The inference time of padding operator is too slow

[dollarser]

What is the difference between padding operator in efficientnet and conventional padding? Why is the inference time of padding for onnx model slow? If I need to run faster, how can I optimize the padding operator? thank you!