University:
Boston University
Course:
MET CS 201 | Introduction to Programming
Academic year:

2023
Views:

82

Pages:

19
Author:

Abeltinohi5s

Mixing blocking and non-blocking sends and receives  MPI_Isend is non-blocking, so there is no deadlock in the following code. int n=5000; // a large data size int my_rank, n_send1[n], n_send2[n], n_recv1[n], n_recv2[n]; MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); MPI_Request send_request; if (my_rank==0){ MPI_Isend( &n_send1, n, MPI_INT, 1, 10, MPI_COMM_WORLD, &send_request); MPI_Recv( &n_recv2, n, MPI_INT, 1, 11, MPI_COMM_WORLD); } else if (my_rank==1){ MPI_Isend( &n_send2, n, MPI_INT, 0, 11, MPI_COMM_WORLD, &send_request); MPI_Recv( &n_recv1, n, MPI_INT, 0, 10, MPI_COMM_WORLD); } Collective Communication Subroutines  Collective:  One to many  Many to one  Many to many Figure from: Practical MPI Programming, IBM Redbook Collective communication: MPI_Scatter  The root process sends chunks of an array to all processes. Each non-root process receives a chunk of the array and stores it in receive buffer. The root process also copies a chunk of the array to its own receive buffer.  Syntax: int MPI_Scatter (const void* send_data, int send_count, MPI_Datatype send_datatype, void* recv_data, int recv_count, MPI_Datatype recv_datatype, int root, MPI_Comm comm)  send_data: The send array that originally resides on the root process.  send_count: Number of elements to be sent to each process (i.e. the chunk size). It is often (approximately) equal to the size of the array divided by the number of processes.  send_datatype: Datatype of the send data.  recv_data: The receive buffer on all processes.  recv_count: Number of elements that the receive buffer can hold (i.e. the chunk size). It should be equal to send_count if send_datatype and recv_datatype are the same.  recv_datatype: Datatype of the receive data.  root: The rank of the root process. Collective communication: MPI_Gather  MPI_Gather is the inverse of MPI_Scatter.  Each non-root process sends a chunk of data to the root process. The root process receives chunks of data and stores them (including its own chunk) in the receive buffer in the order of MPI ranks.  Syntax: int MPI_Gather (const void* send_data, int send_count, MPI_Datatype send_datatype, void* recv_data, int recv_count, MPI_Datatype recv_datatype, int root, MPI_Comm comm)  send_data: The send data on each process.  send_count: Number of elements of the send data (i.e. the chunk size).  send_datatype: Datatype of the send data.  recv_data: The receive buffer on the root process.  recv_count: Number of elements of the receive data (i.e. the chunk size, not the size of the receive buffer) . It should be equal to send_count if send_datatype and recv_datatype are the same.  recv_datatype: Datatype of the receive data.  root: The rank of the root process. An example for MPI_Scatter and MPI_Gather  Compute the average of all elements in an array. int rank, nproc, i, m, n=100; double sub_avg=0., global_avg=0.; double * array = NULL, * sub_avgs = NULL; MPI_Init(&argc, &argv); MPI_Comm_rank(MPI_COMM_WORLD, &rank); MPI_Comm_size(MPI_COMM_WORLD, &nproc); m = (int) n/nproc; // chunk size if(rank==0){ array = (double *) malloc(n*sizeof(double)); for(i=0; i