as some blocking ones used internally are actually implement with nonblocking+wait, use the right blocking tag in this case.
MPI_Datatype recvtype, int root, MPI_Comm comm)
{
MPI_Request request;
- Colls::igatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm, &request);
+ Colls::igatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
MPI_Datatype recvtype, int root, MPI_Comm comm)
{
MPI_Request request;
- Colls::iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, &request);
+ Colls::iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
void *recvbuf, const int *recvcounts, const int *recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm)
{
MPI_Request request;
- Colls::ialltoallw(sendbuf, sendcounts, senddisps, sendtypes, recvbuf, recvcounts, recvdisps, recvtypes, comm, &request);
+ Colls::ialltoallw(sendbuf, sendcounts, senddisps, sendtypes, recvbuf, recvcounts, recvdisps, recvtypes, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm)
{
MPI_Request request;
- Colls::igather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, &request);
+ Colls::igather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
const int *recvcounts, const int *displs, MPI_Datatype recvtype, MPI_Comm comm)
{
MPI_Request request;
- Colls::iallgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm, &request);
+ Colls::iallgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm, &request, 0);
MPI_Request* requests = request->get_nbc_requests();
int count = request->get_nbc_requests_size();
Request::waitall(count, requests, MPI_STATUS_IGNORE);
void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm)
{
MPI_Request request;
- Colls::iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, &request);
+ Colls::iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
return Coll_reduce_ompi_basic_linear::reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
}
MPI_Request request;
- Colls::ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, &request);
+ Colls::ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
void *recvbuf, const int *recvcounts, const int *recvdisps, MPI_Datatype recvtype, MPI_Comm comm)
{
MPI_Request request;
- Colls::ialltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm, &request);
+ Colls::ialltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm, &request, 0);
return Request::wait(&request, MPI_STATUS_IGNORE);
}
namespace smpi{
-int Colls::ibarrier(MPI_Comm comm, MPI_Request* request)
+int Colls::ibarrier(MPI_Comm comm, MPI_Request* request, int external)
{
int size = comm->size();
int rank = comm->rank();
+ int system_tag=COLL_TAG_BARRIER-external;
(*request) = new Request( nullptr, 0, MPI_BYTE,
- rank,rank, COLL_TAG_BARRIER, comm, MPI_REQ_PERSISTENT);
+ rank,rank, system_tag, comm, MPI_REQ_PERSISTENT);
if (rank > 0) {
MPI_Request* requests = new MPI_Request[2];
requests[0] = Request::isend (nullptr, 0, MPI_BYTE, 0,
- COLL_TAG_BARRIER,
+ system_tag,
comm);
requests[1] = Request::irecv (nullptr, 0, MPI_BYTE, 0,
- COLL_TAG_BARRIER,
+ system_tag,
comm);
(*request)->set_nbc_requests(requests, 2);
}
else {
MPI_Request* requests = new MPI_Request[(size - 1) * 2];
for (int i = 1; i < 2 * size - 1; i += 2) {
- requests[i - 1] = Request::irecv(nullptr, 0, MPI_BYTE, MPI_ANY_SOURCE, COLL_TAG_BARRIER, comm);
- requests[i] = Request::isend(nullptr, 0, MPI_BYTE, (i + 1) / 2, COLL_TAG_BARRIER, comm);
+ requests[i - 1] = Request::irecv(nullptr, 0, MPI_BYTE, MPI_ANY_SOURCE, system_tag, comm);
+ requests[i] = Request::isend(nullptr, 0, MPI_BYTE, (i + 1) / 2, system_tag, comm);
}
(*request)->set_nbc_requests(requests, 2*(size-1));
}
return MPI_SUCCESS;
}
-int Colls::ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request)
+int Colls::ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request, int external)
{
int size = comm->size();
int rank = comm->rank();
+ int system_tag=COLL_TAG_BCAST-external;
(*request) = new Request( nullptr, 0, MPI_BYTE,
- rank,rank, COLL_TAG_BCAST, comm, MPI_REQ_PERSISTENT);
+ rank,rank, system_tag, comm, MPI_REQ_PERSISTENT);
if (rank != root) {
MPI_Request* requests = new MPI_Request[1];
requests[0] = Request::irecv (buf, count, datatype, root,
- COLL_TAG_BCAST,
+ system_tag,
comm);
(*request)->set_nbc_requests(requests, 1);
}
for (int i = 0; i < size; i++) {
if(i!=root){
requests[n] = Request::isend(buf, count, datatype, i,
- COLL_TAG_BCAST,
+ system_tag,
comm
);
n++;
}
int Colls::iallgather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf,int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
+ void *recvbuf,int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external)
{
- const int system_tag = COLL_TAG_ALLGATHER;
+ const int system_tag = COLL_TAG_ALLGATHER-external;
