static void big_get(void *buf, ptl_size_t left_to_get, MPIDI_VC_t *vc, ptl_match_bits_t match_bits, MPID_Request *rreq)
{
int ret;
MPID_nem_ptl_vc_area *vc_ptl;
ptl_size_t start, get_sz;
vc_ptl = VC_PTL(vc);
start = (ptl_size_t)buf;
/* we need to handle all events */
REQ_PTL(rreq)->event_handler = handler_recv_big_get;
while (left_to_get > 0) {
/* get up to the maximum allowed by the portals interface */
if (left_to_get > MPIDI_nem_ptl_ni_limits.max_msg_size)
get_sz = MPIDI_nem_ptl_ni_limits.max_msg_size;
else
get_sz = left_to_get;
ret = MPID_nem_ptl_rptl_get(MPIDI_nem_ptl_global_md, start, get_sz, vc_ptl->id, vc_ptl->ptg, match_bits, 0, rreq);
DBG_MSG_GET("global", get_sz, vc->pg_rank, match_bits);
MPIU_DBG_MSG_P(CH3_CHANNEL, VERBOSE, " buf=%p", (char *)start);
MPIU_Assert(ret == 0);
/* account for what has been sent */
start += get_sz;
left_to_get -= get_sz;
REQ_PTL(rreq)->num_gets++;
}
}
int vc_terminate(MPIDI_VC_t *vc)
{
int mpi_errno = MPI_SUCCESS;
int req_errno = MPI_SUCCESS;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_VC_TERMINATE);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_VC_TERMINATE);
if (vc->state != MPIDI_VC_STATE_CLOSED) {
/* VC is terminated as a result of a fault. Complete
outstanding sends with an error and terminate
connection immediately. */
MPIR_ERR_SET1(req_errno, MPIX_ERR_PROC_FAILED, "**comm_fail", "**comm_fail %d", vc->pg_rank);
mpi_errno = MPID_nem_ptl_vc_terminated(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
} else if (vc_ptl->num_queued_sends == 0) {
mpi_errno = MPID_nem_ptl_vc_terminated(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
} else {
/* the send_queued function will call vc_terminated if vc->state is
CLOSED and the last queued send has been sent*/
}
fn_exit:
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_VC_TERMINATE);
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPID_nem_ptl_init_id(MPIDI_VC_t *vc)
{
int mpi_errno = MPI_SUCCESS;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
char *bc;
int pmi_errno;
int val_max_sz;
MPIR_CHKLMEM_DECL(1);
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_MPID_NEM_PTL_INIT_ID);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_MPID_NEM_PTL_INIT_ID);
pmi_errno = PMI_KVS_Get_value_length_max(&val_max_sz);
MPIR_ERR_CHKANDJUMP1(pmi_errno, mpi_errno, MPI_ERR_OTHER, "**fail", "**fail %d", pmi_errno);
MPIR_CHKLMEM_MALLOC(bc, char *, val_max_sz, mpi_errno, "bc");
mpi_errno = vc->pg->getConnInfo(vc->pg_rank, bc, val_max_sz, vc->pg);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
mpi_errno = MPID_nem_ptl_get_id_from_bc(bc, &vc_ptl->id, &vc_ptl->pt, &vc_ptl->ptg, &vc_ptl->ptc, &vc_ptl->ptr, &vc_ptl->ptrg, &vc_ptl->ptrc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
vc_ptl->id_initialized = TRUE;
MPIDI_CHANGE_VC_STATE(vc, ACTIVE);
fn_exit:
MPIR_CHKLMEM_FREEALL();
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_MPID_NEM_PTL_INIT_ID);
return mpi_errno;
fn_fail:
goto fn_exit;
}
static void big_meappend(void *buf, ptl_size_t left_to_send, MPIDI_VC_t *vc, ptl_match_bits_t match_bits, MPID_Request *sreq)
{
int i, ret, was_incomplete;
MPID_nem_ptl_vc_area *vc_ptl;
ptl_me_t me;
vc_ptl = VC_PTL(vc);
me.start = buf;
me.ct_handle = PTL_CT_NONE;
me.uid = PTL_UID_ANY;
me.options = ( PTL_ME_OP_PUT | PTL_ME_OP_GET | PTL_ME_USE_ONCE | PTL_ME_IS_ACCESSIBLE | PTL_ME_EVENT_LINK_DISABLE |
PTL_ME_EVENT_UNLINK_DISABLE );
me.match_id = vc_ptl->id;
me.match_bits = match_bits;
me.ignore_bits = 0;
me.min_free = 0;
/* allocate enough handles to cover all get operations */
REQ_PTL(sreq)->get_me_p = MPIU_Malloc(sizeof(ptl_handle_me_t) *
((left_to_send / MPIDI_nem_ptl_ni_limits.max_msg_size) + 1));
/* queue up as many entries as necessary to describe the entire message */
for (i = 0; left_to_send > 0; i++) {
/* send up to the maximum allowed by the portals interface */
if (left_to_send > MPIDI_nem_ptl_ni_limits.max_msg_size)
me.length = MPIDI_nem_ptl_ni_limits.max_msg_size;
else
me.length = left_to_send;
ret = PtlMEAppend(MPIDI_nem_ptl_ni, MPIDI_nem_ptl_get_pt, &me, PTL_PRIORITY_LIST, sreq,
&REQ_PTL(sreq)->get_me_p[i]);
DBG_MSG_MEAPPEND("CTL", vc->pg_rank, me, sreq);
MPIU_Assert(ret == 0);
/* increment the cc for each get operation */
MPIDI_CH3U_Request_increment_cc(sreq, &was_incomplete);
MPIU_Assert(was_incomplete);
/* account for what has been sent */
me.start = (char *)me.start + me.length;
left_to_send -= me.length;
}
}
static int get_target_info(int rank, ptl_process_t *id, ptl_pt_index_t local_data_pt, ptl_pt_index_t *target_data_pt,
ptl_pt_index_t *target_control_pt)
{
int mpi_errno = MPI_SUCCESS;
struct MPIDI_VC *vc;
MPID_nem_ptl_vc_area *vc_ptl;
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_MPID_NEM_GET_TARGET_INFO);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_MPID_NEM_GET_TARGET_INFO);
MPIDI_PG_Get_vc(MPIDI_Process.my_pg, rank, &vc);
vc_ptl = VC_PTL(vc);
if (!vc_ptl->id_initialized) {
mpi_errno = MPID_nem_ptl_init_id(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
}
*id = vc_ptl->id;
MPIR_Assert(local_data_pt == MPIDI_nem_ptl_pt || local_data_pt == MPIDI_nem_ptl_get_pt ||
local_data_pt == MPIDI_nem_ptl_control_pt);
if (local_data_pt == MPIDI_nem_ptl_pt) {
*target_data_pt = vc_ptl->pt;
*target_control_pt = vc_ptl->ptr;
}
else if (local_data_pt == MPIDI_nem_ptl_get_pt) {
*target_data_pt = vc_ptl->ptg;
*target_control_pt = PTL_PT_ANY;
}
else if (local_data_pt == MPIDI_nem_ptl_control_pt) {
*target_data_pt = vc_ptl->ptc;
*target_control_pt = vc_ptl->ptrc;
}
fn_exit:
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_MPID_NEM_GET_TARGET_INFO);
return mpi_errno;
fn_fail:
goto fn_exit;
}
static int vc_init(MPIDI_VC_t *vc)
{
int mpi_errno = MPI_SUCCESS;
MPIDI_CH3I_VC *const vc_ch = &vc->ch;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_VC_INIT);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_VC_INIT);
vc->sendNoncontig_fn = MPID_nem_ptl_SendNoncontig;
vc_ch->iStartContigMsg = MPID_nem_ptl_iStartContigMsg;
vc_ch->iSendContig = MPID_nem_ptl_iSendContig;
vc_ch->num_pkt_handlers = 2;
vc_ch->pkt_handler = MPID_nem_ptl_pkt_handlers;
MPID_nem_ptl_pkt_handlers[MPIDI_NEM_PTL_PKT_CANCEL_SEND_REQ] =
MPID_nem_ptl_pkt_cancel_send_req_handler;
MPID_nem_ptl_pkt_handlers[MPIDI_NEM_PTL_PKT_CANCEL_SEND_RESP] =
MPID_nem_ptl_pkt_cancel_send_resp_handler;
vc_ch->lmt_initiate_lmt = MPID_nem_ptl_lmt_initiate_lmt;
vc_ch->lmt_start_recv = MPID_nem_ptl_lmt_start_recv;
vc_ch->lmt_start_send = MPID_nem_ptl_lmt_start_send;
vc_ch->lmt_handle_cookie = MPID_nem_ptl_lmt_handle_cookie;
vc_ch->lmt_done_send = MPID_nem_ptl_lmt_done_send;
vc_ch->lmt_done_recv = MPID_nem_ptl_lmt_done_recv;
vc->comm_ops = &comm_ops;
vc_ch->next = NULL;
vc_ch->prev = NULL;
vc_ptl->id_initialized = FALSE;
vc_ptl->num_queued_sends = 0;
mpi_errno = MPID_nem_ptl_init_id(vc);
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_VC_INIT);
return mpi_errno;
}
int MPID_nem_ptl_pkt_cancel_send_req_handler(MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt,
MPIDI_msg_sz_t *buflen, MPID_Request **rreqp)
{
int ret, mpi_errno = MPI_SUCCESS;
MPIDI_nem_ptl_pkt_cancel_send_req_t *req_pkt = (MPIDI_nem_ptl_pkt_cancel_send_req_t *)pkt;
MPID_PKT_DECL_CAST(upkt, MPIDI_nem_ptl_pkt_cancel_send_resp_t, resp_pkt);
MPID_Request *search_req, *resp_req;
ptl_me_t me;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
MPIU_DBG_MSG_FMT(CH3_OTHER,VERBOSE,(MPIU_DBG_FDEST,
"received cancel send req pkt, sreq=0x%08x, rank=%d, tag=%d, context=%d",
req_pkt->sender_req_id, req_pkt->match.parts.rank,
req_pkt->match.parts.tag, req_pkt->match.parts.context_id));
/* create a dummy request and search for the message */
/* create a request */
search_req = MPID_Request_create();
MPID_nem_ptl_init_req(search_req);
MPIR_ERR_CHKANDJUMP1(!search_req, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPID_Request_create");
MPIU_Object_set_ref(search_req, 2); /* 1 ref for progress engine and 1 ref for us */
search_req->kind = MPID_REQUEST_MPROBE;
/* create a dummy ME to use for searching the list */
me.start = NULL;
me.length = 0;
me.ct_handle = PTL_CT_NONE;
me.uid = PTL_UID_ANY;
me.options = ( PTL_ME_OP_PUT | PTL_ME_USE_ONCE );
me.min_free = 0;
me.match_bits = NPTL_MATCH(req_pkt->match.parts.tag, req_pkt->match.parts.context_id, req_pkt->match.parts.rank);
me.match_id = vc_ptl->id;
me.ignore_bits = NPTL_MATCH_IGNORE;
/* FIXME: this should use a custom handler that throws the data away inline */
REQ_PTL(search_req)->event_handler = handle_mprobe;
/* submit a search request */
ret = PtlMESearch(MPIDI_nem_ptl_ni, MPIDI_nem_ptl_pt, &me, PTL_SEARCH_DELETE, search_req);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmesearch", "**ptlmesearch %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_MESearch("REG", vc ? vc->pg_rank : 0, me, search_req);
/* wait for search request to complete */
do {
mpi_errno = MPID_nem_ptl_poll(FALSE);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
} while (!MPID_Request_is_complete(search_req));
/* send response */
resp_pkt->type = MPIDI_NEM_PKT_NETMOD;
resp_pkt->subtype = MPIDI_NEM_PTL_PKT_CANCEL_SEND_RESP;
resp_pkt->ack = REQ_PTL(search_req)->found;
resp_pkt->sender_req_id = req_pkt->sender_req_id;
MPID_nem_ptl_iStartContigMsg(vc, resp_pkt, sizeof(*resp_pkt), NULL,
0, &resp_req);
/* if the message was found, free the temporary buffer used to copy the data */
if (REQ_PTL(search_req)->found)
MPIU_Free(search_req->dev.tmpbuf);
MPID_Request_release(search_req);
if (resp_req != NULL)
MPID_Request_release(resp_req);
fn_exit:
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPID_nem_ptl_improbe(MPIDI_VC_t *vc, int source, int tag, MPID_Comm *comm, int context_offset, int *flag,
MPID_Request **message, MPI_Status *status)
{
int mpi_errno = MPI_SUCCESS;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
int ret;
ptl_process_t id_any;
ptl_me_t me;
MPID_Request *req;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_PTL_IMPROBE);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_PTL_IMPROBE);
id_any.phys.nid = PTL_NID_ANY;
id_any.phys.pid = PTL_PID_ANY;
/* create a request */
req = MPID_Request_create();
MPID_nem_ptl_init_req(req);
MPIR_ERR_CHKANDJUMP1(!req, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPID_Request_create");
MPIU_Object_set_ref(req, 2); /* 1 ref for progress engine and 1 ref for us */
REQ_PTL(req)->event_handler = handle_mprobe;
req->kind = MPID_REQUEST_MPROBE;
/* create a dummy ME to use for searching the list */
me.start = NULL;
me.length = 0;
me.ct_handle = PTL_CT_NONE;
me.uid = PTL_UID_ANY;
me.options = ( PTL_ME_OP_PUT | PTL_ME_USE_ONCE );
me.min_free = 0;
me.match_bits = NPTL_MATCH(tag, comm->context_id + context_offset, source);
if (source == MPI_ANY_SOURCE)
me.match_id = id_any;
else {
if (!vc_ptl->id_initialized) {
mpi_errno = MPID_nem_ptl_init_id(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
}
me.match_id = vc_ptl->id;
}
if (tag == MPI_ANY_TAG)
me.ignore_bits = NPTL_MATCH_IGNORE_ANY_TAG;
else
me.ignore_bits = NPTL_MATCH_IGNORE;
/* submit a search request */
ret = PtlMESearch(MPIDI_nem_ptl_ni, MPIDI_nem_ptl_pt, &me, PTL_SEARCH_DELETE, req);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmesearch", "**ptlmesearch %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_MESearch("REG", vc ? vc->pg_rank : 0, me, req);
/* wait for search request to complete */
do {
mpi_errno = MPID_nem_ptl_poll(FALSE);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
} while (!MPID_Request_is_complete(req));
*flag = REQ_PTL(req)->found;
if (*flag) {
req->comm = comm;
MPIR_Comm_add_ref(comm);
MPIR_Request_extract_status(req, status);
*message = req;
}
else {
MPID_Request_release(req);
}
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_PTL_IMPROBE);
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPID_nem_ptl_recv_posted(MPIDI_VC_t *vc, MPID_Request *rreq)
{
int mpi_errno = MPI_SUCCESS;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
ptl_me_t me;
int dt_contig;
MPIDI_msg_sz_t data_sz;
MPID_Datatype *dt_ptr;
MPI_Aint dt_true_lb;
MPI_Aint last;
ptl_process_t id_any;
int ret;
MPIU_CHKPMEM_DECL(1);
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_PTL_RECV_POSTED);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_PTL_RECV_POSTED);
id_any.phys.nid = PTL_NID_ANY;
id_any.phys.pid = PTL_PID_ANY;
MPID_nem_ptl_init_req(rreq);
me.ct_handle = PTL_CT_NONE;
me.uid = PTL_UID_ANY;
me.options = ( PTL_ME_OP_PUT | PTL_ME_IS_ACCESSIBLE | PTL_ME_EVENT_LINK_DISABLE |
PTL_ME_EVENT_UNLINK_DISABLE | PTL_ME_USE_ONCE );
if (vc == NULL) {
/* MPI_ANY_SOURCE receive */
me.match_id = id_any;
} else {
if (!vc_ptl->id_initialized) {
mpi_errno = MPID_nem_ptl_init_id(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
}
me.match_id = vc_ptl->id;
}
MPIU_DBG_MSG_FMT(CH3_CHANNEL, VERBOSE, (MPIU_DBG_FDEST, "tag=%#x ctx=%#x rank=%#x", rreq->dev.match.parts.tag, rreq->dev.match.parts.context_id, rreq->dev.match.parts.rank));
me.match_bits = NPTL_MATCH(rreq->dev.match.parts.tag, rreq->dev.match.parts.context_id,
rreq->dev.match.parts.rank);
if (rreq->dev.match.parts.tag == MPI_ANY_TAG)
me.ignore_bits = NPTL_MATCH_IGNORE_ANY_TAG;
else
me.ignore_bits = NPTL_MATCH_IGNORE;
me.min_free = 0;
MPIDI_Datatype_get_info(rreq->dev.user_count, rreq->dev.datatype, dt_contig, data_sz, dt_ptr, dt_true_lb);
MPIU_DBG_MSG_FMT(CH3_CHANNEL, VERBOSE, (MPIU_DBG_FDEST, "count="MPI_AINT_FMT_DEC_SPEC" datatype=%#x contig=%d data_sz=%lu", rreq->dev.user_count, rreq->dev.datatype, dt_contig, data_sz));
if (data_sz <= PTL_LARGE_THRESHOLD) {
if (dt_contig) {
/* small contig message */
void *start = (char *)rreq->dev.user_buf + dt_true_lb;
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Small contig message");
if (start == NULL)
me.start = &dummy;
else
me.start = start;
me.length = data_sz;
REQ_PTL(rreq)->event_handler = handler_recv_dequeue_complete;
} else {
/* small noncontig */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Small noncontig message");
rreq->dev.segment_ptr = MPID_Segment_alloc();
MPIR_ERR_CHKANDJUMP1(rreq->dev.segment_ptr == NULL, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPID_Segment_alloc");
MPID_Segment_init(rreq->dev.user_buf, rreq->dev.user_count, rreq->dev.datatype, rreq->dev.segment_ptr, 0);
rreq->dev.segment_first = 0;
rreq->dev.segment_size = data_sz;
last = rreq->dev.segment_size;
rreq->dev.iov_count = MPL_IOV_LIMIT;
MPID_Segment_pack_vector(rreq->dev.segment_ptr, rreq->dev.segment_first, &last, rreq->dev.iov, &rreq->dev.iov_count);
if (last == rreq->dev.segment_size) {
/* entire message fits in IOV */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " entire message fits in IOV");
me.start = rreq->dev.iov;
