ONE - On-device Neural Engine
|
Main purpose of this pass - is to fuse 'Conv->BatchNorm' into 'Conv' Currently 'BatchNorm' split by NNC frontends into 'Scale->Scale->BiasAdd' This optimization performs in two steps (repeated while graph changing): More...
#include <FuseArithmeticOps.h>
Public Member Functions | |
PassData | run (PassData data) override |
run compiler pass | |
std::string | getName () override |
Public Member Functions inherited from nnc::Pass | |
virtual void | cleanup () |
clean compiler pass data | |
virtual | ~Pass ()=default |
Main purpose of this pass - is to fuse 'Conv->BatchNorm' into 'Conv' Currently 'BatchNorm' split by NNC frontends into 'Scale->Scale->BiasAdd' This optimization performs in two steps (repeated while graph changing):
Definition at line 34 of file FuseArithmeticOps.h.
|
inlineoverridevirtual |
Reimplemented from nnc::Pass.
Definition at line 39 of file FuseArithmeticOps.h.
|
overridevirtual |
run compiler pass
data | - data that pass is taken |
PassException | object if errors occured |
Implements nnc::Pass.
Definition at line 235 of file FuseArithmeticOps.cpp.
Referenced by package.infer.session::inference().