/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */ /* * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana * University Research and Technology * Corporation. All rights reserved. * Copyright (c) 2004-2021 The University of Tennessee and The University * of Tennessee Research Foundation. All rights * reserved. * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, * University of Stuttgart. All rights reserved. * Copyright (c) 2004-2005 The Regents of the University of California. * All rights reserved. * Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved. * Copyright (c) 2013 Los Alamos National Security, LLC. All Rights * reserved. * Copyright (c) 2013 FUJITSU LIMITED. All rights reserved. * Copyright (c) 2014-2017 Research Organization for Information Science * and Technology (RIST). All rights reserved. * Copyright (c) 2017 IBM Corporation. All rights reserved. * Copyright (c) 2021 Amazon.com, Inc. or its affiliates. All Rights * reserved. * $COPYRIGHT$ * * Additional copyrights may follow * * $HEADER$ */ #include "ompi_config.h" #include "mpi.h" #include "ompi/constants.h" #include "ompi/datatype/ompi_datatype.h" #include "opal/datatype/opal_convertor_internal.h" #include "ompi/communicator/communicator.h" #include "ompi/mca/coll/coll.h" #include "ompi/mca/coll/base/coll_tags.h" #include "ompi/mca/pml/pml.h" #include "ompi/mca/coll/base/coll_base_functions.h" #include "coll_base_topo.h" #include "coll_base_util.h" #include "opal/util/minmax.h" /* * We want to minimize the amount of temporary memory needed while allowing as many ranks * to exchange data simultaneously. We use a variation of the ring algorithm, where in a * single step a process exchange the data with both neighbors at distance k (on the left * and the right on a logical ring topology). With this approach we need to pack the data * for a single of the two neighbors, as we can then use the original buffer (and datatype * and count) to send the data to the other. */ int mca_coll_base_alltoallv_intra_basic_inplace(const void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module) { int i, size, rank, left, right, err = MPI_SUCCESS, line; ptrdiff_t extent; ompi_request_t *req = MPI_REQUEST_NULL; char *tmp_buffer; size_t packed_size = 0, max_size, type_size; opal_convertor_t convertor; /* Initialize. */ size = ompi_comm_size(comm); rank = ompi_comm_rank(comm); ompi_datatype_type_size(rdtype, &type_size); for (i = 0, max_size = 0 ; i < size ; ++i) { if (i == rank) { continue; } packed_size = rcounts[i] * type_size; max_size = opal_max(packed_size, max_size); } /* Easy way out */ if ((1 == size) || (0 == max_size) ) { return MPI_SUCCESS; } /* Find the largest amount of packed send/recv data among all peers where * we need to pack before the send. */ #if OPAL_ENABLE_HETEROGENEOUS_SUPPORT for (i = 1 ; i <= (size >> 1) ; ++i) { right = (rank + i) % size; ompi_proc_t *ompi_proc = ompi_comm_peer_lookup(comm, right); if( OPAL_UNLIKELY(opal_local_arch != ompi_proc->super.proc_convertor->master->remote_arch)) { packed_size = opal_datatype_compute_remote_size(&rdtype->super, ompi_proc->super.proc_convertor->master->remote_sizes); packed_size *= rcounts[right]; max_size = packed_size > max_size ? packed_size : max_size; } } #endif /* OPAL_ENABLE_HETEROGENEOUS_SUPPORT */ ompi_datatype_type_extent(rdtype, &extent); /* Allocate a temporary buffer */ tmp_buffer = calloc (max_size, 1); if( NULL == tmp_buffer) { err = OMPI_ERR_OUT_OF_RESOURCE; line = __LINE__; goto error_hndl; } for (i = 1 ; i <= (size >> 1) ; ++i) { struct iovec iov = {.iov_base = tmp_buffer, .iov_len = max_size}; uint32_t iov_count = 1; right = (rank + i) % size; left = (rank + size - i) % size; if( 0 != rcounts[right] ) { /* nothing to exchange with the peer on the right */ ompi_proc_t *right_proc = ompi_comm_peer_lookup(comm, right); opal_convertor_clone(right_proc->super.proc_convertor, &convertor, 0); opal_convertor_prepare_for_send(&convertor, &rdtype->super, rcounts[right], (char *) rbuf + rdisps[right] * extent); packed_size = max_size; err = opal_convertor_pack(&convertor, &iov, &iov_count, &packed_size); if (1 != err) { line = __LINE__; goto error_hndl; } /* Receive data from the right */ err = MCA_PML_CALL(irecv ((char *) rbuf + rdisps[right] * extent, rcounts[right], rdtype, right, MCA_COLL_BASE_TAG_ALLTOALLV, comm, &req)); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } } if( (left != right) && (0 != rcounts[left]) ) { /* Send data to the left */ err = MCA_PML_CALL(send ((char *) rbuf + rdisps[left] * extent, rcounts[left], rdtype, left, MCA_COLL_BASE_TAG_ALLTOALLV, MCA_PML_BASE_SEND_STANDARD, comm)); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } err = ompi_request_wait (&req, MPI_STATUSES_IGNORE); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } /* Receive data from the left */ err = MCA_PML_CALL(irecv ((char *) rbuf + rdisps[left] * extent, rcounts[left], rdtype, left, MCA_COLL_BASE_TAG_ALLTOALLV, comm, &req)); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } } if( 0 != rcounts[right] ) { /* nothing to exchange with the peer on the right */ /* Send data to the right */ err = MCA_PML_CALL(send ((char *) tmp_buffer, packed_size, MPI_PACKED, right, MCA_COLL_BASE_TAG_ALLTOALLV, MCA_PML_BASE_SEND_STANDARD, comm)); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } } err = ompi_request_wait (&req, MPI_STATUSES_IGNORE); if (MPI_SUCCESS != err) { line = __LINE__; goto error_hndl; } } error_hndl: /* Free the temporary buffer */ if( NULL != tmp_buffer ) free (tmp_buffer); if( MPI_SUCCESS != err ) { OPAL_OUTPUT((ompi_coll_base_framework.framework_output, "%s:%4d\tError occurred %d, rank %2d", __FILE__, line, err, rank)); (void)line; // silence compiler warning } /* All done */ return err; } int ompi_coll_base_alltoallv_intra_pairwise(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t *sdtype, void* rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module) { int line = -1, err = 0, rank, size, step = 0, sendto, recvfrom; size_t sdtype_size, rdtype_size; void *psnd, *prcv; ompi_request_t *req; ptrdiff_t sext, rext; if (MPI_IN_PLACE == sbuf) { return mca_coll_base_alltoallv_intra_basic_inplace (rbuf, rcounts, rdisps, rdtype, comm, module); } size = ompi_comm_size(comm); rank = ompi_comm_rank(comm); OPAL_OUTPUT((ompi_coll_base_framework.framework_output, "coll:base:alltoallv_intra_pairwise rank %d", rank)); ompi_datatype_type_size(sdtype, &sdtype_size); ompi_datatype_type_size(rdtype, &rdtype_size); ompi_datatype_type_extent(sdtype, &sext); ompi_datatype_type_extent(rdtype, &rext); /* Perform pairwise exchange starting from 1 since local exchange is done */ for (step = 0; step < size; step++) { req = MPI_REQUEST_NULL; /* Determine sender and receiver for this step. */ sendto = (rank + step) % size; recvfrom = (rank + size - step) % size; /* Determine sending and receiving locations */ psnd = (char*)sbuf + (ptrdiff_t)sdisps[sendto] * sext; prcv = (char*)rbuf + (ptrdiff_t)rdisps[recvfrom] * rext; /* send and receive */ if (0 < rcounts[recvfrom] && 0 < rdtype_size) { err = MCA_PML_CALL(irecv(prcv, rcounts[recvfrom], rdtype, recvfrom, MCA_COLL_BASE_TAG_ALLTOALLV, comm, &req)); if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; } } if (0 < scounts[sendto] && 0 < sdtype_size) { err = MCA_PML_CALL(send(psnd, scounts[sendto], sdtype, sendto, MCA_COLL_BASE_TAG_ALLTOALLV, MCA_PML_BASE_SEND_STANDARD, comm)); if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; } } if (MPI_REQUEST_NULL != req) { err = ompi_request_wait(&req, MPI_STATUS_IGNORE); if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; } } } return MPI_SUCCESS; err_hndl: OPAL_OUTPUT((ompi_coll_base_framework.framework_output, "%s:%4d\tError occurred %d, rank %2d at step %d", __FILE__, line, err, rank, step)); (void)line; // silence compiler warning return err; } /** * Linear functions are copied from the basic coll module. For * some small number of nodes and/or small data sizes they are just as * fast as base/tree based segmenting operations and as such may be * selected by the decision functions. These are copied into this module * due to the way we select modules in V1. i.e. in V2 we will handle this * differently and so will not have to duplicate code. */ int ompi_coll_base_alltoallv_intra_basic_linear(const void *sbuf, const int *scounts, const int *sdisps, struct ompi_datatype_t *sdtype, void *rbuf, const int *rcounts, const int *rdisps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module) { int i, size, rank, err, nreqs; size_t sdtype_size = 0, rdtype_size = 0; char *psnd, *prcv; ptrdiff_t sext, rext; ompi_request_t **preq, **reqs; mca_coll_base_module_t *base_module = (mca_coll_base_module_t*) module; mca_coll_base_comm_t *data = base_module->base_data; if (MPI_IN_PLACE == sbuf) { return mca_coll_base_alltoallv_intra_basic_inplace (rbuf, rcounts, rdisps, rdtype, comm, module); } size = ompi_comm_size(comm); rank = ompi_comm_rank(comm); OPAL_OUTPUT((ompi_coll_base_framework.framework_output, "coll:base:alltoallv_intra_basic_linear rank %d", rank)); ompi_datatype_type_size(rdtype, &rdtype_size); ompi_datatype_type_size(sdtype, &sdtype_size); ompi_datatype_type_extent(sdtype, &sext); ompi_datatype_type_extent(rdtype, &rext); /* Simple optimization - handle send to self first */ psnd = ((char *) sbuf) + (ptrdiff_t)sdisps[rank] * sext; prcv = ((char *) rbuf) + (ptrdiff_t)rdisps[rank] * rext; if (0 < scounts[rank] && 0 < sdtype_size) { err = ompi_datatype_sndrcv(psnd, scounts[rank], sdtype, prcv, rcounts[rank], rdtype); if (MPI_SUCCESS != err) { return err; } } /* If only one process, we're done. */ if (1 == size) { return MPI_SUCCESS; } /* Now, initiate all send/recv to/from others. */ nreqs = 0; reqs = preq = ompi_coll_base_comm_get_reqs(data, 2 * size); if( NULL == reqs ) { err = OMPI_ERR_OUT_OF_RESOURCE; goto err_hndl; } /* Post all receives first */ for (i = 0; i < size; ++i) { if (i == rank) { continue; } if (0 < rcounts[i] && 0 < rdtype_size) { ++nreqs; prcv = ((char *) rbuf) + (ptrdiff_t)rdisps[i] * rext; err = MCA_PML_CALL(irecv_init(prcv, rcounts[i], rdtype, i, MCA_COLL_BASE_TAG_ALLTOALLV, comm, preq++)); if (MPI_SUCCESS != err) { goto err_hndl; } } } /* Now post all sends */ for (i = 0; i < size; ++i) { if (i == rank) { continue; } if (0 < scounts[i] && 0 < sdtype_size) { ++nreqs; psnd = ((char *) sbuf) + (ptrdiff_t)sdisps[i] * sext; err = MCA_PML_CALL(isend_init(psnd, scounts[i], sdtype, i, MCA_COLL_BASE_TAG_ALLTOALLV, MCA_PML_BASE_SEND_STANDARD, comm, preq++)); if (MPI_SUCCESS != err) { goto err_hndl; } } } /* Start your engines. This will never return an error. */ MCA_PML_CALL(start(nreqs, reqs)); /* Wait for them all. If there's an error, note that we don't care * what the error was -- just that there *was* an error. The PML * will finish all requests, even if one or more of them fail. * i.e., by the end of this call, all the requests are free-able. * So free them anyway -- even if there was an error, and return the * error after we free everything. */ err = ompi_request_wait_all(nreqs, reqs, MPI_STATUSES_IGNORE); err_hndl: /* find a real error code */ if (MPI_ERR_IN_STATUS == err) { for( i = 0; i < nreqs; i++ ) { if (MPI_REQUEST_NULL == reqs[i]) continue; if (MPI_ERR_PENDING == reqs[i]->req_status.MPI_ERROR) continue; if (reqs[i]->req_status.MPI_ERROR != MPI_SUCCESS) { err = reqs[i]->req_status.MPI_ERROR; break; } } } /* Free the requests in all cases as they are persistent */ ompi_coll_base_free_reqs(reqs, nreqs); return err; }