Skip to content

[C++][Compute] Investigate prefix-aware early rejection for StringView equality kernels #55254

Open

Description

@Awesome06

Describe the enhancement requested

In #55171, direct comparison kernels were added between StringView and standard offset-based string types.

As noted by @dmitry-chirkov-dremio in PR review, equality comparisons involving StringView can leverage the 4-byte inline prefix stored in the 16-byte view descriptor (e.g. via util::EqualBinaryView or inline prefix checks against offset strings) to reject mismatches without dereferencing payload buffers.

A prototype demonstrated roughly 1.8x–2x speedups on workloads with ~50% early mismatches. However, because prefix checking adds overhead when strings match, incorporating this requires broader benchmarking and tuning of adaptive heuristics (such as batch sampling thresholds).

This issue tracks benchmarking and exploring prefix-aware fast paths for equal / not_equal kernels across BinaryView / Utf8View types.

Component(s)

C++

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

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions