int ADD_SUFFIX(MPID_nem_ofi_iprobe_impl)(struct MPIDI_VC *vc,
int source,
int tag,
MPIR_Comm * comm,
int context_offset,
int *flag, MPI_Status * status, MPIR_Request ** rreq_ptr)
{
int ret, mpi_errno = MPI_SUCCESS;
fi_addr_t remote_proc = 0;
uint64_t match_bits, mask_bits;
size_t len;
MPIR_Request rreq_s, *rreq;
BEGIN_FUNC(FCNAME);
if (rreq_ptr) {
MPIDI_CH3I_NM_OFI_RC(MPID_nem_ofi_create_req(&rreq, 1));
rreq->kind = MPIR_REQUEST_KIND__RECV;
*rreq_ptr = rreq;
rreq->comm = comm;
rreq->dev.match.parts.rank = source;
rreq->dev.match.parts.tag = tag;
rreq->dev.match.parts.context_id = comm->context_id;
MPIR_Comm_add_ref(comm);
}
else {
rreq = &rreq_s;
rreq->dev.OnDataAvail = NULL;
}
REQ_OFI(rreq)->pack_buffer = NULL;
REQ_OFI(rreq)->event_callback = ADD_SUFFIX(peek_callback);
REQ_OFI(rreq)->match_state = PEEK_INIT;
OFI_ADDR_INIT(source, vc, remote_proc);
#if API_SET == API_SET_1
match_bits = init_recvtag(&mask_bits, comm->context_id + context_offset, source, tag);
#elif API_SET == API_SET_2
match_bits = init_recvtag_2(&mask_bits, comm->context_id + context_offset, tag);
#endif
/* ------------------------------------------------------------------------- */
/* fi_recvmsg with FI_PEEK: */
/* Initiate a search for a match in the hardware or software queue. */
/* The search can complete immediately with -ENOMSG. */
/* I successful, libfabric will enqueue a context entry into the completion */
/* queue to make the search nonblocking. This code will poll until the */
/* entry is enqueued. */
/* ------------------------------------------------------------------------- */
msg_tagged_t msg;
uint64_t msgflags = FI_PEEK;
msg.msg_iov = NULL;
msg.desc = NULL;
msg.iov_count = 0;
msg.addr = remote_proc;
msg.tag = match_bits;
msg.ignore = mask_bits;
msg.context = (void *) &(REQ_OFI(rreq)->ofi_context);
msg.data = 0;
if(*flag == CLAIM_PEEK)
msgflags|=FI_CLAIM;
ret = fi_trecvmsg(gl_data.endpoint,&msg,msgflags);
if(ret == -ENOMSG) {
if (rreq_ptr) {
MPIR_Request_free(rreq);
*rreq_ptr = NULL;
*flag = 0;
}
MPID_nem_ofi_poll(MPID_NONBLOCKING_POLL);
goto fn_exit;
}
MPIR_ERR_CHKANDJUMP4((ret < 0), mpi_errno, MPI_ERR_OTHER,
"**ofi_peek", "**ofi_peek %s %d %s %s",
__SHORT_FILE__, __LINE__, FCNAME, fi_strerror(-ret));
while (PEEK_INIT == REQ_OFI(rreq)->match_state)
MPID_nem_ofi_poll(MPID_BLOCKING_POLL);
if (PEEK_NOT_FOUND == REQ_OFI(rreq)->match_state) {
if (rreq_ptr) {
MPIR_Request_free(rreq);
*rreq_ptr = NULL;
*flag = 0;
}
MPID_nem_ofi_poll(MPID_NONBLOCKING_POLL);
goto fn_exit;
}
if (status != MPI_STATUS_IGNORE)
*status = rreq->status;
MPIR_Request_add_ref(rreq);
*flag = 1;
END_FUNC_RC(FCNAME);
}
int MPID_nem_ofi_poll(int in_blocking_poll)
{
int complete = 0, mpi_errno = MPI_SUCCESS;
ssize_t ret;
cq_tagged_entry_t wc;
cq_err_entry_t error;
MPIDI_VC_t *vc;
MPID_Request *req;
req_fn reqFn;
BEGIN_FUNC(FCNAME);
do {
/* ----------------------------------------------------- */
/* Poll the completion queue */
/* The strategy here is */
/* ret>0 successfull poll, events returned */
/* ret==0 empty poll, no events/no error */
/* ret<0, error, but some error instances should not */
/* cause MPI to terminate */
/* ----------------------------------------------------- */
ret = fi_cq_read(gl_data.cq, /* Tagged completion queue */
(void *) &wc, /* OUT: Tagged completion entry */
1); /* Number of entries to poll */
if (ret > 0) {
if (NULL != wc.op_context) {
req = context_to_req(wc.op_context);
if (REQ_OFI(req)->event_callback) {
MPIDI_CH3I_NM_OFI_RC(REQ_OFI(req)->event_callback(&wc, req));
continue;
}
reqFn = req->dev.OnDataAvail;
if (reqFn) {
if (REQ_OFI(req)->pack_buffer) {
MPIU_Free(REQ_OFI(req)->pack_buffer);
}
vc = REQ_OFI(req)->vc;
complete = 0;
MPIDI_CH3I_NM_OFI_RC(reqFn(vc, req, &complete));
continue;
}
else {
MPIU_Assert(0);
}
}
else {
MPIU_Assert(0);
}
}
else if (ret == -FI_EAGAIN)
;
else if (ret < 0) {
if (ret == -FI_EAVAIL) {
ret = fi_cq_readerr(gl_data.cq, (void *) &error, 0);
if (error.err == FI_ETRUNC) {
/* ----------------------------------------------------- */
/* This error message should only be delivered on send */
/* events. We want to ignore truncation errors */
/* on the sender side, but complete the request anyway */
/* Other kinds of requests, this is fatal. */
/* ----------------------------------------------------- */
req = context_to_req(error.op_context);
if (req->kind == MPID_REQUEST_SEND) {
mpi_errno = REQ_OFI(req)->event_callback(NULL, req);
}
else if (req->kind == MPID_REQUEST_RECV) {
mpi_errno = REQ_OFI(req)->event_callback(&wc, req);
req->status.MPI_ERROR = MPI_ERR_TRUNCATE;
req->status.MPI_TAG = error.tag;
}
else {
mpi_errno = MPI_ERR_OTHER;
}
}
else if (error.err == FI_ECANCELED) {
req = context_to_req(error.op_context);
MPIR_STATUS_SET_CANCEL_BIT(req->status, TRUE);
}
else {
mpi_errno = MPI_ERR_OTHER;
}
}
else {
MPIR_ERR_CHKANDJUMP4(1, mpi_errno, MPI_ERR_OTHER, "**ofi_poll",
"**ofi_poll %s %d %s %s", __SHORT_FILE__,
__LINE__, FCNAME, fi_strerror(-ret));
}
}
} while (in_blocking_poll && (ret > 0));
END_FUNC_RC(FCNAME);
}