Skip to content
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.

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

1.14-rc1 cherry-pick request: NCCL broadcast bug #29231

Merged
merged 2 commits into from May 31, 2019
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
2 changes: 1 addition & 1 deletion tensorflow/core/kernels/nccl_ops.cc
Expand Up @@ -248,7 +248,7 @@ class NcclBroadcastRecvKernel : public NcclAsyncOpBase {
compute_stream->parent(), compute_stream, gpu_info->event_mgr,
gpu_info->gpu_id, /*input=*/nullptr, output, /*global_rank=*/-1,
std::move(actual_done));
NcclManager::instance()->AddBroadcastSend(
NcclManager::instance()->AddBroadcastRecv(
std::move(participant), {GetCollectiveKey(c),
/*num_local_devices=*/num_devices(),
/*num_global_devices=*/num_devices(),
Expand Down
7 changes: 6 additions & 1 deletion tensorflow/core/nccl/nccl_manager.cc
Expand Up @@ -410,7 +410,12 @@ void NcclManager::AddParticipant(std::unique_ptr<Participant> participant,
CollectiveType collective_type,
ncclRedOp_t reduction_op) {
Collective* to_run = nullptr;
const DataType data_type = participant->input->dtype();
DataType data_type;
if (participant->input != nullptr) {
data_type = participant->input->dtype();
} else {
data_type = participant->output->dtype();
}
{
mutex_lock l(mu_);
auto collective_it = collectives_.find(context.collective_key);
Expand Down