Bring own cuda kernel code to TVM

Questions
#1

How can I directly invoke my existing CUDA code within the TVM framework for an operation?

It seems that I can rewrite the op in the form of relay op like tutorial here. Yet the example in scan.py shows that I probalbly need to rewrite the cuda code into python using te. Is there a way that I can directly invoke my existing CUDA code ?

#2

maybe you are looking for BYOC: Bring Your Own Codegen To TVM — tvm 0.16.dev0 documentation

#3

Thanks. I noticed this after the posting.