Hi, I’m trying to ramp up a TVM environment that uses DNNL - on linux. I followed the “How to Bring Your Own Codegen to TVM” guide but could find anything related to linking the compiled libdnnl.a and its headers to TVM. Is there a special way this should be done? Or do I simply edit TVM’s CMakeLists.txt? If so, what should be the target when using cmake’s target_include_directories command?
Also - Is it possible to use MicroTVM while offloading supported operators to DNNL, or can this only be done in regular TVM?
Thanks!