ONE - On-device Neural Engine
|
Handles input(s) while converting a canonical node to TFL node(s). This class informs DomainConverter how to handle inputs of a specific canonical node. More...
#include <GraphBlock.h>
Public Member Functions | |
virtual void | handover (CanonicalT *origin, TFLT *replacer)=0 |
Assign origin's inputs to replacer's inputs. (This is called when origin belongs in Tensor domain.) | |
virtual std::vector< loco::Node * > | getInputsToConvert (CanonicalT *origin)=0 |
Returns the list of inputs that needs to have FeatureDecode as its input. (This is called when origin belongs in Feature domain.) | |
virtual void | set (TFLT *replacer, std::vector< loco::Node * > &new_inputs)=0 |
Set the inputs of replacer to new_inputs. | |
virtual void | nullify (CanonicalT *origin)=0 |
Set the inputs to nullptr. | |
Handles input(s) while converting a canonical node to TFL node(s). This class informs DomainConverter how to handle inputs of a specific canonical node.
Definition at line 109 of file GraphBlock.h.
|
pure virtual |
Returns the list of inputs that needs to have FeatureDecode as its input. (This is called when origin belongs in Feature domain.)
Referenced by exo::DomainConverter< CanonicalT, TFLT >::convert().
|
pure virtual |
Assign origin's inputs to replacer's inputs. (This is called when origin belongs in Tensor domain.)
Referenced by exo::DomainConverter< CanonicalT, TFLT >::convert().
|
pure virtual |
Set the inputs to nullptr.
Referenced by exo::DomainConverter< CanonicalT, TFLT >::convert().
|
pure virtual |
Set the inputs of replacer to new_inputs.
Referenced by exo::DomainConverter< CanonicalT, TFLT >::convert().