[TVM Discuss] [Questions] Why tensorize rather than call_intrin directly?

2020-06-24 Thread Michael via TVM Discuss
When tensorize some computations, you must write a tensor intrinsic. I realize that we can just use the call_extern or call_intrin directly from the test cases. Why we need to tensorize since it will do a lot of verifications between the intrinsic and body. That causes me some troubles when

[TVM Discuss] [Questions] [relay]where op error:can_dispatch(n):NodeFunctor calls un-registered function on type Any

2020-04-29 Thread Michael via TVM Discuss
I think that's because of the function TransformShape in the file data_layout.cc. It's used to split shape like from NCHW to NCHWc. But when you want to convert your output shape from NCHW to NHWC for example, it will cause error. So you need to add your own shape transform function. ---