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This repository has been archived by the owner on Nov 17, 2023.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

RNNOp only call cuda/cudnn if GPU ctx is requested #16632

Merged
merged 1 commit into from
Oct 27, 2019
+5 −0
Merged

RNNOp only call cuda/cudnn if GPU ctx is requested #16632

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
5 changes: 5 additions & 0 deletions src/operator/rnn-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -422,6 +422,8 @@ class RNNOp {
init_mem_ = false;
reserve_mem_size_ = 0;
#endif

if (ctx_.dev_type == kGPU) {
#if MXNET_USE_CUDNN == 1
init_cudnn_ = false;
dtype_ = mshadow::DataType<DType>::kCudnnFlag;
Expand Down Expand Up @@ -505,6 +507,7 @@ class RNNOp {
LOG(FATAL) << "RNN on GPU is only available for cuDNN at the moment.";
}
#endif // MXNET_USE_CUDNN == 1
}

if (ctx_.dev_type == kCPU) {
this->init_space_ = false;
Expand All @@ -523,6 +526,7 @@ class RNNOp {
}

~RNNOp() {
if (ctx_.dev_type == kGPU) {
#if MXNET_USE_CUDNN == 1
CUDNN_CALL(cudnnDestroyTensorDescriptor(hx_desc_));
CUDNN_CALL(cudnnDestroyTensorDescriptor(cx_desc_));
Expand Down Expand Up @@ -557,6 +561,7 @@ class RNNOp {
CUDNN_CALL(cudnnDestroyRNNDataDescriptor(dy_data_desc_));
#endif // MXNET_USE_CUDNN_GE_7200
#endif // MXNET_USE_CUDNN
}
}

void Forward(const OpContext &ctx, const std::vector<TBlob> &in_data,
Expand Down