Skip to content

fix: synchronize custom collectives before return #4082

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

Merged
junhaha666 merged 1 commit into from
Jul 29, 2026
Merged

fix: synchronize custom collectives before return #4082

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
8 changes: 8 additions & 0 deletions csrc/include/custom_all_reduce.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -709,6 +709,7 @@ __global__ void __launch_bounds__(512, 1) allgather_naive(
int write_idx = warp_id * size + idx;
result[write_idx] = ptrs[warp_id][idx];
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

template <typename T, int ngpus>
Expand Down Expand Up @@ -736,6 +737,7 @@ __global__ void __launch_bounds__(512, 1) allgather_vec(
int write_idx = warp_id * size + idx;
*(reinterpret_cast<P*>(&result[0]) + write_idx) = ptrs[warp_id][idx];
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

template <typename T, int ngpus>
Expand Down Expand Up @@ -772,6 +774,7 @@ __global__ void __launch_bounds__(512, 1) allgather_lastdim(RankData* _dp,
int write_idx = (ngpus * y + warp_id) * last_dim_size + x;
*(reinterpret_cast<P*>(&result[0]) + write_idx) = ptrs[warp_id][idx];
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

// ========== reduce_scatter kernel start ==========
Expand Down Expand Up @@ -813,6 +816,7 @@ __global__ void __launch_bounds__(512, 1) reduce_scatter_split_first_dim(
*(reinterpret_cast<P*>(result) + store_index) =
packed_reduce<P, ngpus, A>(ptrs, load_index);
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

// reduce_scatter, scatter on last dim — naive (non-vectorized) fallback.
Expand Down Expand Up @@ -850,6 +854,7 @@ __global__ void __launch_bounds__(256, 1) reduce_scatter_split_lastdim_naive(
}
result[i] = downcast_s<T>(rslt_reg);
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

// reduce_scatter, scatter on last dim — vectorized (16B / thread).
Expand Down Expand Up @@ -883,6 +888,7 @@ __global__ void __launch_bounds__(256, 1) reduce_scatter_split_lastdim(
int load_index = index_y * packed_dim_n + rank * splited_n + index_x;
*(reinterpret_cast<P*>(result) + i) = packed_reduce<P, ngpus, A>(ptrs, load_index);
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

// reduce_scatter, scatter on middle dim — naive (non-vectorized) fallback.
Expand Down Expand Up @@ -920,6 +926,7 @@ __global__ void __launch_bounds__(256, 1) reduce_scatter_split_middim_naive(
}
result[i] = downcast_s<T>(rslt_reg);
}
end_sync<ngpus, true>(sg, self_sg, rank);
}

// reduce_scatter, scatter on middle dim — vectorized (16B / thread along k).
Expand Down Expand Up @@ -954,6 +961,7 @@ __global__ void __launch_bounds__(256, 1) reduce_scatter_split_middim(
int load_index = index_m * (n * packed_dim_k) + (rank * splited_n + index_n) * packed_dim_k + index_k;
*(reinterpret_cast<P*>(result) + i) = packed_reduce<P, ngpus, A>(ptrs, load_index);
}
end_sync<ngpus, true>(sg, self_sg, rank);
}
// ========== reduce_scatter kernel end ==========

Expand Down
Loading