Thanks for the example. One of our goal is to consolidate the setting into a single target so that the configuration becomes simple. In this case it should be the system target.
I still think it is useful to allow an optionally `target_host`(we can also change the name if we find a better alternative) for the following reasons: - It can be viewed as a "compilation option" for the "virtual target"(about how to generate the driving code). - It ensures the conciseness of the target specification when the we target a single gpu(which is the most common setting). In most cases `target_host` is implied to be the system cpu target, but in the case of cross compilation to a different board we will need to specify it. - We can differentiate from the real CPU + GPU mix setting(E0), as the composite target fits the later case, and we can provide specific optimizations for the programs in the form of E1. Of course, it is important for the code generator to: - validate the target host matches the system cpu target if a composite target is specified - Fill in the target host using the target in the composition if target host is not specified in the multiple device case. --- [Visit Topic](https://discuss.tvm.ai/t/rfc-tvm-target-specification/6844/20) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/f756b5ca4bf789bbe4ff0369af4e41908f226cd5a060756470f1289b66f5c2e2).
