LBANN  0.103.0
LivermoreBigArtificialNeuralNetworkToolkit
lbann::slice_layer< TensorDataType, Layout, Device > Class Template Reference

Slice tensor along a specified dimension. More...

#include <slice.hpp>

Inheritance diagram for lbann::slice_layer< TensorDataType, Layout, Device >:
[legend]
Collaboration diagram for lbann::slice_layer< TensorDataType, Layout, Device >:
[legend]

Public Member Functions

 slice_layer (lbann_comm *comm)
 
 slice_layer (const slice_layer &other)=default
 
slice_layeroperator= (const slice_layer &other)=default
 
slice_layercopy () const override
 Copy function. This function dynamically allocates memory for a layer instance and instantiates a copy. The caller is responsible for deallocating the instance. More...
 
std::string get_type () const override
 Get the layer type's name. More...
 
data_layout get_data_layout () const override
 Get data layout of the data tensors. We assume that the data layouts of the previous activations, activations, previous error signals, and error signals are the same. Each concrete layer that is templated on its data layout should override this function to return its template parameter. More...
 
El::Device get_device_allocation () const override
 Get the device allocation for the data tensors. We assume that the decice allocation of the previous activations, activations, previous error signals, and error signals are the same. Each concrete layer that is templated on its device allocation should override this function to return its template parameter. More...
 
bool can_run_inplace () const override
 If True, the computation can run in-place (feeding each input activations tensor as the corresponding output activations) More...
 
int get_backprop_requirements () const override
 Returns the necessary tensors for computing backpropagation. More...
 
description get_description () const override
 Human-readable description. More...
 
void setup_slice_points (size_t slice_dim, std::vector< size_t > slice_points)
 
void setup_slice_points (size_t slice_dim, bool set_slice_points_from_data_reader, const slice_points_mode var_category)
 
Serialization
template<typename ArchiveT >
void serialize (ArchiveT &ar)
 
- Public Member Functions inherited from lbann::data_type_layer< TensorDataType >
 data_type_layer (lbann_comm *, bool persistent_error_signals=false)
 
virtual ~data_type_layer ()=default
 
std::string get_datatype_name () const override
 
void forward_prop () final
 
void summarize_matrices (lbann_summary &summarizer, int step) override
 
void check_setup () override
 
const OutputAbsDistMatrixTypeget_activations (const Layer &child) const override
 
OutputAbsDistMatrixTypeget_activations (int child_index=0)
 
const OutputAbsDistMatrixTypeget_activations (int child_index=0) const
 
const InputAbsDistMatrixTypeget_error_signals (const Layer &parent) const override
 
InputAbsDistMatrixTypeget_error_signals (int parent_index=0)
 
const InputAbsDistMatrixTypeget_error_signals (int parent_index=0) const
 
El::Int current_output_mini_batch_size () const override
 
El::Int infer_mini_batch_size_from_parents_or_default_to_current () const override
 
OutputAbsDistMatrixTypeget_temp_grad ()
 
InputAbsDistMatrixTypeget_branch_tag_input (int tag)
 
std::vector< std::unique_ptr< InputAbsDistMatrixType > > & get_branch_tag_input_vector ()
 
std::vector< std::unique_ptr< OutputAbsDistMatrixType > > & get_all_activations ()
 
std::vector< std::unique_ptr< InputAbsDistMatrixType > > & get_all_prev_activations ()
 
std::vector< std::unique_ptr< OutputAbsDistMatrixType > > & get_all_prev_error_signals ()
 
std::vector< std::unique_ptr< InputAbsDistMatrixType > > & get_all_error_signals ()
 
OutputAbsMatrixTypeget_local_activations (int child_index=0)
 
const OutputAbsMatrixTypeget_local_activations (int child_index=0) const
 
InputAbsMatrixTypeget_local_error_signals (int parent_index=0)
 
const InputAbsMatrixTypeget_local_error_signals (int parent_index=0) const
 
void set_keep_error_signals (bool) override
 Set whether to keep or dynamically reallocate error signals. More...
 
El::mpi::Comm & get_subgrid_comm ()
 
void serialize (ArchiveT &ar)
 
- Public Member Functions inherited from lbann::Layer
void write_proto (lbann_data::Layer &proto) const
 Write layer to proto file. More...
 
lbann_commget_comm () const
 
int get_grid_tag () const noexcept
 Identifying tag for process grid. More...
 
void set_grid_tag (int tag)
 Set process grid. More...
 
bool runs_inplace () const
 If true, the layer will run in-place (the input and output activations point to the same tensor). Value is set during graph setup (in setup_pointers) based on layer traits and neighboring layers. More...
 
bool distconv_enabled () const
 Indicate whether distconv is enabled. More...
 
 Layer ()
 
virtual ~Layer ()=default
 
void set_name (const std::string name)
 Set the layer instance's name. Each layer in a model should have a unique, preferably human-readable, name. More...
 
void set_model (model *m)
 Set the model that manages this layer. More...
 
std::string get_name () const
 Get the layer instance's name. More...
 
modelget_model () const noexcept
 Get a reference to the model that manages this layer. More...
 
int get_expected_num_parent_layers () const noexcept
 Get expected number of parent layers. A negative value indicates no limit. More...
 
int get_expected_num_child_layers () const noexcept
 Get expected number of child layers. A negative value indicates no limit. More...
 
ParallelStrategyget_parallel_strategy () noexcept
 Get the parallel strategy for the layer. More...
 
ParallelStrategy const & get_parallel_strategy () const noexcept
 Get the parallel strategy for the layer. More...
 
bool using_gpus () const noexcept
 Whether the layer is using a GPU implementation. More...
 
void back_prop ()
 Backward propagation step. Given the objective function gradients w.r.t. the output tensors, compute the gradients w.r.t. the input tensors and w.r.t. the weights. This is essentially an application of the chain rule. More...
 
bool update ()
 Update step. Update the layer's internal members. Note that the optimization step for the weights happens elsewhere. More...
 
virtual void setup (size_t max_mini_batch_size, const std::vector< El::Grid *> &grids)
 Setup layer members. More...
 
void summarize_stats (lbann_summary &summarizer, int step)
 
void reset_counters ()
 Reset layer stat counters. More...
 
void set_communication_flag (SubGraphCommunication type)
 
SubGraphCommunication get_communication_flag ()
 
void set_num_spliting_groups (El::Int spliting_groups)
 
El::Int get_num_spliting_groups () const
 
std::shared_ptr< El::Grid > get_mygrid () const
 
void reset_inter_subgrid_vc_comm (std::shared_ptr< El::mpi::Comm > mpi_comm)
 
void set_subgraph_parallelism_execution ()
 
bool subgraph_parallelism_execution () const noexcept
 
void set_run_layer_in_subgraph ()
 
bool get_run_layer_in_subgraph () const noexcept
 
const Layerget_parent_layer (size_t index=0) const
 
const Layerget_child_layer (size_t index=0) const
 
std::vector< const Layer * > get_parent_layers () const
 
std::vector< const Layer * > get_child_layers () const
 
size_t find_parent_layer_index (const Layer &l) const
 
size_t find_child_layer_index (const Layer &l) const
 
int get_num_parents () const noexcept
 Get number of parent layers. More...
 
int get_num_children () const noexcept
 Get number of child layers. More...
 
void add_parent_layer (ViewingLayerPtr parent)
 Add a parent layer. More...
 
void add_child_layer (ViewingLayerPtr child)
 Add a child layer. More...
 
void replace_parent_layer (ViewingLayerPtr l, size_t index)
 
void replace_child_layer (ViewingLayerPtr l, size_t index)
 
void clear_parent_layers ()
 Remove pointers to parent layers. More...
 
void clear_child_layers ()
 Remove pointers to child layers. More...
 
ViewingLayerPtr get_parent_layer_pointer (size_t index) const
 
ViewingLayerPtr get_child_layer_pointer (size_t index) const
 
virtual std::vector< ViewingLayerPtrget_layer_pointers ()
 List of pointers to other layers. More...
 
virtual void set_layer_pointers (std::vector< ViewingLayerPtr > layers)
 Set list of pointers to other layers. More...
 
std::vector< ViewingWeightsPtrget_weights_pointers () const
 List of pointers to weights. More...
 
void set_weights_pointers (std::vector< ViewingWeightsPtr > ptrs)
 Set list of pointers to weights. More...
 
void replace_weights (Layer const &other_layer)
 Replace weights with another Layer's weights. More...
 
std::vector< int > get_input_dims (size_t input_index=0) const
 Get input tensor dimensions. More...
 
int get_input_size (size_t input_index=0) const
 Get input tensor size. More...
 
std::vector< int > get_output_dims (size_t output_index=0) const
 Get output tensor dimensions. More...
 
int get_output_size (size_t output_index=0) const
 Get output tensor size. More...
 
void set_output_dims (std::vector< int > dims, size_t output_index=0)
 Set output tensor dimensions. More...
 
El::Int infer_mini_batch_size_from_parents () const
 
void set_hint_layer (ViewingLayerPtr l)
 Set hint layer. More...
 
const Layerget_hint_layer () const
 Get hint layer. More...
 
void freeze ()
 
void unfreeze ()
 
bool is_frozen () const
 
template<typename ArchiveT >
void serialize (ArchiveT &ar)
 

Protected Member Functions

void write_specific_proto (lbann_data::Layer &proto) const final
 
 slice_layer ()
 
void setup_dims () override
 Setup tensor dimensions Called by the 'setup' function. If there are any input tensors, the base method sets all uninitialized output tensor dimensions equal to the first input tensor dimensions. More...
 
void fp_setup_outputs () override
 Setup output tensors. Called by the 'forward_prop' function. Each output tensor is resized to match the mini-batch size. More...
 
void bp_setup_gradient_wrt_inputs () override
 Setup gradient w.r.t. input tensors. Called by the 'back_prop' function. Each gradient w.r.t. input tensor is resized to match the mini-batch size. More...
 
void fp_compute () override
 Apply layer operation. Called by the 'forward_prop' function. Given the input tensors, the output tensors are populated with computed values. More...
 
void bp_compute () override
 Compute objective funciton gradients. Called by the 'back_prop' function. Given the input, output, and gradient w.r.t. output tensors, the gradient w.r.t. input tensors are populated with the computed values and the gradients w.r.t. the weights are sent to the appropriate optimizers. More...
 
void fp_compute_subgrid ()
 
void bp_compute_subgrid ()
 
- Protected Member Functions inherited from lbann::data_type_layer< TensorDataType >
InputAbsDistMatrixTypeget_prev_activations (int parent_index=0)
 
const InputAbsDistMatrixTypeget_prev_activations (int parent_index=0) const
 
OutputAbsDistMatrixTypeget_prev_error_signals (int child_index=0)
 
const OutputAbsDistMatrixTypeget_prev_error_signals (int child_index=0) const
 
const InputAbsMatrixTypeget_local_prev_activations (int parent_index=0) const
 
const OutputAbsMatrixTypeget_local_prev_error_signals (int child_index=0) const
 
void setup_matrices (const std::vector< El::Grid * > &grids) override
 
void setup_data (size_t max_mini_batch_size) override
 
void fp_setup_inputs () override
 
void fp_setup_outputs () override
 
void bp_setup_gradient_wrt_inputs () override
 
void bp_compute () override
 
InputAbsDistMatrixType const & weights_values (size_t idx) const
 Get the values matrix for a specific weights object. More...
 
weightsmaster_weights (size_t idx)
 Get a specific master weights object. More...
 
weights const & master_weights (size_t idx) const
 
 data_type_layer (data_type_layer &&other)=default
 Protected lifecycle functions. More...
 
 data_type_layer (data_type_layer const &other)
 
data_type_layeroperator= (data_type_layer &&other)=default
 
data_type_layeroperator= (data_type_layer const &other)
 
- Protected Member Functions inherited from lbann::Layer
void setup_grid ()
 Setup process grid. More...
 
virtual void setup_pointers ()
 Setup layer pointers. Called by the 'setup' function. Pointers to parent/child layers are assumed to be already initialized. More...
 
virtual void setup_gpu ()
 Setup GPU objects. Called by the 'setup' function if the layer is on GPUs. More...
 
virtual bool update_compute ()
 Perform the computation for the update step. Returns false if the layer must reset for a new training epoch. More...
 
