int mca_pml_csum_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
int rc;
mca_pml_csum_recv_request_t *recvreq;
MCA_PML_CSUM_RECV_REQUEST_ALLOC(recvreq, rc);
if (NULL == recvreq)
return rc;
MCA_PML_CSUM_RECV_REQUEST_INIT(recvreq,
addr,
count, datatype, src, tag, comm, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&((recvreq)->req_recv.req_base),
PERUSE_RECV);
MCA_PML_CSUM_RECV_REQUEST_START(recvreq);
ompi_request_wait_completion(&recvreq->req_recv.req_base.req_ompi);
if (NULL != status) { /* return status */
OMPI_STATUS_SET(status, &recvreq->req_recv.req_base.req_ompi.req_status);
}
rc = recvreq->req_recv.req_base.req_ompi.req_status.MPI_ERROR;
ompi_request_free( (ompi_request_t**)&recvreq );
return rc;
}
/**
* Hierarchical blocking barrier
*/
int mca_coll_ml_barrier_intra(struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
int rc;
ompi_request_t *req;
mca_coll_ml_module_t *ml_module = (mca_coll_ml_module_t *) module;
#if OPAL_ENABLE_DEBUG
static int barriers_count = 0;
#endif
ML_VERBOSE(10, ("Barrier num %d start.", ++barriers_count));
rc = mca_coll_ml_barrier_launch(ml_module, &req);
if (OPAL_UNLIKELY(rc != OMPI_SUCCESS)) {
ML_ERROR(("Failed to launch a barrier."));
return rc;
}
/* Blocking barrier */
ompi_request_wait_completion(req);
ompi_request_free(&req);
ML_VERBOSE(10, ("Barrier num %d was done.", barriers_count));
return OMPI_SUCCESS;
}
int
mca_pml_ob1_mprobe(int src,
int tag,
struct ompi_communicator_t *comm,
struct ompi_message_t **message,
ompi_status_public_t * status)
{
int rc = OMPI_SUCCESS;
mca_pml_ob1_recv_request_t *recvreq;
MCA_PML_OB1_RECV_REQUEST_ALLOC(recvreq, rc);
if (NULL == recvreq)
return rc;
recvreq->req_recv.req_base.req_type = MCA_PML_REQUEST_MPROBE;
/* initialize the request enough to probe and get the status */
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, NULL, 0, &ompi_mpi_char.dt,
src, tag, comm, false);
MCA_PML_OB1_RECV_REQUEST_START(recvreq);
ompi_request_wait_completion(&recvreq->req_recv.req_base.req_ompi);
if( NULL != status ) {
*status = recvreq->req_recv.req_base.req_ompi.req_status;
}
*message = ompi_message_alloc();
(*message)->comm = comm;
(*message)->req_ptr = recvreq;
(*message)->count = recvreq->req_recv.req_base.req_ompi.req_status._ucount;
return OMPI_SUCCESS;
}
int mca_pml_ob1_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
mca_pml_ob1_recv_request_t recvreq;
int rc;
OBJ_CONSTRUCT(&recvreq, mca_pml_ob1_recv_request_t);
MCA_PML_OB1_RECV_REQUEST_INIT(&recvreq, addr, count, datatype,
src, tag, comm, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(recvreq.req_recv.req_base),
PERUSE_RECV);
MCA_PML_OB1_RECV_REQUEST_START(&recvreq);
ompi_request_wait_completion(&recvreq.req_recv.req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq.req_recv.req_base.req_ompi.req_status;
}
rc = recvreq.req_recv.req_base.req_ompi.req_status.MPI_ERROR;
MCA_PML_BASE_RECV_REQUEST_FINI(&recvreq.req_recv);
OBJ_DESTRUCT(&recvreq);
return rc;
}
int mca_pml_ob1_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
mca_pml_ob1_recv_request_t *recvreq = NULL;
int rc;
#if !OMPI_ENABLE_THREAD_MULTIPLE
recvreq = mca_pml_ob1_recvreq;
mca_pml_ob1_recvreq = NULL;
if( OPAL_UNLIKELY(NULL == recvreq) )
#endif /* !OMPI_ENABLE_THREAD_MULTIPLE */
{
MCA_PML_OB1_RECV_REQUEST_ALLOC(recvreq);
