *  @author Dongryeol Lee (dongryel@cc.gatech.edu)
 */
@@ -11,6 +14,9 @@

 namespace core {
 namespace table {
+
+/** @brief An abstract class for a list of subtables.
+ */
 template<typename IncomingSubTableType>
 class SubTableList {
  public:
@@ -21,20 +27,36 @@ class SubTableList {
    typedef IncomingSubTableType SubTableType;

  private:
+
+    // For boost serialization.
    friend class boost::serialization::access;

+    /** @brief The list of subtables.
+     */
    std::vector<SubTableType> list_;

  public:

+    /** @brief Resets the subtable list to be an empty list.
+     */
+    void Reset() {
+      list_.resize(0);
+    }
+
+    /** @brief Returns the subtable at a given position.
+     */
    SubTableType &operator[](int pos) {
      return list_[pos];
    }

+    /** @brief The size of the subtable list.
+     */
    unsigned int size() const {
      return list_.size();
    }

+    /** @brief Serializes each element of the subtable list.
+     */
    template<class Archive>
    void serialize(Archive &ar, const unsigned int version) {
      for(unsigned int i = 0; i < list_.size(); i++) {
@@ -42,6 +64,8 @@ class SubTableList {
      }
    }

+    /** @brief Pushes back a subtable for loading.
+     */
    template<typename OldFromNewIndexType>
    void push_back(
      int rank_in, core::table::DenseMatrix &data_alias_in,
@@ -54,6 +78,9 @@ class SubTableList {
        max_num_levels_to_serialize_in);
    }

+    /** @brief Pushes back a starting node and the number of levels to
+     *         serialize under.
+     */
    template<typename TableType, typename TreeType>
    void push_back(
      TableType *table_in, TreeType *start_node_in,
@@ -63,7 +90,7 @@ class SubTableList {
        table_in, start_node_in, max_num_levels_to_serialize_in);
    }
 };
-};
-};
+}
+}

 #endif