[Question] Does TVM Support inplace operator?

Questions
#1

Assume I have a operator, I wanna it’s input and output use the same memory. Could It be done with TVM? And How？

I wanna use this feature with te.exterm so that I cound use my custom operator without output memory allocation.

#2

I think it’s possible. Try not passing any output tensor to te.extern, and directly modify the input tensor.

#3

so if we don’t get the op output, will it be removed when build? @masahi