if (NULL == recvreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, addr, count, datatype,
src, tag, comm, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(recvreq->req_recv.req_base),
PERUSE_RECV);
MCA_PML_OB1_RECV_REQUEST_START(recvreq);
ompi_request_wait_completion(&recvreq->req_recv.req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq->req_recv.req_base.req_ompi.req_status;
}
rc = recvreq->req_recv.req_base.req_ompi.req_status.MPI_ERROR;
#if OMPI_ENABLE_THREAD_MULTIPLE
MCA_PML_OB1_RECV_REQUEST_RETURN(recvreq);
#else
if( NULL != mca_pml_ob1_recvreq ) {
MCA_PML_OB1_RECV_REQUEST_RETURN(recvreq);
} else {
mca_pml_ob1_recv_request_fini (recvreq);
mca_pml_ob1_recvreq = recvreq;
}
#endif
return rc;
}
int mca_pml_ob1_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
mca_pml_ob1_recv_request_t *recvreq = NULL;
int rc;
if (OPAL_LIKELY(!ompi_mpi_thread_multiple)) {
recvreq = mca_pml_ob1_recvreq;
mca_pml_ob1_recvreq = NULL;
}
if( OPAL_UNLIKELY(NULL == recvreq) ) {
MCA_PML_OB1_RECV_REQUEST_ALLOC(recvreq);
if (NULL == recvreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
recvreq->req_recv.req_base.req_type = MCA_PML_REQUEST_RECV;
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, addr, count, datatype,
src, tag, comm, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(recvreq->req_recv.req_base),
PERUSE_RECV);
MCA_PML_OB1_RECV_REQUEST_START(recvreq);
ompi_request_wait_completion(&recvreq->req_recv.req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq->req_recv.req_base.req_ompi.req_status;
}
rc = recvreq->req_recv.req_base.req_ompi.req_status.MPI_ERROR;
if (recvreq->req_recv.req_base.req_pml_complete) {
/* make buffer defined when the request is compeleted,
and before releasing the objects. */
MEMCHECKER(
memchecker_call(&opal_memchecker_base_mem_defined,
recvreq->req_recv.req_base.req_addr,
recvreq->req_recv.req_base.req_count,
recvreq->req_recv.req_base.req_datatype);
);
int mca_pml_ob1_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
mca_pml_ob1_recv_request_t *recvreq = NULL;
int rc;
if (OPAL_LIKELY(!ompi_mpi_thread_multiple)) {
recvreq = mca_pml_ob1_recvreq;
mca_pml_ob1_recvreq = NULL;
}
if( OPAL_UNLIKELY(NULL == recvreq) ) {
MCA_PML_OB1_RECV_REQUEST_ALLOC(recvreq);
if (NULL == recvreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
recvreq->req_recv.req_base.req_type = MCA_PML_REQUEST_RECV;
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, addr, count, datatype,
src, tag, comm, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(recvreq->req_recv.req_base),
PERUSE_RECV);
MCA_PML_OB1_RECV_REQUEST_START(recvreq);
ompi_request_wait_completion(&recvreq->req_recv.req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq->req_recv.req_base.req_ompi.req_status;
}
rc = recvreq->req_recv.req_base.req_ompi.req_status.MPI_ERROR;
if (OPAL_UNLIKELY(ompi_mpi_thread_multiple || NULL != mca_pml_ob1_recvreq)) {
MCA_PML_OB1_RECV_REQUEST_RETURN(recvreq);
} else {
mca_pml_ob1_recv_request_fini (recvreq);
mca_pml_ob1_recvreq = recvreq;
}
return rc;
}
int ompi_request_default_wait(
ompi_request_t ** req_ptr,
ompi_status_public_t * status)
{
ompi_request_t *req = *req_ptr;
ompi_request_wait_completion(req);
#if OPAL_ENABLE_FT_CR == 1
OMPI_CRCP_REQUEST_COMPLETE(req);
#endif
/* return status. If it's a generalized request, we *have* to
invoke the query_fn, even if the user procided STATUS_IGNORE.