 Layer (Layer &&other)=default
 
 Layer (Layer const &other)
 
Layeroperator= (Layer &&other)=default
 
Layeroperator= (Layer const &other)
 
void add_weights (ViewingWeightsPtr w)
 
size_t num_weights () const noexcept
 
bool has_weights () const noexcept
 
bool has_weights (size_t idx) const noexcept
 
void set_num_weights (size_t n)
 
void set_weights (size_t idx, ViewingWeightsPtr w)
 
weights const & get_weights (size_t idx) const
 
weightsget_weights (size_t idx)
 
void add_as_gradient_source ()
 
void remove_as_gradient_source ()
 

Protected Attributes

El::SyncInfo< DevicesyncSubGridCommunication = El::SyncInfo<Device>()
 
- Protected Attributes inherited from lbann::Layer
int m_expected_num_parent_layers = 1
 
int m_expected_num_child_layers = 1
 Expected number of child layers. A negative value indicates no limit. More...
 
modelm_model = nullptr
 Reference to model managing this layer. More...
 
bool m_frozen
 Avoid back prop if frozen. More...
 
EvalType m_fp_time
 Time spent in forward propagation. More...
 
EvalType m_fp_compute_time
 Time spent in the forward propagation computation. More...
 
EvalType m_bp_time
 Time spent in backward propagation. More...
 
EvalType m_bp_compute_time
 Time spent in the backward propagation computation. More...
 
EvalType m_update_time
 Time spent in updates. More...
 
std::string m_name
 Layer instance's name. Each layer in a model should have a unique, preferably human-readable, name. More...
 
bool m_runs_inplace = false
 If true, the layer will run in-place (the input and output activations point to the same tensor). Value is set during graph setup (in setup_pointers) based on layer traits and neighboring layers. More...
 
int m_grid_tag = -1
 Identifying tag for process grid. More...
 
SubGraphCommunication subgraph_communication_method = PT2PT
 
bool m_subgraph_parallelism_execution = false
 
bool run_layer_in_subgraph = false
 
std::unique_ptr< std::set< int > > m_subgrid_ranks
 
El::Int m_num_spliting_groups = 1
 
std::shared_ptr< El::mpi::Comm > m_interSubGridVCComm
 

Private Attributes

size_t m_slice_dim
 
std::vector< size_t > m_slice_points
 
bool m_set_slice_points_from_data_reader
 
slice_points_mode m_var_category
 

Friends

class cereal::access
 
template<typename U , El::Device D>
void fp_setup_outputs_impl (slice_layer< U, Layout, D > &)
 
template<typename U >
void fp_compute_impl (slice_layer< U, Layout, Device > &)
 
template<typename U >
void bp_compute_impl (slice_layer< U, Layout, Device > &)
 

Additional Inherited Members

- Public Types inherited from lbann::data_type_layer< TensorDataType >
using InputAbsDistMatrixType = El::AbstractDistMatrix< TensorDataType >
 The tensor type expected in this object. More...
 
using OutputAbsDistMatrixType = El::AbstractDistMatrix< TensorDataType >
 
using InputAbsDistMatReadProxyType = El::AbstractDistMatrixReadDeviceProxy< TensorDataType, D >
 The proxy tensor type expected in this object. More...
 
using OutputAbsDistMatReadProxyType = El::AbstractDistMatrixReadDeviceProxy< TensorDataType, D >
 
using InputAbsMatrixType = El::AbstractMatrix< TensorDataType >
 The local tensor type expected in this object. More...
 
using OutputAbsMatrixType = El::AbstractMatrix< TensorDataType >
 
using WeightsProxyType = weights_proxy< TensorDataType >
 The proxy type for weights used by this object. More...
 

Detailed Description

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
class lbann::slice_layer< TensorDataType, Layout, Device >

Slice tensor along a specified dimension.

The tensor is split along one dimension at user-specified points, and each child layer recieves one piece.

Definition at line 49 of file slice.hpp.

Constructor & Destructor Documentation

◆ slice_layer() [1/3]

template<typename TensorDataType , data_layout Layout, El::Device Device>
lbann::slice_layer< TensorDataType, Layout, Device >::slice_layer ( lbann_comm comm)

Definition at line 155 of file slice.hpp.