me.length = rreq->dev.iov_count;
me.options |= PTL_IOVEC;
REQ_PTL(rreq)->event_handler = handler_recv_dequeue_complete;
} else {
/* IOV is not long enough to describe entire message: recv into
buffer and unpack later */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " IOV too long: using bounce buffer");
MPIU_CHKPMEM_MALLOC(REQ_PTL(rreq)->chunk_buffer[0], void *, data_sz, mpi_errno, "chunk_buffer");
me.start = REQ_PTL(rreq)->chunk_buffer[0];
me.length = data_sz;
REQ_PTL(rreq)->event_handler = handler_recv_dequeue_unpack_complete;
}
}
} else {
/* Large message: Create an ME for the first chunk of data, then do a GET for the rest */
if (dt_contig) {
static int handler_recv_dequeue_large(const ptl_event_t *e)
{
int mpi_errno = MPI_SUCCESS;
MPID_Request *const rreq = e->user_ptr;
MPIDI_VC_t *vc;
MPID_nem_ptl_vc_area *vc_ptl;
int ret;
int dt_contig;
MPIDI_msg_sz_t data_sz;
MPID_Datatype *dt_ptr;
MPI_Aint dt_true_lb;
MPI_Aint last;
MPIU_CHKPMEM_DECL(1);
MPIDI_STATE_DECL(MPID_STATE_HANDLER_RECV_DEQUEUE_LARGE);
MPIDI_FUNC_ENTER(MPID_STATE_HANDLER_RECV_DEQUEUE_LARGE);
MPIU_Assert(e->type == PTL_EVENT_PUT || e->type == PTL_EVENT_PUT_OVERFLOW);
MPIDI_Comm_get_vc(rreq->comm, NPTL_MATCH_GET_RANK(e->match_bits), &vc);
vc_ptl = VC_PTL(vc);
dequeue_req(e);
MPIDI_Datatype_get_info(rreq->dev.user_count, rreq->dev.datatype, dt_contig, data_sz, dt_ptr, dt_true_lb);
/* unpack data from unexpected buffer first */
if (e->type == PTL_EVENT_PUT_OVERFLOW) {
if (dt_contig) {
MPIU_Memcpy((char *)rreq->dev.user_buf + dt_true_lb, e->start, e->mlength);
} else {
last = e->mlength;
MPID_Segment_unpack(rreq->dev.segment_ptr, 0, &last, e->start);
MPIU_Assert(last == e->mlength);
rreq->dev.segment_first = e->mlength;
}
}
if (!(e->hdr_data & NPTL_LARGE)) {
/* all data has already been received; we're done */
mpi_errno = handler_recv_complete(e);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
goto fn_exit;
}
MPIU_Assert (e->mlength == PTL_LARGE_THRESHOLD);
/* we need to GET the rest of the data from the sender's buffer */
if (dt_contig) {
big_get((char *)rreq->dev.user_buf + dt_true_lb + PTL_LARGE_THRESHOLD, data_sz - PTL_LARGE_THRESHOLD,
vc, e->match_bits, rreq);
goto fn_exit;
}
/* noncontig recv buffer */
last = rreq->dev.segment_size;
rreq->dev.iov_count = MPL_IOV_LIMIT;
MPID_Segment_pack_vector(rreq->dev.segment_ptr, rreq->dev.segment_first, &last, rreq->dev.iov, &rreq->dev.iov_count);
if (last == rreq->dev.segment_size && rreq->dev.segment_size <= MPIDI_nem_ptl_ni_limits.max_msg_size + PTL_LARGE_THRESHOLD) {
/* Rest of message fits in one IOV */
ptl_md_t md;
md.start = rreq->dev.iov;
md.length = rreq->dev.iov_count;
md.options = PTL_IOVEC;
md.eq_handle = MPIDI_nem_ptl_origin_eq;
md.ct_handle = PTL_CT_NONE;
ret = PtlMDBind(MPIDI_nem_ptl_ni, &md, &REQ_PTL(rreq)->md);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmdbind", "**ptlmdbind %s", MPID_nem_ptl_strerror(ret));
REQ_PTL(rreq)->event_handler = handler_recv_complete;
ret = MPID_nem_ptl_rptl_get(REQ_PTL(rreq)->md, 0, rreq->dev.segment_size - rreq->dev.segment_first, vc_ptl->id, vc_ptl->ptg,
e->match_bits, 0, rreq);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlget", "**ptlget %s", MPID_nem_ptl_strerror(ret));
goto fn_exit;
}
/* message won't fit in a single IOV, allocate buffer and unpack when received */
/* FIXME: For now, allocate a single large buffer to hold entire message */
MPIU_CHKPMEM_MALLOC(REQ_PTL(rreq)->chunk_buffer[0], void *, data_sz - PTL_LARGE_THRESHOLD,
mpi_errno, "chunk_buffer");
big_get(REQ_PTL(rreq)->chunk_buffer[0], data_sz - PTL_LARGE_THRESHOLD, vc, e->match_bits, rreq);
fn_exit:
MPIU_CHKPMEM_COMMIT();
fn_exit2:
MPIDI_FUNC_EXIT(MPID_STATE_HANDLER_RECV_DEQUEUE_LARGE);
return mpi_errno;
fn_fail:
MPIU_CHKPMEM_REAP();
goto fn_exit2;
}
static int send_msg(ptl_hdr_data_t ssend_flag, struct MPIDI_VC *vc, const void *buf, MPI_Aint count, MPI_Datatype datatype, int dest,
int tag, MPID_Comm *comm, int context_offset, struct MPID_Request **request)
{
int mpi_errno = MPI_SUCCESS;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
int ret;
MPIDI_msg_sz_t data_sz;
int dt_contig;
MPI_Aint dt_true_lb;
MPID_Datatype *dt_ptr;
MPID_Request *sreq = NULL;
ptl_me_t me;
int initial_iov_count, remaining_iov_count;
ptl_md_t md;
MPI_Aint last;
MPIU_CHKPMEM_DECL(2);
MPIDI_STATE_DECL(MPID_STATE_SEND_MSG);
MPIDI_FUNC_ENTER(MPID_STATE_SEND_MSG);
MPID_nem_ptl_request_create_sreq(sreq, mpi_errno, comm);
sreq->dev.match.parts.rank = dest;
sreq->dev.match.parts.tag = tag;
sreq->dev.match.parts.context_id = comm->context_id + context_offset;
sreq->ch.vc = vc;
if (!vc_ptl->id_initialized) {
mpi_errno = MPID_nem_ptl_init_id(vc);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
}
MPIDI_Datatype_get_info(count, datatype, dt_contig, data_sz, dt_ptr, dt_true_lb);
MPIU_DBG_MSG_FMT(CH3_CHANNEL, VERBOSE, (MPIU_DBG_FDEST, "count="MPI_AINT_FMT_DEC_SPEC" datatype=%#x contig=%d data_sz=%lu", count, datatype, dt_contig, data_sz));
if (data_sz <= PTL_LARGE_THRESHOLD) {
/* Small message. Send all data eagerly */
if (dt_contig) {
void *start = (char *)buf + dt_true_lb;
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Small contig message");
REQ_PTL(sreq)->event_handler = handler_send;
MPIU_DBG_MSG_P(CH3_CHANNEL, VERBOSE, "&REQ_PTL(sreq)->event_handler = %p", &(REQ_PTL(sreq)->event_handler));
if (start == NULL)
ret = MPID_nem_ptl_rptl_put(MPIDI_nem_ptl_global_md, (ptl_size_t)&dummy, data_sz, PTL_NO_ACK_REQ, vc_ptl->id, vc_ptl->pt,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag, data_sz));
else
ret = MPID_nem_ptl_rptl_put(MPIDI_nem_ptl_global_md, (ptl_size_t)start, data_sz, PTL_NO_ACK_REQ, vc_ptl->id, vc_ptl->pt,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag, data_sz));
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlput", "**ptlput %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_PUT("global", data_sz, vc->pg_rank, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), NPTL_HEADER(ssend_flag, data_sz));
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, "id.nid = %#x", vc_ptl->id.phys.nid);
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, "id.pid = %#x", vc_ptl->id.phys.pid);
MPIU_DBG_MSG_P(CH3_CHANNEL, VERBOSE, "sreq = %p", sreq);
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, "vc_ptl->pt = %d", vc_ptl->pt);
MPIU_DBG_MSG_P(CH3_CHANNEL, VERBOSE, "REQ_PTL(sreq)->event_handler = %p", REQ_PTL(sreq)->event_handler);
goto fn_exit;
}
/* noncontig data */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Small noncontig message");
sreq->dev.segment_ptr = MPID_Segment_alloc();
MPIR_ERR_CHKANDJUMP1(sreq->dev.segment_ptr == NULL, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPID_Segment_alloc");
MPID_Segment_init(buf, count, datatype, sreq->dev.segment_ptr, 0);
sreq->dev.segment_first = 0;
sreq->dev.segment_size = data_sz;
last = sreq->dev.segment_size;
sreq->dev.iov_count = MPL_IOV_LIMIT;
MPID_Segment_pack_vector(sreq->dev.segment_ptr, sreq->dev.segment_first, &last, sreq->dev.iov, &sreq->dev.iov_count);
if (last == sreq->dev.segment_size) {
/* IOV is able to describe entire message */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " entire message fits in IOV");
md.start = sreq->dev.iov;
md.length = sreq->dev.iov_count;
md.options = PTL_IOVEC;
md.eq_handle = MPIDI_nem_ptl_origin_eq;