MPI-2:8.2. */
if (OMPI_REQUEST_GEN == req->req_type) {
ompi_grequest_invoke_query(req, &req->req_status);
}
if( MPI_STATUS_IGNORE != status ) {
/* Do *NOT* set status->MPI_ERROR here! See MPI-1.1 doc, sec
3.2.5, p.22 */
status->MPI_TAG = req->req_status.MPI_TAG;
status->MPI_SOURCE = req->req_status.MPI_SOURCE;
status->_ucount = req->req_status._ucount;
status->_cancelled = req->req_status._cancelled;
}
if( req->req_persistent ) {
if( req->req_state == OMPI_REQUEST_INACTIVE ) {
if (MPI_STATUS_IGNORE != status) {
*status = ompi_status_empty;
}
return OMPI_SUCCESS;
}
req->req_state = OMPI_REQUEST_INACTIVE;
return req->req_status.MPI_ERROR;
}
/* If there was an error, don't free the request -- just return
the single error. */
if (MPI_SUCCESS != req->req_status.MPI_ERROR) {
return req->req_status.MPI_ERROR;
}
/* If there's an error while freeing the request, assume that the
request is still there. Otherwise, Bad Things will happen
later! */
return ompi_request_free(req_ptr);
}
int ompi_comm_activate (ompi_communicator_t **newcomm, ompi_communicator_t *comm,
ompi_communicator_t *bridgecomm, const void *arg0,
const void *arg1, bool send_first, int mode)
{
ompi_request_t *req;
int rc;
rc = ompi_comm_activate_nb (newcomm, comm, bridgecomm, arg0, arg1, send_first, mode, &req);
if (OMPI_SUCCESS != rc) {
return rc;
}
ompi_request_wait_completion (req);
rc = req->req_status.MPI_ERROR;
ompi_comm_request_return ((ompi_comm_request_t *) req);
return rc;
}
int
mca_pml_cm_recv(void *addr,
size_t count,
ompi_datatype_t * datatype,
int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
int ret;
mca_pml_cm_thin_recv_request_t *recvreq;
ompi_proc_t* ompi_proc;
MCA_PML_CM_THIN_RECV_REQUEST_ALLOC(recvreq, ret);
if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) return ret;
MCA_PML_CM_THIN_RECV_REQUEST_INIT(recvreq,
ompi_proc,
comm,
src,
datatype,
addr,
count);
MCA_PML_CM_THIN_RECV_REQUEST_START(recvreq, comm, tag, src, ret);
if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) {
/* BWB - XXX - need cleanup of request here */
MCA_PML_CM_THIN_RECV_REQUEST_RETURN(recvreq);
return ret;
}
ompi_request_wait_completion(&recvreq->req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq->req_base.req_ompi.req_status;
}
ret = recvreq->req_base.req_ompi.req_status.MPI_ERROR;
ompi_request_free( (ompi_request_t**)&recvreq );
return ret;
}
int
mca_pml_cm_mrecv(void *buf,
size_t count,
ompi_datatype_t *datatype,
struct ompi_message_t **message,
ompi_status_public_t* status)
{
int ret;
mca_pml_cm_thin_recv_request_t *recvreq;
ompi_proc_t* ompi_proc;
ompi_communicator_t *comm = (*message)->comm;
int peer = (*message)->peer;
MCA_PML_CM_THIN_RECV_REQUEST_ALLOC(recvreq, ret);
if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) return ret;
MCA_PML_CM_THIN_RECV_REQUEST_INIT(recvreq,
ompi_proc,
comm,
peer,
datatype,
buf,
count);
MCA_PML_CM_THIN_RECV_REQUEST_MATCHED_START(recvreq,
message, ret);
if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) {
MCA_PML_CM_THIN_RECV_REQUEST_RETURN(recvreq);
return ret;
}
ompi_request_wait_completion(&recvreq->req_base.req_ompi);
if (NULL != status) { /* return status */
*status = recvreq->req_base.req_ompi.req_status;
}
ret = recvreq->req_base.req_ompi.req_status.MPI_ERROR;
ompi_request_free( (ompi_request_t**)&recvreq );
return ret;
}
int mca_pml_ob1_probe(int src,
int tag,
struct ompi_communicator_t *comm,
ompi_status_public_t * status)
{
mca_pml_ob1_recv_request_t recvreq;
OBJ_CONSTRUCT( &recvreq, mca_pml_ob1_recv_request_t );
recvreq.req_recv.req_base.req_ompi.req_type = OMPI_REQUEST_PML;
recvreq.req_recv.req_base.req_type = MCA_PML_REQUEST_PROBE;
MCA_PML_OB1_RECV_REQUEST_INIT(&recvreq, NULL, 0, &ompi_mpi_char.dt, src, tag, comm, true);
MCA_PML_OB1_RECV_REQUEST_START(&recvreq);
ompi_request_wait_completion(&recvreq.req_recv.req_base.req_ompi);
if (NULL != status) {
OMPI_STATUS_SET(status, &recvreq.req_recv.req_base.req_ompi.req_status);
}
MCA_PML_BASE_RECV_REQUEST_FINI( &recvreq.req_recv );
return OMPI_SUCCESS;
}
int mca_pml_ob1_send(const void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm)
{
mca_pml_ob1_comm_proc_t *ob1_proc = mca_pml_ob1_peer_lookup (comm, dst);
ompi_proc_t *dst_proc = ob1_proc->ompi_proc;
mca_bml_base_endpoint_t* endpoint = mca_bml_base_get_endpoint (dst_proc);
mca_pml_ob1_send_request_t *sendreq = NULL;
int16_t seqn;
int rc;
if (OPAL_UNLIKELY(NULL == endpoint)) {
return OMPI_ERR_UNREACH;
}
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_BUFFERED == sendmode)) {
/* large buffered sends *need* a real request so use isend instead */
ompi_request_t *brequest;
rc = mca_pml_ob1_isend (buf, count, datatype, dst, tag, sendmode, comm, &brequest);
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
return rc;
}
ompi_request_wait_completion (brequest);
ompi_request_free (&brequest);
return OMPI_SUCCESS;
}
seqn = (uint16_t) OPAL_THREAD_ADD32(&ob1_proc->send_sequence, 1);
/**
* The immediate send will not have a request, so they are
* intracable from the point of view of any debugger attached to
* the parallel application.
*/
if (MCA_PML_BASE_SEND_SYNCHRONOUS != sendmode) {
rc = mca_pml_ob1_send_inline (buf, count, datatype, dst, tag, seqn, dst_proc,
endpoint, comm);
if (OPAL_LIKELY(0 <= rc)) {
return OMPI_SUCCESS;
}
}
if (OPAL_LIKELY(!ompi_mpi_thread_multiple)) {
sendreq = mca_pml_ob1_sendreq;
mca_pml_ob1_sendreq = NULL;
}
if( OPAL_UNLIKELY(NULL == sendreq) ) {
MCA_PML_OB1_SEND_REQUEST_ALLOC(comm, dst, sendreq);
if (NULL == sendreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
sendreq->req_send.req_base.req_proc = dst_proc;
sendreq->rdma_frag = NULL;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&sendreq->req_send.req_base,
PERUSE_SEND);
MCA_PML_OB1_SEND_REQUEST_START_W_SEQ(sendreq, endpoint, seqn, rc);
if (OPAL_LIKELY(rc == OMPI_SUCCESS)) {
ompi_request_wait_completion(&sendreq->req_send.req_base.req_ompi);
rc = sendreq->req_send.req_base.req_ompi.req_status.MPI_ERROR;
}
if (OPAL_UNLIKELY(ompi_mpi_thread_multiple || NULL != mca_pml_ob1_sendreq)) {
MCA_PML_OB1_SEND_REQUEST_RETURN(sendreq);
} else {
mca_pml_ob1_send_request_fini (sendreq);
mca_pml_ob1_sendreq = sendreq;
}
return rc;
}
int
mca_pml_bfo_mrecv( void *buf,
size_t count,
ompi_datatype_t *datatype,
struct ompi_message_t **message,
ompi_status_public_t* status )
{
mca_pml_bfo_recv_frag_t* frag;
mca_pml_bfo_recv_request_t *recvreq;
mca_pml_bfo_hdr_t *hdr;
int src, tag, rc;
ompi_communicator_t *comm;
mca_pml_bfo_comm_proc_t* proc;
mca_pml_bfo_comm_t* bfo_comm;
uint64_t seq;
/* get the request from the message and the frag from the request
before we overwrite everything */
comm = (*message)->comm;
recvreq = (mca_pml_bfo_recv_request_t*) (*message)->req_ptr;
frag = (mca_pml_bfo_recv_frag_t*) recvreq->req_recv.req_base.req_addr;
src = recvreq->req_recv.req_base.req_ompi.req_status.MPI_SOURCE;
tag = recvreq->req_recv.req_base.req_ompi.req_status.MPI_TAG;
seq = recvreq->req_recv.req_base.req_sequence;
bfo_comm = recvreq->req_recv.req_base.req_comm->c_pml_comm;
/* make the request a recv request again */
/* The old request kept pointers to comm and the char datatype.
