Vyas Ramasubramani
Vyas Ramasubramani
Another piece of code that belongs here is the [`default_allocator`](https://github.com/rapidsai/cudf/blob/branch-22.12/cpp/src/hash/hash_allocator.cuh#L66). The allocator can also be (and is) used with the `concurrent_unordered_map`, but when the latter is removed the only usage...
@PointKernel in light of the ongoing discussions we're having about other improvements to cuco, are there any helpers in cudf that we might want to move to cuco itself so...
We are skipping the problematic test to avoid CI failures as of #15259, but we don't have a fix for the underlying issue yet.
I can't reproduce the memcheck error right now. I do see the same racecheck issues though.
@davidwendt are items 2 and 3 still something that you plan to pursue?
Do you think it's something that we'll want to pursue again? Was trying to gauge whether these changes would still provide a perf benefit and if it's worth keeping the...
@brandon-b-miller is this issue still valid? I see that this works now, for instance: ``` In [96]: cudf.DataFrame({ ...: "a": cudf.Series([1, 1, 2, 2, 3, 3], dtype="float32"), ...: "b": cudf.Series([2,...
Ah OK. Perhaps we should keep this issue open for the moment, then.
This issue can be closed, 0.8.36 (and 0.8.37) have since been released on conda-forge.
Our current thinking regarding prioritization is that we're willing to eat the temporary disruptions for users in 24.10 if we get rid of Arrow before we have a good alternative...