ONE - On-device Neural Engine
|
Class to infer shape before running kernels. It does the following: More...
#include <StaticBackwardShapeInferer.h>
Public Member Functions | |
StaticBackwardShapeInferer (compiler::train::LoweredTrainableGraph *lowered_subg) | |
void | infer (void) |
Infer shape of operands belonging to ops and set the output shape. If output shape cannot be known without running op, mark it so that it can be allocated when running kernel. | |
void | dump () |
Public Member Functions inherited from onert::ir::train::TrainableOperationVisitor | |
virtual | ~TrainableOperationVisitor ()=default |
Class to infer shape before running kernels. It does the following:
Definition at line 41 of file StaticBackwardShapeInferer.h.
|
inline |
Definition at line 44 of file StaticBackwardShapeInferer.h.
void onert::compiler::train::StaticBackwardShapeInferer::dump | ( | ) |
Definition at line 55 of file StaticBackwardShapeInferer.cc.
void onert::compiler::train::StaticBackwardShapeInferer::infer | ( | void | ) |
Infer shape of operands belonging to ops and set the output shape. If output shape cannot be known without running op, mark it so that it can be allocated when running kernel.
Definition at line 33 of file StaticBackwardShapeInferer.cc.
References onert::compiler::train::LoweredTrainableGraph::graph(), onert::ir::IOperation::name(), onert::ir::train::TrainableGraph::operation(), onert::ir::Graph::topolSortOperations(), and onert::compiler::train::LoweredTrainableGraph::trainable_graph().