PetaVision
Alpha
|
#include <PresynapticPerspectiveConvolveDelivery.hpp>
BaseDelivery Parameters | |
List of parameters needed from the BaseDelivery class | |
virtual void | ioParam_channelCode (enum ParamsIOFlag ioFlag) |
channelCode: Specifies which channel in the post layer this connection is attached to More... | |
HyPerDeliveryFacade Parameters | |
List of parameters needed from the HyPerDeliveryFacade class | |
virtual void | ioParam_convertRateToSpikeCount (enum ParamsIOFlag ioFlag) |
convertRateToSpikeCount: If true, presynaptic activity is converted from a rate (spikes per second) to a count (number of spikes in the timestep). More... | |
PresynapticPerspectiveConvolveDelivery Parameters | |
List of parameters needed from the PresynapticPerspectiveConvolveDelivery class | |
virtual void | ioParam_receiveGpu (enum ParamsIOFlag ioFlag) override |
receiveGpu: PresynapticPerspectiveConvolveDelivery always sets receiveGpu to false. The receiveGpu=true case is handled by the PresynapticPerspectiveGPUDelivery class. | |
Public Types | |
enum | AccumulateType { UNDEFINED, CONVOLVE, STOCHASTIC } |
Public Member Functions | |
PresynapticPerspectiveConvolveDelivery (char const *name, HyPerCol *hc) | |
virtual void | deliver () override |
virtual void | deliverUnitInput (float *recvBuffer) override |
ChannelType | getChannelCode () const |
bool | getDataStructuresAllocatedFlag () const |
std::string const & | getDescription () const |
char const * | getDescription_c () const |
bool | getInitialValuesSetFlag () const |
bool | getInitInfoCommunicatedFlag () const |
MPIBlock const * | getMPIBlock () |
char const * | getName () const |
std::string const & | getObjectType () const |
HyPerLayer * | getPostLayer () const |
HyPerLayer * | getPreLayer () const |
bool | getReceiveGpu () const |
void | ioParams (enum ParamsIOFlag ioFlag, bool printHeader, bool printFooter) |
virtual bool | isAllInputReady () override |
bool | isUsingGPU () const |
char const * | lookupKeyword () const |
void | readParams () |
virtual Response::Status | registerData (Checkpointer *checkpointer) |
virtual Response::Status | respond (std::shared_ptr< BaseMessage const > message) override |
void | setConnectionData (ConnectionData *connectionData) |
void | writeParams () |
Protected Member Functions | |
virtual Response::Status | allocateDataStructures () override |
void | allocateThreadGSyn () |
virtual Response::Status | cleanup () |
virtual Response::Status | communicateInitInfo (std::shared_ptr< CommunicateInitInfoMessage const > message) override |
double | convertToRateDeltaTimeFactor (double timeConstantTau) const |
virtual Response::Status | copyInitialStateToGPU () |
int | initialize (char const *name, HyPerCol *hc) |
virtual Response::Status | initializeState () |
virtual int | ioParamsFillGroup (enum ParamsIOFlag ioFlag) override |
virtual Response::Status | prepareCheckpointWrite () |
virtual Response::Status | processCheckpointRead () |
virtual Response::Status | readStateFromCheckpoint (Checkpointer *checkpointer) override |
Response::Status | respondAllocateData (std::shared_ptr< AllocateDataMessage const > message) |
Response::Status | respondCleanup (std::shared_ptr< CleanupMessage const > message) |
Response::Status | respondCommunicateInitInfo (std::shared_ptr< CommunicateInitInfoMessage const > message) |
Response::Status | respondCopyInitialStateToGPU (std::shared_ptr< CopyInitialStateToGPUMessage const > message) |
Response::Status | respondInitializeState (std::shared_ptr< InitializeStateMessage const > message) |
Response::Status | respondPrepareCheckpointWrite (std::shared_ptr< PrepareCheckpointWriteMessage const > message) |
Response::Status | respondProcessCheckpointRead (std::shared_ptr< ProcessCheckpointReadMessage const > message) |
Response::Status | respondReadStateFromCheckpoint (std::shared_ptr< ReadStateFromCheckpointMessage< Checkpointer > const > message) |
Response::Status | respondRegisterData (std::shared_ptr< RegisterDataMessage< Checkpointer > const > message) |
Response::Status | respondSetCudaDevice (std::shared_ptr< SetCudaDeviceMessage const > message) |
virtual Response::Status | setCudaDevice (std::shared_ptr< SetCudaDeviceMessage const > message) |
void | setDataStructuresAllocatedFlag () |
void | setDescription () |
void | setInitialValuesSetFlag () |
void | setInitInfoCommunicatedFlag () |
int | setName (char const *name) |
virtual void | setObjectType () override |
int | setParent (HyPerCol *hc) |
Protected Attributes | |
std::string | description |
AccumulateType | mAccumulateType = CONVOLVE |
ArborList * | mArborList = nullptr |
ChannelType | mChannelCode = CHANNEL_EXC |
ConnectionData * | mConnectionData = nullptr |
bool | mConvertRateToSpikeCount = false |
PVCuda::CudaDevice * | mCudaDevice = nullptr |
bool | mDataStructuresAllocatedFlag = false |
float | mDeltaTimeFactor = 1.0f |
bool | mInitialValuesSetFlag = false |
bool | mInitInfoCommunicatedFlag = false |
std::string | mObjectType |
HyPerLayer * | mPostLayer = nullptr |
HyPerLayer * | mPreLayer = nullptr |
bool | mReceiveGpu = false |
std::vector< std::vector< float > > | mThreadGSyn |
bool | mUpdateGSynFromPostPerspective = false |
bool | mUsingGPUFlag = false |
WeightsPair * | mWeightsPair = nullptr |
char * | name = nullptr |
HyPerCol * | parent = nullptr |
The delivery class for HyPerConns using the presynaptic perspective on the CPU, with accumulate type "convolve".
