areusch commented on PR #14179: URL: https://github.com/apache/tvm/pull/14179#issuecomment-1468482950
@Mousius i think it's two angles to the same problem: GraphExecutor isn't a great choice with the C runtime. However, you can in practice use AOT with the C++ runtime. I would think in terms of which compiler config is most obvious to an embedded user, the order would be: 1. the CPU setting 2. the runtime 3. the executor since the executor is essentially an implementation detail to new users (they are already expecting to be shown how to interact with TVM and wouldn't know that a priori; whereas they would certainly know they can't link a C++ malloc-laden runtime and they would expect to need to tell the compiler to target their particular class of CPUs). so in this sense it seems like it's better to auto-select executor based on runtime. thoughts? -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
