Codestin Search App

PointKernel · 2026-09-11T23:54:52Z

Description

This PR adds a primitive lexicographic row comparator and uses it for multi-column numeric sorting, including stable sort. It reuses the reduced type map from primitive row equality to keep nonnumeric comparison instantiations out of numeric sort kernels. The comparator uses device views and ordering spans directly, avoiding general table preprocessing, with shared comparison helpers in lexicographic_common.cuh.

On RTX PRO 6000, the representative eight-column INT32/FLOAT32 sort and sorted_order benchmarks show a 1.79x geometric-mean speedup over main. Block-sort and merge register usage drops from 64 to 48 registers per thread.

Checklist

I am familiar with the Contributing Guidelines.
New or existing tests cover these changes.
The documentation is up to date with these changes.

copy-pr-bot · 2026-09-11T23:54:55Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

PointKernel · 2026-09-12T00:05:54Z

/ok to test

PointKernel · 2026-09-12T00:06:44Z

I32/F32 only — 32 cases. Before = main; after = this PR, first paired round. Negative change means faster.

Type	Rows	Columns	Nulls	Operation	Change
I32	262,144	1	0%	sort	-0.35%
I32	262,144	1	10%	sort	+0.19%
I32	2,097,152	1	0%	sort	-0.13%
I32	2,097,152	1	10%	sort	-0.28%
I32	262,144	8	0%	sort	-48.89%
I32	262,144	8	10%	sort	-41.72%
I32	2,097,152	8	0%	sort	-45.28%
I32	2,097,152	8	10%	sort	-36.50%
I32	262,144	1	0%	sorted_order	-0.56%
I32	262,144	1	10%	sorted_order	-0.25%
I32	2,097,152	1	0%	sorted_order	-0.27%
I32	2,097,152	1	10%	sorted_order	-0.25%
I32	262,144	8	0%	sorted_order	-49.28%
I32	262,144	8	10%	sorted_order	-42.72%
I32	2,097,152	8	0%	sorted_order	-45.97%
I32	2,097,152	8	10%	sorted_order	-37.67%
F32	262,144	1	0%	sort	+1.36%
F32	262,144	1	10%	sort	+0.02%
F32	2,097,152	1	0%	sort	-0.13%
F32	2,097,152	1	10%	sort	-0.40%
F32	262,144	8	0%	sort	-48.91%
F32	262,144	8	10%	sort	-40.36%
F32	2,097,152	8	0%	sort	-46.28%
F32	2,097,152	8	10%	sort	-37.79%
F32	262,144	1	0%	sorted_order	-0.25%
F32	262,144	1	10%	sorted_order	-0.17%
F32	2,097,152	1	0%	sorted_order	-0.28%
F32	2,097,152	1	10%	sorted_order	-0.28%
F32	262,144	8	0%	sorted_order	-49.33%
F32	262,144	8	10%	sorted_order	-41.63%
F32	2,097,152	8	0%	sorted_order	-46.98%
F32	2,097,152	8	10%	sorted_order	-39.06%

The primitive comparator delivers a 1.79x geometric-mean speedup for eight-column INT32/FLOAT32 sorting across three rounds on RTX PRO 6000. Block-sort and merge register usage drops from 64 to 48 registers per thread.

coderabbitai · 2026-09-12T02:30:31Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: bb04c834-880f-4cab-8281-0ab4bfd2e9d8

📥 Commits

Reviewing files that changed from the base of the PR and between 11a901b and 9aa85a4.

📒 Files selected for processing (8)

cpp/include/cudf/detail/row_operator/lexicographic.cuh
cpp/include/cudf/detail/row_operator/lexicographic_common.cuh
cpp/include/cudf/detail/row_operator/primitive_lexicographic.cuh
cpp/include/cudf/detail/row_operator/primitive_row_operators.cuh
cpp/src/sort/sort_impl.cuh
cpp/tests/sort/sort_test.cpp
cpp/tests/sort/stable_sort_tests.cpp
cpp/tests/streams/sorting_test.cpp

Included review availability: Your plan provides up to 12 included reviews per hour; 10 remain after this review.

📝 Summary

Summary by CodeRabbit

New Features
- Added optimized lexicographic sorting for tables containing multiple numeric columns.
- Supports per-column ascending or descending order and configurable null precedence.
- Provides consistent ordering for nulls, NaNs, signed zeros, and numeric boundary values.
- Applies to regular, stable, and key-based sorting workflows.
Documentation
- Clarified that primitive row operations support tables composed entirely of numeric columns.
Tests
- Added coverage for mixed-column sorting, sliced tables, nulls, NaNs, ties, and varied ordering options.

