# MPI: point to point messages <div class="center"> [Notes](/phys52015/notes/mpi/point-to-point) </div>

## Main idea - Two-sided communication (sender and receiver) - Processes identified by `(communicator, rank)` pair - Need to "describe" data to the library - How much data to send, where to get it from - How much data to receive, where to put it

## Pointers and datatypes ```c int MPI_Send(const void *buffer, int count, MPI_Datatype dtype, int dest, int tag, MPI_Comm comm); int MPI_Recv(void *buffer, int count, MPI_Datatype dtype, int src, int tag, MPI_Comm comm, MPI_Status *status); ``` - To pass data to library, pass _pointer_ to first element ```c /* Sending from an array */ int *data = ...; MPI_Send(data, ...); /* Sending a scalar value */ double scalar = 10; MPI_Send(&scalar, ...); ```

### How does MPI know what to send? - `count` and `datatype` arguments - `MPI_Datatype` describes single "unit" - `count` how many copies of the unit described by the datatype? | C type | MPI_Datatype | |:---------|:-------------| | `char` | MPI_CHAR | | `int` | MPI_INT | | `float` | MPI_FLOAT | | `double` | MPI_DOUBLE |

### Example ```c /* Single scalar value */ double value = ...; MPI_Send(&value, 1, MPI_DOUBLE, ...); ``` ```c /* Second entry of array */ double *array = ...; MPI_Send(&array[1], 1, MPI_DOUBLE, ...) ``` ```c /* Five entries of the array */ double *array = ...; MPI_Send(array, 5, MPI_DOUBLE, ...) ```

### _No type-checking_ - Datatype description is _independent_ of actual type of data buffer - If you describe your data incorrectly, MPI won't complain - You get to keep both pieces ```c int value = ...; /* MPI will not complain, buffer overflow */ MPI_Send(&value, 5, MPI_INT, ...); ```

## Message matching - Messages sent/received over a communicator: `(rank, comm)` pair - Messages with the same `(rank, comm)` pair can be distinguished by _tags_ - No matching on count or datatype (can differ on sender and receiver) - No message overtaking

### Example ```c if (rank == 0) { /* This message is matched ... */ MPI_Send(..., 1 /* dest */, 0 /* tag */, MPI_COMM_WORLD); } else if (rank == 1) { /* ... here */ MPI_Recv(..., 0 /* src */, 0 /* tag */, MPI_COMM_WORLD, MPI_STATUS_IGNORE); } ```

### Example: mismatch ```c if (rank == 0) { /* This message is not matched ... */ MPI_Send(..., 1 /* dest */, 0 /* tag */, MPI_COMM_WORLD); } else if (rank == 1) { /* ... the tag is different */ MPI_Recv(..., 0 /* src */, 1 /* tag */, MPI_COMM_WORLD, MPI_STATUS_IGNORE); } ```

## Deadlocks - Ordering of sends and receives is important ```c if (rank == 0) { MPI_Send(sendbuf, nentries, MPI_INT, 1, 0, ...); MPI_Recv(recvbuf, nentries, MPI_INT, 1, 0, ...); } else if (rank == 1) { MPI_Send(sendbuf, nentries, MPI_INT, 0, 0, ...); MPI_Recv(recvbuf, nentries, MPI_INT, 0, 0, ...); } ``` - Deadlock (both waiting for a receive to appear) - Exercise: [deadlocks](/phys52015/notes/mpi/point-to-point/#deadlock-exercise) --- ## Over to you - MPI [hello world](/phys52015/exercises/hello/#mpi) - Exercises in [point to point notes](/phys52015/notes/mpi/point-to-point/) - Messages [around a ring](/phys52015/exercises/mpi-ring/)