comaniac opened a new pull request #5701: URL: https://github.com/apache/incubator-tvm/pull/5701
We previously supported multiple outputs in graph partition pass by creating a new tuple node as a single output. However, we missed a point that this tuple output has to be decomposed when generating C code. Fortunately, TVM runtime already flats tuples to multiple tensors, so it's not hard to make runtime working. This PR modifies the C codegen accordingly. Note that although the new added unit test uses C codegen, I've tested locally and make sure this change is also applicable to DNNL. cc @masahi (sorry for the confusions 😓 ), @zhiics ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