MPI_Aint lb = 0;
MPI_Aint recvext = 0;
}
int Colls::iscatter(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request)
+ void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external)
{
- const int system_tag = COLL_TAG_SCATTER;
+ const int system_tag = COLL_TAG_SCATTER-external;
MPI_Aint lb = 0;
MPI_Aint sendext = 0;
}
int Colls::iallgatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf,
- const int *recvcounts, const int *displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
+ const int *recvcounts, const int *displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external)
{
- const int system_tag = COLL_TAG_ALLGATHERV;
+ const int system_tag = COLL_TAG_ALLGATHERV-external;
MPI_Aint lb = 0;
MPI_Aint recvext = 0;
return MPI_SUCCESS;
}
-int Colls::ialltoall( const void *sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request){
- int system_tag = COLL_TAG_ALLTOALL;
+int Colls::ialltoall( const void *sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external){
+ int system_tag = COLL_TAG_ALLTOALL-external;
MPI_Aint lb = 0;
MPI_Aint sendext = 0;
MPI_Aint recvext = 0;
}
int Colls::ialltoallv(const void *sendbuf, const int *sendcounts, const int *senddisps, MPI_Datatype sendtype,
- void *recvbuf, const int *recvcounts, const int *recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request){
- const int system_tag = COLL_TAG_ALLTOALLV;
+ void *recvbuf, const int *recvcounts, const int *recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request, int external){
+ const int system_tag = COLL_TAG_ALLTOALLV-external;
MPI_Aint lb = 0;
MPI_Aint sendext = 0;
MPI_Aint recvext = 0;
}
int Colls::ialltoallw(const void *sendbuf, const int *sendcounts, const int *senddisps, const MPI_Datatype* sendtypes,
- void *recvbuf, const int *recvcounts, const int *recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request *request){
- const int system_tag = COLL_TAG_ALLTOALLV;
+ void *recvbuf, const int *recvcounts, const int *recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request *request, int external){
+ const int system_tag = COLL_TAG_ALLTOALLW-external;
/* Initialize. */
int rank = comm->rank();
}
int Colls::igather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+ void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external)
{
- const int system_tag = COLL_TAG_GATHER;
+ const int system_tag = COLL_TAG_GATHER-external;
MPI_Aint lb = 0;
MPI_Aint recvext = 0;
}
int Colls::igatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int *recvcounts, const int *displs,
- MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+ MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external)
{
- int system_tag = COLL_TAG_GATHERV;
+ int system_tag = COLL_TAG_GATHERV-external;
MPI_Aint lb = 0;
MPI_Aint recvext = 0;
return MPI_SUCCESS;
}
int Colls::iscatterv(const void *sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype, void *recvbuf, int recvcount,
- MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+ MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external)
{
- int system_tag = COLL_TAG_SCATTERV;
+ int system_tag = COLL_TAG_SCATTERV-external;
MPI_Aint lb = 0;
MPI_Aint sendext = 0;
}
int Colls::ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root,
- MPI_Comm comm, MPI_Request* request)
+ MPI_Comm comm, MPI_Request* request, int external)
{
- const int system_tag = COLL_TAG_REDUCE;
+ const int system_tag = COLL_TAG_REDUCE-external;
MPI_Aint lb = 0;
MPI_Aint dataext = 0;
}
int Colls::iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
- MPI_Op op, MPI_Comm comm, MPI_Request* request)
+ MPI_Op op, MPI_Comm comm, MPI_Request* request, int external)
{
- const int system_tag = COLL_TAG_ALLREDUCE;
+ const int system_tag = COLL_TAG_ALLREDUCE-external;
MPI_Aint lb = 0;
MPI_Aint dataext = 0;
return MPI_SUCCESS;
}
-int Colls::iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request)
+int Colls::iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request, int external)
{
- int system_tag = -888;
+ int system_tag = -888-external;
MPI_Aint lb = 0;
MPI_Aint dataext = 0;
return MPI_SUCCESS;
}
-int Colls::iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request)
+int Colls::iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request, int external)
{
- int system_tag = -888;
+ int system_tag = -888-external;
MPI_Aint lb = 0;
MPI_Aint dataext = 0;
int rank = comm->rank();
}
int Colls::ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcounts, MPI_Datatype datatype, MPI_Op op,
- MPI_Comm comm, MPI_Request* request){
+ MPI_Comm comm, MPI_Request* request, int external){
//Version where each process performs the reduce for its own part. Alltoall pattern for comms.