Definition at line 19 of file PresynapticPerspectiveConvolveDelivery.hpp.
|
overridevirtual |
The method that delivers presynaptic activity to the given postsynaptic channel. It loops over presynaptic neurons, skipping over any whose activity is zero (to take advantage of sparsity). Each neuron then modifies the region of the post channel that the weights argument specifies for that pre-synaptic neuron.
If OpenMP is used, we parallelize over the presynaptic neuron. To avoid the possibility of collisions where more than one pre-neuron writes to the same post-neuron, we internally allocate multiple buffers the size of the post channel, and accumulate them at the end.
Implements PV::HyPerDelivery.
Definition at line 74 of file PresynapticPerspectiveConvolveDelivery.cpp.
|
inlineinherited |
Get-method for mDataStructuresAllocatedFlag, which is false on initialization and then becomes true once setDataStructuresAllocatedFlag() is called.
Definition at line 102 of file BaseObject.hpp.
|
inlineinherited |
Get-method for mInitialValuesSetFlag, which is false on initialization and then becomes true once setInitialValuesSetFlag() is called.
Definition at line 108 of file BaseObject.hpp.
|
inlineinherited |
Get-method for mInitInfoCommunicatedFlag, which is false on initialization and then becomes true once setInitInfoCommunicatedFlag() is called.
Definition at line 95 of file BaseObject.hpp.
|
protectedvirtualinherited |
channelCode: Specifies which channel in the post layer this connection is attached to
Channels can be -1 for no update, or >= 0 for channel number.
0 is excitatory, 1 is inhibitory
Definition at line 29 of file BaseDelivery.cpp.
|
protectedvirtualinherited |
convertRateToSpikeCount: If true, presynaptic activity is converted from a rate (spikes per second) to a count (number of spikes in the timestep).
If this flag is true and the presynaptic layer is not spiking, the activity will be interpreted as a spike rate, and will be converted to a spike count when delivering activity to the postsynaptic GSyn buffer. If this flag is false, activity will not be converted.
Definition at line 38 of file HyPerDelivery.cpp.
|
inherited |
Method for reading or writing the params from group in the parent HyPerCol's parameters. The group from params is selected using the name of the connection.
If ioFlag is set to write, the printHeader and printFooter flags control whether a header and footer for the parameter group is produces. These flags are set to true for layers, connections, and probes; and set to false for weight initializers and normalizers. If ioFlag is set to read, the printHeader and printFooter flags are ignored.
Note that ioParams is not virtual. To add parameters in a derived class, override ioParamFillGroup.
Definition at line 74 of file BaseObject.cpp.
|
overrideprotectedvirtual |
The virtual method for reading parameters from the parent HyPerCol's parameters, and writing to the output params file.
Derived classes with additional parameters typically override ioParamsFillGroup to call the base class's ioParamsFillGroup method and then call ioParam_[parametername] for each of their parameters. The ioParam_[parametername] methods should call the parent HyPerCol's ioParamValue() and related methods, to ensure that all parameters that get read also get written to the outputParams-generated file.
Reimplemented from PV::HyPerDelivery.
Definition at line 31 of file PresynapticPerspectiveConvolveDelivery.cpp.
|
overridevirtualinherited |
A virtual method to indicate whether the presynaptic layer's input is ready to be delivered.
Reimplemented from PV::BaseDelivery.
Definition at line 99 of file HyPerDelivery.cpp.
|
inlineinherited |
Returns true if the object requires the GPU; false otherwise. HyPerCol will not initialize the GPU unless one of the objects in its hierarchy returns true
Definition at line 116 of file BaseObject.hpp.
|
inherited |
Look up the keyword of the params group with the same name as the object.
Definition at line 42 of file BaseObject.cpp.
|
inlineinherited |
A method that reads the parameters for the group whose name matches the name of the object. It, along with writeParams(), is a wrapper around ioParams, so that readParams and writeParams automatically run through the same parameters in the same order.
Definition at line 62 of file BaseObject.hpp.
|
inlineprotectedinherited |
This method sets mDataStructuresAllocatedFlag to true.
Definition at line 175 of file BaseObject.hpp.
|
inlineprotectedinherited |
This method sets the flag returned by getInitialValuesSetFlag to true.
Definition at line 180 of file BaseObject.hpp.
|
inlineprotectedinherited |
This method sets mInitInfoCommunicatedFlag to true.
Definition at line 170 of file BaseObject.hpp.
|
inlineinherited |
A method that writes the parameters for the group whose name matches the name of the object. It, along with readParams(), is a wrapper around ioParams, so that readParams and writeParams automatically run through the same parameters in the same order.
Definition at line 69 of file BaseObject.hpp.