Class EnableNCHW4Pass

Inheritance Relationships

Base Type

Class Documentation

class mgb::gopt::EnableNCHW4Pass : public mgb::gopt::TensorReformatPass

convert tensor format to nchw4 to speed up inference on CUDA

Public Functions

const char *name() const override

Public Static Functions

std::unique_ptr<EnableNCHW4Pass> make_nchw4_converter()

make nchw -> nchw4 converter opt pass