Currently, Relay conv2d internally decides whether a Relay Conv2d operator is depthwise or not. This makes code somewhat messy - lots of it conditions and indirections, quite confusing HWOI vs HWOI kernel layout. In addition, it is difficult to understand from the debug runtime if the conv operator was depthwise or not.
As far as I understand, code will be much simpler if we just have a separate Relay depthwise conv2d operator. We already have different compute, schedule, InferType etc. Refactoring might not be too painful. What do other folks think? @kevinthesun @tqchen @yzhliu @FrozenGene @ramana-arm @haichen --- [Visit Topic](https://discuss.tvm.ai/t/separate-relay-depthwise-conv-operator/6114/1) 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/992acf395eca2fa3c8c9ee49d6bfcd2a85c5ec9230c0d61615f5ab966f411c79).
