Skip to content

[Backport branch/3.4.x] Fixes a race condition in DeviceTopK #10682

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
github-actions wants to merge 1 commit into
base: branch/3.4.x
Choose a base branch
from
Open

[Backport branch/3.4.x] Fixes a race condition in DeviceTopK #10682

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
6 changes: 4 additions & 2 deletions cub/cub/agent/agent_topk.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -645,10 +645,12 @@ struct AgentTopK
bool is_last_pass,
CounterUpdateFn counter_update_fn)
{
// Ensure all writes to the global memory-histogram are visible to all threads before
// proceeding to compute the prefix sum over the histogram.
// Make this block's histogram contributions visible device-wide before publishing retirement
__threadfence();

// Wait for all threads in this block to finish merge_histograms() before thread 0 signals completion
__syncthreads();

// Identify the last block in the grid to perform the prefix sum over the histogram
bool is_last_block = false;
if (threadIdx.x == 0)
Expand Down