FrozenGene commented on pull request #5753: URL: https://github.com/apache/incubator-tvm/pull/5753#issuecomment-656066410
@tqchen The structure and implementation has been refactored. Could you spend some time in reviewing it again? Besides the graph runtime factory, there are 2 changes needs to be noticed. 1. In `src/runtime/module.cc`, change the `query_import` logic to allow we could look up function beyond 2 levels. (for example, we could have `x` module -> `y` module -> `cpu` host side -> `gpu` side. This is been exposed in multi models graph runtime factory packaging) 2. In `graph runtime factory`, we need `tvm::Map` to store params and return to python side. However, inside its implementation, it has one `static constexpr array`. Its definition in the `tvm/node/container.cc` (not compiled into `libtvmruntime.so`). This will have problem (`undefined reference`) inside `libtvmruntime.so` if we use `tvm::Map`. So I move its definition in the `tvm/runtime/container.cc` so that we could use it in the runtime part code. ---------------------------------------------------------------- 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]
