MapReduce::MapReduce(MPI_Comm comm) MapReduce::MapReduce() MapReduce::MapReduce(double dummy)
You can create a MapReduce object in any of the three ways shown, as well as via the copy() method. The three creation methods differ slightly in how MPI is initialized and finalized.
In the first case, you pass an MPI communicator to the constructor. This means your program should initialize (and finalize) MPI, which creates the MPI_COMM_WORLD communicator (all the processors you are running on). Normally this is what you pass to the MapReduce constructor, but you can pass a communicator for a subset of your processors if desired. You can also instantiate multiple MapReduce objects, giving them each a communicator for all the processors or communicators for a subset of processors.
The second case can be used if your program does not use MPI at all. The library will initialize MPI if it has not already been initialized. It will not finalize MPI, but this should be fine. Worst case, your program may complain when it exits if MPI has not been finalized.
The third case is the same as the second except that the library will finalize MPI when the last instance of a MapReduce object is destructed. Note that this means your program cannot delete all its MapReduce objects in a early phase of the program and then instantiate more MapReduce objects later. This limitation is why the second case is provided. The third case is invoked by passing a double to the constructor. If this is done for any instantiated MapReduce object, then the library will finalize MPI. The value of the double doesn't matter as it isn't used. The use of a double is simply to make it different than the first case, since MPI_Comm is often implemented by MPI libraries as a type cast to an integer.
As examples, any of these lines of code will create a MapReduce object, where "mr" is either a pointer to the object or the object itself:
MapReduce *mr = new MapReduce(MPI_COMM_WORLD); MapReduce *mr = new MapReduce(); MapReduce *mr = new MapReduce(0.0); MapReduce mr(MPI_COMM_WORLD); MapReduce mr(); MapReduce mr; MapReduce mr(0.0);