◆ slice_layer() [2/3]

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
lbann::slice_layer< TensorDataType, Layout, Device >::slice_layer ( const slice_layer< TensorDataType, Layout, Device > &  other)
default

◆ slice_layer() [3/3]

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
lbann::slice_layer< TensorDataType, Layout, Device >::slice_layer ( )
inlineprotected

Definition at line 96 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

Member Function Documentation

◆ bp_compute()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::bp_compute ( )
overrideprotectedvirtual

Compute objective funciton gradients. Called by the 'back_prop' function. Given the input, output, and gradient w.r.t. output tensors, the gradient w.r.t. input tensors are populated with the computed values and the gradients w.r.t. the weights are sent to the appropriate optimizers.

Reimplemented from lbann::Layer.

Definition at line 364 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ bp_compute_subgrid()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::bp_compute_subgrid ( )
protected

Definition at line 332 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ bp_setup_gradient_wrt_inputs()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::bp_setup_gradient_wrt_inputs ( )
overrideprotectedvirtual

Setup gradient w.r.t. input tensors. Called by the 'back_prop' function. Each gradient w.r.t. input tensor is resized to match the mini-batch size.

Implements lbann::Layer.

Definition at line 354 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ can_run_inplace()

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
bool lbann::slice_layer< TensorDataType, Layout, Device >::can_run_inplace ( ) const
inlineoverridevirtual

If True, the computation can run in-place (feeding each input activations tensor as the corresponding output activations)

Reimplemented from lbann::Layer.

Definition at line 69 of file slice.hpp.

◆ copy()

template<typename TensorDataType , data_layout Layout, El::Device Device>
slice_layer< TensorDataType, Layout, Device > * lbann::slice_layer< TensorDataType, Layout, Device >::copy ( ) const
overridevirtual

Copy function. This function dynamically allocates memory for a layer instance and instantiates a copy. The caller is responsible for deallocating the instance.

Implements lbann::Layer.

Definition at line 173 of file slice.hpp.

Here is the call graph for this function:

◆ fp_compute()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::fp_compute ( )
overrideprotectedvirtual

Apply layer operation. Called by the 'forward_prop' function. Given the input tensors, the output tensors are populated with computed values.

Implements lbann::Layer.

Definition at line 317 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ fp_compute_subgrid()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::fp_compute_subgrid ( )
protected

Definition at line 252 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ fp_setup_outputs()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::fp_setup_outputs ( )
overrideprotectedvirtual

Setup output tensors. Called by the 'forward_prop' function. Each output tensor is resized to match the mini-batch size.

Implements lbann::Layer.

Definition at line 246 of file slice.hpp.

Here is the caller graph for this function:

◆ get_backprop_requirements()

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
int lbann::slice_layer< TensorDataType, Layout, Device >::get_backprop_requirements ( ) const
inlineoverridevirtual

Returns the necessary tensors for computing backpropagation.

Reimplemented from lbann::Layer.

Definition at line 70 of file slice.hpp.

Here is the call graph for this function:

◆ get_data_layout()

template<typename TensorDataType , data_layout Layout, El::Device Device>
data_layout lbann::slice_layer< TensorDataType, Layout, Device >::get_data_layout ( ) const
overridevirtual

Get data layout of the data tensors. We assume that the data layouts of the previous activations, activations, previous error signals, and error signals are the same. Each concrete layer that is templated on its data layout should override this function to return its template parameter.

Implements lbann::Layer.

Definition at line 185 of file slice.hpp.

◆ get_description()

template<typename TensorDataType , data_layout Layout, El::Device Device>
description lbann::slice_layer< TensorDataType, Layout, Device >::get_description ( ) const
overridevirtual

Human-readable description.

Reimplemented from lbann::Layer.

Definition at line 198 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ get_device_allocation()

template<typename TensorDataType , data_layout Layout, El::Device Device>
El::Device lbann::slice_layer< TensorDataType, Layout, Device >::get_device_allocation ( ) const
overridevirtual

Get the device allocation for the data tensors. We assume that the decice allocation of the previous activations, activations, previous error signals, and error signals are the same. Each concrete layer that is templated on its device allocation should override this function to return its template parameter.

