search

apache / mxnet

Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
https://mxnet.apache.org
Apache License 2.0
20.75k stars 6.8k forks source link

Unable to cross compile with USE_TVM_OP=ON #20656

Open jasaw opened 2 years ago

jasaw commented 2 years ago

Description

I can cross compile successfully without TVM_OP, but when I turn on TVM_OP, the post build step calls contrib/tvmop/compile.py where the python script imports tvm which loads the libtvm.so file. Problem here is the cross compiled libtvm.so file is for Aarch64 target which won't load on x64_86 host that does the cross compilation. This contrib/tvmop/compile.py file basically pulls all the supported operators and generates a my_kernel.cu cuda source file and uses nvcc to compile and g++ to link into libtvmop.so file. contrib/tvmop/compile.py also generates tvmop.conf.

Question is, how do I cross compile this step?

github-actions[bot] commented 2 years ago

Welcome to Apache MXNet (incubating)! We are on a mission to democratize AI, and we are glad that you are contributing to it by opening this issue. Please make sure to include all the relevant context, and one of the @apache/mxnet-committers will be here shortly. If you are interested in contributing to our project, let us know! Also, be sure to check out our guide on contributing to MXNet and our development guides wiki.