cudf icon indicating copy to clipboard operation
cudf copied to clipboard

Published 20 hours ago • verified publisher icon rapidsai

[FEA] Update to CCCL 2.3 or 2.4

Open bdice opened this issue 1 year ago • 3 comments

Currently RAPIDS is built with CCCL 2.2. This issue lists tasks that we can follow up on once we have upgraded to CCCL 2.3 or 2.4. (We haven't decided on an exact timeline for updating, so RAPIDS could target CCCL 2.3 or 2.4 depending on that timing.)

CCCL 2.3

  • If we upgrade to 2.3, we will need a patch for https://github.com/NVIDIA/cccl/pull/1499 which will be fixed in 2.4.
  • CCCL 2.3 performance may be a motivating factor:
    • Up to 60% performance improvements of cub::DeviceSelect::UniqueByKey, cub::DeviceScan::ExclusiveSumByKey, and cub::DeviceReduce::ReduceByKey on A100. cub::DeviceSegmentedReduce now supports 64-bit indexing.
  • Replace device uses of thrust::optional with cuda::std::optional
    • https://github.com/rapidsai/cudf/pull/15091#issuecomment-2004286213

CCCL 2.4

  • See notes on patch above

Additional Context

  • https://github.com/NVIDIA/cccl/releases
  • Test PR: #14704

cc: @miscco @jrhemstad @robertmaynard

bdice avatar Mar 18 '24 15:03 bdice

CUDA 12.4 ships with CCCL 2.3.

The CCCL support matrix states that CCCL 2.2 is officially untested/unsupported when using 12.4+, so moving to a newer version will be needed.

robertmaynard avatar Apr 02 '24 20:04 robertmaynard

CCCL 2.3 also requires changes to projects due to thrust::tuple constructor changes.

robertmaynard avatar Apr 16 '24 13:04 robertmaynard

We can also tackle https://github.com/NVIDIA/cuCollections/issues/469 after this migration is complete.

bdice avatar May 02 '24 18:05 bdice

Are we now considering going straight to 12.5? See https://github.com/rapidsai/rapids-cmake/pull/607 (CC @trxcllnt).

vyasr avatar May 15 '24 19:05 vyasr

RAPIDS has been updated to CCCL 2.5 (a pre-release commit for now, and eventually the 2.5.0 release).

bdice avatar Jun 11 '24 22:06 bdice