Merge pull request #1741 from H27CK/shrink-buffers

Shrink MPI buffers for spike communication

nestkernel/event_delivery_manager.cpp

@@ -65,6 +65,7 @@ EventDeliveryManager::EventDeliveryManager()
   , recv_buffer_target_data_()
   , buffer_size_target_data_has_changed_( false )
   , buffer_size_spike_data_has_changed_( false )
+  , decrease_buffer_size_spike_data_( true )
   , gather_completed_checker_()
 {
 }
@@ -88,6 +89,7 @@ EventDeliveryManager::initialize()
   off_grid_spiking_ = false;
   buffer_size_target_data_has_changed_ = false;
   buffer_size_spike_data_has_changed_ = false;
+  decrease_buffer_size_spike_data_ = true;
 
 #pragma omp parallel
   {
@@ -143,10 +145,13 @@ EventDeliveryManager::resize_send_recv_buffers_target_data()
 void
 EventDeliveryManager::resize_send_recv_buffers_spike_data_()
 {
-  send_buffer_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
-  recv_buffer_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
-  send_buffer_off_grid_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
-  recv_buffer_off_grid_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
+  if ( kernel().mpi_manager.get_buffer_size_spike_data() > send_buffer_spike_data_.size() )
+  {
+    send_buffer_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
+    recv_buffer_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
+    send_buffer_off_grid_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
+    recv_buffer_off_grid_spike_data_.resize( kernel().mpi_manager.get_buffer_size_spike_data() );
+  }
 }
 
 void
@@ -312,6 +317,9 @@ EventDeliveryManager::gather_spike_data_( const thread tid,
 
   const AssignedRanks assigned_ranks = kernel().vp_manager.get_assigned_ranks( tid );
 
+  // Assume a single gather round
+  decrease_buffer_size_spike_data_ = true;
+
   while ( gather_completed_checker_.any_false() )
   {
     // Assume this is the last gather round and change to false
@@ -384,12 +392,21 @@ EventDeliveryManager::gather_spike_data_( const thread tid,
 #pragma omp single
       {
         buffer_size_spike_data_has_changed_ = kernel().mpi_manager.increase_buffer_size_spike_data();
+        decrease_buffer_size_spike_data_ = false;
       }
     }
 #pragma omp barrier
 
   } // of while
 
+#pragma omp single
+  {
+    if ( decrease_buffer_size_spike_data_ and kernel().mpi_manager.adaptive_spike_buffers() )
+    {
+      kernel().mpi_manager.decrease_buffer_size_spike_data();
+    }
+  } // of omp single; implicit barrier
+
   reset_spike_register_( tid );
 }
 

nestkernel/event_delivery_manager.h

@@ -431,6 +431,8 @@ class EventDeliveryManager : public ManagerInterface
   bool buffer_size_target_data_has_changed_;
   //!< whether size of MPI buffer for communication of spikes was changed
   bool buffer_size_spike_data_has_changed_;
+  //!< whether size of MPI buffer for communication of spikes can be decreased
+  bool decrease_buffer_size_spike_data_;
 
   PerThreadBoolIndicator gather_completed_checker_;
 };

nestkernel/mpi_manager.cpp

@@ -68,6 +68,7 @@ nest::MPIManager::MPIManager()
   , adaptive_spike_buffers_( true )
   , growth_factor_buffer_spike_data_( 1.5 )
   , growth_factor_buffer_target_data_( 1.5 )
+  , shrink_factor_buffer_spike_data_( 1.1 )
   , send_recv_count_spike_data_per_rank_( 0 )
   , send_recv_count_target_data_per_rank_( 0 )
 #ifdef HAVE_MPI
@@ -218,6 +219,8 @@ nest::MPIManager::set_status( const DictionaryDatum& dict )
 
   updateValue< long >( dict, names::max_buffer_size_target_data, max_buffer_size_target_data_ );
   updateValue< long >( dict, names::max_buffer_size_spike_data, max_buffer_size_spike_data_ );
+
+  updateValue< double >( dict, names::shrink_factor_buffer_spike_data, shrink_factor_buffer_spike_data_ );
 }
 
 void

nestkernel/mpi_manager.h

@@ -258,6 +258,12 @@ class MPIManager : public ManagerInterface
    */
   bool increase_buffer_size_spike_data();
 
+  /**
+   * Decreases the size of the MPI buffer for communication of spikes if it
+   * can be decreased.
+   */
+  void decrease_buffer_size_spike_data();
+
   /**
    * Returns whether MPI buffers for communication of connections are adaptive.
    */
@@ -338,6 +344,8 @@ class MPIManager : public ManagerInterface
   double growth_factor_buffer_spike_data_;
   double growth_factor_buffer_target_data_;
 
+  double shrink_factor_buffer_spike_data_;
+
   unsigned int send_recv_count_spike_data_per_rank_;
   unsigned int send_recv_count_target_data_per_rank_;
 
@@ -646,6 +654,17 @@ MPIManager::increase_buffer_size_spike_data()
   }
 }
 
+inline void
+MPIManager::decrease_buffer_size_spike_data()
+{
+  assert( adaptive_spike_buffers_ );
+  // the minimum is set to 4.0 * get_num_processes() to differentiate the initial size
+  if ( buffer_size_spike_data_ / shrink_factor_buffer_spike_data_ > 4.0 * get_num_processes() )
+  {
+    set_buffer_size_spike_data( floor( buffer_size_spike_data_ / shrink_factor_buffer_spike_data_ ) );
+  }
+}
+
 inline bool
 MPIManager::adaptive_target_buffers() const
 {

nestkernel/nest_names.cpp

@@ -394,6 +394,7 @@ const Name send_buffer_size_secondary_events( "send_buffer_size_secondary_events
 const Name senders( "senders" );
 const Name shape( "shape" );
 const Name shift_now_spikes( "shift_now_spikes" );
+const Name shrink_factor_buffer_spike_data( "shrink_factor_buffer_spike_data" );
 const Name sigma( "sigma" );
 const Name sigmoid( "sigmoid" );
 const Name sion_chunksize( "sion_chunksize" );

nestkernel/nest_names.h

@@ -415,6 +415,7 @@ extern const Name senders;
 extern const Name send_buffer_size_secondary_events;
 extern const Name shape;
 extern const Name shift_now_spikes;
+extern const Name shrink_factor_buffer_spike_data;
 extern const Name sigma;
 extern const Name sigmoid;
 extern const Name sion_chunksize;