#include "simix/smx_private.h"
#include "surf/surf.h"
#include "simgrid/sg_config.h"
-
+#include "colls/colls.h"
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_base, smpi, "Logging specific to SMPI (base)");
xbt_assert(ref, "Cannot match recv against null reference");
xbt_assert(req, "Cannot match recv against null request");
if((ref->src == MPI_ANY_SOURCE || req->src == ref->src)
- && (ref->tag == MPI_ANY_TAG || req->tag == ref->tag)){
+ && ((ref->tag == MPI_ANY_TAG && req->tag >=0) || req->tag == ref->tag)){
//we match, we can transfer some values
// FIXME : move this to the copy function ?
if(ref->src == MPI_ANY_SOURCE)ref->real_src = req->src;
if(req->detached==1){
ref->detached_sender=req; //tie the sender to the receiver, as it is detached and has to be freed in the receiver
}
+ XBT_DEBUG("match succeeded");
return 1;
}else return 0;
}
xbt_assert(req, "Cannot match send against null request");
if((req->src == MPI_ANY_SOURCE || req->src == ref->src)
- && (req->tag == MPI_ANY_TAG || req->tag == ref->tag))
+ && ((req->tag == MPI_ANY_TAG && ref->tag >=0)|| req->tag == ref->tag))
{
if(req->src == MPI_ANY_SOURCE)req->real_src = ref->src;
if(req->tag == MPI_ANY_TAG)req->real_tag = ref->tag;
if(ref->detached==1){
req->detached_sender=ref; //tie the sender to the receiver, as it is detached and has to be freed in the receiver
}
-
+ XBT_DEBUG("match succeeded");
return 1;
} else return 0;
}
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, PERSISTENT | SEND);
request->refcount++;
return request;
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, PERSISTENT | SSEND | SEND);
request->refcount++;
return request;
int src, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, src, smpi_comm_rank(comm), tag,
+ build_request(buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag,
comm, PERSISTENT | RECV);
request->refcount++;
return request;
} else {
- int receiver = smpi_group_index(smpi_comm_group(request->comm), request->dst);
+ int receiver = request->dst;//smpi_group_index(smpi_comm_group(request->comm), request->dst);
#ifdef HAVE_TRACING
int rank = smpi_process_index();
if((*request)->refcount<0) xbt_die("wrong refcount");
if((*request)->refcount==0){
+ print_request("Destroying", (*request));
xbt_free(*request);
*request = MPI_REQUEST_NULL;
+ }else{
+ print_request("Decrementing", (*request));
+
}
}else{
xbt_die("freeing an already free request");
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, NON_PERSISTENT | SEND);
return request;
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, NON_PERSISTENT | ISEND | SEND);
smpi_mpi_start(request);
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, NON_PERSISTENT | ISEND | SSEND | SEND);
smpi_mpi_start(request);
return request;
int src, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, src, smpi_comm_rank(comm), tag,
+ build_request(buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag,
comm, NON_PERSISTENT | RECV);
return request;
}
int src, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, src, smpi_comm_rank(comm), tag,
+ build_request(buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag,
comm, NON_PERSISTENT | RECV);
smpi_mpi_start(request);
int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, NON_PERSISTENT | SEND);
-
smpi_mpi_start(request);
smpi_mpi_wait(&request, MPI_STATUS_IGNORE);
int dst, int tag, MPI_Comm comm)
{
MPI_Request request =
- build_request(buf, count, datatype, smpi_comm_rank(comm), dst, tag,
+ build_request(buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag,
comm, NON_PERSISTENT | SSEND | SEND);
- smpi_mpi_start(request); smpi_mpi_wait(&request, MPI_STATUS_IGNORE);
+ smpi_mpi_start(request);
+ smpi_mpi_wait(&request, MPI_STATUS_IGNORE);
}
void smpi_mpi_sendrecv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
{
MPI_Request requests[2];
MPI_Status stats[2];
-
+ int myid=smpi_process_index();
+ if ((smpi_group_index(smpi_comm_group(comm), dst) == myid) && (smpi_group_index(smpi_comm_group(comm), src) == myid)) {
+ smpi_datatype_copy(sendbuf, sendcount, sendtype,
+ recvbuf, recvcount, recvtype);
+ return;
+ }
requests[0] =
smpi_isend_init(sendbuf, sendcount, sendtype, dst, sendtag, comm);
requests[1] =
if(!(req->detached && req->flags & SEND)){
if(status != MPI_STATUS_IGNORE) {
- status->MPI_SOURCE = req->src == MPI_ANY_SOURCE ? req->real_src : req->src;
+ int src = req->src == MPI_ANY_SOURCE ? req->real_src : req->src;
+ status->MPI_SOURCE = smpi_group_rank(smpi_comm_group(req->comm), src);
status->MPI_TAG = req->tag == MPI_ANY_TAG ? req->real_tag : req->tag;
status->MPI_ERROR = req->truncated ? MPI_ERR_TRUNCATE : MPI_SUCCESS;
// this handles the case were size in receive differs from size in send
if (TRACE_smpi_view_internals()) {
if(req->flags & RECV){
int rank = smpi_process_index();
- int src_traced = smpi_group_index(smpi_comm_group(req->comm), req->src == MPI_ANY_SOURCE ? req->real_src : req->src);
+ int src_traced = (req->src == MPI_ANY_SOURCE ? req->real_src : req->src);
TRACE_smpi_recv(rank, src_traced, rank);
}
}
while(flag==0){
smpi_mpi_iprobe(source, tag, comm, &flag, status);
XBT_DEBUG("Busy Waiting on probing : %d", flag);
- if(!flag) {
- simcall_process_sleep(0.0001);
- }
}
}
void smpi_mpi_iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status* status){
- MPI_Request request =build_request(NULL, 0, MPI_CHAR, source, smpi_comm_rank(comm), tag,
+
+ MPI_Request request =build_request(NULL, 0, MPI_CHAR, source == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), source), smpi_comm_rank(comm), tag,
comm, NON_PERSISTENT | RECV);
+ //to avoid deadlock, we have to sleep some time here, or the timer won't advance and we will only do iprobe simcalls
+ double sleeptime= sg_cfg_get_double("smpi/iprobe");
+ //multiplier to the sleeptime, to increase speed of execution, each failed iprobe will increase it
+ static int nsleeps = 1;
+
+ simcall_process_sleep(sleeptime);
+
// behave like a receive, but don't do it
smx_rdv_t mailbox;
MPI_Request req = (MPI_Request)SIMIX_comm_get_src_data(request->action);
*flag = 1;
if(status != MPI_STATUS_IGNORE) {
- status->MPI_SOURCE = req->src;
+ status->MPI_SOURCE = smpi_group_rank(smpi_comm_group(comm), req->src);
status->MPI_TAG = req->tag;
status->MPI_ERROR = MPI_SUCCESS;
status->count = req->real_size;
}
+ nsleeps=1;//reset the number of sleeps we will do next time
+ }
+ else {
+ *flag = 0;
+ nsleeps++;
}
- else *flag = 0;
smpi_mpi_request_free(&request);
return;
void *recvbuf, int recvcount, MPI_Datatype recvtype,
int root, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_GATHER;
int rank, size, src, index;
MPI_Aint lb = 0, recvext = 0;
MPI_Request *requests;
}
}
+
+void smpi_mpi_reduce_scatter(void *sendbuf, void *recvbuf, int *recvcounts,
+ MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
+{
+ int i, size, count;
+ int *displs;
+ int rank = smpi_process_index();
+ void *tmpbuf;
+
+ /* arbitrarily choose root as rank 0 */
+ size = smpi_comm_size(comm);
+ count = 0;
+ displs = xbt_new(int, size);
+ for (i = 0; i < size; i++) {
+ displs[i] = count;
+ count += recvcounts[i];
+ }
+ tmpbuf=(void*)xbt_malloc(count*smpi_datatype_get_extent(datatype));
+ mpi_coll_reduce_fun(sendbuf, tmpbuf, count, datatype, op, 0, comm);
+ smpi_mpi_scatterv(tmpbuf, recvcounts, displs, datatype, recvbuf,
+ recvcounts[rank], datatype, 0, comm);
+ xbt_free(displs);
+ xbt_free(tmpbuf);
+}
+
void smpi_mpi_gatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
void *recvbuf, int *recvcounts, int *displs,
MPI_Datatype recvtype, int root, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_GATHERV;
int rank, size, src, index;
MPI_Aint lb = 0, recvext = 0;
MPI_Request *requests;
int recvcount, MPI_Datatype recvtype,
MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_ALLGATHER;
int rank, size, other, index;
MPI_Aint lb = 0, recvext = 0;
MPI_Request *requests;
int *recvcounts, int *displs,
MPI_Datatype recvtype, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_ALLGATHERV;
int rank, size, other, index;
MPI_Aint lb = 0, recvext = 0;
MPI_Request *requests;
void *recvbuf, int recvcount, MPI_Datatype recvtype,
int root, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_SCATTER;
int rank, size, dst, index;
MPI_Aint lb = 0, sendext = 0;
MPI_Request *requests;
MPI_Datatype sendtype, void *recvbuf, int recvcount,
MPI_Datatype recvtype, int root, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_SCATTERV;
int rank, size, dst, index;
MPI_Aint lb = 0, sendext = 0;
MPI_Request *requests;
MPI_Datatype datatype, MPI_Op op, int root,
MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = COLL_TAG_REDUCE;
int rank, size, src, index;
MPI_Aint lb = 0, dataext = 0;
MPI_Request *requests;
void **tmpbufs;
+ char* sendtmpbuf = (char*) sendbuf;
+ if( sendbuf == MPI_IN_PLACE ) {
+ sendtmpbuf = (char *)recvbuf;
+ }
+
rank = smpi_comm_rank(comm);
size = smpi_comm_size(comm);
+ //non commutative case, use a working algo from openmpi
+ if(!smpi_op_is_commute(op)){
+ smpi_coll_tuned_reduce_ompi_basic_linear(sendtmpbuf, recvbuf, count,
+ datatype, op, root, comm);
+ return;
+ }
+
if(rank != root) {
// Send buffer to root
- smpi_mpi_send(sendbuf, count, datatype, root, system_tag, comm);
+ smpi_mpi_send(sendtmpbuf, count, datatype, root, system_tag, comm);
} else {
// FIXME: check for errors
smpi_datatype_extent(datatype, &lb, &dataext);
// Local copy from root
- if (sendbuf && recvbuf)
- smpi_datatype_copy(sendbuf, count, datatype, recvbuf, count, datatype);
+ if (sendtmpbuf && recvbuf)
+ smpi_datatype_copy(sendtmpbuf, count, datatype, recvbuf, count, datatype);
// Receive buffers from senders
//TODO: make a MPI_barrier here ?
requests = xbt_new(MPI_Request, size - 1);
void smpi_mpi_scan(void *sendbuf, void *recvbuf, int count,
MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
{
- int system_tag = 666;
+ int system_tag = 888;
int rank, size, other, index;
MPI_Aint lb = 0, dataext = 0;
MPI_Request *requests;
