Skip to content

Stop requiring argument_type in bloom filter policy hashers - #825

Merged
PointKernel merged 1 commit into
NVIDIA:devfrom
PointKernel:remove-hasher-argument-type-requirement
Jul 2, 2026
Merged

Stop requiring argument_type in bloom filter policy hashers#825
PointKernel merged 1 commit into
NVIDIA:devfrom
PointKernel:remove-hasher-argument-type-requirement

Conversation

@PointKernel

Copy link
Copy Markdown
Member

Per rapidsai/cudf#23049 (comment), parametric_filter_policy no longer reads typename Hash::argument_type: split_hash is templated on the key type and deduces everything from the actual call, so hashers without that alias (e.g. cudf's XXHash_64) work directly. hash_result_type is fixed to uint64_t with the 64-bit assert moved into split_hash.

@PointKernel
PointKernel requested a review from sleeepyjack as a code owner July 2, 2026 18:32
@PointKernel PointKernel added topic: bloom_filter Issues related to bloom_filter type: improvement Improvement / enhancement to an existing function labels Jul 2, 2026
@PointKernel
PointKernel merged commit 0883368 into NVIDIA:dev Jul 2, 2026
23 checks passed
@PointKernel
PointKernel deleted the remove-hasher-argument-type-requirement branch July 2, 2026 18:58
rapids-bot Bot pushed a commit to NVIDIA/cudf that referenced this pull request Jul 6, 2026
…23049)

This PR migrates the Parquet reader and the streaming bloom filter onto cuco's new `parametric_filter_policy` (bumped via rapidsai/rapids-cmake#1044). The Parquet reader defines a local `arrow_filter_policy` alias hashing keys with `XXHash_64`, and streaming uses the policy directly with `identity_hash`; no shared cudf policy header is added. Depends on NVIDIA/cuCollections#825, which drops the hasher `argument_type` requirement so `XXHash_64` can be used without an adapter.

Authors:
  - Yunsong Wang (https://github.com/PointKernel)

Approvers:
  - Muhammad Haseeb (https://github.com/mhaseeb123)
  - Lawrence Mitchell (https://github.com/wence-)

URL: #23049
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

topic: bloom_filter Issues related to bloom_filter type: improvement Improvement / enhancement to an existing function

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants