Skip to content

Commit

Permalink
Merge pull request #4622 from brminich/hwtm/sw_rndv_thresh_fix
Browse files Browse the repository at this point in the history 
UCP: Fix minimum rndv thresh for tag offload
  • Loading branch information
@brminich
brminich authored Jan 10, 2020
2 parents cda6f1c + 9919b12 commit 373bd29
Show file tree
Hide file tree
Showing 4 changed files with 75 additions and 18 deletions.
63 changes: 45 additions & 18 deletions src/ucp/core/ucp_ep.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@
#include <ucp/wireup/wireup_cm.h>
#include <ucp/tag/eager.h>
#include <ucp/tag/offload.h>
#include <ucp/tag/rndv.h>
#include <ucp/stream/stream.h>
#include <ucp/core/ucp_listener.h>
#include <ucs/datastruct/queue.h>
Expand Down Expand Up @@ -1072,12 +1073,24 @@ static void ucp_ep_config_adjust_max_short(ssize_t *max_short,
ucs_assert(*max_short >= -1);
}

static void ucp_ep_config_set_am_rndv_thresh(ucp_worker_h worker, uct_iface_attr_t *iface_attr,
uct_md_attr_t *md_attr, ucp_ep_config_t *config,
/* With tag offload, SW RNDV requests are temporarily stored in the receiver
* user buffer when matched. Thus, minimum message size allowed to be sent with
* RNDV protocol should be bigger than maximal possible SW RNDV request
* (i.e. header plus packed keys size). */
size_t ucp_ep_tag_offload_min_rndv_thresh(ucp_ep_config_t *config)
{
return sizeof(ucp_rndv_rts_hdr_t) + config->tag.rndv.rkey_size;
}

static void ucp_ep_config_set_am_rndv_thresh(ucp_worker_h worker,
uct_iface_attr_t *iface_attr,
uct_md_attr_t *md_attr,
ucp_ep_config_t *config,
size_t min_rndv_thresh,
size_t max_rndv_thresh)
{
ucp_context_h context = worker->context;
size_t rndv_thresh, rndv_nbr_thresh;
size_t rndv_thresh, rndv_nbr_thresh, min_thresh;

ucs_assert(config->key.am_lane != UCP_NULL_LANE);
ucs_assert(config->key.lanes[config->key.am_lane].rsc_index != UCP_NULL_RESOURCE);
Expand All @@ -1102,12 +1115,14 @@ static void ucp_ep_config_set_am_rndv_thresh(ucp_worker_h worker, uct_iface_attr
rndv_thresh);
}

min_thresh = ucs_max(iface_attr->cap.am.min_zcopy, min_rndv_thresh);

config->tag.rndv.am_thresh = ucp_ep_thresh(rndv_thresh,
iface_attr->cap.am.min_zcopy,
min_thresh,
max_rndv_thresh);

config->tag.rndv_send_nbr.am_thresh = ucp_ep_thresh(rndv_nbr_thresh,
iface_attr->cap.am.min_zcopy,
min_thresh,
max_rndv_thresh);

ucs_trace("Active Message rndv threshold is %zu (send_nbr: %zu)",
Expand All @@ -1117,12 +1132,13 @@ static void ucp_ep_config_set_am_rndv_thresh(ucp_worker_h worker, uct_iface_attr
static void ucp_ep_config_set_rndv_thresh(ucp_worker_t *worker,
ucp_ep_config_t *config,
ucp_lane_index_t *lanes,
size_t min_rndv_thresh,
size_t max_rndv_thresh)
{
ucp_context_t *context = worker->context;
ucp_lane_index_t lane = lanes[0];
ucp_rsc_index_t rsc_index;
size_t rndv_thresh, rndv_nbr_thresh;
size_t rndv_thresh, rndv_nbr_thresh, min_thresh;
uct_iface_attr_t *iface_attr;

if (lane == UCP_NULL_LANE) {
Expand Down Expand Up @@ -1155,13 +1171,15 @@ static void ucp_ep_config_set_rndv_thresh(ucp_worker_t *worker,
rndv_thresh);
}

min_thresh = ucs_max(iface_attr->cap.get.min_zcopy, min_rndv_thresh);

/* TODO: need to check minimal PUT Zcopy */
config->tag.rndv.rma_thresh = ucp_ep_thresh(rndv_thresh,
iface_attr->cap.get.min_zcopy,
min_thresh,
max_rndv_thresh);

config->tag.rndv_send_nbr.rma_thresh = ucp_ep_thresh(rndv_nbr_thresh,
iface_attr->cap.get.min_zcopy,
min_thresh,
max_rndv_thresh);

ucs_trace("rndv threshold is %zu (send_nbr: %zu)",
Expand Down Expand Up @@ -1278,6 +1296,8 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
size_t it;
size_t max_rndv_thresh;
size_t max_am_rndv_thresh;
size_t min_rndv_thresh;
size_t min_am_rndv_thresh;
ucs_status_t status;
double rndv_max_bw;
int i;
Expand Down Expand Up @@ -1325,6 +1345,7 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
config->am_u.reply_proto = &ucp_am_reply_proto;
max_rndv_thresh = SIZE_MAX;
max_am_rndv_thresh = SIZE_MAX;
min_am_rndv_thresh = 0;

config->tag.offload.max_eager_short.memtype_on = -1;
config->tag.offload.max_eager_short.memtype_off = -1;
Expand Down Expand Up @@ -1400,13 +1421,15 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
UCT_IFACE_FLAG_TAG_EAGER_ZCOPY, 0,
iface_attr->cap.tag.eager.max_bcopy);

config->tag.offload.max_rndv_iov = iface_attr->cap.tag.rndv.max_iov;
config->tag.offload.max_rndv_zcopy = iface_attr->cap.tag.rndv.max_zcopy;
config->tag.sync_proto = &ucp_tag_offload_sync_proto;
config->tag.proto = &ucp_tag_offload_proto;
config->tag.lane = lane;
max_rndv_thresh = iface_attr->cap.tag.eager.max_zcopy;
max_am_rndv_thresh = iface_attr->cap.tag.eager.max_bcopy;
config->tag.offload.max_rndv_iov = iface_attr->cap.tag.rndv.max_iov;
config->tag.offload.max_rndv_zcopy = iface_attr->cap.tag.rndv.max_zcopy;
config->tag.sync_proto = &ucp_tag_offload_sync_proto;
config->tag.proto = &ucp_tag_offload_proto;
config->tag.lane = lane;
max_rndv_thresh = iface_attr->cap.tag.eager.max_zcopy;
max_am_rndv_thresh = iface_attr->cap.tag.eager.max_bcopy;
min_rndv_thresh = ucp_ep_tag_offload_min_rndv_thresh(config);
min_am_rndv_thresh = min_rndv_thresh;

ucs_assert_always(iface_attr->cap.tag.rndv.max_hdr >=
sizeof(ucp_tag_offload_unexp_rndv_hdr_t));
Expand All @@ -1415,7 +1438,7 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
/* Must have active messages for using rendezvous */
tag_lanes[0] = lane;
ucp_ep_config_set_rndv_thresh(worker, config, tag_lanes,
max_rndv_thresh);
min_rndv_thresh, max_rndv_thresh);
}

/* Max Eager short has to be set after Zcopy and RNDV thresholds */
Expand Down Expand Up @@ -1452,11 +1475,14 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
* tag-matching protocols */
/* TODO: set threshold level based on all available lanes */

config->tag.eager = config->am;
config->tag.lane = lane;
config->tag.eager = config->am;
config->tag.lane = lane;
min_rndv_thresh = iface_attr->cap.get.min_zcopy;
min_am_rndv_thresh = iface_attr->cap.am.min_zcopy;

ucp_ep_config_set_rndv_thresh(worker, config,
config->key.rma_bw_lanes,
min_rndv_thresh,
max_rndv_thresh);

/* Max Eager short has to be set after Zcopy and RNDV thresholds */
Expand All @@ -1468,6 +1494,7 @@ ucs_status_t ucp_ep_config_init(ucp_worker_h worker, ucp_ep_config_t *config,
/* Calculate rndv threshold for AM Rendezvous, which may be used by
* any tag-matching protocol (AM and offload). */
ucp_ep_config_set_am_rndv_thresh(worker, iface_attr, md_attr, config,
min_am_rndv_thresh,
max_am_rndv_thresh);
} else {
/* Stub endpoint */
Expand Down
2 changes: 2 additions & 0 deletions src/ucp/core/ucp_ep.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -491,4 +491,6 @@ int ucp_ep_is_cm_local_connected(ucp_ep_h ep);

unsigned ucp_ep_local_disconnect_progress(void *arg);

size_t ucp_ep_tag_offload_min_rndv_thresh(ucp_ep_config_t *config);

#endif
14 changes: 14 additions & 0 deletions test/gtest/ucp/test_ucp_tag.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -381,6 +381,20 @@ UCS_TEST_P(test_ucp_tag_limits, check_max_short_rndv_thresh_zero, "RNDV_THRESH=0
ucp_ep_config(sender().ep())->tag.rndv_send_nbr.am_thresh);
EXPECT_LE(max_short,
ucp_ep_config(sender().ep())->tag.rndv_send_nbr.rma_thresh);

if (m_test_offload) {
// There is a lower bound for rndv threshold with tag offload. We should
// not send messages smaller than SW RNDV request size, because receiver
// may temporarily store this request in the user buffer (which will
// result in crash if the request does not fit user buffer).
size_t min_rndv = ucp_ep_tag_offload_min_rndv_thresh(ucp_ep_config(sender().ep()));

EXPECT_GT(min_rndv, 0ul); // min_rndv should be RTS size at least
EXPECT_GE(min_rndv,
ucp_ep_config(sender().ep())->tag.rndv_send_nbr.am_thresh);
EXPECT_GE(min_rndv,
ucp_ep_config(sender().ep())->tag.rndv_send_nbr.rma_thresh);
}
}

UCS_TEST_P(test_ucp_tag_limits, check_max_short_zcopy_thresh_zero, "ZCOPY_THRESH=0") {
Expand Down
14 changes: 14 additions & 0 deletions test/gtest/ucp/test_ucp_tag_offload.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -305,6 +305,20 @@ UCS_TEST_P(test_ucp_tag_offload, connect)
e->connect(&receiver(), get_ep_params());
}

UCS_TEST_P(test_ucp_tag_offload, small_rndv, "RNDV_THRESH=0", "TM_THRESH=0")
{
activate_offload(sender());
send_recv(sender(), 0x11ul, 0ul);
send_recv(sender(), 0x11ul, 1ul);
}

UCS_TEST_P(test_ucp_tag_offload, small_sw_rndv, "RNDV_THRESH=0", "TM_THRESH=0",
"TM_SW_RNDV=y")
{
activate_offload(sender());
send_recv(sender(), 0x11ul, 0ul);
send_recv(sender(), 0x11ul, 1ul);
}

UCP_INSTANTIATE_TEST_CASE(test_ucp_tag_offload)

Expand Down

0 comments on commit 373bd29

Please sign in to comment.