[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-03-02 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-593648767 @jroesch Let's bring this in to unblock @mbaret. We will move to the pattern infra once it is done.

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-03-02 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-593534309 @mbaret Please take another look and see if the comment makes sense. Let's bring this in soon so that your algorithm PR can be

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-26 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591497528 @jroesch thanks for commenting. Could you please comment on the discussion forum and share more alternative solutions so that we can

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-25 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591105578 @mbaret Thanks for all discussions. I think this is actually a separate issue to this PR. We probably want to have an RFC for it. We

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-25 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591062207 I probably missed something, but we already have a function because the composite pass has been executed, right?

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-25 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591051281 no need to register if we don't have an op for it. They are annotated already.

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-25 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591009487 @mbaret We shouldn't even get to that line because the op in the call is not a primitive op, but a FunctionNode. We need to check

[GitHub] [incubator-tvm] zhiics commented on issue #4933: [Relay] Target annotation for external codegen

2020-02-25 Thread GitBox
zhiics commented on issue #4933: [Relay] Target annotation for external codegen URL: https://github.com/apache/incubator-tvm/pull/4933#issuecomment-591005365 @mbaret We can check composite in the op register only if it is an `op`. Otherwise, we can check the CallNode whose op is a function