+++ /dev/null
-/* Copyright (c) 2007-2015. The SimGrid Team.
- * All rights reserved. */
-
-/* This program is free software; you can redistribute it and/or modify it
- * under the terms of the license (GNU LGPL) which comes with this package. */
-
-#include "private.h"
-#include <vector>
-
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_rma, smpi, "Logging specific to SMPI (RMA operations)");
-
-typedef struct s_smpi_mpi_win{
- void* base;
- MPI_Aint size;
- int disp_unit;
- MPI_Comm comm;
- MPI_Info info;
- int assert;
- std::vector<MPI_Request> *requests;
- xbt_mutex_t mut;
- msg_bar_t bar;
- MPI_Win* connected_wins;
- char* name;
- int opened;
- MPI_Group group;
- int count; //for ordering the accs
-} s_smpi_mpi_win_t;
-
-
-MPI_Win smpi_mpi_win_create( void *base, MPI_Aint size, int disp_unit, MPI_Info info, MPI_Comm comm){
- int comm_size = comm->size();
- int rank = comm->rank();
- XBT_DEBUG("Creating window");
-
- MPI_Win win = xbt_new(s_smpi_mpi_win_t, 1);
- win->base = base;
- win->size = size;
- win->disp_unit = disp_unit;
- win->assert = 0;
- win->info = info;
- if(info!=MPI_INFO_NULL)
- info->refcount++;
- win->comm = comm;
- win->name = nullptr;
- win->opened = 0;
- win->group = MPI_GROUP_NULL;
- win->requests = new std::vector<MPI_Request>();
- win->mut=xbt_mutex_init();
- win->connected_wins = xbt_new0(MPI_Win, comm_size);
- win->connected_wins[rank] = win;
- win->count = 0;
- if(rank==0){
- win->bar = MSG_barrier_init(comm_size);
- }
- mpi_coll_allgather_fun(&(win->connected_wins[rank]), sizeof(MPI_Win), MPI_BYTE, win->connected_wins, sizeof(MPI_Win),
- MPI_BYTE, comm);
-
- mpi_coll_bcast_fun(&(win->bar), sizeof(msg_bar_t), MPI_BYTE, 0, comm);
-
- mpi_coll_barrier_fun(comm);
-
- return win;
-}
-
-int smpi_mpi_win_free( MPI_Win* win){
- //As per the standard, perform a barrier to ensure every async comm is finished
- MSG_barrier_wait((*win)->bar);
- xbt_mutex_acquire((*win)->mut);
- delete (*win)->requests;
- xbt_mutex_release((*win)->mut);
- xbt_free((*win)->connected_wins);
- if ((*win)->name != nullptr){
- xbt_free((*win)->name);
- }
- if((*win)->info!=MPI_INFO_NULL){
- MPI_Info_free(&(*win)->info);
- }
-
- mpi_coll_barrier_fun((*win)->comm);
- int rank=(*win)->comm->rank();
- if(rank == 0)
- MSG_barrier_destroy((*win)->bar);
- xbt_mutex_destroy((*win)->mut);
- xbt_free(*win);
- *win = MPI_WIN_NULL;
- return MPI_SUCCESS;
-}
-
-void smpi_mpi_win_get_name(MPI_Win win, char* name, int* length){
- if(win->name==nullptr){
- *length=0;
- name=nullptr;
- return;
- }
- *length = strlen(win->name);
- strncpy(name, win->name, *length+1);
-}
-
-void smpi_mpi_win_get_group(MPI_Win win, MPI_Group* group){
- if(win->comm != MPI_COMM_NULL){
- *group = win->comm->group();
- } else {
- *group = MPI_GROUP_NULL;
- }
-}
-
-void smpi_mpi_win_set_name(MPI_Win win, char* name){
- win->name = xbt_strdup(name);
-}
-
-int smpi_mpi_win_fence(int assert, MPI_Win win)
-{
- XBT_DEBUG("Entering fence");
- if (win->opened == 0)
- win->opened=1;
- if (assert != MPI_MODE_NOPRECEDE) {
- // This is not the first fence => finalize what came before
- MSG_barrier_wait(win->bar);
- xbt_mutex_acquire(win->mut);
- // This (simulated) mutex ensures that no process pushes to the vector of requests during the waitall.
