Skip to content

UCT/DEVICE: ring db in atomic way #10994

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

UCT/DEVICE: ring db in atomic way #10994

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 16 additions & 15 deletions src/uct/ib/mlx5/gdaki/gdaki.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,24 +258,25 @@ UCS_F_DEVICE void uct_rc_mlx5_gda_db(uct_rc_gdaki_dev_ep_t *ep,
{
cuda::atomic_ref<uint64_t, cuda::thread_scope_device> ref(
ep->sq_ready_index);
uint64_t wqe_base_orig = wqe_base;
const uint64_t wqe_next = wqe_base + count;
const bool skip_db = !(flags & UCT_DEVICE_FLAG_NODELAY) &&
!((wqe_base ^ wqe_next) & 128);

__threadfence();
while (!ref.compare_exchange_strong(wqe_base, wqe_base + count,
cuda::std::memory_order_relaxed)) {
wqe_base = wqe_base_orig;
}

if (!(flags & UCT_DEVICE_FLAG_NODELAY) &&
!((wqe_base ^ (wqe_base + count)) & 128)) {
return;
if (skip_db) {
const uint64_t wqe_base_orig = wqe_base;
while (!ref.compare_exchange_strong(wqe_base, wqe_next,
cuda::std::memory_order_relaxed)) {
wqe_base = wqe_base_orig;
}
} else {
while (READ_ONCE(ep->sq_ready_index) != wqe_base) {
}
uct_rc_mlx5_gda_ring_db(ep, wqe_next);
uct_rc_mlx5_gda_update_dbr(ep, wqe_next);
uct_rc_mlx5_gda_ring_db(ep, wqe_next);
ref.store(wqe_next, cuda::std::memory_order_release);
}

uct_rc_mlx5_gda_lock(&ep->sq_lock);
uct_rc_mlx5_gda_ring_db(ep, ep->sq_ready_index);
uct_rc_mlx5_gda_update_dbr(ep, ep->sq_ready_index);
uct_rc_mlx5_gda_ring_db(ep, ep->sq_ready_index);
uct_rc_mlx5_gda_unlock(&ep->sq_lock);
}

UCS_F_DEVICE bool
Expand Down