Walkthrough

Changes

Primitive lexicographic sorting

Layer / File(s)	Summary
Shared comparison utilities `cpp/include/cudf/detail/row_operator/lexicographic*.cuh`	Moves physical element, NaN, weak-ordering, and convenience comparator definitions into `lexicographic_common.cuh`.
Primitive row comparison and sort integration `cpp/include/cudf/detail/row_operator/primitive_lexicographic.cuh`, `cpp/include/cudf/detail/row_operator/primitive_row_operators.cuh`, `cpp/src/sort/sort_impl.cuh`	Adds numeric multi-column row comparison with null precedence and sort direction, then uses it for compatible tables.
Sorting regression coverage `cpp/tests/sort/sort_test.cpp`, `cpp/tests/sort/stable_sort_tests.cpp`, `cpp/tests/streams/sorting_test.cpp`	Adds coverage for extrema, nulls, NaNs, signed zeros, sliced columns, ties, and opposing column policies across sort APIs.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~30 minutes

Change: Feature

Suggested reviewers: davidwendt

Merge Risk: ⚪ Minimal · up to 9aa85

This change adds a faster comparison path for sorting tables whose columns are all numeric, with the previous general implementation retained for every other input. New tests cover nulls, NaN and signed-zero values, extreme values, sliced inputs, and tie ordering, and no unresolved correctness concerns remain, so the change looks ready to merge.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 3 files. (5 skipped: 5 …	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: using primitive lexicographic comparison for numeric sorting.
Description check	✅ Passed	The description directly explains the primitive row comparator, numeric sorting support, performance results, shared helpers, and test coverage.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Docstring Coverage

Explanation

Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 7 functions across 3 files. (5 skipped: 5 unsupported.)

Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands.}

bdice · 2026-09-12T13:12:39Z

+      make_device_uvector_async(null_precedence, stream, cudf::get_current_device_resource_ref());
+    if (not column_order.empty() or not null_precedence.empty()) {
+      // Finish copying the host policies before the caller can release them after this call.
+      cudf::detail::sync_stream(stream);


I think we can defer this sync further, until after do_sort.

Also why should the stream sync responsibility be handled in different parts of the call stack for primitive and non-primitive? The other comparator below must be handling this internally, because there’s no corresponding stream sync for it.

Can we structure this as an if/else (rather than an early exit) and lift the stream syncs accordingly? Perhaps that means shifting the cudf::detail::row::lexicographic::self_comparator to take device views of the column order and null precedence, and doing the H->D copy and stream sync in this function.

PointKernel added 3 commits September 11, 2026 16:05

Use primitive lexicographic row comparison for numeric sorting

41bd299

Extract shared lexicographic row operator definitions

d90904e

Remove preprocessing from primitive lexicographic sorting

9aa85a4

github-actions Bot assigned PointKernel Sep 11, 2026

github-actions Bot added the libcudf Affects libcudf (C++/CUDA) code. label Sep 11, 2026

PointKernel added improvement Improvement / enhancement to an existing function Performance Performance related issue non-breaking Non-breaking change labels Sep 11, 2026

PointKernel mentioned this pull request Sep 12, 2026

[FEA] Split primitive row equality and hashing into separate headers #24140

Open

PointKernel marked this pull request as ready for review September 12, 2026 02:22

PointKernel requested a review from a team as a code owner September 12, 2026 02:22

PointKernel requested review from mhaseeb123 and simoneves September 12, 2026 02:22

bdice reviewed Sep 12, 2026

View reviewed changes

Navigation Menu

Uh oh!

Use primitive lexicographic comparison for numeric sorting - #24138

Use primitive lexicographic comparison for numeric sorting#24138
PointKernel wants to merge 3 commits into
NVIDIA:mainfrom
PointKernel:primitive-lexicographic-sort

PointKernel commented Sep 11, 2026

Uh oh!

copy-pr-bot Bot commented Sep 11, 2026

Uh oh!

PointKernel commented Sep 12, 2026

Uh oh!

PointKernel commented Sep 12, 2026

Uh oh!

coderabbitai Bot commented Sep 12, 2026

Summary by CodeRabbit

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

bdice Sep 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

PointKernel commented Sep 11, 2026

Description

Checklist

Uh oh!

copy-pr-bot Bot commented Sep 11, 2026

Uh oh!

PointKernel commented Sep 12, 2026

Uh oh!

PointKernel commented Sep 12, 2026

Uh oh!

coderabbitai Bot commented Sep 12, 2026

Summary by CodeRabbit

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

bdice Sep 12, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants