caffe2::CUDARecurrentNetworkExecutor Class Reference

Inheritance diagram for caffe2::CUDARecurrentNetworkExecutor:

Public Member Functions
	CUDARecurrentNetworkExecutor (const NetDef &step_net_def, std::map< string, string > &recurrent_input_map, std::string timestep_blob)
void	setMaxStreams (int n)
Public Member Functions inherited from caffe2::RecurrentNetworkExecutorBase
void	EnsureTimestepInitialized (int t, Workspace *ws, const std::vector< std::unique_ptr< ObserverBase< OperatorBase >>> &observers_list)
	Callers must call EnsureTimestepInitialized before starting execution for each of the relevant timesteps. More...
void	SetMaxParallelTimesteps (int p)
	Set limit for the number of timesteps that run in parallel. More...
size_t	NumObserversStepNet ()

Protected Member Functions
bool	Run (int T) override
bool	RunBackwards (int T) override
bool	ignoreLinkDependencies () override
void	AnalyzeOps () override
Protected Member Functions inherited from caffe2::RecurrentNetworkExecutorBase
	RecurrentNetworkExecutorBase (const NetDef &step_net_def, std::map< string, string > &recurrent_input_map, std::string timestep_blob)
void	PrintInfo (int t)
	For debug purposes, print the dependency structure. More...

Additional Inherited Members
Data Fields inherited from caffe2::RecurrentNetworkExecutorBase
bool	debug_ = false
Protected Attributes inherited from caffe2::RecurrentNetworkExecutorBase
std::vector< std::vector< RNNNetOperator > >	timestep_ops_
std::vector< OperatorBase * >	op_ptrs_
std::vector< RNNNetOperator >	timestep_ops_template_
NetDef	step_net_def_
std::vector< std::vector< string > >	op_deps_
std::vector< Workspace * >	workspaces_
std::map< string, string >	recurrent_input_map_
std::string	timestep_blob_
int	max_parallel_timesteps_ = -1

Detailed Description

Definition at line 12 of file recurrent_network_executor_gpu.h.

Member Function Documentation

void caffe2::CUDARecurrentNetworkExecutor::AnalyzeOps ( )

inlineoverrideprotectedvirtual

Check if there is an op that only depends on ops from previous timestep, and that ops is not the last op. Then we can start computation in subsequent timesteps before the whole previous timestep has finished. If there is no parallelism, we can avoid overhead of event-based dependency management.

Reimplemented from caffe2::RecurrentNetworkExecutorBase.

Definition at line 31 of file recurrent_network_executor_gpu.h.

---

The documentation for this class was generated from the following files:

• caffe2/operators/rnn/recurrent_network_executor_gpu.h
• caffe2/operators/rnn/recurrent_network_executor_gpu.cc