Implements lbann::Layer.

Definition at line 192 of file slice.hpp.

◆ get_type()

template<typename TensorDataType , data_layout Layout, El::Device Device>
std::string lbann::slice_layer< TensorDataType, Layout, Device >::get_type ( ) const
overridevirtual

Get the layer type's name.

A layer type name should be brief, unique, and human-readable description of the layer's mathematical operation that is recognizable to ML practitioners (e.g., "Convolution", "ReLU")

Implements lbann::Layer.

Definition at line 179 of file slice.hpp.

Here is the caller graph for this function:

◆ operator=()

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
slice_layer& lbann::slice_layer< TensorDataType, Layout, Device >::operator= ( const slice_layer< TensorDataType, Layout, Device > &  other)
default

◆ serialize()

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
template<typename ArchiveT >
void lbann::slice_layer< TensorDataType, Layout, Device >::serialize ( ArchiveT &  ar)

◆ setup_dims()

template<typename TensorDataType , data_layout Layout, El::Device Device>
void lbann::slice_layer< TensorDataType, Layout, Device >::setup_dims ( )
overrideprotectedvirtual

Setup tensor dimensions Called by the 'setup' function. If there are any input tensors, the base method sets all uninitialized output tensor dimensions equal to the first input tensor dimensions.

Reimplemented from lbann::Layer.

Definition at line 36 of file slice_impl.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ setup_slice_points() [1/2]

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
void lbann::slice_layer< TensorDataType, Layout, Device >::setup_slice_points ( size_t  slice_dim,
std::vector< size_t >  slice_points 
)
inline

Definition at line 74 of file slice.hpp.

◆ setup_slice_points() [2/2]

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
void lbann::slice_layer< TensorDataType, Layout, Device >::setup_slice_points ( size_t  slice_dim,
bool  set_slice_points_from_data_reader,
const slice_points_mode  var_category 
)
inline

Definition at line 80 of file slice.hpp.

Here is the call graph for this function:

◆ write_specific_proto()

template<typename T , data_layout L, El::Device D>
void lbann::slice_layer< T, L, D >::write_specific_proto ( lbann_data::Layer &  proto) const
finalprotectedvirtual

Add layer specific data to prototext

Implements lbann::Layer.

Definition at line 146 of file slice.hpp.

Here is the call graph for this function:
Here is the caller graph for this function:

Friends And Related Function Documentation

◆ bp_compute_impl

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
template<typename U >
void bp_compute_impl ( slice_layer< U, Layout, Device > &  )
friend

◆ cereal::access

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
friend class cereal::access
friend

Definition at line 95 of file slice.hpp.

◆ fp_compute_impl

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
template<typename U >
void fp_compute_impl ( slice_layer< U, Layout, Device > &  )
friend

◆ fp_setup_outputs_impl

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
template<typename U , El::Device D>
void fp_setup_outputs_impl ( slice_layer< U, Layout, D > &  )
friend

Member Data Documentation

◆ m_set_slice_points_from_data_reader

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
bool lbann::slice_layer< TensorDataType, Layout, Device >::m_set_slice_points_from_data_reader
private

Slice points are automatically defined by the data reader

Definition at line 113 of file slice.hpp.

◆ m_slice_dim

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
size_t lbann::slice_layer< TensorDataType, Layout, Device >::m_slice_dim
private

Tensor dimension to slice.

Definition at line 109 of file slice.hpp.

◆ m_slice_points

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
std::vector<size_t> lbann::slice_layer< TensorDataType, Layout, Device >::m_slice_points
private

Slice points for each child layer.

Definition at line 111 of file slice.hpp.

◆ m_var_category

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
slice_points_mode lbann::slice_layer< TensorDataType, Layout, Device >::m_var_category
private

Category for retrieving slice points from data reader

Definition at line 115 of file slice.hpp.

◆ syncSubGridCommunication

template<typename TensorDataType, data_layout Layout = data_layout::DATA_PARALLEL, El::Device Device = El::Device::CPU>
El::SyncInfo<Device> lbann::slice_layer< TensorDataType, Layout, Device >::syncSubGridCommunication = El::SyncInfo<Device>()
protected

Definition at line 93 of file slice.hpp.


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