- const int system_tag = COLL_TAG_REDUCE_SCATTER;
+ const int system_tag = COLL_TAG_REDUCE_SCATTER-external;
MPI_Aint lb = 0;
MPI_Aint dataext = 0;
constexpr int COLL_TAG_BARRIER = -778;
constexpr int COLL_TAG_REDUCE_SCATTER = -889;
constexpr int COLL_TAG_ALLTOALLV = -1000;
+constexpr int COLL_TAG_ALLTOALLW = -1020;
constexpr int COLL_TAG_ALLTOALL = -1112;
constexpr int COLL_TAG_GATHERV = -2223;
constexpr int COLL_TAG_BCAST = -3334;
constexpr int COLL_TAG_ALLREDUCE = -4445;
+
// SMPI_RMA_TAG has to be the smallest one, as it will be decremented for accumulate ordering.
constexpr int SMPI_RMA_TAG = -6666;
const int* recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm);
//async collectives
- static int ibarrier(MPI_Comm comm, MPI_Request* request);
+ static int ibarrier(MPI_Comm comm, MPI_Request* request, int external=1);
static int ibcast(void *buf, int count, MPI_Datatype datatype,
- int root, MPI_Comm comm, MPI_Request* request);
+ int root, MPI_Comm comm, MPI_Request* request, int external=1);
static int igather (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
- MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request);
+ MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external=1);
static int igatherv (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
- const int* recvcounts, const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request);
+ const int* recvcounts, const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external=1);
static int iallgather (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
- int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request);
+ int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request, int external=1);
static int iallgatherv (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
- const int* recvcounts, const int* displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request);
+ const int* recvcounts, const int* displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request, int external=1);
static int iscatter (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
- int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request);
+ int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external=1);
static int iscatterv (const void* sendbuf, const int* sendcounts, const int* displs, MPI_Datatype sendtype,
- void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request);
+ void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request, int external=1);
static int ireduce
- (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, MPI_Request *request, int external=1);
static int iallreduce
- (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request, int external=1);
static int iscan
- (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request, int external=1);
static int iexscan
- (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request, int external=1);
static int ireduce_scatter
- (const void* sendbuf, void* recvbuf, const int* recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, const int* recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request, int external=1);
static int ireduce_scatter_block
- (const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request);
+ (const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request, int external=1);
static int ialltoall (const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
- int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request);
+ int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request, int external=1);
static int ialltoallv
(const void* sendbuf, const int* sendcounts, const int* senddisps, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts,
- const int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request);
+ const int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request, int external=1);
static int ialltoallw
(const void* sendbuf, const int* sendcounts, const int* senddisps, const MPI_Datatype* sendtypes, void* recvbuf, const int* recvcounts,
- const int* recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request *request);
+ const int* recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request *request, int external=1);
static void (*smpi_coll_cleanup_callback)();