Question aboutf use QuantizeLinear Node with my custom op

[AnnaTrainingG]

how to fuse QuantizeLinear Node with my custom op when convert onnx to trtengine
I find that the QuantizeLinear will be called like:

it cost long time, How to fuse it with my custom kernel before QuantizeLinear node. the compute in QuantizeLinear is what
like:
my_custom_op - Q -- DQ --- Conv. fuse to => my_custom_op_with_Q --- (DQ_Conv)

1.what the format of QuantizeLinear's output in this image
2.the compute is QuantizeLinear Kernel is that clamp((input*scale).round, -127, 127) ?

[cjluo-omniml]

Is this question in the context of trt engine building phase?