[EPIC] RAPIDS Should not need to patch CCCL

[jrhemstad]

RAPIDS currently has to apply a variety of patches to CCCL source code to work around various issues.

This issue is to track eliminating the need for RAPIDS to apply any patches to CCCL source code.

To start, for each patch we need a separate issue created with the following information:

• Link to patch file
• Summary of what the patch is for
• Link to PR that added patch and any other relevant context
• High level of proposed solution, or TBD

All issues should be added to this task list:

Tasks

Beta Give feedback

1. test

[trxcllnt]

The patches we apply to all RAPIDS libraries:

• revert_pr_211.diff (PR)

The patches we apply exclusively to cuDF:

• thrust_disable_64bit_dispatching.diff (PR)
• thrust_faster_sort_compile_times.diff (PR)
• thrust_faster_scan_compile_times.diff (PR)

[bdice]

I made some progress here. The patch reverting PR 211 is actually masking a bug in cuSpatial, which I fixed. This patch can be removed from rapids-cmake and cuDF. See these PRs I just opened. They should merge in this order:

1. Fix multipolygon geometry iterator. rapidsai/cuspatial#1402
2. Remove CCCL patch for PR 211. rapidsai/cudf#16207
3. Remove CCCL patch for PR 211. rapidsai/rapids-cmake#640

Next steps would be to investigate the patches for only cuDF, specifically macros to reduce compile time (64 bit dispatch and fewer CUB arch policies).

[jrhemstad]

Next steps would be to investigate the patches for only cuDF, specifically macros to reduce compile time (64 bit dispatch and fewer CUB arch policies).

I just filed: #1958

For fewer CUB arch policies, I'm still a bit confused tbh. My understanding of ChainedPolicy in CUB is that this patch should have no impact on compile time because CUB only instantiates kernels for the architectures you build for. I need @gevtushenko to look at that patch because he knows ChainedPolicy way better than I do.

[bdice]

All three PRs above have merged, which should fix up the RAPIDS CCCL devcontainer builds. The remaining cuDF patches should apply cleanly over the current CCCL. To continue removing those cuDF patches, I think #1958 is the next step.