We're about to release those, but need to make sure comm
doesn't go out of scope (we don't care about the char datatype
anymore). So retain comm, then release the frag, then reinit
the frag (which will retain comm), then release comm (but the
frag still has it's ref, so it'll stay in scope). Make
sense? */
OBJ_RETAIN(comm);
MCA_PML_BASE_RECV_REQUEST_FINI(&recvreq->req_recv);
recvreq->req_recv.req_base.req_type = MCA_PML_REQUEST_RECV;
MCA_PML_BFO_RECV_REQUEST_INIT(recvreq,
buf,
count, datatype,
src, tag, comm, false);
OBJ_RELEASE(comm);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&((recvreq)->req_recv.req_base),
PERUSE_RECV);
/* init/re-init the request */
recvreq->req_lock = 0;
recvreq->req_pipeline_depth = 0;
recvreq->req_bytes_received = 0;
recvreq->req_rdma_cnt = 0;
recvreq->req_rdma_idx = 0;
recvreq->req_pending = false;
MCA_PML_BASE_RECV_START(&recvreq->req_recv.req_base);
/* Note - sequence number already assigned */
recvreq->req_recv.req_base.req_sequence = seq;
proc = &bfo_comm->procs[recvreq->req_recv.req_base.req_peer];
recvreq->req_recv.req_base.req_proc = proc->ompi_proc;
prepare_recv_req_converter(recvreq);
/* we can't go through the match, since we already have the match.
Cheat and do what REQUEST_START does, but without the frag
search */
hdr = (mca_pml_bfo_hdr_t*)frag->segments->seg_addr.pval;
switch(hdr->hdr_common.hdr_type) {
case MCA_PML_BFO_HDR_TYPE_MATCH:
mca_pml_bfo_recv_request_progress_match(recvreq, frag->btl, frag->segments,
frag->num_segments);
break;
case MCA_PML_BFO_HDR_TYPE_RNDV:
mca_pml_bfo_recv_request_progress_rndv(recvreq, frag->btl, frag->segments,
frag->num_segments);
break;
case MCA_PML_BFO_HDR_TYPE_RGET:
mca_pml_bfo_recv_request_progress_rget(recvreq, frag->btl, frag->segments,
frag->num_segments);
break;
default:
assert(0);
}
ompi_message_return(*message);
*message = MPI_MESSAGE_NULL;
ompi_request_wait_completion(&(recvreq->req_recv.req_base.req_ompi));
MCA_PML_BFO_RECV_FRAG_RETURN(frag);
if (NULL != status) { /* return status */
*status = recvreq->req_recv.req_base.req_ompi.req_status;
}
rc = recvreq->req_recv.req_base.req_ompi.req_status.MPI_ERROR;
ompi_request_free( (ompi_request_t**)&recvreq );
return rc;
}
int mca_pml_ob1_send(void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm)
{
mca_pml_ob1_comm_t* ob1_comm = comm->c_pml_comm;
ompi_proc_t *dst_proc = ompi_comm_peer_lookup (comm, dst);
mca_bml_base_endpoint_t* endpoint = (mca_bml_base_endpoint_t*)
dst_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML];
mca_pml_ob1_send_request_t *sendreq =
alloca(mca_pml_base_send_requests.fl_frag_size);
int16_t seqn;
int rc;
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_BUFFERED == sendmode)) {
/* large buffered sends *need* a real request so use isend instead */
ompi_request_t *brequest;
rc = mca_pml_ob1_isend (buf, count, datatype, dst, tag, sendmode, comm, &brequest);
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
return rc;
}
/* free the request and return. don't care if it completes now */
ompi_request_free (&brequest);
return OMPI_SUCCESS;
}
if (OPAL_UNLIKELY(NULL == endpoint)) {
