prov/efa: Remove peer lookup from efa_rdm_pke_sendv()

Now that all handshake requests, and responses use a txe, we can rely on
all send packets having a txe and we no longer need to look up the peer
in the send path.  This is an optimization that will make all EFA sends
faster.

Signed-off-by: Seth Zegelstein <szegel@amazon.com>

prov/efa/src/rdm/efa_rdm_pke.c

@@ -390,17 +390,18 @@ ssize_t efa_rdm_pke_sendv(struct efa_rdm_pke **pkt_entry_vec,
 	struct efa_conn *conn;
 	struct efa_rdm_ep *ep;
 	struct efa_rdm_pke *pkt_entry;
-	struct efa_rdm_peer *peer;
+	struct efa_rdm_ope *txe;
 	struct ibv_sge sg_list[2];  /* efa device support up to 2 iov */
 	struct ibv_data_buf inline_data_list[2];
 	int ret, pkt_idx, iov_cnt;
 
 	assert(pkt_entry_cnt);
 	ep = pkt_entry_vec[0]->ep;
 	assert(ep);
-	peer = efa_rdm_ep_get_peer(ep, pkt_entry_vec[0]->addr);
-	assert(peer);
-	if (peer->flags & EFA_RDM_PEER_IN_BACKOFF)
+
+	txe = pkt_entry_vec[0]->ope;
+	assert(txe);
+	if (txe->peer->flags & EFA_RDM_PEER_IN_BACKOFF)
 		return -FI_EAGAIN;
 
 	conn = efa_av_addr_to_conn(ep->base_ep.av, pkt_entry_vec[0]->addr);
@@ -411,7 +412,7 @@ ssize_t efa_rdm_pke_sendv(struct efa_rdm_pke **pkt_entry_vec,
 	for (pkt_idx = 0; pkt_idx < pkt_entry_cnt; ++pkt_idx) {
 		pkt_entry = pkt_entry_vec[pkt_idx];
 		assert(pkt_entry->pkt_size);
-		assert(efa_rdm_ep_get_peer(ep, pkt_entry->addr) == peer);
+		assert(efa_rdm_ep_get_peer(ep, pkt_entry->addr) == txe->peer);
 
 		qp->ibv_qp_ex->wr_id = (uintptr_t)pkt_entry;
 		ibv_wr_send(qp->ibv_qp_ex);