- // Without this, the vector could get redimensionned when another process pushes.
- // This would result in the array used by smpi_mpi_waitall() to be invalidated.
- // Another solution would be to copy the data and cleanup the vector *before* smpi_mpi_waitall
- std::vector<MPI_Request> *reqs = win->requests;
- int size = static_cast<int>(reqs->size());
- // start all requests that have been prepared by another process
- if (size > 0) {
- for (const auto& req : *reqs) {
- if (req && (req->flags & PREPARED))
- smpi_mpi_start(req);
- }
-
- MPI_Request* treqs = &(*reqs)[0];
-
- smpi_mpi_waitall(size, treqs, MPI_STATUSES_IGNORE);
- }
- win->count=0;
- xbt_mutex_release(win->mut);
- }
- win->assert = assert;
-
- MSG_barrier_wait(win->bar);
- XBT_DEBUG("Leaving fence");
-
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_put( void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
- MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Win win)
-{
- if(win->opened==0)//check that post/start has been done
- return MPI_ERR_WIN;
- //get receiver pointer
- MPI_Win recv_win = win->connected_wins[target_rank];
-
- void* recv_addr = static_cast<void*> ( static_cast<char*>(recv_win->base) + target_disp * recv_win->disp_unit);
- XBT_DEBUG("Entering MPI_Put to %d", target_rank);
-
- if(target_rank != win->comm->rank()){
- //prepare send_request
- MPI_Request sreq = smpi_rma_send_init(origin_addr, origin_count, origin_datatype, smpi_process_index(),
- win->comm->group()->index(target_rank), SMPI_RMA_TAG+1, win->comm, MPI_OP_NULL);
-
- //prepare receiver request
- MPI_Request rreq = smpi_rma_recv_init(recv_addr, target_count, target_datatype, smpi_process_index(),
- win->comm->group()->index(target_rank), SMPI_RMA_TAG+1, recv_win->comm, MPI_OP_NULL);
-
- //push request to receiver's win
- xbt_mutex_acquire(recv_win->mut);
- recv_win->requests->push_back(rreq);
- xbt_mutex_release(recv_win->mut);
- //start send
- smpi_mpi_start(sreq);
-
- //push request to sender's win
- xbt_mutex_acquire(win->mut);
- win->requests->push_back(sreq);
- xbt_mutex_release(win->mut);
- }else{
- smpi_datatype_copy(origin_addr, origin_count, origin_datatype, recv_addr, target_count, target_datatype);
- }
-
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_get( void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
- MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Win win)
-{
- if(win->opened==0)//check that post/start has been done
- return MPI_ERR_WIN;
- //get sender pointer
- MPI_Win send_win = win->connected_wins[target_rank];
-
- void* send_addr = static_cast<void*>(static_cast<char*>(send_win->base) + target_disp * send_win->disp_unit);
- XBT_DEBUG("Entering MPI_Get from %d", target_rank);
-
- if(target_rank != win->comm->rank()){
- //prepare send_request
- MPI_Request sreq = smpi_rma_send_init(send_addr, target_count, target_datatype,
- win->comm->group()->index(target_rank), smpi_process_index(), SMPI_RMA_TAG+2, send_win->comm,
- MPI_OP_NULL);
-
- //prepare receiver request
- MPI_Request rreq = smpi_rma_recv_init(origin_addr, origin_count, origin_datatype,
- win->comm->group()->index(target_rank), smpi_process_index(), SMPI_RMA_TAG+2, win->comm,
- MPI_OP_NULL);
-
- //start the send, with another process than us as sender.
- smpi_mpi_start(sreq);
- //push request to receiver's win
- xbt_mutex_acquire(send_win->mut);
- send_win->requests->push_back(sreq);
- xbt_mutex_release(send_win->mut);
-
- //start recv
- smpi_mpi_start(rreq);
- //push request to sender's win
- xbt_mutex_acquire(win->mut);
- win->requests->push_back(rreq);
- xbt_mutex_release(win->mut);
- }else{
- smpi_datatype_copy(send_addr, target_count, target_datatype, origin_addr, origin_count, origin_datatype);
- }
-
- return MPI_SUCCESS;
-}
-
-
-int smpi_mpi_accumulate( void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
- MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Op op, MPI_Win win)
-{
- if(win->opened==0)//check that post/start has been done
- return MPI_ERR_WIN;
- //FIXME: local version
- //get receiver pointer
- MPI_Win recv_win = win->connected_wins[target_rank];
-
- void* recv_addr = static_cast<void*>(static_cast<char*>(recv_win->base) + target_disp * recv_win->disp_unit);
- XBT_DEBUG("Entering MPI_Accumulate to %d", target_rank);
- //As the tag will be used for ordering of the operations, add count to it
- //prepare send_request
- MPI_Request sreq = smpi_rma_send_init(origin_addr, origin_count, origin_datatype,
- smpi_process_index(), win->comm->group()->index(target_rank), SMPI_RMA_TAG+3+win->count, win->comm, op);
-
- //prepare receiver request
- MPI_Request rreq = smpi_rma_recv_init(recv_addr, target_count, target_datatype,
- smpi_process_index(), win->comm->group()->index(target_rank), SMPI_RMA_TAG+3+win->count, recv_win->comm, op);
-
- win->count++;
- //push request to receiver's win
- xbt_mutex_acquire(recv_win->mut);
- recv_win->requests->push_back(rreq);
- xbt_mutex_release(recv_win->mut);
- //start send
- smpi_mpi_start(sreq);
-
- //push request to sender's win
- xbt_mutex_acquire(win->mut);
- win->requests->push_back(sreq);
- xbt_mutex_release(win->mut);
-
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_win_start(MPI_Group group, int assert, MPI_Win win){
- /* From MPI forum advices
- The call to MPI_WIN_COMPLETE does not return until the put call has completed at the origin; and the target window
- will be accessed by the put operation only after the call to MPI_WIN_START has matched a call to MPI_WIN_POST by
- the target process. This still leaves much choice to implementors. The call to MPI_WIN_START can block until the
- matching call to MPI_WIN_POST occurs at all target processes. One can also have implementations where the call to
- MPI_WIN_START is nonblocking, but the call to MPI_PUT blocks until the matching call to MPI_WIN_POST occurred; or
- implementations where the first two calls are nonblocking, but the call to MPI_WIN_COMPLETE blocks until the call
- to MPI_WIN_POST occurred; or even implementations where all three calls can complete before any target process
- called MPI_WIN_POST --- the data put must be buffered, in this last case, so as to allow the put to complete at the
- origin ahead of its completion at the target. However, once the call to MPI_WIN_POST is issued, the sequence above
- must complete, without further dependencies. */
-
- //naive, blocking implementation.
- int i = 0;
- int j = 0;
- int size = group->size();
- MPI_Request* reqs = xbt_new0(MPI_Request, size);
-
- while (j != size) {
- int src = group->index(j);
- if (src != smpi_process_index() && src != MPI_UNDEFINED) {
- reqs[i] = smpi_irecv_init(nullptr, 0, MPI_CHAR, src, SMPI_RMA_TAG + 4, MPI_COMM_WORLD);
- i++;
- }
- j++;
- }
- size=i;
- smpi_mpi_startall(size, reqs);
- smpi_mpi_waitall(size, reqs, MPI_STATUSES_IGNORE);
- for(i=0;i<size;i++){
- smpi_mpi_request_free(&reqs[i]);
- }
- xbt_free(reqs);
- win->opened++; //we're open for business !
- win->group=group;
- group->use();
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_win_post(MPI_Group group, int assert, MPI_Win win){
- //let's make a synchronous send here
- int i = 0;
- int j = 0;
- int size = group->size();
- MPI_Request* reqs = xbt_new0(MPI_Request, size);
-
- while(j!=size){
- int dst=group->index(j);
- if(dst!=smpi_process_index() && dst!=MPI_UNDEFINED){
- reqs[i]=smpi_mpi_send_init(nullptr, 0, MPI_CHAR, dst, SMPI_RMA_TAG+4, MPI_COMM_WORLD);
- i++;
- }
- j++;
- }
- size=i;
-
- smpi_mpi_startall(size, reqs);
- smpi_mpi_waitall(size, reqs, MPI_STATUSES_IGNORE);
- for(i=0;i<size;i++){
- smpi_mpi_request_free(&reqs[i]);
- }
- xbt_free(reqs);
- win->opened++; //we're open for business !
