saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Run backward gru in main thread for bidirection (#118)

Browse source
This commit is contained in:
Yufeng Li 2018-12-07 10:24:57 -08:00 committed by GitHub
parent 6169f3c619
commit 23e122f304
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
onnxruntime/core/providers/cpu/rnn/deep_cpu_gru.cc
View file

@ -374,21 +374,6 @@ Status DeepCpuGruOp::ComputeImpl(OpKernelContext& context) const {
gsl::span<T> hidden_output_2 = hidden_output.subspan(hidden_output_size_per_direction,
hidden_output_size_per_direction);
// run backward first as it needs an extra reverse sequence operation
std::packaged_task<void()> task_bw{
[&]() {
std::unique_ptr<detail::UniDirectionalGru<T>> bw = std::make_unique<detail::UniDirectionalGru<T>>(
alloc, logger,
seq_length, batch_size, input_size, hidden_size_, linear_before_reset_, Direction::kReverse,
bias_2, initial_hidden_2,
activation_funcs_.Entries()[2],
activation_funcs_.Entries()[3],
clip_, ttp_);
bw->Compute(input, sequence_lens_span, num_directions_, input_weights_2, recurrent_weights_2, output_2, hidden_output_2);
}};
auto task_results_bw = task_bw.get_future();
ttp_.RunTask(std::move(task_bw));
std::packaged_task<void()> task_fw{
[&]() {
std::unique_ptr<detail::UniDirectionalGru<T>> fw = std::make_unique<detail::UniDirectionalGru<T>>(
@ -403,8 +388,16 @@ Status DeepCpuGruOp::ComputeImpl(OpKernelContext& context) const {
auto task_results_fw = task_fw.get_future();
ttp_.RunTask(std::move(task_fw));
std::unique_ptr<detail::UniDirectionalGru<T>> bw = std::make_unique<detail::UniDirectionalGru<T>>(
alloc, logger,
seq_length, batch_size, input_size, hidden_size_, linear_before_reset_, Direction::kReverse,
bias_2, initial_hidden_2,
activation_funcs_.Entries()[2],
activation_funcs_.Entries()[3],
clip_, ttp_);
bw->Compute(input, sequence_lens_span, num_directions_, input_weights_2, recurrent_weights_2, output_2, hidden_output_2);
task_results_fw.get();
task_results_bw.get();
} else {
std::unique_ptr<detail::UniDirectionalGru<T>> gru_p = std::make_unique<detail::UniDirectionalGru<T>>(
alloc, logger,