Skip to content

Run the NN-descent per-iteration host update on the calling thread #2748

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

Closed
bdice wants to merge 1 commit into from
Closed

Run the NN-descent per-iteration host update on the calling thread #2748

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
9 changes: 4 additions & 5 deletions cpp/src/neighbors/detail/nn_descent.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2751,8 +2751,6 @@ void GNND<Data_t, Index_t>::build(Data_t* data,
raft::copy(res, d_list_sizes_old_.view(), graph_.h_list_sizes_old.view());
raft::resource::sync_stream(res);

std::thread update_and_sample_thread(update_and_sample, it);

RAFT_LOG_DEBUG("# GNND iteration: %lu / %lu", it + 1, build_config_.max_iterations);

// Reuse dists_buffer_ to save GPU memory. graph_buffer_ cannot be reused, because it
Expand Down Expand Up @@ -2786,7 +2784,8 @@ void GNND<Data_t, Index_t>::build(Data_t* data,
THROW("NN_DESCENT cannot be run for __CUDA_ARCH__ < 700");
}

update_and_sample_thread.join();
// Overlaps with the device work enqueued above, which uses disjoint buffers.
update_and_sample(it > 0);

if (update_counter_ == -1) { break; }
raft::copy(res, graph_host_buffer_.view(), graph_buffer_.view());
Expand Down Expand Up @@ -2897,7 +2896,6 @@ void GNND<Data_t, Index_t>::build(
raft::copy(res, d_list_sizes_old_.view(), graph_.h_list_sizes_old.view());
raft::resource::sync_stream(res);

std::thread update_and_sample_thread(update_and_sample, it);
RAFT_LOG_DEBUG("# GNND iteration: %lu / %lu", it + 1, build_config_.max_iterations);

static_assert(DEGREE_ON_DEVICE * sizeof(*(dists_buffer_.data_handle())) >=
Expand All @@ -2914,7 +2912,8 @@ void GNND<Data_t, Index_t>::build(
stream);

local_join(stream, dataset, dist_epilogue);
update_and_sample_thread.join();
// Overlaps with the device work enqueued above, which uses disjoint buffers.
update_and_sample(it > 0);
if (update_counter_ == -1) { break; }
raft::copy(res, graph_host_buffer_.view(), graph_buffer_.view());
raft::copy(res, dists_host_buffer_.view(), dists_buffer_.view());
Expand Down
Loading