Keith Kraus

I agree that this is kinda the exact use case that #14926 is designed for. Along with something like a PyCapsule based protocol.

Just a note that #14926 will first yield the C++ level functions and C structs, and there would likely need to be a follow up in implementing the Python protocol...

This seemed stale so I have created a new PR based on this PR to try to help it over the finish line: https://github.com/gabime/spdlog/pull/2901

~Indeed, using `CPM_LOCAL_PACKAGES_ONLY` seems to have the behavior I'm looking for, albeit with an extra call to `cpm_find_package` underneath the `CPMAddPackage` call, but that shouldn't be an issue. Apologies for...

Unfortunately not. We have a scenario that looks roughly like the following: ``` ... # Always want to fetch these as we apply some patches on top of them or...

I'm one of the cuDF maintainers where I'm happy to answer any questions that you have about cuDF. I'd also add that cuDF integrates with Dask as well, so if...

> The current version has Dask support already, could you elaborate how to use cuDF under Dask? Essentially instead of using Pandas dataframes for the partitions in a `dask.dataframe.DataFrame` object,...

~Confirmed that it works if using `cudaMallocAsync` / `cudaMallocFromPoolAsync` directly and synchronizing the stream when an allocation fails. Here's the reproducer code I used (tweaked from the snippet you provided):~...

Here's an updated reproducer that allows controlling whether to run using `cudaMalloc`, `cudaMallocAsync`, `cudaMallocFromPoolAsync`, or `rmm::mr::cuda_async_view_memory_resource::allocate`: ```c++ #include #include #include #include #include #include #include #include #include #ifdef RMM_SUPPORT #include #include...

> Thanks @kkraus14. I passed your updated repro along to the driver team and asked that they test on the same driver version you're using to make sure it isn't...