return OMPI_ERR_UNREACH;
}
seqn = (uint16_t) OPAL_THREAD_ADD32(&ob1_comm->procs[dst].send_sequence, 1);
if (MCA_PML_BASE_SEND_SYNCHRONOUS != sendmode) {
rc = mca_pml_ob1_send_inline (buf, count, datatype, dst, tag, seqn, dst_proc,
endpoint, comm);
if (OPAL_LIKELY(0 <= rc)) {
return OMPI_SUCCESS;
}
}
OBJ_CONSTRUCT(sendreq, mca_pml_ob1_send_request_t);
sendreq->req_send.req_base.req_proc = dst_proc;
sendreq->src_des = NULL;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&sendreq->req_send.req_base,
PERUSE_SEND);
MCA_PML_OB1_SEND_REQUEST_START_W_SEQ(sendreq, endpoint, seqn, rc);
if (rc != OMPI_SUCCESS) {
return rc;
}
ompi_request_wait_completion(&sendreq->req_send.req_base.req_ompi);
rc = sendreq->req_send.req_base.req_ompi.req_status.MPI_ERROR;
MCA_PML_BASE_SEND_REQUEST_FINI(&sendreq->req_send);
OBJ_DESTRUCT(sendreq);
return rc;
}
int mca_pml_ob1_send(void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm)
{
mca_pml_ob1_comm_t* ob1_comm = comm->c_pml_comm;
ompi_proc_t *dst_proc = ompi_comm_peer_lookup (comm, dst);
mca_bml_base_endpoint_t* endpoint = (mca_bml_base_endpoint_t*)
dst_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML];
mca_pml_ob1_send_request_t *sendreq = NULL;
int16_t seqn;
int rc;
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_BUFFERED == sendmode)) {
/* large buffered sends *need* a real request so use isend instead */
ompi_request_t *brequest;
rc = mca_pml_ob1_isend (buf, count, datatype, dst, tag, sendmode, comm, &brequest);
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
return rc;
}
/* free the request and return. don't care if it completes now */
ompi_request_free (&brequest);
return OMPI_SUCCESS;
}
if (OPAL_UNLIKELY(NULL == endpoint)) {
return OMPI_ERR_UNREACH;
}
seqn = (uint16_t) OPAL_THREAD_ADD32(&ob1_comm->procs[dst].send_sequence, 1);
/**
* The immediate send will not have a request, so they are
* intracable from the point of view of any debugger attached to
* the parallel application.
*/
if (MCA_PML_BASE_SEND_SYNCHRONOUS != sendmode) {
rc = mca_pml_ob1_send_inline (buf, count, datatype, dst, tag, seqn, dst_proc,
endpoint, comm);
if (OPAL_LIKELY(0 <= rc)) {
return OMPI_SUCCESS;
}
}
#if !OMPI_ENABLE_THREAD_MULTIPLE
sendreq = mca_pml_ob1_sendreq;
if( OPAL_UNLIKELY(NULL == sendreq) )
#endif /* !OMPI_ENABLE_THREAD_MULTIPLE */
{
MCA_PML_OB1_SEND_REQUEST_ALLOC(comm, dst, sendreq);
if (NULL == sendreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
#if !OMPI_ENABLE_THREAD_MULTIPLE
mca_pml_ob1_sendreq = sendreq;
#endif /* !OMPI_ENABLE_THREAD_MULTIPLE */
}
sendreq->req_send.req_base.req_proc = dst_proc;
sendreq->rdma_frag = NULL;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&sendreq->req_send.req_base,
PERUSE_SEND);
MCA_PML_OB1_SEND_REQUEST_START_W_SEQ(sendreq, endpoint, seqn, rc);
if (OPAL_LIKELY(rc == OMPI_SUCCESS)) {
ompi_request_wait_completion(&sendreq->req_send.req_base.req_ompi);
rc = sendreq->req_send.req_base.req_ompi.req_status.MPI_ERROR;
}
#if OMPI_ENABLE_THREAD_MULTIPLE
MCA_PML_OB1_SEND_REQUEST_RETURN(sendreq);
#else
mca_pml_ob1_send_request_fini (sendreq);
#endif
return rc;
}
