Tianqi Chen

This is the intended behavior of RPC for now to restrict the object that we can support (so we can support minimum cases like uTVM). Would be a good starting...

FYI the conda nightly is already available https://anaconda.org/tlcpack That does mean a dependency of a conda env instead of normal python though

This is certainly something that worth discussion. To give background of the original rationale in DLPack. Such requirement is not necessary for host/device copies of any kind as we rightfully...

Larger alignment generally have an advantage of making vectorization simple(in the presence of avx2 or 512) in the case of CPU. Similarly in the case of GPU can inform a...

Thanks @seberg. In the specific case of DLPack, there is also a need of supporting opaque data ptr (in the case of vulkan and opencl), in those cases the `byte_offset`...

Based on the plaintext interpretation of the header comment, if the data is not aligned, a usage of `byte_offset` is certainly encouraged to unify that with opaque ptr situation. Of...

cc @rgommers @wookey @tirthasheshpatel @junrushao1994 @jermainewang @leofang @kkraus14 @szha @hawkinsp

Thanks @areusch this is indeed something that worth covering. The compatibility implication section covers some of that - The frameworks are "safe" to simply use the new struct without changing...

I believe what you said is covered by S1 part, which I agree that some of the migration function(likely filling in the value) would be useful

@wjakob thanks for the suggestion! That is indeed the goal of this RFC(to ensure a long enough period of time before we commit as we value stability), I made it...