NVIDIA / NVIDIA/cudf

[FEA] Update the default hasher in libcudf

Open
#17,531 1 comment 0 reactions 0 assignees View on GitHub
feature request libcudf
Dominant language
C++
Stars
9.8k
Forks
1.1k
Avg merge
3d 6m
Merged PRs (30d)
278

Description

**Is your feature request related to a problem? Please describe.**
Now that `xxhash64` is available, https://github.com/rapidsai/cudf/issues/12829 we should evaluate the default hash function. The default hash function is [`MurmurHash3_x86_32` as of 25.02](https://github.com/rapidsai/cudf/blob/fbc32563809f509c0186081e6012f72a0e83ebcd/cpp/include/cudf/hashing/detail/default_hash.cuh#L33) .

This came up in https://github.com/rapidsai/cudf/issues/12829#issuecomment-1441081594 where we might see performance benefits and register pressure reductions from using a different hash function.

**Describe the solution you'd like**
Compile results from hash join, mixed join and groupby aggregation benchmarks across the available hash functions.

**Additional context**
We will want to add support for `XXHash_32` before running this investigation.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.