lsy643 opened a new pull request #6076: URL: https://github.com/apache/incubator-tvm/pull/6076
Currently, when we want to compile and execute a network with dynamic input, only using `llvm` target can work as expected. If the target is as `CUDA`, the compilation part works fine, but the run part fails. The reason why this happens is that the shape functions dealing with dynamic inputs are normally built on target host like `llvm`, while the output tensors of shape functions need to locate on gpu target. Therefore, in the PR, I try to fix this bug by creating temporary output tensors on cpu for results of shape functions and coping them to the output tensors created by the vm located on gpu. ---------------------------------------------------------------- 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: us...@infra.apache.org