md.ct_handle = PTL_CT_NONE;
ret = PtlMDBind(MPIDI_nem_ptl_ni, &md, &REQ_PTL(sreq)->md);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmdbind", "**ptlmdbind %s", MPID_nem_ptl_strerror(ret));
REQ_PTL(sreq)->event_handler = handler_send;
ret = MPID_nem_ptl_rptl_put(REQ_PTL(sreq)->md, 0, data_sz, PTL_NO_ACK_REQ, vc_ptl->id, vc_ptl->pt,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag, data_sz));
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlput", "**ptlput %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_PUT("sreq", data_sz, vc->pg_rank, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), NPTL_HEADER(ssend_flag, data_sz));
goto fn_exit;
}
/* IOV is not long enough to describe entire message */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " IOV too long: using bounce buffer");
MPIU_CHKPMEM_MALLOC(REQ_PTL(sreq)->chunk_buffer[0], void *, data_sz, mpi_errno, "chunk_buffer");
MPID_Segment_init(buf, count, datatype, sreq->dev.segment_ptr, 0);
sreq->dev.segment_first = 0;
last = data_sz;
MPID_Segment_pack(sreq->dev.segment_ptr, sreq->dev.segment_first, &last, REQ_PTL(sreq)->chunk_buffer[0]);
MPIU_Assert(last == sreq->dev.segment_size);
REQ_PTL(sreq)->event_handler = handler_send;
ret = MPID_nem_ptl_rptl_put(MPIDI_nem_ptl_global_md, (ptl_size_t)REQ_PTL(sreq)->chunk_buffer[0], data_sz, PTL_NO_ACK_REQ,
vc_ptl->id, vc_ptl->pt, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag, data_sz));
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlput", "**ptlput %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_PUT("global", data_sz, vc->pg_rank, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), NPTL_HEADER(ssend_flag, data_sz));
goto fn_exit;
}
/* Large message. Send first chunk of data and let receiver get the rest */
if (dt_contig) {
/* create ME for buffer so receiver can issue a GET for the data */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Large contig message");
big_meappend((char *)buf + dt_true_lb + PTL_LARGE_THRESHOLD, data_sz - PTL_LARGE_THRESHOLD, vc,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), sreq);
REQ_PTL(sreq)->event_handler = handler_send;
ret = MPID_nem_ptl_rptl_put(MPIDI_nem_ptl_global_md, (ptl_size_t)((char *)buf + dt_true_lb), PTL_LARGE_THRESHOLD, PTL_NO_ACK_REQ, vc_ptl->id, vc_ptl->pt,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag | NPTL_LARGE, data_sz));
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlput", "**ptlput %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_PUT("global", PTL_LARGE_THRESHOLD, vc->pg_rank, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), NPTL_HEADER(ssend_flag | NPTL_LARGE, data_sz));
goto fn_exit;
}
/* Large noncontig data */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "Large noncontig message");
sreq->dev.segment_ptr = MPID_Segment_alloc();
MPIR_ERR_CHKANDJUMP1(sreq->dev.segment_ptr == NULL, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPID_Segment_alloc");
MPID_Segment_init(buf, count, datatype, sreq->dev.segment_ptr, 0);
sreq->dev.segment_first = 0;
sreq->dev.segment_size = data_sz;
last = PTL_LARGE_THRESHOLD;
sreq->dev.iov_count = MPL_IOV_LIMIT;
MPID_Segment_pack_vector(sreq->dev.segment_ptr, sreq->dev.segment_first, &last, sreq->dev.iov, &sreq->dev.iov_count);
initial_iov_count = sreq->dev.iov_count;
sreq->dev.segment_first = last;
if (last == PTL_LARGE_THRESHOLD) {
/* first chunk of message fits into IOV */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " first chunk fits in IOV");
if (initial_iov_count < MPL_IOV_LIMIT) {
/* There may be space for the rest of the message in this IOV */
sreq->dev.iov_count = MPL_IOV_LIMIT - sreq->dev.iov_count;
last = sreq->dev.segment_size;
MPID_Segment_pack_vector(sreq->dev.segment_ptr, sreq->dev.segment_first, &last,
&sreq->dev.iov[initial_iov_count], &sreq->dev.iov_count);
remaining_iov_count = sreq->dev.iov_count;
if (last == sreq->dev.segment_size && last <= MPIDI_nem_ptl_ni_limits.max_msg_size + PTL_LARGE_THRESHOLD) {
/* Entire message fit in one IOV */
int was_incomplete;
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, " rest of message fits in one IOV");
/* Create ME for remaining data */
me.start = &sreq->dev.iov[initial_iov_count];
me.length = remaining_iov_count;
me.ct_handle = PTL_CT_NONE;
me.uid = PTL_UID_ANY;
me.options = ( PTL_ME_OP_PUT | PTL_ME_OP_GET | PTL_ME_USE_ONCE | PTL_ME_IS_ACCESSIBLE | PTL_ME_EVENT_LINK_DISABLE |
PTL_ME_EVENT_UNLINK_DISABLE | PTL_IOVEC );
me.match_id = vc_ptl->id;
me.match_bits = NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank);
me.ignore_bits = 0;
me.min_free = 0;
MPIU_CHKPMEM_MALLOC(REQ_PTL(sreq)->get_me_p, ptl_handle_me_t *, sizeof(ptl_handle_me_t), mpi_errno, "get_me_p");
ret = PtlMEAppend(MPIDI_nem_ptl_ni, MPIDI_nem_ptl_get_pt, &me, PTL_PRIORITY_LIST, sreq,
&REQ_PTL(sreq)->get_me_p[0]);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmeappend", "**ptlmeappend %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_MEAPPEND("CTL", vc->pg_rank, me, sreq);
/* increment the cc for the get operation */
MPIDI_CH3U_Request_increment_cc(sreq, &was_incomplete);
MPIU_Assert(was_incomplete);
/* Create MD for first chunk */
md.start = sreq->dev.iov;
md.length = initial_iov_count;
md.options = PTL_IOVEC;
md.eq_handle = MPIDI_nem_ptl_origin_eq;
md.ct_handle = PTL_CT_NONE;
ret = PtlMDBind(MPIDI_nem_ptl_ni, &md, &REQ_PTL(sreq)->md);
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlmdbind", "**ptlmdbind %s", MPID_nem_ptl_strerror(ret));
REQ_PTL(sreq)->event_handler = handler_send;
ret = MPID_nem_ptl_rptl_put(REQ_PTL(sreq)->md, 0, PTL_LARGE_THRESHOLD, PTL_NO_ACK_REQ, vc_ptl->id, vc_ptl->pt,
NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), 0, sreq,
NPTL_HEADER(ssend_flag | NPTL_LARGE, data_sz));
MPIR_ERR_CHKANDJUMP1(ret, mpi_errno, MPI_ERR_OTHER, "**ptlput", "**ptlput %s", MPID_nem_ptl_strerror(ret));
DBG_MSG_PUT("req", PTL_LARGE_THRESHOLD, vc->pg_rank, NPTL_MATCH(tag, comm->context_id + context_offset, comm->rank), NPTL_HEADER(ssend_flag | NPTL_LARGE, data_sz));
goto fn_exit;
}
}
#undef FUNCNAME
#define FUNCNAME MPID_nem_ptl_pkt_cancel_send_req_handler
#undef FCNAME
#define FCNAME MPL_QUOTE(FUNCNAME)
int MPID_nem_ptl_pkt_cancel_send_req_handler(MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt,
void *data ATTRIBUTE((unused)),
intptr_t *buflen, MPIR_Request **rreqp)
{
int ret, mpi_errno = MPI_SUCCESS;
MPIDI_nem_ptl_pkt_cancel_send_req_t *req_pkt = (MPIDI_nem_ptl_pkt_cancel_send_req_t *)pkt;
MPID_PKT_DECL_CAST(upkt, MPIDI_nem_ptl_pkt_cancel_send_resp_t, resp_pkt);
MPIR_Request *search_req, *resp_req;
ptl_me_t me;
MPID_nem_ptl_vc_area *const vc_ptl = VC_PTL(vc);
MPL_DBG_MSG_FMT(MPIDI_CH3_DBG_OTHER,VERBOSE,(MPL_DBG_FDEST,
"received cancel send req pkt, sreq=0x%08x, rank=%d, tag=%d, context=%d",
req_pkt->sender_req_id, req_pkt->match.parts.rank,
req_pkt->match.parts.tag, req_pkt->match.parts.context_id));
/* create a dummy request and search for the message */
/* create a request */
search_req = MPIR_Request_create(MPIR_REQUEST_KIND__MPROBE);
MPID_nem_ptl_init_req(search_req);
MPIR_ERR_CHKANDJUMP1(!search_req, mpi_errno, MPI_ERR_OTHER, "**nomem", "**nomem %s", "MPIR_Request_create");
MPIR_Object_set_ref(search_req, 2); /* 1 ref for progress engine and 1 ref for us */
/* create a dummy ME to use for searching the list */
me.start = NULL;
