Developer Guide

Contents

async_progress_sample.c

#define PROGRESS_THREAD_COUNT 4 MPI_Comm comms[PROGRESS_THREAD_COUNT]; MPI_Request requests[PROGRESS_THREAD_COUNT]; MPI_Info info; int idx; /* create “per-thread” communicators and assign thread id for each communicator */ for (idx = 0; idx < PROGRESS_THREAD_COUNT; idx++) { MPI_Comm_dup(MPI_COMM_WORLD, &comms[idx]); char thread_id_str[256] = { 0 }; sprintf(thread_id_str, "%d", idx); MPI_Info_create(&info); MPI_Info_set(info, "thread_id", thread_id_str); MPI_Comm_set_info(comms[idx], info); MPI_Info_free(&info); } /* distribute MPI operations between communicators – i.e. between progress threads */ for (idx = 0; idx < PROGRESS_THREAD_COUNT; idx++) { MPI_Iallreduce(…, comms[idx], &requests[idx]); } MPI_Waitall(PROGRESS_THREAD_COUNT, requests, …)

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.