Vyas Ramasubramani
Vyas Ramasubramani
I think @galipremsagar or @wence- can probably give you a better answer about how best we want this to look in dask.
@NeilGeorge1 have you had any luck here? Do you need any pointers?
CC @kkraus14 @jorisvandenbossche @paleolimbot @jrhemstad @davidwendt
The API I'm envisioning would be ```c++ std::unique_ptr from_arrow(ArrowSchema const* schema, ArrowDeviceArray *input, rmm::cuda_stream_view stream, rmm::mr::device_memory_resource mr); ``` The object would look roughly like this: ```c++ class arrow_column { arrow_column(ArrowDeviceArray...
Whoops also obviously CC @zeroshade
Here is a more complete sketch of what I'm imagining. I haven't thought all the way through how the `to_arrow` side of things should look, but here's one proposal: ```c++...
@zeroshade #18402 switches the existing host Arrow conversion code paths to use the new types from #18084 and adds support for the host arrow capsule protocol. I am currently working...
The device support is in #15370.
The fix is in #16101, you're right that this is just a test for that behavior. I'd be fine skipping the test and only backporting the actual fix. CC @bdice.
I'll add that the profiler output is probably confusing in this case because the profiler's GPU vs CPU time columns really says "did we run a cudf or a pandas...