Allow setting MPI port names of MPI recording and stimulation backends from device labels

nestkernel/recording_backend_mpi.cpp

@@ -357,7 +357,27 @@ nest::RecordingBackendMPI::set_status( const DictionaryDatum& )
 void
 nest::RecordingBackendMPI::get_port( const RecordingDevice* device, std::string* port_name )
 {
-  get_port( device->get_node_id(), device->get_label(), port_name );
+  const std::string& label = device->get_label();
+  std::size_t colon_position;
+
+  // The MPI address can be provided by two different means:
+  // a) the address is taken from the device's label, which then has to be in the format: endpoint_address:address
+  // b) the address is provided via a file: {data_path}/{data_prefix}{device_label}/{node_id}.txt
+
+  // Case a: device label is the endpoint address in the following format:
+  // endpoint_address:address string
+  colon_position = label.find( ":", 0 );
+
+  // check if string start with "endpoint_address:" and a colon is found
+  if ( label.find( "endpoint_address:" ) == 0 )
+  {
+    *port_name = label.substr( colon_position + 1 );
+  }
+  // Case b: fallback to get_port implementation that reads the address from file
+  else
+  {
+    get_port( device->get_node_id(), label, port_name );
+  }
 }
 
 void

nestkernel/recording_backend_mpi.h

@@ -46,7 +46,16 @@ Description
 
 The `mpi` recording backend sends collected data to a remote process
 using MPI.
+There are two ways to set the MPI port (if both are set, option A has precedence):
 
+A)
+The label of the recording device has the pattern:
+
+::
+
+   endpoint_address:port_name
+
+B)
 The name of the MPI port to send data to is read from a file for each
 device configured to use this backend. The file needs to be named
 according to the following pattern:

nestkernel/stimulation_backend_mpi.cpp

@@ -285,7 +285,27 @@ nest::StimulationBackendMPI::cleanup()
 void
 nest::StimulationBackendMPI::get_port( nest::StimulationDevice* device, std::string* port_name )
 {
-  get_port( device->get_node_id(), device->get_label(), port_name );
+  const std::string& label = device->get_label();
+  std::size_t colon_position;
+
+  // The MPI address can be provided by two different means:
+  // a) the address is given by the label in the format: endpoint_address:address
+  // b) the file is provided via a file: {data_path}/{data_prefix}{label}/{node_id}.txt
+
+  // Case a: label is the endpoint address in the following format:
+  // endpoint_address:address string
+  colon_position = label.find( ":", 0 );
+
+  // check if string start with "endpoint_address:" and a colon is found
+  if ( label.find( "endpoint_address:" ) == 0 )
+  {
+    *port_name = label.substr( colon_position + 1 );
+  }
+  // Case b: fallback to get_port implementation that reads the address from file
+  else
+  {
+    get_port( device->get_node_id(), label, port_name );
+  }
 }
 
 void

nestkernel/stimulation_backend_mpi.h

@@ -47,6 +47,15 @@ The `mpi` stimulation backend collects data from MPI channels and
 updates stimulation devices just before each run. This is useful for
 co-simulation or for receiving stimuli from external software.
 
+There are two ways to set the MPI port (if both are set, option A has precedence):
+
+A)
+The label of the recording device has the pattern:
+
+::
+   endpoint_address:port_name
+
+B)
 The name of the MPI port to receive data on is read from a file for
 each device configured to use this backend.  The file needs to be
 named according to the following pattern: