Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

RNNOp to call cudaEventCreate lazily #16584

Merged
merged 2 commits into from
Oct 23, 2019
Merged

RNNOp to call cudaEventCreate lazily #16584

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8ab2d3b
RNNOp to call cudaEventCreate lazily
DickJC123 Oct 22, 2019
edd5cc9
Saw unrelated failure. Trigger CI.
DickJC123 Oct 23, 2019
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: 7 additions & 2 deletions src/operator/rnn-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -493,7 +493,6 @@ class RNNOp {

CUDNN_CALL(cudnnCreateRNNDescriptor(&rnn_desc_));
CUDNN_CALL(cudnnCreateDropoutDescriptor(&dropout_desc_));
CUDA_CALL(cudaEventCreateWithFlags(&dgrad_sync_event_, cudaEventDisableTiming));

#if MXNET_USE_CUDNN_GE_7200
CUDNN_CALL(cudnnCreateRNNDataDescriptor(&x_data_desc_));
Expand Down Expand Up @@ -538,7 +537,8 @@ class RNNOp {
CUDNN_CALL(cudnnDestroyFilterDescriptor(dw_desc_));
CUDNN_CALL(cudnnDestroyRNNDescriptor(rnn_desc_));
CUDNN_CALL(cudnnDestroyDropoutDescriptor(dropout_desc_));
CUDA_CALL(cudaEventDestroy(dgrad_sync_event_));
if (dgrad_sync_event_created_)
CUDA_CALL(cudaEventDestroy(dgrad_sync_event_));

if (init_cudnn_) {
for (size_t i = 0; i < x_desc_vec_.size(); ++i) {
Expand Down Expand Up @@ -1502,6 +1502,10 @@ class RNNOp {
if (CUDNN_VERSION <= 7604 && dgrad_sync_needed_) {
// Without blocking the CPU, create a synchronization point of all current GPU activity. No
// need to call cudaStreamWaitEvent- cudaEventRecord on the legacy default stream suffices.
if (!dgrad_sync_event_created_) {
CUDA_CALL(cudaEventCreateWithFlags(&dgrad_sync_event_, cudaEventDisableTiming));
dgrad_sync_event_created_ = true;
}
CUDA_CALL(cudaEventRecord(dgrad_sync_event_, cudaStreamLegacy));
}
}
Expand Down Expand Up @@ -1535,6 +1539,7 @@ class RNNOp {

cudnnTensorFormat_t format_;
cudaEvent_t dgrad_sync_event_;
bool dgrad_sync_event_created_ = false;
bool dgrad_sync_needed_ = false;
#endif // MXNET_USE_CUDNN
bool init_space_, temp_init_space_;
Expand Down