ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #332 from hjelmn/openib_updates

Просмотр исходного кода 
Openib updates
Этот коммит содержится в:
Nathan Hjelm 2015-01-15 15:05:18 -06:00
родитель d13c14ec82 cde79bfa60
Коммит 006074c48d
6 изменённых файлов: 222 добавлений и 232 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

241
opal/mca/btl/openib/btl_openib.c
Просмотреть файл

@ -531,10 +531,12 @@ static int mca_btl_openib_tune_endpoint(mca_btl_openib_module_t* openib_btl,
if a user distributes different INI files or parameters for different node/procs, if a user distributes different INI files or parameters for different node/procs,
it is on his own responsibility */ it is on his own responsibility */
switch(mca_btl_openib_component.receive_queues_source) { switch(mca_btl_openib_component.receive_queues_source) {
case MCA_BASE_VAR_SOURCE_COMMAND_LINE: case MCA_BASE_VAR_SOURCE_COMMAND_LINE:
case MCA_BASE_VAR_SOURCE_ENV: case MCA_BASE_VAR_SOURCE_ENV:
case MCA_BASE_VAR_SOURCE_MAX: case MCA_BASE_VAR_SOURCE_FILE:
break; case MCA_BASE_VAR_SOURCE_SET:
case MCA_BASE_VAR_SOURCE_OVERRIDE:
break;
/* If the queues configuration was set from command line /* If the queues configuration was set from command line
(with --mca btl_openib_receive_queues parameter) => both sides have a same configuration */ (with --mca btl_openib_receive_queues parameter) => both sides have a same configuration */
@ -543,40 +545,38 @@ static int mca_btl_openib_tune_endpoint(mca_btl_openib_module_t* openib_btl,
not possible that remote side got its queues configuration from command line => not possible that remote side got its queues configuration from command line =>
(by prio) the configuration was set from INI file or (if not configure) (by prio) the configuration was set from INI file or (if not configure)
by default queues configuration */ by default queues configuration */
case MCA_BASE_VAR_SOURCE_FILE: case BTL_OPENIB_RQ_SOURCE_DEVICE_INI:
case MCA_BASE_VAR_SOURCE_SET: if(NULL != values.receive_queues) {
case MCA_BASE_VAR_SOURCE_OVERRIDE: recv_qps = values.receive_queues;
if(NULL != values.receive_queues) { } else {
recv_qps = values.receive_queues; recv_qps = mca_btl_openib_component.default_recv_qps;
} else { }
recv_qps = mca_btl_openib_component.default_recv_qps;
}
if(0 != strcmp(mca_btl_openib_component.receive_queues, if(0 != strcmp(mca_btl_openib_component.receive_queues,
recv_qps)) { recv_qps)) {
opal_show_help("help-mpi-btl-openib.txt", opal_show_help("help-mpi-btl-openib.txt",
"unsupported queues configuration", true, "unsupported queues configuration", true,
opal_process_info.nodename, opal_process_info.nodename,
ibv_get_device_name(openib_btl->device->ib_dev), ibv_get_device_name(openib_btl->device->ib_dev),
(openib_btl->device->ib_dev_attr).vendor_id, (openib_btl->device->ib_dev_attr).vendor_id,
(openib_btl->device->ib_dev_attr).vendor_part_id, (openib_btl->device->ib_dev_attr).vendor_part_id,
mca_btl_openib_component.receive_queues, mca_btl_openib_component.receive_queues,
opal_get_proc_hostname(endpoint->endpoint_proc->proc_opal), opal_get_proc_hostname(endpoint->endpoint_proc->proc_opal),
endpoint->rem_info.rem_vendor_id, endpoint->rem_info.rem_vendor_id,
endpoint->rem_info.rem_vendor_part_id, endpoint->rem_info.rem_vendor_part_id,
recv_qps); recv_qps);
return OPAL_ERROR; return OPAL_ERROR;
} }
break; break;
/* If the local queues configuration was set /* If the local queues configuration was set
by default queues => check all possible cases for remote side and compare */ by default queues => check all possible cases for remote side and compare */
case MCA_BASE_VAR_SOURCE_DEFAULT: case MCA_BASE_VAR_SOURCE_DEFAULT:
if(NULL != values.receive_queues) { if(NULL != values.receive_queues) {
if(0 != strcmp(mca_btl_openib_component.receive_queues, if(0 != strcmp(mca_btl_openib_component.receive_queues,
values.receive_queues)) { values.receive_queues)) {
opal_show_help("help-mpi-btl-openib.txt", opal_show_help("help-mpi-btl-openib.txt",
"unsupported queues configuration", true, "unsupported queues configuration", true,
opal_process_info.nodename, opal_process_info.nodename,
ibv_get_device_name(openib_btl->device->ib_dev), ibv_get_device_name(openib_btl->device->ib_dev),
@ -588,10 +588,10 @@ static int mca_btl_openib_tune_endpoint(mca_btl_openib_module_t* openib_btl,
endpoint->rem_info.rem_vendor_part_id, endpoint->rem_info.rem_vendor_part_id,
values.receive_queues); values.receive_queues);
return OPAL_ERROR; return OPAL_ERROR;
}
} }
break; }
break;
} }
return OPAL_SUCCESS; return OPAL_SUCCESS;
@ -634,7 +634,7 @@ static int prepare_device_for_use (mca_btl_openib_device_t *device)
#if OPAL_HAVE_THREADS #if OPAL_HAVE_THREADS
if(mca_btl_openib_component.use_async_event_thread) { if(mca_btl_openib_component.use_async_event_thread) {
mca_btl_openib_async_cmd_t async_command; mca_btl_openib_async_cmd_t async_command;
/* start the async even thread if it is not already started */ /* start the async even thread if it is not already started */
if (start_async_event_thread() != OPAL_SUCCESS) if (start_async_event_thread() != OPAL_SUCCESS)
@ -642,8 +642,8 @@ static int prepare_device_for_use (mca_btl_openib_device_t *device)
device->got_fatal_event = false; device->got_fatal_event = false;
device->got_port_event = false; device->got_port_event = false;
async_command.a_cmd = OPENIB_ASYNC_CMD_FD_ADD; async_command.a_cmd = OPENIB_ASYNC_CMD_FD_ADD;
async_command.fd = device->ib_dev_context->async_fd; async_command.fd = device->ib_dev_context->async_fd;
if (write(mca_btl_openib_component.async_pipe[1], if (write(mca_btl_openib_component.async_pipe[1],
&async_command, sizeof(mca_btl_openib_async_cmd_t))<0){ &async_command, sizeof(mca_btl_openib_async_cmd_t))<0){
BTL_ERROR(("Failed to write to pipe [%d]",errno)); BTL_ERROR(("Failed to write to pipe [%d]",errno));
@ -1133,15 +1133,16 @@ ib_frag_alloc(mca_btl_openib_module_t *btl, size_t size, uint8_t order,
/* check if pending fragment has enough space for coalescing */ /* check if pending fragment has enough space for coalescing */
static mca_btl_openib_send_frag_t *check_coalescing(opal_list_t *frag_list, static mca_btl_openib_send_frag_t *check_coalescing(opal_list_t *frag_list,
opal_mutex_t *lock, mca_btl_base_endpoint_t *ep, size_t size) opal_mutex_t *lock, struct mca_btl_base_endpoint_t *ep, size_t size,
mca_btl_openib_coalesced_frag_t **cfrag)
{ {
mca_btl_openib_send_frag_t *frag = NULL; mca_btl_openib_send_frag_t *frag = NULL;
if(opal_list_is_empty(frag_list)) if (opal_list_is_empty(frag_list))
return NULL; return NULL;
OPAL_THREAD_LOCK(lock); OPAL_THREAD_LOCK(lock);
if(!opal_list_is_empty(frag_list)) { if (!opal_list_is_empty(frag_list)) {
int qp; int qp;
size_t total_length; size_t total_length;
opal_list_item_t *i = opal_list_get_first(frag_list); opal_list_item_t *i = opal_list_get_first(frag_list);
@ -1158,10 +1159,20 @@ static mca_btl_openib_send_frag_t *check_coalescing(opal_list_t *frag_list,
qp = to_base_frag(frag)->base.order; qp = to_base_frag(frag)->base.order;
if(total_length <= mca_btl_openib_component.qp_infos[qp].size) if(total_length <= mca_btl_openib_component.qp_infos[qp].size) {
opal_list_remove_first(frag_list); /* make sure we can allocate a coalescing frag before returning success */
else *cfrag = alloc_coalesced_frag();
if (OPAL_LIKELY(NULL != cfrag)) {
(*cfrag)->send_frag = frag;
(*cfrag)->sent = false;
opal_list_remove_first(frag_list);
} else {
frag = NULL;
}
} else {
frag = NULL; frag = NULL;
}
} }
OPAL_THREAD_UNLOCK(lock); OPAL_THREAD_UNLOCK(lock);
@ -1188,7 +1199,7 @@ mca_btl_base_descriptor_t* mca_btl_openib_alloc(
mca_btl_openib_module_t *obtl = (mca_btl_openib_module_t*)btl; mca_btl_openib_module_t *obtl = (mca_btl_openib_module_t*)btl;
int qp = frag_size_to_order(obtl, size); int qp = frag_size_to_order(obtl, size);
mca_btl_openib_send_frag_t *sfrag = NULL; mca_btl_openib_send_frag_t *sfrag = NULL;
mca_btl_openib_coalesced_frag_t *cfrag; mca_btl_openib_coalesced_frag_t *cfrag = NULL;
assert(qp != MCA_BTL_NO_ORDER); assert(qp != MCA_BTL_NO_ORDER);
@ -1197,26 +1208,25 @@ mca_btl_base_descriptor_t* mca_btl_openib_alloc(
int prio = !(flags & MCA_BTL_DES_FLAGS_PRIORITY); int prio = !(flags & MCA_BTL_DES_FLAGS_PRIORITY);
sfrag = check_coalescing(&ep->qps[qp].no_wqe_pending_frags[prio], sfrag = check_coalescing(&ep->qps[qp].no_wqe_pending_frags[prio],
&ep->endpoint_lock, ep, size); &ep->endpoint_lock, ep, size, &cfrag);
if(NULL == sfrag) { if (NULL == sfrag) {
if(BTL_OPENIB_QP_TYPE_PP(qp)) { if(BTL_OPENIB_QP_TYPE_PP(qp)) {
sfrag = check_coalescing(&ep->qps[qp].no_credits_pending_frags[prio], sfrag = check_coalescing(&ep->qps[qp].no_credits_pending_frags[prio],
&ep->endpoint_lock, ep, size); &ep->endpoint_lock, ep, size, &cfrag);
} else { } else {
sfrag = check_coalescing( sfrag = check_coalescing(
&obtl->qps[qp].u.srq_qp.pending_frags[prio], &obtl->qps[qp].u.srq_qp.pending_frags[prio],
&obtl->ib_lock, ep, size); &obtl->ib_lock, ep, size, &cfrag);
} }
} }
} }
if(NULL == sfrag) if (NULL == sfrag) {
return ib_frag_alloc((mca_btl_openib_module_t*)btl, size, order, flags); return ib_frag_alloc((mca_btl_openib_module_t*)btl, size, order, flags);
}
/* begin coalescing message */ /* begin coalescing message */
cfrag = alloc_coalesced_frag();
cfrag->send_frag = sfrag;
/* fix up new coalescing header if this is the first coalesced frag */ /* fix up new coalescing header if this is the first coalesced frag */
if(sfrag->hdr != sfrag->chdr) { if(sfrag->hdr != sfrag->chdr) {
@ -1250,10 +1260,9 @@ mca_btl_base_descriptor_t* mca_btl_openib_alloc(
to_base_frag(cfrag)->segment.base.seg_addr.pval = cfrag->hdr + 1; to_base_frag(cfrag)->segment.base.seg_addr.pval = cfrag->hdr + 1;
to_base_frag(cfrag)->segment.base.seg_len = size; to_base_frag(cfrag)->segment.base.seg_len = size;
/* save coalesced fragment on a main fragment; we will need it after send /* NTH: there is no reason to append the coalesced fragment here. No more
* completion to free it and to call upper layer callback */ * fragments will be added until either send or free has been called on
opal_list_append(&sfrag->coalesced_frags, (opal_list_item_t*)cfrag); * the coalesced frag. */
sfrag->coalesced_length += (size+sizeof(mca_btl_openib_header_coalesced_t));
to_base_frag(cfrag)->base.des_flags = flags; to_base_frag(cfrag)->base.des_flags = flags;
@ -1306,6 +1315,15 @@ int mca_btl_openib_free(
default: default:
break; break;
} }
if (openib_frag_type(des) == MCA_BTL_OPENIB_FRAG_COALESCED && !to_coalesced_frag(des)->sent) {
mca_btl_openib_send_frag_t *sfrag = to_coalesced_frag(des)->send_frag;
/* the coalesced fragment would have sent the original fragment but that
* will not happen so send the fragment now */
mca_btl_openib_endpoint_send(to_com_frag(sfrag)->endpoint, sfrag);
}
MCA_BTL_IB_FRAG_RETURN(des); MCA_BTL_IB_FRAG_RETURN(des);
return OPAL_SUCCESS; return OPAL_SUCCESS;
@ -1703,16 +1721,15 @@ int mca_btl_openib_sendi( struct mca_btl_base_module_t* btl,
{ {
mca_btl_openib_module_t *obtl = (mca_btl_openib_module_t*)btl; mca_btl_openib_module_t *obtl = (mca_btl_openib_module_t*)btl;
size_t size = payload_size + header_size; size_t size = payload_size + header_size;
size_t eager_limit;
int qp = frag_size_to_order(obtl, size), int qp = frag_size_to_order(obtl, size),
prio = !(flags & MCA_BTL_DES_FLAGS_PRIORITY), prio = !(flags & MCA_BTL_DES_FLAGS_PRIORITY),
ib_rc; ib_rc;
int32_t cm_return;
bool do_rdma = false; bool do_rdma = false;
ompi_free_list_item_t* item = NULL; ompi_free_list_item_t* item = NULL;
mca_btl_openib_frag_t *frag; mca_btl_openib_frag_t *frag;
mca_btl_openib_header_t *hdr; mca_btl_openib_header_t *hdr;
int send_signaled; int send_signaled;
int rc;
OPAL_THREAD_LOCK(&ep->endpoint_lock); OPAL_THREAD_LOCK(&ep->endpoint_lock);
@ -1734,45 +1751,26 @@ int mca_btl_openib_sendi( struct mca_btl_base_module_t* btl,
/* Allocate WQE */ /* Allocate WQE */
if(OPAL_UNLIKELY(qp_get_wqe(ep, qp) < 0)) { if(OPAL_UNLIKELY(qp_get_wqe(ep, qp) < 0)) {
goto no_credits_or_wqe; goto cant_send_wqe;
}
/* eager rdma or send ? Check eager rdma credits */
/* Note: Maybe we want to implement isend only for eager rdma ?*/
eager_limit = mca_btl_openib_component.eager_limit +
sizeof(mca_btl_openib_header_coalesced_t) +
sizeof(mca_btl_openib_control_header_t);
if(OPAL_LIKELY(size <= eager_limit)) {
if(acquire_eager_rdma_send_credit(ep) == OPAL_SUCCESS) {
do_rdma = true;
}
}
/* if(!do_rdma && acquire_send_credit(ep, frag) != OPAL_SUCCESS) { */
/* Check send credits if it is no rdma */
if(!do_rdma) {
if(BTL_OPENIB_QP_TYPE_PP(qp)) {
if(OPAL_UNLIKELY(OPAL_THREAD_ADD32(&ep->qps[qp].u.pp_qp.sd_credits, -1) < 0)){
OPAL_THREAD_ADD32(&ep->qps[qp].u.pp_qp.sd_credits, 1);
goto no_credits_or_wqe;
}
} else {
if(OPAL_UNLIKELY(OPAL_THREAD_ADD32(&obtl->qps[qp].u.srq_qp.sd_credits, -1) < 0)){
OPAL_THREAD_ADD32(&obtl->qps[qp].u.srq_qp.sd_credits, 1);
goto no_credits_or_wqe;
}
}
} }
/* Allocate fragment */ /* Allocate fragment */
OMPI_FREE_LIST_GET_MT(&obtl->device->qps[qp].send_free, item); OMPI_FREE_LIST_GET_MT(&obtl->device->qps[qp].send_free, item);
if(OPAL_UNLIKELY(NULL == item)) { if(OPAL_UNLIKELY(NULL == item)) {
/* we don't return NULL because maybe later we will try to coalesce */ /* we don't return NULL because maybe later we will try to coalesce */
goto no_frags; goto cant_send_wqe;
} }
frag = to_base_frag(item); frag = to_base_frag(item);
hdr = to_send_frag(item)->hdr; hdr = to_send_frag(item)->hdr;
/* eager rdma or send ? Check eager rdma credits */
/* Note: Maybe we want to implement isend only for eager rdma ?*/
rc = mca_btl_openib_endpoint_credit_acquire (ep, qp, prio, size, &do_rdma,
to_send_frag(frag), false);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
goto cant_send_frag;
}
frag->segment.base.seg_len = size; frag->segment.base.seg_len = size;
frag->base.order = qp; frag->base.order = qp;
frag->base.des_flags = flags; frag->base.des_flags = flags;
@ -1797,29 +1795,6 @@ int mca_btl_openib_sendi( struct mca_btl_base_module_t* btl,
assert(max_data == payload_size); assert(max_data == payload_size);
} }
/* Set all credits */
BTL_OPENIB_GET_CREDITS(ep->eager_rdma_local.credits, hdr->credits);
if(hdr->credits)
hdr->credits |= BTL_OPENIB_RDMA_CREDITS_FLAG;
if(!do_rdma) {
if(BTL_OPENIB_QP_TYPE_PP(qp) && 0 == hdr->credits) {
BTL_OPENIB_GET_CREDITS(ep->qps[qp].u.pp_qp.rd_credits, hdr->credits);
}
} else {
hdr->credits |= (qp << 11);
}
BTL_OPENIB_GET_CREDITS(ep->qps[qp].u.pp_qp.cm_return, cm_return);
/* cm_seen is only 8 bytes, but cm_return is 32 bytes */
if(cm_return > 255) {
hdr->cm_seen = 255;
cm_return -= 255;
OPAL_THREAD_ADD32(&ep->qps[qp].u.pp_qp.cm_return, cm_return);
} else {
hdr->cm_seen = cm_return;
}
#if BTL_OPENIB_FAILOVER_ENABLED #if BTL_OPENIB_FAILOVER_ENABLED
send_signaled = 1; send_signaled = 1;
#else #else
@ -1827,7 +1802,7 @@ int mca_btl_openib_sendi( struct mca_btl_base_module_t* btl,
#endif #endif
ib_rc = post_send(ep, to_send_frag(item), do_rdma, send_signaled); ib_rc = post_send(ep, to_send_frag(item), do_rdma, send_signaled);
if(!ib_rc) { if (!ib_rc) {
if (0 == send_signaled) { if (0 == send_signaled) {
MCA_BTL_IB_FRAG_RETURN(frag); MCA_BTL_IB_FRAG_RETURN(frag);
} }
@ -1838,33 +1813,21 @@ int mca_btl_openib_sendi( struct mca_btl_base_module_t* btl,
} }
#endif #endif
OPAL_THREAD_UNLOCK(&ep->endpoint_lock); OPAL_THREAD_UNLOCK(&ep->endpoint_lock);
return OPAL_SUCCESS; return OPAL_SUCCESS;
} }
/* Failed to send, do clean up all allocated resources */ /* Failed to send, do clean up all allocated resources */
if(ep->nbo) { if (ep->nbo) {
BTL_OPENIB_HEADER_NTOH(*hdr); BTL_OPENIB_HEADER_NTOH(*hdr);
} }
if(BTL_OPENIB_IS_RDMA_CREDITS(hdr->credits)) {
OPAL_THREAD_ADD32(&ep->eager_rdma_local.credits, mca_btl_openib_endpoint_credit_release (ep, qp, do_rdma, to_send_frag(frag));
BTL_OPENIB_CREDITS(hdr->credits));
} cant_send_frag:
if (!do_rdma && BTL_OPENIB_QP_TYPE_PP(qp)) { MCA_BTL_IB_FRAG_RETURN(frag);
OPAL_THREAD_ADD32(&ep->qps[qp].u.pp_qp.rd_credits, cant_send_wqe:
hdr->credits); qp_put_wqe (ep, qp);
}
no_frags:
if(do_rdma) {
OPAL_THREAD_ADD32(&ep->eager_rdma_remote.tokens, 1);
} else {
if(BTL_OPENIB_QP_TYPE_PP(qp)) {
OPAL_THREAD_ADD32(&ep->qps[qp].u.pp_qp.sd_credits, 1);
} else if BTL_OPENIB_QP_TYPE_SRQ(qp){
OPAL_THREAD_ADD32(&obtl->qps[qp].u.srq_qp.sd_credits, 1);
}
}
no_credits_or_wqe:
qp_put_wqe(ep, qp);
cant_send: cant_send:
OPAL_THREAD_UNLOCK(&ep->endpoint_lock); OPAL_THREAD_UNLOCK(&ep->endpoint_lock);
/* We can not send the data directly, so we just return descriptor */ /* We can not send the data directly, so we just return descriptor */
@ -1888,11 +1851,19 @@ int mca_btl_openib_send(
openib_frag_type(des) == MCA_BTL_OPENIB_FRAG_COALESCED); openib_frag_type(des) == MCA_BTL_OPENIB_FRAG_COALESCED);
if(openib_frag_type(des) == MCA_BTL_OPENIB_FRAG_COALESCED) { if(openib_frag_type(des) == MCA_BTL_OPENIB_FRAG_COALESCED) {
frag = to_coalesced_frag(des)->send_frag;
/* save coalesced fragment on a main fragment; we will need it after send
* completion to free it and to call upper layer callback */
opal_list_append(&frag->coalesced_frags, (opal_list_item_t*) des);
frag->coalesced_length += to_coalesced_frag(des)->hdr->alloc_size +
sizeof(mca_btl_openib_header_coalesced_t);
to_coalesced_frag(des)->sent = true;
to_coalesced_frag(des)->hdr->tag = tag; to_coalesced_frag(des)->hdr->tag = tag;
to_coalesced_frag(des)->hdr->size = des->des_local->seg_len; to_coalesced_frag(des)->hdr->size = des->des_local->seg_len;
if(ep->nbo) if(ep->nbo)
BTL_OPENIB_HEADER_COALESCED_HTON(*to_coalesced_frag(des)->hdr); BTL_OPENIB_HEADER_COALESCED_HTON(*to_coalesced_frag(des)->hdr);
frag = to_coalesced_frag(des)->send_frag;
} else { } else {
frag = to_send_frag(des); frag = to_send_frag(des);
to_com_frag(des)->endpoint = ep; to_com_frag(des)->endpoint = ep;

5
opal/mca/btl/openib/btl_openib.h
Просмотреть файл

@ -129,10 +129,7 @@ struct mca_btl_openib_qp_info_t {
(BTL_OPENIB_QP_TYPE(Q) == MCA_BTL_OPENIB_XRC_QP) (BTL_OPENIB_QP_TYPE(Q) == MCA_BTL_OPENIB_XRC_QP)
typedef enum { typedef enum {
BTL_OPENIB_RQ_SOURCE_DEFAULT, BTL_OPENIB_RQ_SOURCE_DEVICE_INI = MCA_BASE_VAR_SOURCE_MAX,
BTL_OPENIB_RQ_SOURCE_MCA,
BTL_OPENIB_RQ_SOURCE_DEVICE_INI,
BTL_OPENIB_RQ_SOURCE_MAX
} btl_openib_receive_queues_source_t; } btl_openib_receive_queues_source_t;
typedef enum { typedef enum {

8
opal/mca/btl/openib/btl_openib_component.c
Просмотреть файл

@ -2082,9 +2082,7 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
} }
/* If the MCA param was specified, skip all the checks */ /* If the MCA param was specified, skip all the checks */
if ( MCA_BASE_VAR_SOURCE_COMMAND_LINE || if (MCA_BASE_VAR_SOURCE_DEFAULT != mca_btl_openib_component.receive_queues_source) {
MCA_BASE_VAR_SOURCE_ENV ==
mca_btl_openib_component.receive_queues_source) {
goto good; goto good;
} }
@ -2102,7 +2100,7 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
mca_btl_openib_component.receive_queues = mca_btl_openib_component.receive_queues =
strdup(values.receive_queues); strdup(values.receive_queues);
mca_btl_openib_component.receive_queues_source = mca_btl_openib_component.receive_queues_source =
MCA_BASE_VAR_SOURCE_FILE; BTL_OPENIB_RQ_SOURCE_DEVICE_INI;
} }
} }
@ -2138,7 +2136,7 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
from the default, we're ok. But if the from the default, we're ok. But if the
component.receive_queues value came from the 1st component.receive_queues value came from the 1st
device's INI file, we must error. */ device's INI file, we must error. */
else if (BTL_OPENIB_RQ_SOURCE_DEVICE_INI == else if ((mca_base_var_source_t) BTL_OPENIB_RQ_SOURCE_DEVICE_INI ==
mca_btl_openib_component.receive_queues_source) { mca_btl_openib_component.receive_queues_source) {
opal_show_help("help-mpi-btl-openib.txt", opal_show_help("help-mpi-btl-openib.txt",
"locally conflicting receive_queues", true, "locally conflicting receive_queues", true,

104
opal/mca/btl/openib/btl_openib_endpoint.c
Просмотреть файл

@ -1,4 +1,4 @@
/* -*- Mode: C; c-basic-offset:4 ; -*- */ /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/* /*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology * University Research and Technology
@ -11,7 +11,7 @@
* Copyright (c) 2004-2005 The Regents of the University of California. * Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved. * All rights reserved.
* Copyright (c) 2006-2013 Cisco Systems, Inc. All rights reserved. * Copyright (c) 2006-2013 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2006-2012 Los Alamos National Security, LLC. All rights * Copyright (c) 2006-2014 Los Alamos National Security, LLC. All rights
* reserved. * reserved.
* Copyright (c) 2006-2007 Voltaire All rights reserved. * Copyright (c) 2006-2007 Voltaire All rights reserved.
* Copyright (c) 2006-2009 Mellanox Technologies, Inc. All rights reserved. * Copyright (c) 2006-2009 Mellanox Technologies, Inc. All rights reserved.
@ -54,7 +54,7 @@
static void mca_btl_openib_endpoint_construct(mca_btl_base_endpoint_t* endpoint); static void mca_btl_openib_endpoint_construct(mca_btl_base_endpoint_t* endpoint);
static void mca_btl_openib_endpoint_destruct(mca_btl_base_endpoint_t* endpoint); static void mca_btl_openib_endpoint_destruct(mca_btl_base_endpoint_t* endpoint);
static inline int acruire_wqe(mca_btl_openib_endpoint_t *ep, static inline int acquire_wqe(mca_btl_openib_endpoint_t *ep,
mca_btl_openib_send_frag_t *frag) mca_btl_openib_send_frag_t *frag)
{ {
int qp = to_base_frag(frag)->base.order; int qp = to_base_frag(frag)->base.order;
@ -70,91 +70,34 @@ static inline int acruire_wqe(mca_btl_openib_endpoint_t *ep,
return OPAL_SUCCESS; return OPAL_SUCCESS;
} }
static int acquire_send_credit(mca_btl_openib_endpoint_t *endpoint,
mca_btl_openib_send_frag_t *frag)
{
mca_btl_openib_module_t *openib_btl = endpoint->endpoint_btl;
int qp = to_base_frag(frag)->base.order;
int prio = !(to_base_frag(frag)->base.des_flags & MCA_BTL_DES_FLAGS_PRIORITY);
if(BTL_OPENIB_QP_TYPE_PP(qp)) {
if(OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, 1);
opal_list_append(&endpoint->qps[qp].no_credits_pending_frags[prio],
(opal_list_item_t *)frag);
return OPAL_ERR_OUT_OF_RESOURCE;
}
} else {
if(OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, -1) < 0)
{
OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, 1);
OPAL_THREAD_LOCK(&openib_btl->ib_lock);
opal_list_append(&openib_btl->qps[qp].u.srq_qp.pending_frags[prio],
(opal_list_item_t *)frag);
OPAL_THREAD_UNLOCK(&openib_btl->ib_lock);
return OPAL_ERR_OUT_OF_RESOURCE;
}
}
return OPAL_SUCCESS;
}
/* this function is called with endpoint->endpoint_lock held */ /* this function is called with endpoint->endpoint_lock held */
int mca_btl_openib_endpoint_post_send(mca_btl_openib_endpoint_t *endpoint, int mca_btl_openib_endpoint_post_send(mca_btl_openib_endpoint_t *endpoint,
mca_btl_openib_send_frag_t *frag) mca_btl_openib_send_frag_t *frag)
{ {
int prio = !(to_base_frag(frag)->base.des_flags & MCA_BTL_DES_FLAGS_PRIORITY);
mca_btl_openib_header_t *hdr = frag->hdr; mca_btl_openib_header_t *hdr = frag->hdr;
mca_btl_base_descriptor_t *des = &to_base_frag(frag)->base; mca_btl_base_descriptor_t *des = &to_base_frag(frag)->base;
int qp, ib_rc; int qp, ib_rc, rc;
int32_t cm_return;
bool do_rdma = false; bool do_rdma = false;
size_t eager_limit; size_t size;
if(OPAL_LIKELY(des->order == MCA_BTL_NO_ORDER)) if(OPAL_LIKELY(des->order == MCA_BTL_NO_ORDER))
des->order = frag->qp_idx; des->order = frag->qp_idx;
qp = des->order; qp = des->order;
if(acruire_wqe(endpoint, frag) != OPAL_SUCCESS) if(acquire_wqe(endpoint, frag) != OPAL_SUCCESS)
return OPAL_ERR_RESOURCE_BUSY; return OPAL_ERR_RESOURCE_BUSY;
eager_limit = mca_btl_openib_component.eager_limit + size = des->des_local->seg_len + frag->coalesced_length;
sizeof(mca_btl_openib_header_coalesced_t) +
sizeof(mca_btl_openib_control_header_t);
if(des->des_local->seg_len + frag->coalesced_length <= eager_limit &&
(des->des_flags & MCA_BTL_DES_FLAGS_PRIORITY)) {
/* High priority frag. Try to send over eager RDMA */
if(acquire_eager_rdma_send_credit(endpoint) == OPAL_SUCCESS)
do_rdma = true;
}
if(!do_rdma && acquire_send_credit(endpoint, frag) != OPAL_SUCCESS) { rc = mca_btl_openib_endpoint_credit_acquire (endpoint, qp, prio, size,
&do_rdma, frag, true);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
qp_put_wqe(endpoint, qp); qp_put_wqe(endpoint, qp);
return OPAL_ERR_RESOURCE_BUSY; return OPAL_ERR_RESOURCE_BUSY;
} }
BTL_OPENIB_GET_CREDITS(endpoint->eager_rdma_local.credits, hdr->credits);
if(hdr->credits)
hdr->credits |= BTL_OPENIB_RDMA_CREDITS_FLAG;
if(!do_rdma) {
if(BTL_OPENIB_QP_TYPE_PP(qp) && 0 == hdr->credits) {
BTL_OPENIB_GET_CREDITS(endpoint->qps[qp].u.pp_qp.rd_credits, hdr->credits);
}
} else {
hdr->credits |= (qp << 11);
}
BTL_OPENIB_GET_CREDITS(endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
/* cm_seen is only 8 bytes, but cm_return is 32 bytes */
if(cm_return > 255) {
hdr->cm_seen = 255;
cm_return -= 255;
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
} else {
hdr->cm_seen = cm_return;
}
qp_reset_signal_count(endpoint, qp); qp_reset_signal_count(endpoint, qp);
ib_rc = post_send(endpoint, frag, do_rdma, 1); ib_rc = post_send(endpoint, frag, do_rdma, 1);
@ -164,27 +107,12 @@ int mca_btl_openib_endpoint_post_send(mca_btl_openib_endpoint_t *endpoint,
if(endpoint->nbo) if(endpoint->nbo)
BTL_OPENIB_HEADER_NTOH(*hdr); BTL_OPENIB_HEADER_NTOH(*hdr);
if(BTL_OPENIB_IS_RDMA_CREDITS(hdr->credits)) { mca_btl_openib_endpoint_credit_release (endpoint, qp, do_rdma, frag);
OPAL_THREAD_ADD32(&endpoint->eager_rdma_local.credits,
BTL_OPENIB_CREDITS(hdr->credits));
}
qp_put_wqe(endpoint, qp); qp_put_wqe(endpoint, qp);
if(do_rdma) { BTL_ERROR(("error posting send request error %d: %s. size = %lu\n",
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1); ib_rc, strerror(ib_rc), size));
} else {
if(BTL_OPENIB_QP_TYPE_PP(qp)) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.rd_credits,
hdr->credits);
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, 1);
} else if BTL_OPENIB_QP_TYPE_SRQ(qp){
mca_btl_openib_module_t *openib_btl = endpoint->endpoint_btl;
OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, 1);
}
}
BTL_ERROR(("error posting send request error %d: %s\n",
ib_rc, strerror(ib_rc)));
return OPAL_ERROR; return OPAL_ERROR;
} }
@ -709,8 +637,8 @@ void mca_btl_openib_endpoint_connected(mca_btl_openib_endpoint_t *endpoint)
/* We need to post this one */ /* We need to post this one */
if (OPAL_ERROR == mca_btl_openib_endpoint_post_send(endpoint, frag)) { if (OPAL_ERROR == mca_btl_openib_endpoint_post_send(endpoint, frag)) {
BTL_ERROR(("Error posting send")); BTL_ERROR(("Error posting send"));
} }
} }
OPAL_THREAD_UNLOCK(&endpoint->endpoint_lock); OPAL_THREAD_UNLOCK(&endpoint->endpoint_lock);

95
opal/mca/btl/openib/btl_openib_endpoint.h
Просмотреть файл

@ -625,6 +625,101 @@ static inline int post_send(mca_btl_openib_endpoint_t *ep,
return ibv_post_send(ep->qps[qp].qp->lcl_qp, sr_desc, &bad_wr); return ibv_post_send(ep->qps[qp].qp->lcl_qp, sr_desc, &bad_wr);
} }
/* called with the endpoint lock held */
static inline int mca_btl_openib_endpoint_credit_acquire (struct mca_btl_base_endpoint_t *endpoint, int qp,
int prio, size_t size, bool *do_rdma,
mca_btl_openib_send_frag_t *frag, bool queue_frag)
{
mca_btl_openib_module_t *openib_btl = endpoint->endpoint_btl;
mca_btl_openib_header_t *hdr = frag->hdr;
size_t eager_limit;
int32_t cm_return;
eager_limit = mca_btl_openib_component.eager_limit +
sizeof(mca_btl_openib_header_coalesced_t) +
sizeof(mca_btl_openib_control_header_t);
if (!(prio && size < eager_limit && acquire_eager_rdma_send_credit(endpoint) == OPAL_SUCCESS)) {
*do_rdma = false;
if (BTL_OPENIB_QP_TYPE_PP(qp)) {
if (OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, -1) < 0) {
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, 1);
if (queue_frag) {
opal_list_append(&endpoint->qps[qp].no_credits_pending_frags[prio],
(opal_list_item_t *)frag);
}
return OPAL_ERR_OUT_OF_RESOURCE;
}
} else {
if(OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, -1) < 0) {
OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, 1);
if (queue_frag) {
OPAL_THREAD_LOCK(&openib_btl->ib_lock);
opal_list_append(&openib_btl->qps[qp].u.srq_qp.pending_frags[prio],
(opal_list_item_t *)frag);
OPAL_THREAD_UNLOCK(&openib_btl->ib_lock);
}
return OPAL_ERR_OUT_OF_RESOURCE;
}
}
} else {
/* High priority frag. Try to send over eager RDMA */
*do_rdma = true;
}
/* Set all credits */
BTL_OPENIB_GET_CREDITS(endpoint->eager_rdma_local.credits, hdr->credits);
if (hdr->credits) {
hdr->credits |= BTL_OPENIB_RDMA_CREDITS_FLAG;
}
if (!*do_rdma) {
if (BTL_OPENIB_QP_TYPE_PP(qp) && 0 == hdr->credits) {
BTL_OPENIB_GET_CREDITS(endpoint->qps[qp].u.pp_qp.rd_credits, hdr->credits);
}
} else {
hdr->credits |= (qp << 11);
}
BTL_OPENIB_GET_CREDITS(endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
/* cm_seen is only 8 bytes, but cm_return is 32 bytes */
if(cm_return > 255) {
hdr->cm_seen = 255;
cm_return -= 255;
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.cm_return, cm_return);
} else {
hdr->cm_seen = cm_return;
}
return OPAL_SUCCESS;
}
/* called with the endpoint lock held. */
static inline void mca_btl_openib_endpoint_credit_release (struct mca_btl_base_endpoint_t *endpoint, int qp,
bool do_rdma, mca_btl_openib_send_frag_t *frag)
{
mca_btl_openib_header_t *hdr = frag->hdr;
if (BTL_OPENIB_IS_RDMA_CREDITS(hdr->credits)) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_local.credits, BTL_OPENIB_CREDITS(hdr->credits));
}
if (do_rdma) {
OPAL_THREAD_ADD32(&endpoint->eager_rdma_remote.tokens, 1);
} else {
if(BTL_OPENIB_QP_TYPE_PP(qp)) {
OPAL_THREAD_ADD32 (&endpoint->qps[qp].u.pp_qp.rd_credits, hdr->credits);
OPAL_THREAD_ADD32(&endpoint->qps[qp].u.pp_qp.sd_credits, 1);
} else if BTL_OPENIB_QP_TYPE_SRQ(qp){
mca_btl_openib_module_t *openib_btl = endpoint->endpoint_btl;
OPAL_THREAD_ADD32(&openib_btl->qps[qp].u.srq_qp.sd_credits, 1);
}
}
}
END_C_DECLS END_C_DECLS
#endif #endif

1
opal/mca/btl/openib/btl_openib_frag.h
Просмотреть файл

@ -371,6 +371,7 @@ typedef struct mca_btl_openib_coalesced_frag_t {
mca_btl_openib_frag_t super; mca_btl_openib_frag_t super;
mca_btl_openib_send_frag_t *send_frag; mca_btl_openib_send_frag_t *send_frag;
mca_btl_openib_header_coalesced_t *hdr; mca_btl_openib_header_coalesced_t *hdr;
bool sent;
} mca_btl_openib_coalesced_frag_t; } mca_btl_openib_coalesced_frag_t;
OBJ_CLASS_DECLARATION(mca_btl_openib_coalesced_frag_t); OBJ_CLASS_DECLARATION(mca_btl_openib_coalesced_frag_t);