/* MPIDI_CH3I_SendNoncontig - Sends a message by packing
directly into cells. The caller must initialize sreq->dev.segment
as well as segment_first and segment_size. */
int MPIDI_CH3I_SendNoncontig( MPIDI_VC_t *vc, MPID_Request *sreq, void *header, MPIDI_msg_sz_t hdr_sz )
{
int mpi_errno = MPI_SUCCESS;
int again = 0;
MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
MPIDI_DBG_Print_packet((MPIDI_CH3_Pkt_t *)header);
MPIU_THREAD_CS_ENTER(MPIDCOMM,);
if (!MPIDI_CH3I_Sendq_empty(MPIDI_CH3I_shm_sendq)) /* MT */
{
/* send queue is not empty, enqueue the request then check to
see if we can send any now */
MPIDI_DBG_PRINTF((55, FCNAME, "enqueuing"));
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *)header;
sreq->ch.noncontig = TRUE;
sreq->ch.header_sz = hdr_sz;
sreq->ch.vc = vc;
MPIDI_CH3I_Sendq_enqueue(&MPIDI_CH3I_shm_sendq, sreq);
mpi_errno = MPIDI_CH3I_Shm_send_progress();
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
goto fn_exit;
}
/* send as many cells of data as you can */
MPID_nem_mpich_send_seg_header(sreq->dev.segment_ptr, &sreq->dev.segment_first, sreq->dev.segment_size, header, hdr_sz, vc, &again);
while(!again && sreq->dev.segment_first < sreq->dev.segment_size)
MPID_nem_mpich_send_seg(sreq->dev.segment_ptr, &sreq->dev.segment_first, sreq->dev.segment_size, vc, &again);
if (again)
{
/* we didn't finish sending everything */
sreq->ch.noncontig = TRUE;
sreq->ch.vc = vc;
if (sreq->dev.segment_first == 0) /* nothing was sent, save header */
{
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *)header;
sreq->ch.header_sz = hdr_sz;
}
else
{
/* part of message was sent, make this req an active send */
MPIU_Assert(MPIDI_CH3I_shm_active_send == NULL);
MPIDI_CH3I_shm_active_send = sreq;
}
MPIDI_CH3I_Sendq_enqueue(&MPIDI_CH3I_shm_sendq, sreq);
goto fn_exit;
}
/* finished sending all data, complete the request */
if (!sreq->dev.OnDataAvail)
{
MPIU_Assert(MPIDI_Request_get_type(sreq) != MPIDI_REQUEST_TYPE_GET_RESP);
MPIDI_CH3U_Request_complete(sreq);
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, ".... complete %d bytes", (int) (sreq->dev.segment_size));
}
else
{
int complete = 0;
mpi_errno = sreq->dev.OnDataAvail(vc, sreq, &complete);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
MPIU_Assert(complete); /* all data has been sent, we should always complete */
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, ".... complete %d bytes", (int) (sreq->dev.segment_size));
}
fn_exit:
MPIU_THREAD_CS_EXIT(MPIDCOMM,);
MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPID_nem_scif_SendNoncontig(MPIDI_VC_t * vc, MPID_Request * sreq, void *header,
MPIDI_msg_sz_t hdr_sz)
{
int mpi_errno = MPI_SUCCESS;
int iov_n;
MPID_IOV iov[MPID_IOV_LIMIT];
MPID_IOV *iov_p;
MPIDI_msg_sz_t offset = 0;
int complete;
uint64_t seqno = 0;
MPID_nem_scif_vc_area *vc_scif = VC_SCIF(vc);
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_SCIF_SENDNONCONTIG);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_SCIF_SENDNONCONTIG);
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "scif_SendNoncontig");
MPIU_Assert(hdr_sz <= sizeof(MPIDI_CH3_Pkt_t));
iov[0].MPID_IOV_BUF = header;
iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t);
iov_n = MPID_IOV_LIMIT - 1;
mpi_errno = MPIDI_CH3U_Request_load_send_iov(sreq, &iov[1], &iov_n);
MPIU_ERR_CHKANDJUMP(mpi_errno, mpi_errno, MPI_ERR_OTHER, "**loadsendiov");
iov_n += 1;
offset = 0;
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue) &&
MPID_nem_scif_poll_send(vc_scif->sc->fd, &vc_scif->sc->csend)) {
offset =
MPID_nem_scif_writev(vc_scif->sc->fd, &vc_scif->sc->csend, iov, iov_n, &seqno);
MPIU_ERR_CHKANDJUMP1(offset <= 0, mpi_errno, MPI_ERR_OTHER,
"**scif_writev", "**scif_writev %s", MPIU_Strerror(errno));
MPIU_DBG_MSG_D(CH3_CHANNEL, VERBOSE, "scif_send noncontig " MPIDI_MSG_SZ_FMT, offset);
}
if (offset < iov[0].MPID_IOV_LEN) {
/* header was not yet sent, save it in req */
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *) header;
iov[0].MPID_IOV_BUF = (MPID_IOV_BUF_CAST) & sreq->dev.pending_pkt;
iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t);
}
/* check if whole iov was sent, and save any unsent portion of
* iov */
sreq->dev.iov_count = 0;
complete = 1;
for (iov_p = &iov[0]; iov_p < &iov[iov_n]; ++iov_p) {
if (offset < iov_p->MPID_IOV_LEN) {
sreq->dev.iov[sreq->dev.iov_count].MPID_IOV_BUF =
(MPID_IOV_BUF_CAST) ((char *) iov_p->MPID_IOV_BUF + offset);
sreq->dev.iov[sreq->dev.iov_count].MPID_IOV_LEN = iov_p->MPID_IOV_LEN - offset;
offset = 0;
++sreq->dev.iov_count;
complete = 0;
seqno = 0;
}
else
offset -= iov_p->MPID_IOV_LEN;
}
if (seqno)
complete = MPID_nem_scif_chk_seqno(&vc_scif->sc->csend, seqno);
if (complete) {
/* sent whole iov */
int (*reqFn) (MPIDI_VC_t *, MPID_Request *, int *);
reqFn = sreq->dev.OnDataAvail;
if (!reqFn) {
MPIDI_CH3U_Request_complete(sreq);
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, ".... complete");
goto fn_exit;
}
complete = 0;
mpi_errno = reqFn(vc, sreq, &complete);
if (mpi_errno)
MPIU_ERR_POP(mpi_errno);
if (complete) {
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, ".... complete");
goto fn_exit;
}
seqno = 0;
}
/* enqueue request */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "enqueuing");
MPIU_Assert(seqno || (sreq->dev.iov_count >= 1 && sreq->dev.iov[0].MPID_IOV_LEN > 0));
RQ_SCIF(sreq)->seqno = seqno;
sreq->ch.vc = vc;
sreq->dev.iov_offset = 0;
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue)) {
/* this will be the first send on the queue: queue it and set
* the write flag on the pollfd */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
}
else {
/* there are other sends in the queue before this one: try to
* send from the queue */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
mpi_errno = MPID_nem_scif_send_queued(vc, &vc_scif->send_queue);
if (mpi_errno)
MPIU_ERR_POP(mpi_errno);
}
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_SCIF_SENDNONCONTIG);
return mpi_errno;
fn_fail:
MPIU_Object_set_ref(sreq, 0);
MPIDI_CH3_Request_destroy(sreq);
goto fn_exit;
}
int MPID_nem_scif_iStartContigMsg(MPIDI_VC_t * vc, void *hdr, MPIDI_msg_sz_t hdr_sz,
void *data, MPIDI_msg_sz_t data_sz, MPID_Request ** sreq_ptr)
{
int mpi_errno = MPI_SUCCESS;
MPID_Request *sreq = NULL;
MPIDI_msg_sz_t offset = 0;
MPID_nem_scif_vc_area *vc_scif = VC_SCIF(vc);
scifconn_t *sc = vc_scif->sc;
uint64_t seqno = 0;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_SCIF_ISTARTCONTIGMSG);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_SCIF_ISTARTCONTIGMSG);
MPIU_Assert(hdr_sz <= sizeof(MPIDI_CH3_Pkt_t));
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "scif_iStartContigMsg");
MPIDI_DBG_Print_packet((MPIDI_CH3_Pkt_t *) hdr);
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue) &&
MPID_nem_scif_poll_send(sc->fd, &sc->csend)) {
MPID_IOV iov[2];
iov[0].MPID_IOV_BUF = hdr;
iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t);
iov[1].MPID_IOV_BUF = data;
iov[1].MPID_IOV_LEN = data_sz;
offset = MPID_nem_scif_writev(sc->fd, &sc->csend, iov, 2, &seqno);
MPIU_ERR_CHKANDJUMP1(offset <= 0, mpi_errno, MPI_ERR_OTHER,
"**scif_writev", "**scif_writev %s", MPIU_Strerror(errno));
MPIU_DBG_MSG_FMT(CH3_CHANNEL, VERBOSE,
(MPIU_DBG_FDEST, "scif_send " MPIDI_MSG_SZ_FMT " fd=%d",
offset, sc->fd));
if (offset == sizeof(MPIDI_CH3_Pkt_t) + data_sz) {
if (!seqno) {
/* sent whole message */
*sreq_ptr = NULL;
goto fn_exit;
}
}
}
/* create and enqueue request */
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "enqueuing");
/* create a request */
sreq = MPID_Request_create();
RQ_SCIF(sreq)->seqno = seqno;
MPIU_Assert(sreq != NULL);
MPIU_Object_set_ref(sreq, 2);
sreq->kind = MPID_REQUEST_SEND;
sreq->dev.OnDataAvail = 0;
sreq->ch.vc = vc;
sreq->dev.iov_offset = 0;
if (!seqno) {
if (offset < sizeof(MPIDI_CH3_Pkt_t)) {
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *) hdr;
sreq->dev.iov[0].MPID_IOV_BUF = (char *) &sreq->dev.pending_pkt + offset;
sreq->dev.iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t) - offset;
if (data_sz) {
sreq->dev.iov[1].MPID_IOV_BUF = data;
sreq->dev.iov[1].MPID_IOV_LEN = data_sz;
sreq->dev.iov_count = 2;
}
else
sreq->dev.iov_count = 1;
}
else {
sreq->dev.iov[0].MPID_IOV_BUF = (char *) data + (offset - sizeof(MPIDI_CH3_Pkt_t));
sreq->dev.iov[0].MPID_IOV_LEN = data_sz - (offset - sizeof(MPIDI_CH3_Pkt_t));
sreq->dev.iov_count = 1;
}
MPIU_Assert(sreq->dev.iov_count >= 1 && sreq->dev.iov[0].MPID_IOV_LEN > 0);
}
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue)) {
/* this will be the first send on the queue: queue it and set
* the write flag on the pollfd */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
}
else {
/* there are other sends in the queue before this one: try to
* send from the queue */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
mpi_errno = MPID_nem_scif_send_queued(vc, &vc_scif->send_queue);
if (mpi_errno)
MPIU_ERR_POP(mpi_errno);
}
*sreq_ptr = sreq;
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_SCIF_ISTARTCONTIGMSG);
return mpi_errno;
fn_fail:
goto fn_exit;
}
int MPID_nem_scif_iSendContig(MPIDI_VC_t * vc, MPID_Request * sreq, void *hdr,
MPIDI_msg_sz_t hdr_sz, void *data, MPIDI_msg_sz_t data_sz)
{
int mpi_errno = MPI_SUCCESS;
MPIDI_msg_sz_t offset = 0;
MPID_nem_scif_vc_area *vc_scif = VC_SCIF(vc);
scifconn_t *sc = vc_scif->sc;
uint64_t seqno = 0;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_SCIF_ISENDCONTIGMSG);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_SCIF_ISENDCONTIGMSG);
MPIU_Assert(hdr_sz <= sizeof(MPIDI_CH3_Pkt_t));
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "scif_iSendContig");
MPIDI_DBG_Print_packet((MPIDI_CH3_Pkt_t *) hdr);
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue) &&
MPID_nem_scif_poll_send(sc->fd, &sc->csend)) {
MPID_IOV iov[2];
iov[0].MPID_IOV_BUF = hdr;
iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t);
iov[1].MPID_IOV_BUF = data;
iov[1].MPID_IOV_LEN = data_sz;
offset = MPID_nem_scif_writev(sc->fd, &sc->csend, iov, 2, &seqno);
MPIU_ERR_CHKANDJUMP1(offset <= 0, mpi_errno, MPI_ERR_OTHER,
"**scif_writev", "**scif_writev %s", MPIU_Strerror(errno));
MPIU_DBG_MSG_FMT(CH3_CHANNEL, VERBOSE,
(MPIU_DBG_FDEST, "scif_send " MPIDI_MSG_SZ_FMT " fd=%d",
offset, sc->fd));
if (offset == sizeof(MPIDI_CH3_Pkt_t) + data_sz) {
/* sent whole message */
int (*reqFn) (MPIDI_VC_t *, MPID_Request *, int *);
if (seqno)
goto enqueue_request;
reqFn = sreq->dev.OnDataAvail;
if (!reqFn) {
MPIU_Assert(MPIDI_Request_get_type(sreq) != MPIDI_REQUEST_TYPE_GET_RESP);
MPIDI_CH3U_Request_complete(sreq);
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, ".... complete");
goto fn_exit;
}
else {
int complete = 0;
mpi_errno = reqFn(vc, sreq, &complete);
if (mpi_errno)
MPIU_ERR_POP(mpi_errno);
if (complete) {
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, ".... complete");
goto fn_exit;
}
/* not completed: more to send */
goto enqueue_request;
}
}
}
/* save iov */
if (offset < sizeof(MPIDI_CH3_Pkt_t)) {
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *) hdr;
sreq->dev.iov[0].MPID_IOV_BUF = (char *) &sreq->dev.pending_pkt + offset;
sreq->dev.iov[0].MPID_IOV_LEN = sizeof(MPIDI_CH3_Pkt_t) - offset;
if (data_sz) {
sreq->dev.iov[1].MPID_IOV_BUF = data;
sreq->dev.iov[1].MPID_IOV_LEN = data_sz;
sreq->dev.iov_count = 2;
}
else
sreq->dev.iov_count = 1;
seqno = 0;
}
else {
sreq->dev.iov[0].MPID_IOV_BUF = (char *) data + (offset - sizeof(MPIDI_CH3_Pkt_t));
sreq->dev.iov[0].MPID_IOV_LEN = data_sz - (offset - sizeof(MPIDI_CH3_Pkt_t));
sreq->dev.iov_count = 1;
seqno = 0;
}
enqueue_request:
/* enqueue request */
RQ_SCIF(sreq)->seqno = seqno;
MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "enqueuing");
MPIU_Assert(seqno || (sreq->dev.iov_count >= 1 && sreq->dev.iov[0].MPID_IOV_LEN > 0));
sreq->ch.vc = vc;
sreq->dev.iov_offset = 0;
if (MPIDI_CH3I_Sendq_empty(vc_scif->send_queue)) {
/* this will be the first send on the queue: queue it and set
* the write flag on the pollfd */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
}
else {
/* there are other sends in the queue before this one: try to
* send from the queue */
MPIDI_CH3I_Sendq_enqueue(&vc_scif->send_queue, sreq);
mpi_errno = MPID_nem_scif_send_queued(vc, &vc_scif->send_queue);
if (mpi_errno)
MPIU_ERR_POP(mpi_errno);
}
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_SCIF_ISENDCONTIGMSG);
return mpi_errno;
fn_fail:
goto fn_exit;
}
/* MPIDI_CH3I_SendNoncontig - Sends a message by packing
directly into cells. The caller must initialize sreq->dev.segment
as well as segment_first and segment_size. */
int MPIDI_CH3I_SendNoncontig( MPIDI_VC_t *vc, MPIR_Request *sreq, void *header, intptr_t hdr_sz,
MPL_IOV *hdr_iov, int n_hdr_iov)
{
int mpi_errno = MPI_SUCCESS;
int again = 0;
intptr_t orig_segment_first = sreq->dev.segment_first;
MPIR_FUNC_VERBOSE_STATE_DECL(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
MPIR_FUNC_VERBOSE_ENTER(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
MPIDI_DBG_Print_packet((MPIDI_CH3_Pkt_t *)header);
MPID_THREAD_CS_ENTER(POBJ, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
if (n_hdr_iov > 0) {
/* translate segments to iovs and combine with the extended header iov. */
mpi_errno = MPIDI_CH3_SendNoncontig_iov(vc, sreq, header, hdr_sz,
hdr_iov, n_hdr_iov);
if (mpi_errno)
MPIR_ERR_POP(mpi_errno);
goto fn_exit;
}
if (!MPIDI_CH3I_Sendq_empty(MPIDI_CH3I_shm_sendq)) /* MT */
{
/* send queue is not empty, enqueue the request then check to
see if we can send any now */
MPL_DBG_MSG(MPIDI_CH3_DBG_OTHER, TERSE, "enqueuing");
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *)header;
sreq->ch.noncontig = TRUE;
sreq->ch.header_sz = hdr_sz;
sreq->ch.vc = vc;
MPIDI_CH3I_Sendq_enqueue(&MPIDI_CH3I_shm_sendq, sreq);
mpi_errno = MPIDI_CH3I_Shm_send_progress();
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
goto fn_exit;
}
/* send as many cells of data as you can */
MPID_nem_mpich_send_seg_header(sreq->dev.segment_ptr, &sreq->dev.segment_first, sreq->dev.segment_size,
header, hdr_sz, vc, &again);
while(!again && sreq->dev.segment_first < sreq->dev.segment_size)
MPID_nem_mpich_send_seg(sreq->dev.segment_ptr, &sreq->dev.segment_first, sreq->dev.segment_size, vc, &again);
if (again)
{
/* we didn't finish sending everything */
sreq->ch.noncontig = TRUE;
sreq->ch.vc = vc;
if (sreq->dev.segment_first == orig_segment_first) /* nothing was sent, save header */
{
sreq->dev.pending_pkt = *(MPIDI_CH3_Pkt_t *)header;
sreq->ch.header_sz = hdr_sz;
}
else
{
/* part of message was sent, make this req an active send */
MPIR_Assert(MPIDI_CH3I_shm_active_send == NULL);
MPIDI_CH3I_shm_active_send = sreq;
}
MPIDI_CH3I_Sendq_enqueue(&MPIDI_CH3I_shm_sendq, sreq);
goto fn_exit;
}
/* finished sending all data, complete the request */
if (!sreq->dev.OnDataAvail)
{
MPIR_Assert(MPIDI_Request_get_type(sreq) != MPIDI_REQUEST_TYPE_GET_RESP);
mpi_errno = MPID_Request_complete(sreq);
if (mpi_errno != MPI_SUCCESS) {
MPIR_ERR_POP(mpi_errno);
}
MPL_DBG_MSG_D(MPIDI_CH3_DBG_CHANNEL, VERBOSE, ".... complete %d bytes", (int) (sreq->dev.segment_size));
}
else
{
int complete = 0;
mpi_errno = sreq->dev.OnDataAvail(vc, sreq, &complete);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
MPIR_Assert(complete); /* all data has been sent, we should always complete */
MPL_DBG_MSG_D(MPIDI_CH3_DBG_CHANNEL, VERBOSE, ".... complete %d bytes", (int) (sreq->dev.segment_size));
}
fn_exit:
MPID_THREAD_CS_EXIT(POBJ, MPIR_THREAD_GLOBAL_ALLFUNC_MUTEX);
MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_MPIDI_CH3I_SENDNONCONTIG);
return mpi_errno;
fn_fail:
goto fn_exit;
}
