@masahi Sorry, I have been busy at some other things recently. I'd like to be
back if I could have more free time again. Thanks!
--
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/apache/incubator-tvm/issues
@FrozenGene, I'd like to try to implement this. I am thinking if I could
follow @Huyuwei proposal by doing this with backward compatible way.
The padding would accept 1 value, or 2 values, or 4 values ( 3 and 6 for coming
3d ops).
For 1 and 2 case, the existing logic would be kept, which woul
@masahi This should be the correct approach. How about we will try to implement
new functions with the unified manner but only being used by 3D. And keep the
existing 2D ones untouched. Then we could transform them to be unified easily
when we could have enough experience and data in the future
@masahi, for other ops, pad can support any dimentions. I am looking at
pooling, it seems there would be some works, current pooling [implementation
](https://github.com/apache/incubator-tvm/blob/ba9d96bcb585557d4d3dbee40112367b08b28084/topi/include/topi/nn/pooling.h#L61)only
supports 2D. I am