- win->group=group;
- group->use();
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_win_complete(MPI_Win win){
- if(win->opened==0)
- xbt_die("Complete called on already opened MPI_Win");
-
- XBT_DEBUG("Entering MPI_Win_Complete");
- int i = 0;
- int j = 0;
- int size = win->group->size();
- MPI_Request* reqs = xbt_new0(MPI_Request, size);
-
- while(j!=size){
- int dst=win->group->index(j);
- if(dst!=smpi_process_index() && dst!=MPI_UNDEFINED){
- reqs[i]=smpi_mpi_send_init(nullptr, 0, MPI_CHAR, dst, SMPI_RMA_TAG+5, MPI_COMM_WORLD);
- i++;
- }
- j++;
- }
- size=i;
- XBT_DEBUG("Win_complete - Sending sync messages to %d processes", size);
- smpi_mpi_startall(size, reqs);
- smpi_mpi_waitall(size, reqs, MPI_STATUSES_IGNORE);
-
- for(i=0;i<size;i++){
- smpi_mpi_request_free(&reqs[i]);
- }
- xbt_free(reqs);
-
- //now we can finish RMA calls
- xbt_mutex_acquire(win->mut);
- std::vector<MPI_Request> *reqqs = win->requests;
- size = static_cast<int>(reqqs->size());
-
- XBT_DEBUG("Win_complete - Finishing %d RMA calls", size);
- if (size > 0) {
- // start all requests that have been prepared by another process
- for (const auto& req : *reqqs) {
- if (req && (req->flags & PREPARED))
- smpi_mpi_start(req);
- }
-
- MPI_Request* treqs = &(*reqqs)[0];
- smpi_mpi_waitall(size, treqs, MPI_STATUSES_IGNORE);
- reqqs->clear();
- }
- xbt_mutex_release(win->mut);
-
- win->group->unuse();
- win->opened--; //we're closed for business !
- return MPI_SUCCESS;
-}
-
-int smpi_mpi_win_wait(MPI_Win win){
- //naive, blocking implementation.
- XBT_DEBUG("Entering MPI_Win_Wait");
- int i=0,j=0;
- int size = win->group->size();
- MPI_Request* reqs = xbt_new0(MPI_Request, size);
-
- while(j!=size){
- int src=win->group->index(j);
- if(src!=smpi_process_index() && src!=MPI_UNDEFINED){
- reqs[i]=smpi_irecv_init(nullptr, 0, MPI_CHAR, src,SMPI_RMA_TAG+5, MPI_COMM_WORLD);
- i++;
- }
- j++;
- }
- size=i;
- XBT_DEBUG("Win_wait - Receiving sync messages from %d processes", size);
- smpi_mpi_startall(size, reqs);
- smpi_mpi_waitall(size, reqs, MPI_STATUSES_IGNORE);
- for(i=0;i<size;i++){
- smpi_mpi_request_free(&reqs[i]);
- }
- xbt_free(reqs);
- xbt_mutex_acquire(win->mut);
- std::vector<MPI_Request> *reqqs = win->requests;
- size = static_cast<int>(reqqs->size());
-
- XBT_DEBUG("Win_wait - Finishing %d RMA calls", size);
- if (size > 0) {
- // start all requests that have been prepared by another process
- for (const auto& req : *reqqs) {
- if (req && (req->flags & PREPARED))
- smpi_mpi_start(req);
- }
-
- MPI_Request* treqs = &(*reqqs)[0];
- smpi_mpi_waitall(size, treqs, MPI_STATUSES_IGNORE);
- reqqs->clear();
- }
- xbt_mutex_release(win->mut);
-
- win->group->unuse();
- win->opened--; //we're opened for business !
- return MPI_SUCCESS;
-}
