cloudhan

I start to see some blind spots that was ignore by me previously. The very important one is in `zipped_divide(target, tiler).compose(layout_tv, _)` that I always want to **derive** the `tiler`...

For rank-sensitivity of raked_product, I'd imagine an utility `record hierarchy -> flatten -> product -> unflatten with recorded hierarchy`, but may not be useful...

@mnicely Sorry, but I think this is still relevant. Inactivity does not imply it is solved.

I manually verified with lastest main@ bbe579a9e3beb6ea6626d9227ec32d0dae119a49, The case 1 works now (aka, for 1d data Thr and Val). The case 3 works as expected now (aka, for 2d data...

Off topic: Just came across this issue (as a github-mancer). Based on your recent questions I assume you want to write gemm from ground up. And not to be offense,...

@Borkason Have you tried `concfg clean` https://github.com/lukesampson/concfg

@imawizard Great thanks for the sharing for config. Do you have any suggestion on the command mode disparity, say `:wq`, is there any equivalent for the similar commonly used commands?...

@tdegeus https://github.com/romeric/Fastor would be a good reference.

Great idea. And after careful thinking about `cuda_objects` and `cuda_library` with `rdc=True`, The currect design is flawed. In rdc case, objects propagation for dlink follows the table, and in non-rdc...

That is because cuda_objects only provides a CcInfo with only compilation_context, but no linking context. As the original design, cuda_objects is just compilation actions, whose actifacts are to be fed...