Provide explicit pool sizes and avoid RMM detail APIs #4086
Conversation
harrism
added
improvement
Not your fault, cuDF did it first. :) And anyway, these alignment utilities should probably have been made public. The real thing I'm fixing here is the change in behavior of |
|
@ChuckHastings do you know if the wheel test failures here are a known issue? |
Every day around this time I tend to see failures in local builds and in CI builds like this. They appear to be transient. I suspect some sort of race condition in building/installing artifacts into the conda repositories, and since we depend on many RAPIDS artifacts that we are more susceptible to this than other packages. They generally resolve themselves by rerunning a few hours later. |
|
I don't have permission to rerun actions on this repo. But I can merge latest which will require rerunning all. |
|
/merge |
…ilities, and optional pool_memory_resource initial size (#1424) Follow-on to #1417, this PR deprecates the following: - `rmm::detail::available_device_memory` in favor of rmm::available_device_memory - `rmm::detail::is_aligned`, `rmm::detail::align_up` and related alignment utility functions in favor of the `rmm::` top level namespace versions. - The `rmm::pool_memory_resource` constructors that take an optional initial size parameter. Should be merged after the following: - rapidsai/cugraph#4086 - rapidsai/cudf#14741 - rapidsai/raft#2088 Authors: - Mark Harris (https://github.com/harrism) Approvers: - Michael Schellenberger Costa (https://github.com/miscco) - Rong Ou (https://github.com/rongou) URL: #1424
This PR fixes up cuGraph to avoid usage that will soon be deprecated in RMM.
Depends on rapidsai/rmm#1417
Fixes #4066